linux/fs/btrfs/disk-io.c

3130 lines
84 KiB
C
Raw Normal View History

/*
* Copyright (C) 2007 Oracle. All rights reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License v2 as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this program; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 021110-1307, USA.
*/
#include <linux/fs.h>
#include <linux/blkdev.h>
#include <linux/scatterlist.h>
#include <linux/swap.h>
#include <linux/radix-tree.h>
#include <linux/writeback.h>
#include <linux/buffer_head.h>
#include <linux/workqueue.h>
#include <linux/kthread.h>
#include <linux/freezer.h>
#include <linux/crc32c.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 09:04:11 +01:00
#include <linux/slab.h>
#include <linux/migrate.h>
#include <linux/ratelimit.h>
#include <asm/unaligned.h>
#include "compat.h"
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "btrfs_inode.h"
#include "volumes.h"
#include "print-tree.h"
#include "async-thread.h"
#include "locking.h"
#include "tree-log.h"
#include "free-space-cache.h"
#include "inode-map.h"
static struct extent_io_ops btree_extent_io_ops;
static void end_workqueue_fn(struct btrfs_work *work);
static void free_fs_root(struct btrfs_root *root);
static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
int read_only);
static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
struct btrfs_root *root);
static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
static int btrfs_destroy_marked_extents(struct btrfs_root *root,
struct extent_io_tree *dirty_pages,
int mark);
static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
struct extent_io_tree *pinned_extents);
static int btrfs_cleanup_transaction(struct btrfs_root *root);
/*
* end_io_wq structs are used to do processing in task context when an IO is
* complete. This is used during reads to verify checksums, and it is used
* by writes to insert metadata for new file extents after IO is complete.
*/
struct end_io_wq {
struct bio *bio;
bio_end_io_t *end_io;
void *private;
struct btrfs_fs_info *info;
int error;
int metadata;
struct list_head list;
struct btrfs_work work;
};
/*
* async submit bios are used to offload expensive checksumming
* onto the worker threads. They checksum file and metadata bios
* just before they are sent down the IO stack.
*/
struct async_submit_bio {
struct inode *inode;
struct bio *bio;
struct list_head list;
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
extent_submit_bio_hook_t *submit_bio_start;
extent_submit_bio_hook_t *submit_bio_done;
int rw;
int mirror_num;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
unsigned long bio_flags;
/*
* bio_offset is optional, can be used if the pages in the bio
* can't tell us where in the file the bio should go
*/
u64 bio_offset;
struct btrfs_work work;
};
/*
* Lockdep class keys for extent_buffer->lock's in this root. For a given
* eb, the lockdep key is determined by the btrfs_root it belongs to and
* the level the eb occupies in the tree.
*
* Different roots are used for different purposes and may nest inside each
* other and they require separate keysets. As lockdep keys should be
* static, assign keysets according to the purpose of the root as indicated
* by btrfs_root->objectid. This ensures that all special purpose roots
* have separate keysets.
*
* Lock-nesting across peer nodes is always done with the immediate parent
* node locked thus preventing deadlock. As lockdep doesn't know this, use
* subclass to avoid triggering lockdep warning in such cases.
*
* The key is set by the readpage_end_io_hook after the buffer has passed
* csum validation but before the pages are unlocked. It is also set by
* btrfs_init_new_buffer on freshly allocated blocks.
*
* We also add a check to make sure the highest level of the tree is the
* same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
* needs update as well.
*/
#ifdef CONFIG_DEBUG_LOCK_ALLOC
# if BTRFS_MAX_LEVEL != 8
# error
# endif
static struct btrfs_lockdep_keyset {
u64 id; /* root objectid */
const char *name_stem; /* lock name stem */
char names[BTRFS_MAX_LEVEL + 1][20];
struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
} btrfs_lockdep_keysets[] = {
{ .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
{ .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
{ .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
{ .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
{ .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
{ .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
{ .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
{ .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
{ .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
{ .id = 0, .name_stem = "tree" },
};
void __init btrfs_init_lockdep(void)
{
int i, j;
/* initialize lockdep class names */
for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
for (j = 0; j < ARRAY_SIZE(ks->names); j++)
snprintf(ks->names[j], sizeof(ks->names[j]),
"btrfs-%s-%02d", ks->name_stem, j);
}
}
void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
int level)
{
struct btrfs_lockdep_keyset *ks;
BUG_ON(level >= ARRAY_SIZE(ks->keys));
/* find the matching keyset, id 0 is the default entry */
for (ks = btrfs_lockdep_keysets; ks->id; ks++)
if (ks->id == objectid)
break;
lockdep_set_class_and_name(&eb->lock,
&ks->keys[level], ks->names[level]);
}
#endif
/*
* extents on the btree inode are pretty simple, there's one extent
* that covers the entire device
*/
static struct extent_map *btree_get_extent(struct inode *inode,
struct page *page, size_t pg_offset, u64 start, u64 len,
int create)
{
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
struct extent_map *em;
int ret;
read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, start, len);
if (em) {
em->bdev =
BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
read_unlock(&em_tree->lock);
goto out;
}
read_unlock(&em_tree->lock);
em = alloc_extent_map();
if (!em) {
em = ERR_PTR(-ENOMEM);
goto out;
}
em->start = 0;
em->len = (u64)-1;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
em->block_len = (u64)-1;
em->block_start = 0;
em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
if (ret == -EEXIST) {
u64 failed_start = em->start;
u64 failed_len = em->len;
free_extent_map(em);
em = lookup_extent_mapping(em_tree, start, len);
if (em) {
ret = 0;
} else {
em = lookup_extent_mapping(em_tree, failed_start,
failed_len);
ret = -EIO;
}
} else if (ret) {
free_extent_map(em);
em = NULL;
}
write_unlock(&em_tree->lock);
if (ret)
em = ERR_PTR(ret);
out:
return em;
}
u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
{
return crc32c(seed, data, len);
}
void btrfs_csum_final(u32 crc, char *result)
{
put_unaligned_le32(~crc, result);
}
/*
* compute the csum for a btree block, and either verify it or write it
* into the csum field of the block.
*/
static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
int verify)
{
u16 csum_size =
btrfs_super_csum_size(&root->fs_info->super_copy);
char *result = NULL;
unsigned long len;
unsigned long cur_len;
unsigned long offset = BTRFS_CSUM_SIZE;
char *kaddr;
unsigned long map_start;
unsigned long map_len;
int err;
u32 crc = ~(u32)0;
unsigned long inline_result;
len = buf->len - offset;
while (len > 0) {
err = map_private_extent_buffer(buf, offset, 32,
&kaddr, &map_start, &map_len);
if (err)
return 1;
cur_len = min(len, map_len - (offset - map_start));
crc = btrfs_csum_data(root, kaddr + offset - map_start,
crc, cur_len);
len -= cur_len;
offset += cur_len;
}
if (csum_size > sizeof(inline_result)) {
result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
if (!result)
return 1;
} else {
result = (char *)&inline_result;
}
btrfs_csum_final(crc, result);
if (verify) {
if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
u32 val;
u32 found = 0;
memcpy(&found, result, csum_size);
read_extent_buffer(buf, &val, 0, csum_size);
printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
"failed on %llu wanted %X found %X "
"level %d\n",
root->fs_info->sb->s_id,
(unsigned long long)buf->start, val, found,
btrfs_header_level(buf));
if (result != (char *)&inline_result)
kfree(result);
return 1;
}
} else {
write_extent_buffer(buf, result, 0, csum_size);
}
if (result != (char *)&inline_result)
kfree(result);
return 0;
}
/*
* we can't consider a given block up to date unless the transid of the
* block matches the transid in the parent node's pointer. This is how we
* detect blocks that either didn't get written at all or got written
* in the wrong place.
*/
static int verify_parent_transid(struct extent_io_tree *io_tree,
struct extent_buffer *eb, u64 parent_transid)
{
struct extent_state *cached_state = NULL;
int ret;
if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
return 0;
lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
0, &cached_state, GFP_NOFS);
if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
btrfs_header_generation(eb) == parent_transid) {
ret = 0;
goto out;
}
printk_ratelimited("parent transid verify failed on %llu wanted %llu "
"found %llu\n",
(unsigned long long)eb->start,
(unsigned long long)parent_transid,
(unsigned long long)btrfs_header_generation(eb));
ret = 1;
clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
out:
unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
&cached_state, GFP_NOFS);
return ret;
}
/*
* helper to read a given tree block, doing retries as required when
* the checksums don't match and we have alternate mirrors to try.
*/
static int btree_read_extent_buffer_pages(struct btrfs_root *root,
struct extent_buffer *eb,
u64 start, u64 parent_transid)
{
struct extent_io_tree *io_tree;
int ret;
int num_copies = 0;
int mirror_num = 0;
clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
while (1) {
ret = read_extent_buffer_pages(io_tree, eb, start, 1,
btree_get_extent, mirror_num);
if (!ret &&
!verify_parent_transid(io_tree, eb, parent_transid))
return ret;
/*
* This buffer's crc is fine, but its contents are corrupted, so
* there is no reason to read the other copies, they won't be
* any less wrong.
*/
if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
return ret;
num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
eb->start, eb->len);
if (num_copies == 1)
return ret;
mirror_num++;
if (mirror_num > num_copies)
return ret;
}
return -EIO;
}
/*
* checksum a dirty tree block before IO. This has extra checks to make sure
* we only fill in the checksum field in the first page of a multi-page block
*/
static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
{
struct extent_io_tree *tree;
u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
u64 found_start;
unsigned long len;
struct extent_buffer *eb;
int ret;
tree = &BTRFS_I(page->mapping->host)->io_tree;
if (page->private == EXTENT_PAGE_PRIVATE) {
WARN_ON(1);
goto out;
}
if (!page->private) {
WARN_ON(1);
goto out;
}
len = page->private >> 2;
WARN_ON(len == 0);
eb = alloc_extent_buffer(tree, start, len, page);
if (eb == NULL) {
WARN_ON(1);
goto out;
}
ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
btrfs_header_generation(eb));
BUG_ON(ret);
WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
found_start = btrfs_header_bytenr(eb);
if (found_start != start) {
WARN_ON(1);
goto err;
}
if (eb->first_page != page) {
WARN_ON(1);
goto err;
}
if (!PageUptodate(page)) {
WARN_ON(1);
goto err;
}
csum_tree_block(root, eb, 0);
err:
free_extent_buffer(eb);
out:
return 0;
}
static int check_tree_block_fsid(struct btrfs_root *root,
struct extent_buffer *eb)
{
struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
u8 fsid[BTRFS_UUID_SIZE];
int ret = 1;
read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
BTRFS_FSID_SIZE);
while (fs_devices) {
if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
ret = 0;
break;
}
fs_devices = fs_devices->seed;
}
return ret;
}
#define CORRUPT(reason, eb, root, slot) \
printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
"root=%llu, slot=%d\n", reason, \
(unsigned long long)btrfs_header_bytenr(eb), \
(unsigned long long)root->objectid, slot)
static noinline int check_leaf(struct btrfs_root *root,
struct extent_buffer *leaf)
{
struct btrfs_key key;
struct btrfs_key leaf_key;
u32 nritems = btrfs_header_nritems(leaf);
int slot;
if (nritems == 0)
return 0;
/* Check the 0 item */
if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
BTRFS_LEAF_DATA_SIZE(root)) {
CORRUPT("invalid item offset size pair", leaf, root, 0);
return -EIO;
}
/*
* Check to make sure each items keys are in the correct order and their
* offsets make sense. We only have to loop through nritems-1 because
* we check the current slot against the next slot, which verifies the
* next slot's offset+size makes sense and that the current's slot
* offset is correct.
*/
for (slot = 0; slot < nritems - 1; slot++) {
btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
btrfs_item_key_to_cpu(leaf, &key, slot + 1);
/* Make sure the keys are in the right order */
if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
CORRUPT("bad key order", leaf, root, slot);
return -EIO;
}
/*
* Make sure the offset and ends are right, remember that the
* item data starts at the end of the leaf and grows towards the
* front.
*/
if (btrfs_item_offset_nr(leaf, slot) !=
btrfs_item_end_nr(leaf, slot + 1)) {
CORRUPT("slot offset bad", leaf, root, slot);
return -EIO;
}
/*
* Check to make sure that we don't point outside of the leaf,
* just incase all the items are consistent to eachother, but
* all point outside of the leaf.
*/
if (btrfs_item_end_nr(leaf, slot) >
BTRFS_LEAF_DATA_SIZE(root)) {
CORRUPT("slot end outside of leaf", leaf, root, slot);
return -EIO;
}
}
return 0;
}
static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
struct extent_state *state)
{
struct extent_io_tree *tree;
u64 found_start;
int found_level;
unsigned long len;
struct extent_buffer *eb;
struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
int ret = 0;
tree = &BTRFS_I(page->mapping->host)->io_tree;
if (page->private == EXTENT_PAGE_PRIVATE)
goto out;
if (!page->private)
goto out;
len = page->private >> 2;
WARN_ON(len == 0);
eb = alloc_extent_buffer(tree, start, len, page);
if (eb == NULL) {
ret = -EIO;
goto out;
}
found_start = btrfs_header_bytenr(eb);
if (found_start != start) {
printk_ratelimited(KERN_INFO "btrfs bad tree block start "
"%llu %llu\n",
(unsigned long long)found_start,
(unsigned long long)eb->start);
ret = -EIO;
goto err;
}
if (eb->first_page != page) {
printk(KERN_INFO "btrfs bad first page %lu %lu\n",
eb->first_page->index, page->index);
WARN_ON(1);
ret = -EIO;
goto err;
}
if (check_tree_block_fsid(root, eb)) {
printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
(unsigned long long)eb->start);
ret = -EIO;
goto err;
}
found_level = btrfs_header_level(eb);
btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
eb, found_level);
ret = csum_tree_block(root, eb, 1);
if (ret) {
ret = -EIO;
goto err;
}
/*
* If this is a leaf block and it is corrupt, set the corrupt bit so
* that we don't try and read the other copies of this block, just
* return -EIO.
*/
if (found_level == 0 && check_leaf(root, eb)) {
set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
ret = -EIO;
}
end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
end = eb->start + end - 1;
err:
free_extent_buffer(eb);
out:
return ret;
}
static void end_workqueue_bio(struct bio *bio, int err)
{
struct end_io_wq *end_io_wq = bio->bi_private;
struct btrfs_fs_info *fs_info;
fs_info = end_io_wq->info;
end_io_wq->error = err;
end_io_wq->work.func = end_workqueue_fn;
end_io_wq->work.flags = 0;
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
if (bio->bi_rw & REQ_WRITE) {
if (end_io_wq->metadata == 1)
btrfs_queue_worker(&fs_info->endio_meta_write_workers,
&end_io_wq->work);
else if (end_io_wq->metadata == 2)
btrfs_queue_worker(&fs_info->endio_freespace_worker,
&end_io_wq->work);
else
btrfs_queue_worker(&fs_info->endio_write_workers,
&end_io_wq->work);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
} else {
if (end_io_wq->metadata)
btrfs_queue_worker(&fs_info->endio_meta_workers,
&end_io_wq->work);
else
btrfs_queue_worker(&fs_info->endio_workers,
&end_io_wq->work);
}
}
/*
* For the metadata arg you want
*
* 0 - if data
* 1 - if normal metadta
* 2 - if writing to the free space cache area
*/
int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
int metadata)
{
struct end_io_wq *end_io_wq;
end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
if (!end_io_wq)
return -ENOMEM;
end_io_wq->private = bio->bi_private;
end_io_wq->end_io = bio->bi_end_io;
end_io_wq->info = info;
end_io_wq->error = 0;
end_io_wq->bio = bio;
end_io_wq->metadata = metadata;
bio->bi_private = end_io_wq;
bio->bi_end_io = end_workqueue_bio;
return 0;
}
unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
{
unsigned long limit = min_t(unsigned long,
info->workers.max_workers,
info->fs_devices->open_devices);
return 256 * limit;
}
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
static void run_one_async_start(struct btrfs_work *work)
{
struct async_submit_bio *async;
async = container_of(work, struct async_submit_bio, work);
async->submit_bio_start(async->inode, async->rw, async->bio,
async->mirror_num, async->bio_flags,
async->bio_offset);
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
}
static void run_one_async_done(struct btrfs_work *work)
{
struct btrfs_fs_info *fs_info;
struct async_submit_bio *async;
int limit;
async = container_of(work, struct async_submit_bio, work);
fs_info = BTRFS_I(async->inode)->root->fs_info;
limit = btrfs_async_submit_limit(fs_info);
limit = limit * 2 / 3;
atomic_dec(&fs_info->nr_async_submits);
if (atomic_read(&fs_info->nr_async_submits) < limit &&
waitqueue_active(&fs_info->async_submit_wait))
wake_up(&fs_info->async_submit_wait);
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
async->submit_bio_done(async->inode, async->rw, async->bio,
async->mirror_num, async->bio_flags,
async->bio_offset);
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
}
static void run_one_async_free(struct btrfs_work *work)
{
struct async_submit_bio *async;
async = container_of(work, struct async_submit_bio, work);
kfree(async);
}
int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
int rw, struct bio *bio, int mirror_num,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
unsigned long bio_flags,
u64 bio_offset,
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
extent_submit_bio_hook_t *submit_bio_start,
extent_submit_bio_hook_t *submit_bio_done)
{
struct async_submit_bio *async;
async = kmalloc(sizeof(*async), GFP_NOFS);
if (!async)
return -ENOMEM;
async->inode = inode;
async->rw = rw;
async->bio = bio;
async->mirror_num = mirror_num;
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
async->submit_bio_start = submit_bio_start;
async->submit_bio_done = submit_bio_done;
async->work.func = run_one_async_start;
async->work.ordered_func = run_one_async_done;
async->work.ordered_free = run_one_async_free;
async->work.flags = 0;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
async->bio_flags = bio_flags;
async->bio_offset = bio_offset;
atomic_inc(&fs_info->nr_async_submits);
if (rw & REQ_SYNC)
btrfs_set_work_high_prio(&async->work);
btrfs_queue_worker(&fs_info->workers, &async->work);
while (atomic_read(&fs_info->async_submit_draining) &&
atomic_read(&fs_info->nr_async_submits)) {
wait_event(fs_info->async_submit_wait,
(atomic_read(&fs_info->nr_async_submits) == 0));
}
return 0;
}
static int btree_csum_one_bio(struct bio *bio)
{
struct bio_vec *bvec = bio->bi_io_vec;
int bio_index = 0;
struct btrfs_root *root;
WARN_ON(bio->bi_vcnt <= 0);
while (bio_index < bio->bi_vcnt) {
root = BTRFS_I(bvec->bv_page->mapping->host)->root;
csum_dirty_buffer(root, bvec->bv_page);
bio_index++;
bvec++;
}
return 0;
}
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
static int __btree_submit_bio_start(struct inode *inode, int rw,
struct bio *bio, int mirror_num,
unsigned long bio_flags,
u64 bio_offset)
{
/*
* when we're called for a write, we're already in the async
* submission context. Just jump into btrfs_map_bio
*/
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
btree_csum_one_bio(bio);
return 0;
}
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
int mirror_num, unsigned long bio_flags,
u64 bio_offset)
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
{
/*
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
* when we're called for a write, we're already in the async
* submission context. Just jump into btrfs_map_bio
*/
return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
}
static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
int mirror_num, unsigned long bio_flags,
u64 bio_offset)
{
int ret;
ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
bio, 1);
BUG_ON(ret);
if (!(rw & REQ_WRITE)) {
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
/*
* called for a read, do the setup so that checksum validation
* can happen in the async kernel threads
*/
return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
mirror_num, 0);
}
/*
* kthread helpers are used to submit writes so that checksumming
* can happen in parallel across all CPUs
*/
return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
inode, rw, bio, mirror_num, 0,
bio_offset,
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
__btree_submit_bio_start,
__btree_submit_bio_done);
}
#ifdef CONFIG_MIGRATION
static int btree_migratepage(struct address_space *mapping,
struct page *newpage, struct page *page)
{
/*
* we can't safely write a btree page from here,
* we haven't done the locking hook
*/
if (PageDirty(page))
return -EAGAIN;
/*
* Buffers may be managed in a filesystem specific way.
* We must have no buffers or drop them.
*/
if (page_has_private(page) &&
!try_to_release_page(page, GFP_KERNEL))
return -EAGAIN;
return migrate_page(mapping, newpage, page);
}
#endif
static int btree_writepage(struct page *page, struct writeback_control *wbc)
{
struct extent_io_tree *tree;
struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
struct extent_buffer *eb;
int was_dirty;
tree = &BTRFS_I(page->mapping->host)->io_tree;
if (!(current->flags & PF_MEMALLOC)) {
return extent_write_full_page(tree, page,
btree_get_extent, wbc);
}
redirty_page_for_writepage(wbc, page);
eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
WARN_ON(!eb);
was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
if (!was_dirty) {
spin_lock(&root->fs_info->delalloc_lock);
root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
spin_unlock(&root->fs_info->delalloc_lock);
}
free_extent_buffer(eb);
unlock_page(page);
return 0;
}
static int btree_writepages(struct address_space *mapping,
struct writeback_control *wbc)
{
struct extent_io_tree *tree;
tree = &BTRFS_I(mapping->host)->io_tree;
if (wbc->sync_mode == WB_SYNC_NONE) {
struct btrfs_root *root = BTRFS_I(mapping->host)->root;
u64 num_dirty;
unsigned long thresh = 32 * 1024 * 1024;
if (wbc->for_kupdate)
return 0;
/* this is a bit racy, but that's ok */
num_dirty = root->fs_info->dirty_metadata_bytes;
if (num_dirty < thresh)
return 0;
}
return extent_writepages(tree, mapping, btree_get_extent, wbc);
}
static int btree_readpage(struct file *file, struct page *page)
{
struct extent_io_tree *tree;
tree = &BTRFS_I(page->mapping->host)->io_tree;
return extent_read_full_page(tree, page, btree_get_extent);
}
static int btree_releasepage(struct page *page, gfp_t gfp_flags)
{
struct extent_io_tree *tree;
struct extent_map_tree *map;
int ret;
if (PageWriteback(page) || PageDirty(page))
return 0;
tree = &BTRFS_I(page->mapping->host)->io_tree;
map = &BTRFS_I(page->mapping->host)->extent_tree;
ret = try_release_extent_state(map, tree, page, gfp_flags);
if (!ret)
return 0;
ret = try_release_extent_buffer(tree, page);
if (ret == 1) {
ClearPagePrivate(page);
set_page_private(page, 0);
page_cache_release(page);
}
return ret;
}
static void btree_invalidatepage(struct page *page, unsigned long offset)
{
struct extent_io_tree *tree;
tree = &BTRFS_I(page->mapping->host)->io_tree;
extent_invalidatepage(tree, page, offset);
btree_releasepage(page, GFP_NOFS);
if (PagePrivate(page)) {
printk(KERN_WARNING "btrfs warning page private not zero "
"on page %llu\n", (unsigned long long)page_offset(page));
ClearPagePrivate(page);
set_page_private(page, 0);
page_cache_release(page);
}
}
static const struct address_space_operations btree_aops = {
.readpage = btree_readpage,
.writepage = btree_writepage,
.writepages = btree_writepages,
.releasepage = btree_releasepage,
.invalidatepage = btree_invalidatepage,
#ifdef CONFIG_MIGRATION
.migratepage = btree_migratepage,
#endif
};
int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
u64 parent_transid)
{
struct extent_buffer *buf = NULL;
struct inode *btree_inode = root->fs_info->btree_inode;
int ret = 0;
buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
if (!buf)
return 0;
read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
buf, 0, 0, btree_get_extent, 0);
free_extent_buffer(buf);
return ret;
}
struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
u64 bytenr, u32 blocksize)
{
struct inode *btree_inode = root->fs_info->btree_inode;
struct extent_buffer *eb;
eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
bytenr, blocksize);
return eb;
}
struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
u64 bytenr, u32 blocksize)
{
struct inode *btree_inode = root->fs_info->btree_inode;
struct extent_buffer *eb;
eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
bytenr, blocksize, NULL);
return eb;
}
int btrfs_write_tree_block(struct extent_buffer *buf)
{
return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
buf->start + buf->len - 1);
}
int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
{
return filemap_fdatawait_range(buf->first_page->mapping,
buf->start, buf->start + buf->len - 1);
}
struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
u32 blocksize, u64 parent_transid)
{
struct extent_buffer *buf = NULL;
int ret;
buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
if (!buf)
return NULL;
ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
if (ret == 0)
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 15:25:08 +01:00
set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
return buf;
}
int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
struct extent_buffer *buf)
{
struct inode *btree_inode = root->fs_info->btree_inode;
if (btrfs_header_generation(buf) ==
root->fs_info->running_transaction->transid) {
btrfs_assert_tree_locked(buf);
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 15:25:08 +01:00
if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
spin_lock(&root->fs_info->delalloc_lock);
if (root->fs_info->dirty_metadata_bytes >= buf->len)
root->fs_info->dirty_metadata_bytes -= buf->len;
else
WARN_ON(1);
spin_unlock(&root->fs_info->delalloc_lock);
}
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 15:25:08 +01:00
/* ugh, clear_extent_buffer_dirty needs to lock the page */
btrfs_set_lock_blocking(buf);
clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
buf);
}
return 0;
}
static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
u32 stripesize, struct btrfs_root *root,
struct btrfs_fs_info *fs_info,
u64 objectid)
{
root->node = NULL;
root->commit_root = NULL;
root->sectorsize = sectorsize;
root->nodesize = nodesize;
root->leafsize = leafsize;
root->stripesize = stripesize;
root->ref_cows = 0;
root->track_dirty = 0;
root->in_radix = 0;
root->orphan_item_inserted = 0;
root->orphan_cleanup_state = 0;
root->fs_info = fs_info;
root->objectid = objectid;
root->last_trans = 0;
root->highest_objectid = 0;
root->name = NULL;
root->inode_tree = RB_ROOT;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
root->block_rsv = NULL;
root->orphan_block_rsv = NULL;
INIT_LIST_HEAD(&root->dirty_list);
INIT_LIST_HEAD(&root->orphan_list);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
INIT_LIST_HEAD(&root->root_list);
spin_lock_init(&root->orphan_lock);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
spin_lock_init(&root->inode_lock);
spin_lock_init(&root->accounting_lock);
mutex_init(&root->objectid_mutex);
mutex_init(&root->log_mutex);
init_waitqueue_head(&root->log_writer_wait);
init_waitqueue_head(&root->log_commit_wait[0]);
init_waitqueue_head(&root->log_commit_wait[1]);
atomic_set(&root->log_commit[0], 0);
atomic_set(&root->log_commit[1], 0);
atomic_set(&root->log_writers, 0);
root->log_batch = 0;
root->log_transid = 0;
root->last_log_commit = 0;
extent_io_tree_init(&root->dirty_log_pages,
fs_info->btree_inode->i_mapping);
memset(&root->root_key, 0, sizeof(root->root_key));
memset(&root->root_item, 0, sizeof(root->root_item));
memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
memset(&root->root_kobj, 0, sizeof(root->root_kobj));
root->defrag_trans_start = fs_info->generation;
init_completion(&root->kobj_unregister);
root->defrag_running = 0;
root->root_key.objectid = objectid;
root->anon_dev = 0;
return 0;
}
static int find_and_setup_root(struct btrfs_root *tree_root,
struct btrfs_fs_info *fs_info,
u64 objectid,
struct btrfs_root *root)
{
int ret;
u32 blocksize;
u64 generation;
__setup_root(tree_root->nodesize, tree_root->leafsize,
tree_root->sectorsize, tree_root->stripesize,
root, fs_info, objectid);
ret = btrfs_find_last_root(tree_root, objectid,
&root->root_item, &root->root_key);
if (ret > 0)
return -ENOENT;
BUG_ON(ret);
generation = btrfs_root_generation(&root->root_item);
blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
blocksize, generation);
if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
free_extent_buffer(root->node);
return -EIO;
}
root->commit_root = btrfs_root_node(root);
return 0;
}
static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info)
{
struct btrfs_root *root;
struct btrfs_root *tree_root = fs_info->tree_root;
struct extent_buffer *leaf;
root = kzalloc(sizeof(*root), GFP_NOFS);
if (!root)
return ERR_PTR(-ENOMEM);
__setup_root(tree_root->nodesize, tree_root->leafsize,
tree_root->sectorsize, tree_root->stripesize,
root, fs_info, BTRFS_TREE_LOG_OBJECTID);
root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
root->root_key.type = BTRFS_ROOT_ITEM_KEY;
root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
/*
* log trees do not get reference counted because they go away
* before a real commit is actually done. They do store pointers
* to file data extents, and those reference counts still get
* updated (along with back refs to the log tree).
*/
root->ref_cows = 0;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
if (IS_ERR(leaf)) {
kfree(root);
return ERR_CAST(leaf);
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
btrfs_set_header_bytenr(leaf, leaf->start);
btrfs_set_header_generation(leaf, trans->transid);
btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
root->node = leaf;
write_extent_buffer(root->node, root->fs_info->fsid,
(unsigned long)btrfs_header_fsid(root->node),
BTRFS_FSID_SIZE);
btrfs_mark_buffer_dirty(root->node);
btrfs_tree_unlock(root->node);
return root;
}
int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info)
{
struct btrfs_root *log_root;
log_root = alloc_log_tree(trans, fs_info);
if (IS_ERR(log_root))
return PTR_ERR(log_root);
WARN_ON(fs_info->log_root_tree);
fs_info->log_root_tree = log_root;
return 0;
}
int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
struct btrfs_root *log_root;
struct btrfs_inode_item *inode_item;
log_root = alloc_log_tree(trans, root->fs_info);
if (IS_ERR(log_root))
return PTR_ERR(log_root);
log_root->last_trans = trans->transid;
log_root->root_key.offset = root->root_key.objectid;
inode_item = &log_root->root_item.inode;
inode_item->generation = cpu_to_le64(1);
inode_item->size = cpu_to_le64(3);
inode_item->nlink = cpu_to_le32(1);
inode_item->nbytes = cpu_to_le64(root->leafsize);
inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
btrfs_set_root_node(&log_root->root_item, log_root->node);
WARN_ON(root->log_root);
root->log_root = log_root;
root->log_transid = 0;
root->last_log_commit = 0;
return 0;
}
struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
struct btrfs_key *location)
{
struct btrfs_root *root;
struct btrfs_fs_info *fs_info = tree_root->fs_info;
struct btrfs_path *path;
struct extent_buffer *l;
u64 generation;
u32 blocksize;
int ret = 0;
root = kzalloc(sizeof(*root), GFP_NOFS);
if (!root)
return ERR_PTR(-ENOMEM);
if (location->offset == (u64)-1) {
ret = find_and_setup_root(tree_root, fs_info,
location->objectid, root);
if (ret) {
kfree(root);
return ERR_PTR(ret);
}
goto out;
}
__setup_root(tree_root->nodesize, tree_root->leafsize,
tree_root->sectorsize, tree_root->stripesize,
root, fs_info, location->objectid);
path = btrfs_alloc_path();
if (!path) {
kfree(root);
return ERR_PTR(-ENOMEM);
}
ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
if (ret == 0) {
l = path->nodes[0];
read_extent_buffer(l, &root->root_item,
btrfs_item_ptr_offset(l, path->slots[0]),
sizeof(root->root_item));
memcpy(&root->root_key, location, sizeof(*location));
}
btrfs_free_path(path);
if (ret) {
kfree(root);
if (ret > 0)
ret = -ENOENT;
return ERR_PTR(ret);
}
generation = btrfs_root_generation(&root->root_item);
blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
blocksize, generation);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
root->commit_root = btrfs_root_node(root);
BUG_ON(!root->node);
out:
if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
root->ref_cows = 1;
btrfs_check_and_init_root_item(&root->root_item);
}
return root;
}
struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
struct btrfs_key *location)
{
struct btrfs_root *root;
int ret;
if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
return fs_info->tree_root;
if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
return fs_info->extent_root;
if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
return fs_info->chunk_root;
if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
return fs_info->dev_root;
if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
return fs_info->csum_root;
again:
spin_lock(&fs_info->fs_roots_radix_lock);
root = radix_tree_lookup(&fs_info->fs_roots_radix,
(unsigned long)location->objectid);
spin_unlock(&fs_info->fs_roots_radix_lock);
if (root)
return root;
root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
if (IS_ERR(root))
return root;
root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
GFP_NOFS);
if (!root->free_ino_pinned || !root->free_ino_ctl) {
ret = -ENOMEM;
goto fail;
}
btrfs_init_free_ino_ctl(root);
mutex_init(&root->fs_commit_mutex);
spin_lock_init(&root->cache_lock);
init_waitqueue_head(&root->cache_wait);
ret = get_anon_bdev(&root->anon_dev);
if (ret)
goto fail;
if (btrfs_root_refs(&root->root_item) == 0) {
ret = -ENOENT;
goto fail;
}
ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
if (ret < 0)
goto fail;
if (ret == 0)
root->orphan_item_inserted = 1;
ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
if (ret)
goto fail;
spin_lock(&fs_info->fs_roots_radix_lock);
ret = radix_tree_insert(&fs_info->fs_roots_radix,
(unsigned long)root->root_key.objectid,
root);
if (ret == 0)
root->in_radix = 1;
spin_unlock(&fs_info->fs_roots_radix_lock);
radix_tree_preload_end();
if (ret) {
if (ret == -EEXIST) {
free_fs_root(root);
goto again;
}
goto fail;
}
ret = btrfs_find_dead_roots(fs_info->tree_root,
root->root_key.objectid);
WARN_ON(ret);
return root;
fail:
free_fs_root(root);
return ERR_PTR(ret);
}
static int btrfs_congested_fn(void *congested_data, int bdi_bits)
{
struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
int ret = 0;
struct btrfs_device *device;
struct backing_dev_info *bdi;
rcu_read_lock();
list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
if (!device->bdev)
continue;
bdi = blk_get_backing_dev_info(device->bdev);
if (bdi && bdi_congested(bdi, bdi_bits)) {
ret = 1;
break;
}
}
rcu_read_unlock();
return ret;
}
/*
* If this fails, caller must call bdi_destroy() to get rid of the
* bdi again.
*/
static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
{
int err;
bdi->capabilities = BDI_CAP_MAP_COPY;
err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
if (err)
return err;
bdi->ra_pages = default_backing_dev_info.ra_pages;
bdi->congested_fn = btrfs_congested_fn;
bdi->congested_data = info;
return 0;
}
static int bio_ready_for_csum(struct bio *bio)
{
u64 length = 0;
u64 buf_len = 0;
u64 start = 0;
struct page *page;
struct extent_io_tree *io_tree = NULL;
struct bio_vec *bvec;
int i;
int ret;
bio_for_each_segment(bvec, bio, i) {
page = bvec->bv_page;
if (page->private == EXTENT_PAGE_PRIVATE) {
length += bvec->bv_len;
continue;
}
if (!page->private) {
length += bvec->bv_len;
continue;
}
length = bvec->bv_len;
buf_len = page->private >> 2;
start = page_offset(page) + bvec->bv_offset;
io_tree = &BTRFS_I(page->mapping->host)->io_tree;
}
/* are we fully contained in this bio? */
if (buf_len <= length)
return 1;
ret = extent_range_uptodate(io_tree, start + length,
start + buf_len - 1);
return ret;
}
/*
* called by the kthread helper functions to finally call the bio end_io
* functions. This is where read checksum verification actually happens
*/
static void end_workqueue_fn(struct btrfs_work *work)
{
struct bio *bio;
struct end_io_wq *end_io_wq;
struct btrfs_fs_info *fs_info;
int error;
end_io_wq = container_of(work, struct end_io_wq, work);
bio = end_io_wq->bio;
fs_info = end_io_wq->info;
/* metadata bio reads are special because the whole tree block must
* be checksummed at once. This makes sure the entire block is in
* ram and up to date before trying to verify things. For
* blocksize <= pagesize, it is basically a noop
*/
if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
!bio_ready_for_csum(bio)) {
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
btrfs_queue_worker(&fs_info->endio_meta_workers,
&end_io_wq->work);
return;
}
error = end_io_wq->error;
bio->bi_private = end_io_wq->private;
bio->bi_end_io = end_io_wq->end_io;
kfree(end_io_wq);
bio_endio(bio, error);
}
static int cleaner_kthread(void *arg)
{
struct btrfs_root *root = arg;
do {
vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
mutex_trylock(&root->fs_info->cleaner_mutex)) {
btrfs_run_delayed_iputs(root);
btrfs_clean_old_snapshots(root);
mutex_unlock(&root->fs_info->cleaner_mutex);
btrfs_run_defrag_inodes(root->fs_info);
}
if (freezing(current)) {
refrigerator();
} else {
set_current_state(TASK_INTERRUPTIBLE);
if (!kthread_should_stop())
schedule();
__set_current_state(TASK_RUNNING);
}
} while (!kthread_should_stop());
return 0;
}
static int transaction_kthread(void *arg)
{
struct btrfs_root *root = arg;
struct btrfs_trans_handle *trans;
struct btrfs_transaction *cur;
u64 transid;
unsigned long now;
unsigned long delay;
int ret;
do {
delay = HZ * 30;
vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
mutex_lock(&root->fs_info->transaction_kthread_mutex);
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_lock(&root->fs_info->trans_lock);
cur = root->fs_info->running_transaction;
if (!cur) {
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_unlock(&root->fs_info->trans_lock);
goto sleep;
}
now = get_seconds();
if (!cur->blocked &&
(now < cur->start_time || now - cur->start_time < 30)) {
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_unlock(&root->fs_info->trans_lock);
delay = HZ * 5;
goto sleep;
}
transid = cur->transid;
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_unlock(&root->fs_info->trans_lock);
Btrfs: do extent allocation and reference count updates in the background The extent allocation tree maintains a reference count and full back reference information for every extent allocated in the filesystem. For subvolume and snapshot trees, every time a block goes through COW, the new copy of the block adds a reference on every block it points to. If a btree node points to 150 leaves, then the COW code needs to go and add backrefs on 150 different extents, which might be spread all over the extent allocation tree. These updates currently happen during btrfs_cow_block, and most COWs happen during btrfs_search_slot. btrfs_search_slot has locks held on both the parent and the node we are COWing, and so we really want to avoid IO during the COW if we can. This commit adds an rbtree of pending reference count updates and extent allocations. The tree is ordered by byte number of the extent and byte number of the parent for the back reference. The tree allows us to: 1) Modify back references in something close to disk order, reducing seeks 2) Significantly reduce the number of modifications made as block pointers are balanced around 3) Do all of the extent insertion and back reference modifications outside of the performance critical btrfs_search_slot code. #3 has the added benefit of greatly reducing the btrfs stack footprint. The extent allocation tree modifications are done without the deep (and somewhat recursive) call chains used in the past. These delayed back reference updates must be done before the transaction commits, and so the rbtree is tied to the transaction. Throttling is implemented to help keep the queue of backrefs at a reasonable size. Since there was a similar mechanism in place for the extent tree extents, that is removed and replaced by the delayed reference tree. Yan Zheng <yan.zheng@oracle.com> helped review and fixup this code. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-13 15:10:06 +01:00
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
if (transid == trans->transid) {
ret = btrfs_commit_transaction(trans, root);
BUG_ON(ret);
} else {
btrfs_end_transaction(trans, root);
}
sleep:
wake_up_process(root->fs_info->cleaner_kthread);
mutex_unlock(&root->fs_info->transaction_kthread_mutex);
if (freezing(current)) {
refrigerator();
} else {
set_current_state(TASK_INTERRUPTIBLE);
if (!kthread_should_stop() &&
!btrfs_transaction_blocked(root->fs_info))
schedule_timeout(delay);
__set_current_state(TASK_RUNNING);
}
} while (!kthread_should_stop());
return 0;
}
struct btrfs_root *open_ctree(struct super_block *sb,
struct btrfs_fs_devices *fs_devices,
char *options)
{
u32 sectorsize;
u32 nodesize;
u32 leafsize;
u32 blocksize;
u32 stripesize;
u64 generation;
u64 features;
struct btrfs_key location;
struct buffer_head *bh;
struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
GFP_NOFS);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
GFP_NOFS);
struct btrfs_root *tree_root = btrfs_sb(sb);
struct btrfs_fs_info *fs_info = NULL;
struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
GFP_NOFS);
struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
GFP_NOFS);
struct btrfs_root *log_tree_root;
int ret;
int err = -EINVAL;
struct btrfs_super_block *disk_super;
if (!extent_root || !tree_root || !tree_root->fs_info ||
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
!chunk_root || !dev_root || !csum_root) {
err = -ENOMEM;
goto fail;
}
fs_info = tree_root->fs_info;
ret = init_srcu_struct(&fs_info->subvol_srcu);
if (ret) {
err = ret;
goto fail;
}
ret = setup_bdi(fs_info, &fs_info->bdi);
if (ret) {
err = ret;
goto fail_srcu;
}
fs_info->btree_inode = new_inode(sb);
if (!fs_info->btree_inode) {
err = -ENOMEM;
goto fail_bdi;
}
mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
INIT_LIST_HEAD(&fs_info->trans_list);
INIT_LIST_HEAD(&fs_info->dead_roots);
INIT_LIST_HEAD(&fs_info->delayed_iputs);
INIT_LIST_HEAD(&fs_info->hashers);
INIT_LIST_HEAD(&fs_info->delalloc_inodes);
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 19:27:11 +02:00
INIT_LIST_HEAD(&fs_info->ordered_operations);
INIT_LIST_HEAD(&fs_info->caching_block_groups);
spin_lock_init(&fs_info->delalloc_lock);
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_lock_init(&fs_info->trans_lock);
spin_lock_init(&fs_info->ref_cache_lock);
spin_lock_init(&fs_info->fs_roots_radix_lock);
spin_lock_init(&fs_info->delayed_iput_lock);
spin_lock_init(&fs_info->defrag_inodes_lock);
mutex_init(&fs_info->reloc_mutex);
init_completion(&fs_info->kobj_unregister);
fs_info->tree_root = tree_root;
fs_info->extent_root = extent_root;
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
fs_info->csum_root = csum_root;
fs_info->chunk_root = chunk_root;
fs_info->dev_root = dev_root;
fs_info->fs_devices = fs_devices;
INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
INIT_LIST_HEAD(&fs_info->space_info);
btrfs_mapping_init(&fs_info->mapping_tree);
btrfs_init_block_rsv(&fs_info->global_block_rsv);
btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
btrfs_init_block_rsv(&fs_info->trans_block_rsv);
btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
btrfs_init_block_rsv(&fs_info->empty_block_rsv);
INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
mutex_init(&fs_info->durable_block_rsv_mutex);
atomic_set(&fs_info->nr_async_submits, 0);
atomic_set(&fs_info->async_delalloc_pages, 0);
atomic_set(&fs_info->async_submit_draining, 0);
atomic_set(&fs_info->nr_async_bios, 0);
atomic_set(&fs_info->defrag_running, 0);
fs_info->sb = sb;
fs_info->max_inline = 8192 * 1024;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 22:12:44 +02:00
fs_info->metadata_ratio = 0;
fs_info->defrag_inodes = RB_ROOT;
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
fs_info->trans_no_join = 0;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
fs_info->thread_pool_size = min_t(unsigned long,
num_online_cpus() + 2, 8);
INIT_LIST_HEAD(&fs_info->ordered_extents);
spin_lock_init(&fs_info->ordered_extent_lock);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
GFP_NOFS);
if (!fs_info->delayed_root) {
err = -ENOMEM;
goto fail_iput;
}
btrfs_init_delayed_root(fs_info->delayed_root);
mutex_init(&fs_info->scrub_lock);
atomic_set(&fs_info->scrubs_running, 0);
atomic_set(&fs_info->scrub_pause_req, 0);
atomic_set(&fs_info->scrubs_paused, 0);
atomic_set(&fs_info->scrub_cancel_req, 0);
init_waitqueue_head(&fs_info->scrub_pause_wait);
init_rwsem(&fs_info->scrub_super_lock);
fs_info->scrub_workers_refcnt = 0;
sb->s_blocksize = 4096;
sb->s_blocksize_bits = blksize_bits(4096);
sb->s_bdi = &fs_info->bdi;
fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
fs_info->btree_inode->i_nlink = 1;
/*
* we set the i_size on the btree inode to the max possible int.
* the real end of the address space is determined by all of
* the devices in the system
*/
fs_info->btree_inode->i_size = OFFSET_MAX;
fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
fs_info->btree_inode->i_mapping);
extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
BTRFS_I(fs_info->btree_inode)->root = tree_root;
memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
sizeof(struct btrfs_key));
BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
insert_inode_hash(fs_info->btree_inode);
Btrfs: free space accounting redo 1) replace the per fs_info extent_io_tree that tracked free space with two rb-trees per block group to track free space areas via offset and size. The reason to do this is because most allocations come with a hint byte where to start, so we can usually find a chunk of free space at that hint byte to satisfy the allocation and get good space packing. If we cannot find free space at or after the given offset we fall back on looking for a chunk of the given size as close to that given offset as possible. When we fall back on the size search we also try to find a slot as close to the size we want as possible, to avoid breaking small chunks off of huge areas if possible. 2) remove the extent_io_tree that tracked the block group cache from fs_info and replaced it with an rb-tree thats tracks block group cache via offset. also added a per space_info list that tracks the block group cache for the particular space so we can lookup related block groups easily. 3) cleaned up the allocation code to make it a little easier to read and a little less complicated. Basically there are 3 steps, first look from our provided hint. If we couldn't find from that given hint, start back at our original search start and look for space from there. If that fails try to allocate space if we can and start looking again. If not we're screwed and need to start over again. 4) small fixes. there were some issues in volumes.c where we wouldn't allocate the rest of the disk. fixed cow_file_range to actually pass the alloc_hint, which has helped a good bit in making the fs_mark test I run have semi-normal results as we run out of space. Generally with data allocations we don't track where we last allocated from, so everytime we did a data allocation we'd search through every block group that we have looking for free space. Now searching a block group with no free space isn't terribly time consuming, it was causing a slight degradation as we got more data block groups. The alloc_hint has fixed this slight degredation and made things semi-normal. There is still one nagging problem I'm working on where we will get ENOSPC when there is definitely plenty of space. This only happens with metadata allocations, and only when we are almost full. So you generally hit the 85% mark first, but sometimes you'll hit the BUG before you hit the 85% wall. I'm still tracking it down, but until then this seems to be pretty stable and make a significant performance gain. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-23 19:14:11 +02:00
spin_lock_init(&fs_info->block_group_cache_lock);
fs_info->block_group_cache_tree = RB_ROOT;
Btrfs: free space accounting redo 1) replace the per fs_info extent_io_tree that tracked free space with two rb-trees per block group to track free space areas via offset and size. The reason to do this is because most allocations come with a hint byte where to start, so we can usually find a chunk of free space at that hint byte to satisfy the allocation and get good space packing. If we cannot find free space at or after the given offset we fall back on looking for a chunk of the given size as close to that given offset as possible. When we fall back on the size search we also try to find a slot as close to the size we want as possible, to avoid breaking small chunks off of huge areas if possible. 2) remove the extent_io_tree that tracked the block group cache from fs_info and replaced it with an rb-tree thats tracks block group cache via offset. also added a per space_info list that tracks the block group cache for the particular space so we can lookup related block groups easily. 3) cleaned up the allocation code to make it a little easier to read and a little less complicated. Basically there are 3 steps, first look from our provided hint. If we couldn't find from that given hint, start back at our original search start and look for space from there. If that fails try to allocate space if we can and start looking again. If not we're screwed and need to start over again. 4) small fixes. there were some issues in volumes.c where we wouldn't allocate the rest of the disk. fixed cow_file_range to actually pass the alloc_hint, which has helped a good bit in making the fs_mark test I run have semi-normal results as we run out of space. Generally with data allocations we don't track where we last allocated from, so everytime we did a data allocation we'd search through every block group that we have looking for free space. Now searching a block group with no free space isn't terribly time consuming, it was causing a slight degradation as we got more data block groups. The alloc_hint has fixed this slight degredation and made things semi-normal. There is still one nagging problem I'm working on where we will get ENOSPC when there is definitely plenty of space. This only happens with metadata allocations, and only when we are almost full. So you generally hit the 85% mark first, but sometimes you'll hit the BUG before you hit the 85% wall. I'm still tracking it down, but until then this seems to be pretty stable and make a significant performance gain. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-09-23 19:14:11 +02:00
extent_io_tree_init(&fs_info->freed_extents[0],
fs_info->btree_inode->i_mapping);
extent_io_tree_init(&fs_info->freed_extents[1],
fs_info->btree_inode->i_mapping);
fs_info->pinned_extents = &fs_info->freed_extents[0];
fs_info->do_barriers = 1;
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 19:27:11 +02:00
mutex_init(&fs_info->ordered_operations_mutex);
mutex_init(&fs_info->tree_log_mutex);
mutex_init(&fs_info->chunk_mutex);
mutex_init(&fs_info->transaction_kthread_mutex);
mutex_init(&fs_info->cleaner_mutex);
mutex_init(&fs_info->volume_mutex);
init_rwsem(&fs_info->extent_commit_sem);
init_rwsem(&fs_info->cleanup_work_sem);
init_rwsem(&fs_info->subvol_sem);
btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
init_waitqueue_head(&fs_info->transaction_throttle);
init_waitqueue_head(&fs_info->transaction_wait);
init_waitqueue_head(&fs_info->transaction_blocked_wait);
init_waitqueue_head(&fs_info->async_submit_wait);
__setup_root(4096, 4096, 4096, 4096, tree_root,
fs_info, BTRFS_ROOT_TREE_OBJECTID);
bh = btrfs_read_dev_super(fs_devices->latest_bdev);
if (!bh) {
err = -EINVAL;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
goto fail_alloc;
}
memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
sizeof(fs_info->super_for_commit));
brelse(bh);
memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
disk_super = &fs_info->super_copy;
if (!btrfs_super_root(disk_super))
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
goto fail_alloc;
/* check FS state, whether FS is broken. */
fs_info->fs_state |= btrfs_super_flags(disk_super);
btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
/*
* In the long term, we'll store the compression type in the super
* block, and it'll be used for per file compression control.
*/
fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
ret = btrfs_parse_options(tree_root, options);
if (ret) {
err = ret;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
goto fail_alloc;
}
features = btrfs_super_incompat_flags(disk_super) &
~BTRFS_FEATURE_INCOMPAT_SUPP;
if (features) {
printk(KERN_ERR "BTRFS: couldn't mount because of "
"unsupported optional features (%Lx).\n",
(unsigned long long)features);
err = -EINVAL;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
goto fail_alloc;
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
features = btrfs_super_incompat_flags(disk_super);
features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
btrfs_set_super_incompat_flags(disk_super, features);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
features = btrfs_super_compat_ro_flags(disk_super) &
~BTRFS_FEATURE_COMPAT_RO_SUPP;
if (!(sb->s_flags & MS_RDONLY) && features) {
printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
"unsupported option features (%Lx).\n",
(unsigned long long)features);
err = -EINVAL;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
goto fail_alloc;
}
btrfs_init_workers(&fs_info->generic_worker,
"genwork", 1, NULL);
btrfs_init_workers(&fs_info->workers, "worker",
fs_info->thread_pool_size,
&fs_info->generic_worker);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
fs_info->thread_pool_size,
&fs_info->generic_worker);
btrfs_init_workers(&fs_info->submit_workers, "submit",
min_t(u64, fs_devices->num_devices,
fs_info->thread_pool_size),
&fs_info->generic_worker);
btrfs_init_workers(&fs_info->caching_workers, "cache",
2, &fs_info->generic_worker);
/* a higher idle thresh on the submit workers makes it much more
* likely that bios will be send down in a sane order to the
* devices
*/
fs_info->submit_workers.idle_thresh = 64;
fs_info->workers.idle_thresh = 16;
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 04:03:00 +01:00
fs_info->workers.ordered = 1;
fs_info->delalloc_workers.idle_thresh = 2;
fs_info->delalloc_workers.ordered = 1;
btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
&fs_info->generic_worker);
btrfs_init_workers(&fs_info->endio_workers, "endio",
fs_info->thread_pool_size,
&fs_info->generic_worker);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
fs_info->thread_pool_size,
&fs_info->generic_worker);
btrfs_init_workers(&fs_info->endio_meta_write_workers,
"endio-meta-write", fs_info->thread_pool_size,
&fs_info->generic_worker);
btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
fs_info->thread_pool_size,
&fs_info->generic_worker);
btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1, &fs_info->generic_worker);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
fs_info->thread_pool_size,
&fs_info->generic_worker);
/*
* endios are largely parallel and should have a very
* low idle thresh
*/
fs_info->endio_workers.idle_thresh = 4;
fs_info->endio_meta_workers.idle_thresh = 4;
fs_info->endio_write_workers.idle_thresh = 2;
fs_info->endio_meta_write_workers.idle_thresh = 2;
btrfs_start_workers(&fs_info->workers, 1);
btrfs_start_workers(&fs_info->generic_worker, 1);
btrfs_start_workers(&fs_info->submit_workers, 1);
btrfs_start_workers(&fs_info->delalloc_workers, 1);
btrfs_start_workers(&fs_info->fixup_workers, 1);
btrfs_start_workers(&fs_info->endio_workers, 1);
btrfs_start_workers(&fs_info->endio_meta_workers, 1);
btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
btrfs_start_workers(&fs_info->endio_write_workers, 1);
btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
btrfs_start_workers(&fs_info->delayed_workers, 1);
btrfs_start_workers(&fs_info->caching_workers, 1);
fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 19:49:59 +01:00
fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
4 * 1024 * 1024 / PAGE_CACHE_SIZE);
nodesize = btrfs_super_nodesize(disk_super);
leafsize = btrfs_super_leafsize(disk_super);
sectorsize = btrfs_super_sectorsize(disk_super);
stripesize = btrfs_super_stripesize(disk_super);
tree_root->nodesize = nodesize;
tree_root->leafsize = leafsize;
tree_root->sectorsize = sectorsize;
tree_root->stripesize = stripesize;
sb->s_blocksize = sectorsize;
sb->s_blocksize_bits = blksize_bits(sectorsize);
if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
sizeof(disk_super->magic))) {
printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
goto fail_sb_buffer;
}
mutex_lock(&fs_info->chunk_mutex);
ret = btrfs_read_sys_array(tree_root);
mutex_unlock(&fs_info->chunk_mutex);
if (ret) {
printk(KERN_WARNING "btrfs: failed to read the system "
"array on %s\n", sb->s_id);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
goto fail_sb_buffer;
}
blocksize = btrfs_level_size(tree_root,
btrfs_super_chunk_root_level(disk_super));
generation = btrfs_super_chunk_root_generation(disk_super);
__setup_root(nodesize, leafsize, sectorsize, stripesize,
chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
chunk_root->node = read_tree_block(chunk_root,
btrfs_super_chunk_root(disk_super),
blocksize, generation);
BUG_ON(!chunk_root->node);
if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
sb->s_id);
goto fail_chunk_root;
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
chunk_root->commit_root = btrfs_root_node(chunk_root);
read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
(unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
BTRFS_UUID_SIZE);
mutex_lock(&fs_info->chunk_mutex);
ret = btrfs_read_chunk_tree(chunk_root);
mutex_unlock(&fs_info->chunk_mutex);
if (ret) {
printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
sb->s_id);
goto fail_chunk_root;
}
btrfs_close_extra_devices(fs_devices);
blocksize = btrfs_level_size(tree_root,
btrfs_super_root_level(disk_super));
generation = btrfs_super_generation(disk_super);
tree_root->node = read_tree_block(tree_root,
btrfs_super_root(disk_super),
blocksize, generation);
if (!tree_root->node)
goto fail_chunk_root;
if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
sb->s_id);
goto fail_tree_root;
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
btrfs_set_root_node(&tree_root->root_item, tree_root->node);
tree_root->commit_root = btrfs_root_node(tree_root);
ret = find_and_setup_root(tree_root, fs_info,
BTRFS_EXTENT_TREE_OBJECTID, extent_root);
if (ret)
goto fail_tree_root;
extent_root->track_dirty = 1;
ret = find_and_setup_root(tree_root, fs_info,
BTRFS_DEV_TREE_OBJECTID, dev_root);
if (ret)
goto fail_extent_root;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
dev_root->track_dirty = 1;
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
ret = find_and_setup_root(tree_root, fs_info,
BTRFS_CSUM_TREE_OBJECTID, csum_root);
if (ret)
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
goto fail_dev_root;
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
csum_root->track_dirty = 1;
fs_info->generation = generation;
fs_info->last_trans_committed = generation;
fs_info->data_alloc_profile = (u64)-1;
fs_info->metadata_alloc_profile = (u64)-1;
fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
ret = btrfs_init_space_info(fs_info);
if (ret) {
printk(KERN_ERR "Failed to initial space info: %d\n", ret);
goto fail_block_groups;
}
ret = btrfs_read_block_groups(extent_root);
if (ret) {
printk(KERN_ERR "Failed to read block groups: %d\n", ret);
goto fail_block_groups;
}
fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
"btrfs-cleaner");
if (IS_ERR(fs_info->cleaner_kthread))
goto fail_block_groups;
fs_info->transaction_kthread = kthread_run(transaction_kthread,
tree_root,
"btrfs-transaction");
if (IS_ERR(fs_info->transaction_kthread))
goto fail_cleaner;
if (!btrfs_test_opt(tree_root, SSD) &&
!btrfs_test_opt(tree_root, NOSSD) &&
!fs_info->fs_devices->rotating) {
printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
"mode\n");
btrfs_set_opt(fs_info->mount_opt, SSD);
}
/* do not make disk changes in broken FS */
if (btrfs_super_log_root(disk_super) != 0 &&
!(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
u64 bytenr = btrfs_super_log_root(disk_super);
if (fs_devices->rw_devices == 0) {
printk(KERN_WARNING "Btrfs log replay required "
"on RO media\n");
err = -EIO;
goto fail_trans_kthread;
}
blocksize =
btrfs_level_size(tree_root,
btrfs_super_log_root_level(disk_super));
log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
if (!log_tree_root) {
err = -ENOMEM;
goto fail_trans_kthread;
}
__setup_root(nodesize, leafsize, sectorsize, stripesize,
log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
log_tree_root->node = read_tree_block(tree_root, bytenr,
blocksize,
generation + 1);
ret = btrfs_recover_log_trees(log_tree_root);
BUG_ON(ret);
if (sb->s_flags & MS_RDONLY) {
ret = btrfs_commit_super(tree_root);
BUG_ON(ret);
}
}
2008-09-26 16:09:34 +02:00
ret = btrfs_find_orphan_roots(tree_root);
BUG_ON(ret);
if (!(sb->s_flags & MS_RDONLY)) {
ret = btrfs_cleanup_fs_roots(fs_info);
BUG_ON(ret);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
ret = btrfs_recover_relocation(tree_root);
if (ret < 0) {
printk(KERN_WARNING
"btrfs: failed to recover relocation\n");
err = -EINVAL;
goto fail_trans_kthread;
}
}
2008-09-26 16:09:34 +02:00
location.objectid = BTRFS_FS_TREE_OBJECTID;
location.type = BTRFS_ROOT_ITEM_KEY;
location.offset = (u64)-1;
fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
if (!fs_info->fs_root)
goto fail_trans_kthread;
if (IS_ERR(fs_info->fs_root)) {
err = PTR_ERR(fs_info->fs_root);
goto fail_trans_kthread;
}
if (!(sb->s_flags & MS_RDONLY)) {
down_read(&fs_info->cleanup_work_sem);
err = btrfs_orphan_cleanup(fs_info->fs_root);
if (!err)
err = btrfs_orphan_cleanup(fs_info->tree_root);
up_read(&fs_info->cleanup_work_sem);
if (err) {
close_ctree(tree_root);
return ERR_PTR(err);
}
}
return tree_root;
fail_trans_kthread:
kthread_stop(fs_info->transaction_kthread);
fail_cleaner:
kthread_stop(fs_info->cleaner_kthread);
/*
* make sure we're done with the btree inode before we stop our
* kthreads
*/
filemap_write_and_wait(fs_info->btree_inode->i_mapping);
invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
fail_block_groups:
btrfs_free_block_groups(fs_info);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
free_extent_buffer(csum_root->node);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
free_extent_buffer(csum_root->commit_root);
fail_dev_root:
free_extent_buffer(dev_root->node);
free_extent_buffer(dev_root->commit_root);
fail_extent_root:
free_extent_buffer(extent_root->node);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
free_extent_buffer(extent_root->commit_root);
fail_tree_root:
free_extent_buffer(tree_root->node);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
free_extent_buffer(tree_root->commit_root);
fail_chunk_root:
free_extent_buffer(chunk_root->node);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
free_extent_buffer(chunk_root->commit_root);
fail_sb_buffer:
btrfs_stop_workers(&fs_info->generic_worker);
btrfs_stop_workers(&fs_info->fixup_workers);
btrfs_stop_workers(&fs_info->delalloc_workers);
btrfs_stop_workers(&fs_info->workers);
btrfs_stop_workers(&fs_info->endio_workers);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
btrfs_stop_workers(&fs_info->endio_meta_workers);
btrfs_stop_workers(&fs_info->endio_meta_write_workers);
btrfs_stop_workers(&fs_info->endio_write_workers);
btrfs_stop_workers(&fs_info->endio_freespace_worker);
btrfs_stop_workers(&fs_info->submit_workers);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
btrfs_stop_workers(&fs_info->delayed_workers);
btrfs_stop_workers(&fs_info->caching_workers);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
fail_alloc:
kfree(fs_info->delayed_root);
fail_iput:
invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
iput(fs_info->btree_inode);
btrfs_close_devices(fs_info->fs_devices);
btrfs_mapping_tree_free(&fs_info->mapping_tree);
fail_bdi:
bdi_destroy(&fs_info->bdi);
fail_srcu:
cleanup_srcu_struct(&fs_info->subvol_srcu);
fail:
kfree(extent_root);
kfree(tree_root);
kfree(fs_info);
kfree(chunk_root);
kfree(dev_root);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
kfree(csum_root);
return ERR_PTR(err);
}
static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
{
char b[BDEVNAME_SIZE];
if (uptodate) {
set_buffer_uptodate(bh);
} else {
printk_ratelimited(KERN_WARNING "lost page write due to "
"I/O error on %s\n",
bdevname(bh->b_bdev, b));
/* note, we dont' set_buffer_write_io_error because we have
* our own ways of dealing with the IO errors
*/
clear_buffer_uptodate(bh);
}
unlock_buffer(bh);
put_bh(bh);
}
struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
{
struct buffer_head *bh;
struct buffer_head *latest = NULL;
struct btrfs_super_block *super;
int i;
u64 transid = 0;
u64 bytenr;
/* we would like to check all the supers, but that would make
* a btrfs mount succeed after a mkfs from a different FS.
* So, we need to add a special mount option to scan for
* later supers, using BTRFS_SUPER_MIRROR_MAX instead
*/
for (i = 0; i < 1; i++) {
bytenr = btrfs_sb_offset(i);
if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
break;
bh = __bread(bdev, bytenr / 4096, 4096);
if (!bh)
continue;
super = (struct btrfs_super_block *)bh->b_data;
if (btrfs_super_bytenr(super) != bytenr ||
strncmp((char *)(&super->magic), BTRFS_MAGIC,
sizeof(super->magic))) {
brelse(bh);
continue;
}
if (!latest || btrfs_super_generation(super) > transid) {
brelse(latest);
latest = bh;
transid = btrfs_super_generation(super);
} else {
brelse(bh);
}
}
return latest;
}
/*
* this should be called twice, once with wait == 0 and
* once with wait == 1. When wait == 0 is done, all the buffer heads
* we write are pinned.
*
* They are released when wait == 1 is done.
* max_mirrors must be the same for both runs, and it indicates how
* many supers on this one device should be written.
*
* max_mirrors == 0 means to write them all.
*/
static int write_dev_supers(struct btrfs_device *device,
struct btrfs_super_block *sb,
int do_barriers, int wait, int max_mirrors)
{
struct buffer_head *bh;
int i;
int ret;
int errors = 0;
u32 crc;
u64 bytenr;
int last_barrier = 0;
if (max_mirrors == 0)
max_mirrors = BTRFS_SUPER_MIRROR_MAX;
/* make sure only the last submit_bh does a barrier */
if (do_barriers) {
for (i = 0; i < max_mirrors; i++) {
bytenr = btrfs_sb_offset(i);
if (bytenr + BTRFS_SUPER_INFO_SIZE >=
device->total_bytes)
break;
last_barrier = i;
}
}
for (i = 0; i < max_mirrors; i++) {
bytenr = btrfs_sb_offset(i);
if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
break;
if (wait) {
bh = __find_get_block(device->bdev, bytenr / 4096,
BTRFS_SUPER_INFO_SIZE);
BUG_ON(!bh);
wait_on_buffer(bh);
if (!buffer_uptodate(bh))
errors++;
/* drop our reference */
brelse(bh);
/* drop the reference from the wait == 0 run */
brelse(bh);
continue;
} else {
btrfs_set_super_bytenr(sb, bytenr);
crc = ~(u32)0;
crc = btrfs_csum_data(NULL, (char *)sb +
BTRFS_CSUM_SIZE, crc,
BTRFS_SUPER_INFO_SIZE -
BTRFS_CSUM_SIZE);
btrfs_csum_final(crc, sb->csum);
/*
* one reference for us, and we leave it for the
* caller
*/
bh = __getblk(device->bdev, bytenr / 4096,
BTRFS_SUPER_INFO_SIZE);
memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
/* one reference for submit_bh */
get_bh(bh);
set_buffer_uptodate(bh);
lock_buffer(bh);
bh->b_end_io = btrfs_end_buffer_write_sync;
}
if (i == last_barrier && do_barriers)
ret = submit_bh(WRITE_FLUSH_FUA, bh);
else
ret = submit_bh(WRITE_SYNC, bh);
if (ret)
errors++;
}
return errors < i ? 0 : -1;
}
int write_all_supers(struct btrfs_root *root, int max_mirrors)
{
struct list_head *head;
struct btrfs_device *dev;
struct btrfs_super_block *sb;
struct btrfs_dev_item *dev_item;
int ret;
int do_barriers;
int max_errors;
int total_errors = 0;
u64 flags;
max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
do_barriers = !btrfs_test_opt(root, NOBARRIER);
sb = &root->fs_info->super_for_commit;
dev_item = &sb->dev_item;
mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
head = &root->fs_info->fs_devices->devices;
list_for_each_entry_rcu(dev, head, dev_list) {
if (!dev->bdev) {
total_errors++;
continue;
}
if (!dev->in_fs_metadata || !dev->writeable)
continue;
btrfs_set_stack_device_generation(dev_item, 0);
btrfs_set_stack_device_type(dev_item, dev->type);
btrfs_set_stack_device_id(dev_item, dev->devid);
btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
btrfs_set_stack_device_io_align(dev_item, dev->io_align);
btrfs_set_stack_device_io_width(dev_item, dev->io_width);
btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
flags = btrfs_super_flags(sb);
btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
if (ret)
total_errors++;
}
if (total_errors > max_errors) {
printk(KERN_ERR "btrfs: %d errors while writing supers\n",
total_errors);
BUG();
}
total_errors = 0;
list_for_each_entry_rcu(dev, head, dev_list) {
if (!dev->bdev)
continue;
if (!dev->in_fs_metadata || !dev->writeable)
continue;
ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
if (ret)
total_errors++;
}
mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
if (total_errors > max_errors) {
printk(KERN_ERR "btrfs: %d errors while writing supers\n",
total_errors);
BUG();
}
return 0;
}
int write_ctree_super(struct btrfs_trans_handle *trans,
struct btrfs_root *root, int max_mirrors)
{
int ret;
ret = write_all_supers(root, max_mirrors);
return ret;
}
int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
{
spin_lock(&fs_info->fs_roots_radix_lock);
radix_tree_delete(&fs_info->fs_roots_radix,
(unsigned long)root->root_key.objectid);
spin_unlock(&fs_info->fs_roots_radix_lock);
if (btrfs_root_refs(&root->root_item) == 0)
synchronize_srcu(&fs_info->subvol_srcu);
__btrfs_remove_free_space_cache(root->free_ino_pinned);
__btrfs_remove_free_space_cache(root->free_ino_ctl);
free_fs_root(root);
return 0;
}
static void free_fs_root(struct btrfs_root *root)
{
iput(root->cache_inode);
WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
if (root->anon_dev)
free_anon_bdev(root->anon_dev);
free_extent_buffer(root->node);
free_extent_buffer(root->commit_root);
kfree(root->free_ino_ctl);
kfree(root->free_ino_pinned);
kfree(root->name);
kfree(root);
}
static int del_fs_roots(struct btrfs_fs_info *fs_info)
{
int ret;
struct btrfs_root *gang[8];
int i;
while (!list_empty(&fs_info->dead_roots)) {
gang[0] = list_entry(fs_info->dead_roots.next,
struct btrfs_root, root_list);
list_del(&gang[0]->root_list);
if (gang[0]->in_radix) {
btrfs_free_fs_root(fs_info, gang[0]);
} else {
free_extent_buffer(gang[0]->node);
free_extent_buffer(gang[0]->commit_root);
kfree(gang[0]);
}
}
while (1) {
ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
(void **)gang, 0,
ARRAY_SIZE(gang));
if (!ret)
break;
for (i = 0; i < ret; i++)
btrfs_free_fs_root(fs_info, gang[i]);
}
return 0;
}
int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
{
u64 root_objectid = 0;
struct btrfs_root *gang[8];
int i;
int ret;
while (1) {
ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
(void **)gang, root_objectid,
ARRAY_SIZE(gang));
if (!ret)
break;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
root_objectid = gang[ret - 1]->root_key.objectid + 1;
for (i = 0; i < ret; i++) {
int err;
root_objectid = gang[i]->root_key.objectid;
err = btrfs_orphan_cleanup(gang[i]);
if (err)
return err;
}
root_objectid++;
}
return 0;
}
int btrfs_commit_super(struct btrfs_root *root)
{
struct btrfs_trans_handle *trans;
int ret;
mutex_lock(&root->fs_info->cleaner_mutex);
btrfs_run_delayed_iputs(root);
btrfs_clean_old_snapshots(root);
mutex_unlock(&root->fs_info->cleaner_mutex);
/* wait until ongoing cleanup work done */
down_write(&root->fs_info->cleanup_work_sem);
up_write(&root->fs_info->cleanup_work_sem);
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
return PTR_ERR(trans);
ret = btrfs_commit_transaction(trans, root);
BUG_ON(ret);
/* run commit again to drop the original snapshot */
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
return PTR_ERR(trans);
btrfs_commit_transaction(trans, root);
ret = btrfs_write_and_wait_transaction(NULL, root);
BUG_ON(ret);
ret = write_ctree_super(NULL, root, 0);
return ret;
}
int close_ctree(struct btrfs_root *root)
{
struct btrfs_fs_info *fs_info = root->fs_info;
int ret;
fs_info->closing = 1;
smp_mb();
btrfs_scrub_cancel(root);
/* wait for any defraggers to finish */
wait_event(fs_info->transaction_wait,
(atomic_read(&fs_info->defrag_running) == 0));
/* clear out the rbtree of defraggable inodes */
btrfs_run_defrag_inodes(root->fs_info);
btrfs_put_block_group_cache(fs_info);
/*
* Here come 2 situations when btrfs is broken to flip readonly:
*
* 1. when btrfs flips readonly somewhere else before
* btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
* and btrfs will skip to write sb directly to keep
* ERROR state on disk.
*
* 2. when btrfs flips readonly just in btrfs_commit_super,
* and in such case, btrfs cannot write sb via btrfs_commit_super,
* and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
* btrfs will cleanup all FS resources first and write sb then.
*/
if (!(fs_info->sb->s_flags & MS_RDONLY)) {
ret = btrfs_commit_super(root);
if (ret)
printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
}
if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
ret = btrfs_error_commit_super(root);
if (ret)
printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
}
kthread_stop(root->fs_info->transaction_kthread);
kthread_stop(root->fs_info->cleaner_kthread);
fs_info->closing = 2;
smp_mb();
if (fs_info->delalloc_bytes) {
printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
(unsigned long long)fs_info->delalloc_bytes);
}
if (fs_info->total_ref_cache_size) {
printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
(unsigned long long)fs_info->total_ref_cache_size);
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 16:45:14 +02:00
free_extent_buffer(fs_info->extent_root->node);
free_extent_buffer(fs_info->extent_root->commit_root);
free_extent_buffer(fs_info->tree_root->node);
free_extent_buffer(fs_info->tree_root->commit_root);
free_extent_buffer(root->fs_info->chunk_root->node);
free_extent_buffer(root->fs_info->chunk_root->commit_root);
free_extent_buffer(root->fs_info->dev_root->node);
free_extent_buffer(root->fs_info->dev_root->commit_root);
free_extent_buffer(root->fs_info->csum_root->node);
free_extent_buffer(root->fs_info->csum_root->commit_root);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
btrfs_free_block_groups(root->fs_info);
del_fs_roots(fs_info);
iput(fs_info->btree_inode);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
kfree(fs_info->delayed_root);
btrfs_stop_workers(&fs_info->generic_worker);
btrfs_stop_workers(&fs_info->fixup_workers);
btrfs_stop_workers(&fs_info->delalloc_workers);
btrfs_stop_workers(&fs_info->workers);
btrfs_stop_workers(&fs_info->endio_workers);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
btrfs_stop_workers(&fs_info->endio_meta_workers);
btrfs_stop_workers(&fs_info->endio_meta_write_workers);
btrfs_stop_workers(&fs_info->endio_write_workers);
btrfs_stop_workers(&fs_info->endio_freespace_worker);
btrfs_stop_workers(&fs_info->submit_workers);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
btrfs_stop_workers(&fs_info->delayed_workers);
btrfs_stop_workers(&fs_info->caching_workers);
btrfs_close_devices(fs_info->fs_devices);
btrfs_mapping_tree_free(&fs_info->mapping_tree);
bdi_destroy(&fs_info->bdi);
cleanup_srcu_struct(&fs_info->subvol_srcu);
kfree(fs_info->extent_root);
kfree(fs_info->tree_root);
kfree(fs_info->chunk_root);
kfree(fs_info->dev_root);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 22:58:54 +01:00
kfree(fs_info->csum_root);
kfree(fs_info);
return 0;
}
int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
{
int ret;
struct inode *btree_inode = buf->first_page->mapping->host;
ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
NULL);
if (!ret)
return ret;
ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
parent_transid);
return !ret;
}
int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
{
struct inode *btree_inode = buf->first_page->mapping->host;
return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
buf);
}
void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
{
struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
u64 transid = btrfs_header_generation(buf);
struct inode *btree_inode = root->fs_info->btree_inode;
int was_dirty;
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 15:25:08 +01:00
btrfs_assert_tree_locked(buf);
if (transid != root->fs_info->generation) {
printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
"found %llu running %llu\n",
(unsigned long long)buf->start,
(unsigned long long)transid,
(unsigned long long)root->fs_info->generation);
WARN_ON(1);
}
was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
buf);
if (!was_dirty) {
spin_lock(&root->fs_info->delalloc_lock);
root->fs_info->dirty_metadata_bytes += buf->len;
spin_unlock(&root->fs_info->delalloc_lock);
}
}
void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 12:12:22 +02:00
{
/*
* looks as though older kernels can get into trouble with
* this code, they end up stuck in balance_dirty_pages forever
*/
u64 num_dirty;
unsigned long thresh = 32 * 1024 * 1024;
if (current->flags & PF_MEMALLOC)
return;
btrfs_balance_delayed_items(root);
num_dirty = root->fs_info->dirty_metadata_bytes;
if (num_dirty > thresh) {
balance_dirty_pages_ratelimited_nr(
root->fs_info->btree_inode->i_mapping, 1);
}
return;
}
void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
{
/*
* looks as though older kernels can get into trouble with
* this code, they end up stuck in balance_dirty_pages forever
*/
u64 num_dirty;
unsigned long thresh = 32 * 1024 * 1024;
if (current->flags & PF_MEMALLOC)
return;
num_dirty = root->fs_info->dirty_metadata_bytes;
if (num_dirty > thresh) {
balance_dirty_pages_ratelimited_nr(
root->fs_info->btree_inode->i_mapping, 1);
}
return;
}
int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
{
struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
int ret;
ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
if (ret == 0)
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 15:25:08 +01:00
set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
return ret;
}
int btree_lock_page_hook(struct page *page)
{
struct inode *inode = page->mapping->host;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct extent_buffer *eb;
unsigned long len;
u64 bytenr = page_offset(page);
if (page->private == EXTENT_PAGE_PRIVATE)
goto out;
len = page->private >> 2;
eb = find_extent_buffer(io_tree, bytenr, len);
if (!eb)
goto out;
btrfs_tree_lock(eb);
btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
spin_lock(&root->fs_info->delalloc_lock);
if (root->fs_info->dirty_metadata_bytes >= eb->len)
root->fs_info->dirty_metadata_bytes -= eb->len;
else
WARN_ON(1);
spin_unlock(&root->fs_info->delalloc_lock);
}
btrfs_tree_unlock(eb);
free_extent_buffer(eb);
out:
lock_page(page);
return 0;
}
static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
int read_only)
{
if (read_only)
return;
if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
printk(KERN_WARNING "warning: mount fs with errors, "
"running btrfsck is recommended\n");
}
int btrfs_error_commit_super(struct btrfs_root *root)
{
int ret;
mutex_lock(&root->fs_info->cleaner_mutex);
btrfs_run_delayed_iputs(root);
mutex_unlock(&root->fs_info->cleaner_mutex);
down_write(&root->fs_info->cleanup_work_sem);
up_write(&root->fs_info->cleanup_work_sem);
/* cleanup FS via transaction */
btrfs_cleanup_transaction(root);
ret = write_ctree_super(NULL, root, 0);
return ret;
}
static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
{
struct btrfs_inode *btrfs_inode;
struct list_head splice;
INIT_LIST_HEAD(&splice);
mutex_lock(&root->fs_info->ordered_operations_mutex);
spin_lock(&root->fs_info->ordered_extent_lock);
list_splice_init(&root->fs_info->ordered_operations, &splice);
while (!list_empty(&splice)) {
btrfs_inode = list_entry(splice.next, struct btrfs_inode,
ordered_operations);
list_del_init(&btrfs_inode->ordered_operations);
btrfs_invalidate_inodes(btrfs_inode->root);
}
spin_unlock(&root->fs_info->ordered_extent_lock);
mutex_unlock(&root->fs_info->ordered_operations_mutex);
return 0;
}
static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
{
struct list_head splice;
struct btrfs_ordered_extent *ordered;
struct inode *inode;
INIT_LIST_HEAD(&splice);
spin_lock(&root->fs_info->ordered_extent_lock);
list_splice_init(&root->fs_info->ordered_extents, &splice);
while (!list_empty(&splice)) {
ordered = list_entry(splice.next, struct btrfs_ordered_extent,
root_extent_list);
list_del_init(&ordered->root_extent_list);
atomic_inc(&ordered->refs);
/* the inode may be getting freed (in sys_unlink path). */
inode = igrab(ordered->inode);
spin_unlock(&root->fs_info->ordered_extent_lock);
if (inode)
iput(inode);
atomic_set(&ordered->refs, 1);
btrfs_put_ordered_extent(ordered);
spin_lock(&root->fs_info->ordered_extent_lock);
}
spin_unlock(&root->fs_info->ordered_extent_lock);
return 0;
}
static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
struct btrfs_root *root)
{
struct rb_node *node;
struct btrfs_delayed_ref_root *delayed_refs;
struct btrfs_delayed_ref_node *ref;
int ret = 0;
delayed_refs = &trans->delayed_refs;
spin_lock(&delayed_refs->lock);
if (delayed_refs->num_entries == 0) {
spin_unlock(&delayed_refs->lock);
printk(KERN_INFO "delayed_refs has NO entry\n");
return ret;
}
node = rb_first(&delayed_refs->root);
while (node) {
ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
node = rb_next(node);
ref->in_tree = 0;
rb_erase(&ref->rb_node, &delayed_refs->root);
delayed_refs->num_entries--;
atomic_set(&ref->refs, 1);
if (btrfs_delayed_ref_is_head(ref)) {
struct btrfs_delayed_ref_head *head;
head = btrfs_delayed_node_to_head(ref);
mutex_lock(&head->mutex);
kfree(head->extent_op);
delayed_refs->num_heads--;
if (list_empty(&head->cluster))
delayed_refs->num_heads_ready--;
list_del_init(&head->cluster);
mutex_unlock(&head->mutex);
}
spin_unlock(&delayed_refs->lock);
btrfs_put_delayed_ref(ref);
cond_resched();
spin_lock(&delayed_refs->lock);
}
spin_unlock(&delayed_refs->lock);
return ret;
}
static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
{
struct btrfs_pending_snapshot *snapshot;
struct list_head splice;
INIT_LIST_HEAD(&splice);
list_splice_init(&t->pending_snapshots, &splice);
while (!list_empty(&splice)) {
snapshot = list_entry(splice.next,
struct btrfs_pending_snapshot,
list);
list_del_init(&snapshot->list);
kfree(snapshot);
}
return 0;
}
static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
{
struct btrfs_inode *btrfs_inode;
struct list_head splice;
INIT_LIST_HEAD(&splice);
spin_lock(&root->fs_info->delalloc_lock);
list_splice_init(&root->fs_info->delalloc_inodes, &splice);
while (!list_empty(&splice)) {
btrfs_inode = list_entry(splice.next, struct btrfs_inode,
delalloc_inodes);
list_del_init(&btrfs_inode->delalloc_inodes);
btrfs_invalidate_inodes(btrfs_inode->root);
}
spin_unlock(&root->fs_info->delalloc_lock);
return 0;
}
static int btrfs_destroy_marked_extents(struct btrfs_root *root,
struct extent_io_tree *dirty_pages,
int mark)
{
int ret;
struct page *page;
struct inode *btree_inode = root->fs_info->btree_inode;
struct extent_buffer *eb;
u64 start = 0;
u64 end;
u64 offset;
unsigned long index;
while (1) {
ret = find_first_extent_bit(dirty_pages, start, &start, &end,
mark);
if (ret)
break;
clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
while (start <= end) {
index = start >> PAGE_CACHE_SHIFT;
start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
page = find_get_page(btree_inode->i_mapping, index);
if (!page)
continue;
offset = page_offset(page);
spin_lock(&dirty_pages->buffer_lock);
eb = radix_tree_lookup(
&(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
offset >> PAGE_CACHE_SHIFT);
spin_unlock(&dirty_pages->buffer_lock);
if (eb) {
ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
&eb->bflags);
atomic_set(&eb->refs, 1);
}
if (PageWriteback(page))
end_page_writeback(page);
lock_page(page);
if (PageDirty(page)) {
clear_page_dirty_for_io(page);
spin_lock_irq(&page->mapping->tree_lock);
radix_tree_tag_clear(&page->mapping->page_tree,
page_index(page),
PAGECACHE_TAG_DIRTY);
spin_unlock_irq(&page->mapping->tree_lock);
}
page->mapping->a_ops->invalidatepage(page, 0);
unlock_page(page);
}
}
return ret;
}
static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
struct extent_io_tree *pinned_extents)
{
struct extent_io_tree *unpin;
u64 start;
u64 end;
int ret;
unpin = pinned_extents;
while (1) {
ret = find_first_extent_bit(unpin, 0, &start, &end,
EXTENT_DIRTY);
if (ret)
break;
/* opt_discard */
if (btrfs_test_opt(root, DISCARD))
ret = btrfs_error_discard_extent(root, start,
end + 1 - start,
NULL);
clear_extent_dirty(unpin, start, end, GFP_NOFS);
btrfs_error_unpin_extent_range(root, start, end);
cond_resched();
}
return 0;
}
static int btrfs_cleanup_transaction(struct btrfs_root *root)
{
struct btrfs_transaction *t;
LIST_HEAD(list);
WARN_ON(1);
mutex_lock(&root->fs_info->transaction_kthread_mutex);
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_lock(&root->fs_info->trans_lock);
list_splice_init(&root->fs_info->trans_list, &list);
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
root->fs_info->trans_no_join = 1;
spin_unlock(&root->fs_info->trans_lock);
while (!list_empty(&list)) {
t = list_entry(list.next, struct btrfs_transaction, list);
if (!t)
break;
btrfs_destroy_ordered_operations(root);
btrfs_destroy_ordered_extents(root);
btrfs_destroy_delayed_refs(t, root);
btrfs_block_rsv_release(root,
&root->fs_info->trans_block_rsv,
t->dirty_pages.dirty_bytes);
/* FIXME: cleanup wait for commit */
t->in_commit = 1;
t->blocked = 1;
if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
wake_up(&root->fs_info->transaction_blocked_wait);
t->blocked = 0;
if (waitqueue_active(&root->fs_info->transaction_wait))
wake_up(&root->fs_info->transaction_wait);
t->commit_done = 1;
if (waitqueue_active(&t->commit_wait))
wake_up(&t->commit_wait);
btrfs_destroy_pending_snapshots(t);
btrfs_destroy_delalloc_inodes(root);
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_lock(&root->fs_info->trans_lock);
root->fs_info->running_transaction = NULL;
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_unlock(&root->fs_info->trans_lock);
btrfs_destroy_marked_extents(root, &t->dirty_pages,
EXTENT_DIRTY);
btrfs_destroy_pinned_extent(root,
root->fs_info->pinned_extents);
atomic_set(&t->use_count, 0);
list_del_init(&t->list);
memset(t, 0, sizeof(*t));
kmem_cache_free(btrfs_transaction_cachep, t);
}
Btrfs: kill trans_mutex We use trans_mutex for lots of things, here's a basic list 1) To serialize trans_handles joining the currently running transaction 2) To make sure that no new trans handles are started while we are committing 3) To protect the dead_roots list and the transaction lists Really the serializing trans_handles joining is not too hard, and can really get bogged down in acquiring a reference to the transaction. So replace the trans_mutex with a trans_lock spinlock and use it to do the following 1) Protect fs_info->running_transaction. All trans handles have to do is check this, and then take a reference of the transaction and keep on going. 2) Protect the fs_info->trans_list. This doesn't get used too much, basically it just holds the current transactions, which will usually just be the currently committing transaction and the currently running transaction at most. 3) Protect the dead roots list. This is only ever processed by splicing the list so this is relatively simple. 4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using the trans_mutex before, so this is a pretty straightforward change. 5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over the entirety of the commit we need to have a way to block new people from creating a new transaction while we're doing our work. So we set no_trans_join and in join_transaction we test to see if that is set, and if it is we do a wait_on_commit. 6) Make the transaction use count atomic so we don't need to take locks to modify it when we're dropping references. 7) Add a commit_lock to the transaction to make sure multiple people trying to commit the same transaction don't race and commit at the same time. 8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl trans. I have tested this with xfstests, but obviously it is a pretty hairy change so lots of testing is greatly appreciated. Thanks, Signed-off-by: Josef Bacik <josef@redhat.com>
2011-04-11 23:25:13 +02:00
spin_lock(&root->fs_info->trans_lock);
root->fs_info->trans_no_join = 0;
spin_unlock(&root->fs_info->trans_lock);
mutex_unlock(&root->fs_info->transaction_kthread_mutex);
return 0;
}
static struct extent_io_ops btree_extent_io_ops = {
.write_cache_pages_lock_hook = btree_lock_page_hook,
.readpage_end_io_hook = btree_readpage_end_io_hook,
.submit_bio_hook = btree_submit_bio_hook,
/* note we're sharing with inode.c for the merge bio hook */
.merge_bio_hook = btrfs_merge_bio_hook,
};