2008-01-08 21:46:30 +01:00
|
|
|
/*
|
|
|
|
* 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/slab.h>
|
2008-04-30 19:59:35 +02:00
|
|
|
#include <linux/blkdev.h>
|
2008-07-22 17:18:09 +02:00
|
|
|
#include <linux/writeback.h>
|
|
|
|
#include <linux/pagevec.h>
|
2008-01-08 21:46:30 +01:00
|
|
|
#include "ctree.h"
|
|
|
|
#include "transaction.h"
|
|
|
|
#include "btrfs_inode.h"
|
2008-07-17 18:53:50 +02:00
|
|
|
#include "extent_io.h"
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
static u64 entry_end(struct btrfs_ordered_extent *entry)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
2008-07-17 18:53:50 +02:00
|
|
|
if (entry->file_offset + entry->len < entry->file_offset)
|
|
|
|
return (u64)-1;
|
|
|
|
return entry->file_offset + entry->len;
|
2008-01-08 21:46:30 +01:00
|
|
|
}
|
|
|
|
|
2008-09-29 21:18:18 +02:00
|
|
|
/* returns NULL if the insertion worked, or it returns the node it did find
|
|
|
|
* in the tree
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
|
|
|
|
struct rb_node *node)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
2009-01-06 03:25:51 +01:00
|
|
|
struct rb_node **p = &root->rb_node;
|
|
|
|
struct rb_node *parent = NULL;
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_extent *entry;
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2009-01-06 03:25:51 +01:00
|
|
|
while (*p) {
|
2008-01-08 21:46:30 +01:00
|
|
|
parent = *p;
|
2008-07-17 18:53:50 +02:00
|
|
|
entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
if (file_offset < entry->file_offset)
|
2008-01-08 21:46:30 +01:00
|
|
|
p = &(*p)->rb_left;
|
2008-07-17 18:53:50 +02:00
|
|
|
else if (file_offset >= entry_end(entry))
|
2008-01-08 21:46:30 +01:00
|
|
|
p = &(*p)->rb_right;
|
|
|
|
else
|
|
|
|
return parent;
|
|
|
|
}
|
|
|
|
|
|
|
|
rb_link_node(node, parent, p);
|
|
|
|
rb_insert_color(node, root);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2008-09-29 21:18:18 +02:00
|
|
|
/*
|
|
|
|
* look for a given offset in the tree, and if it can't be found return the
|
|
|
|
* first lesser offset
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
|
|
|
|
struct rb_node **prev_ret)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
2009-01-06 03:25:51 +01:00
|
|
|
struct rb_node *n = root->rb_node;
|
2008-01-08 21:46:30 +01:00
|
|
|
struct rb_node *prev = NULL;
|
2008-07-17 18:53:50 +02:00
|
|
|
struct rb_node *test;
|
|
|
|
struct btrfs_ordered_extent *entry;
|
|
|
|
struct btrfs_ordered_extent *prev_entry = NULL;
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2009-01-06 03:25:51 +01:00
|
|
|
while (n) {
|
2008-07-17 18:53:50 +02:00
|
|
|
entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
|
2008-01-08 21:46:30 +01:00
|
|
|
prev = n;
|
|
|
|
prev_entry = entry;
|
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
if (file_offset < entry->file_offset)
|
2008-01-08 21:46:30 +01:00
|
|
|
n = n->rb_left;
|
2008-07-17 18:53:50 +02:00
|
|
|
else if (file_offset >= entry_end(entry))
|
2008-01-08 21:46:30 +01:00
|
|
|
n = n->rb_right;
|
|
|
|
else
|
|
|
|
return n;
|
|
|
|
}
|
|
|
|
if (!prev_ret)
|
|
|
|
return NULL;
|
|
|
|
|
2009-01-06 03:25:51 +01:00
|
|
|
while (prev && file_offset >= entry_end(prev_entry)) {
|
2008-07-17 18:53:50 +02:00
|
|
|
test = rb_next(prev);
|
|
|
|
if (!test)
|
|
|
|
break;
|
|
|
|
prev_entry = rb_entry(test, struct btrfs_ordered_extent,
|
|
|
|
rb_node);
|
|
|
|
if (file_offset < entry_end(prev_entry))
|
|
|
|
break;
|
|
|
|
|
|
|
|
prev = test;
|
|
|
|
}
|
|
|
|
if (prev)
|
|
|
|
prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
|
|
|
|
rb_node);
|
2009-01-06 03:25:51 +01:00
|
|
|
while (prev && file_offset < entry_end(prev_entry)) {
|
2008-07-17 18:53:50 +02:00
|
|
|
test = rb_prev(prev);
|
|
|
|
if (!test)
|
|
|
|
break;
|
|
|
|
prev_entry = rb_entry(test, struct btrfs_ordered_extent,
|
|
|
|
rb_node);
|
|
|
|
prev = test;
|
2008-01-08 21:46:30 +01:00
|
|
|
}
|
|
|
|
*prev_ret = prev;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2008-09-29 21:18:18 +02:00
|
|
|
/*
|
|
|
|
* helper to check if a given offset is inside a given entry
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
|
|
|
|
{
|
|
|
|
if (file_offset < entry->file_offset ||
|
|
|
|
entry->file_offset + entry->len <= file_offset)
|
|
|
|
return 0;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2008-09-29 21:18:18 +02:00
|
|
|
/*
|
|
|
|
* look find the first ordered struct that has this offset, otherwise
|
|
|
|
* the first one less than this offset
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
|
|
|
|
u64 file_offset)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
2008-07-17 18:53:50 +02:00
|
|
|
struct rb_root *root = &tree->tree;
|
2008-01-08 21:46:30 +01:00
|
|
|
struct rb_node *prev;
|
|
|
|
struct rb_node *ret;
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_extent *entry;
|
|
|
|
|
|
|
|
if (tree->last) {
|
|
|
|
entry = rb_entry(tree->last, struct btrfs_ordered_extent,
|
|
|
|
rb_node);
|
|
|
|
if (offset_in_entry(entry, file_offset))
|
|
|
|
return tree->last;
|
|
|
|
}
|
|
|
|
ret = __tree_search(root, file_offset, &prev);
|
2008-01-08 21:46:30 +01:00
|
|
|
if (!ret)
|
2008-07-17 18:53:50 +02:00
|
|
|
ret = prev;
|
|
|
|
if (ret)
|
|
|
|
tree->last = ret;
|
2008-01-08 21:46:30 +01:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/* allocate and add a new ordered_extent into the per-inode tree.
|
|
|
|
* file_offset is the logical offset in the file
|
|
|
|
*
|
|
|
|
* start is the disk block number of an extent already reserved in the
|
|
|
|
* extent allocation tree
|
|
|
|
*
|
|
|
|
* len is the length of the extent
|
|
|
|
*
|
|
|
|
* The tree is given a single reference on the ordered extent that was
|
|
|
|
* inserted.
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
|
2008-10-30 19:20:02 +01:00
|
|
|
u64 start, u64 len, u64 disk_len, int type)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
2008-07-17 18:53:50 +02:00
|
|
|
struct rb_node *node;
|
|
|
|
struct btrfs_ordered_extent *entry;
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
|
|
|
entry = kzalloc(sizeof(*entry), GFP_NOFS);
|
2008-01-08 21:46:30 +01:00
|
|
|
if (!entry)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
entry->file_offset = file_offset;
|
|
|
|
entry->start = start;
|
|
|
|
entry->len = len;
|
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
|
|
|
entry->disk_len = disk_len;
|
2009-09-02 22:53:46 +02:00
|
|
|
entry->bytes_left = len;
|
2008-07-24 17:57:52 +02:00
|
|
|
entry->inode = inode;
|
2008-10-30 19:25:28 +01:00
|
|
|
if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
|
2008-10-30 19:20:02 +01:00
|
|
|
set_bit(type, &entry->flags);
|
2008-07-24 17:57:52 +02:00
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
/* one ref for the tree */
|
|
|
|
atomic_set(&entry->refs, 1);
|
|
|
|
init_waitqueue_head(&entry->wait);
|
|
|
|
INIT_LIST_HEAD(&entry->list);
|
2008-07-24 17:57:52 +02:00
|
|
|
INIT_LIST_HEAD(&entry->root_extent_list);
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
node = tree_insert(&tree->tree, file_offset,
|
|
|
|
&entry->rb_node);
|
2009-01-06 03:25:51 +01:00
|
|
|
BUG_ON(node);
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2009-01-06 03:25:51 +01:00
|
|
|
|
2008-07-24 17:57:52 +02:00
|
|
|
spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
|
|
|
|
list_add_tail(&entry->root_extent_list,
|
|
|
|
&BTRFS_I(inode)->root->fs_info->ordered_extents);
|
|
|
|
spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
|
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
BUG_ON(node);
|
2008-01-08 21:46:30 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* Add a struct btrfs_ordered_sum into the list of checksums to be inserted
|
2008-07-18 12:17:13 +02:00
|
|
|
* when an ordered extent is finished. If the list covers more than one
|
|
|
|
* ordered extent, it is split across multiples.
|
2008-07-17 19:53:27 +02:00
|
|
|
*/
|
2008-07-18 12:17:13 +02:00
|
|
|
int btrfs_add_ordered_sum(struct inode *inode,
|
|
|
|
struct btrfs_ordered_extent *entry,
|
|
|
|
struct btrfs_ordered_sum *sum)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
list_add_tail(&sum->list, &entry->list);
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
return 0;
|
2008-01-08 21:46:30 +01:00
|
|
|
}
|
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* this is used to account for finished IO across a given range
|
|
|
|
* of the file. The IO should not span ordered extents. If
|
|
|
|
* a given ordered_extent is completely done, 1 is returned, otherwise
|
|
|
|
* 0.
|
|
|
|
*
|
|
|
|
* test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
|
|
|
|
* to make sure this function only returns 1 once for a given ordered extent.
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
int btrfs_dec_test_ordered_pending(struct inode *inode,
|
2010-02-02 21:51:14 +01:00
|
|
|
struct btrfs_ordered_extent **cached,
|
2008-07-17 18:53:50 +02:00
|
|
|
u64 file_offset, u64 io_size)
|
2008-01-08 21:46:30 +01:00
|
|
|
{
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
2008-01-08 21:46:30 +01:00
|
|
|
struct rb_node *node;
|
2010-02-02 21:51:14 +01:00
|
|
|
struct btrfs_ordered_extent *entry = NULL;
|
2008-07-17 18:53:50 +02:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
node = tree_search(tree, file_offset);
|
2008-01-08 21:46:30 +01:00
|
|
|
if (!node) {
|
2008-07-17 18:53:50 +02:00
|
|
|
ret = 1;
|
|
|
|
goto out;
|
2008-01-08 21:46:30 +01:00
|
|
|
}
|
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
|
|
|
|
if (!offset_in_entry(entry, file_offset)) {
|
|
|
|
ret = 1;
|
|
|
|
goto out;
|
2008-01-08 21:46:30 +01:00
|
|
|
}
|
2008-07-17 18:53:50 +02:00
|
|
|
|
2009-09-02 22:53:46 +02:00
|
|
|
if (io_size > entry->bytes_left) {
|
|
|
|
printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
|
|
|
|
(unsigned long long)entry->bytes_left,
|
|
|
|
(unsigned long long)io_size);
|
|
|
|
}
|
|
|
|
entry->bytes_left -= io_size;
|
|
|
|
if (entry->bytes_left == 0)
|
2008-07-17 18:53:50 +02:00
|
|
|
ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
|
2009-09-02 22:53:46 +02:00
|
|
|
else
|
|
|
|
ret = 1;
|
2008-07-17 18:53:50 +02:00
|
|
|
out:
|
2010-02-02 21:51:14 +01:00
|
|
|
if (!ret && cached && entry) {
|
|
|
|
*cached = entry;
|
|
|
|
atomic_inc(&entry->refs);
|
|
|
|
}
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
return ret == 0;
|
|
|
|
}
|
2008-01-08 21:46:30 +01:00
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* used to drop a reference on an ordered extent. This will free
|
|
|
|
* the extent if the last reference is dropped
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
|
|
|
|
{
|
2008-07-17 18:54:15 +02:00
|
|
|
struct list_head *cur;
|
|
|
|
struct btrfs_ordered_sum *sum;
|
|
|
|
|
|
|
|
if (atomic_dec_and_test(&entry->refs)) {
|
2009-01-06 03:25:51 +01:00
|
|
|
while (!list_empty(&entry->list)) {
|
2008-07-17 18:54:15 +02:00
|
|
|
cur = entry->list.next;
|
|
|
|
sum = list_entry(cur, struct btrfs_ordered_sum, list);
|
|
|
|
list_del(&sum->list);
|
|
|
|
kfree(sum);
|
|
|
|
}
|
2008-07-17 18:53:50 +02:00
|
|
|
kfree(entry);
|
2008-07-17 18:54:15 +02:00
|
|
|
}
|
2008-07-17 18:53:50 +02:00
|
|
|
return 0;
|
2008-01-08 21:46:30 +01:00
|
|
|
}
|
2008-01-15 14:40:48 +01:00
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* remove an ordered extent from the tree. No references are dropped
|
2010-02-02 22:48:28 +01:00
|
|
|
* and you must wake_up entry->wait. You must hold the tree lock
|
2009-11-12 10:34:21 +01:00
|
|
|
* while you call this function.
|
2008-07-17 19:53:27 +02:00
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
static int __btrfs_remove_ordered_extent(struct inode *inode,
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_extent *entry)
|
2008-01-15 14:40:48 +01:00
|
|
|
{
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
2010-03-19 19:07:23 +01:00
|
|
|
struct btrfs_root *root = BTRFS_I(inode)->root;
|
2008-01-15 14:40:48 +01:00
|
|
|
struct rb_node *node;
|
|
|
|
|
2008-07-17 18:53:50 +02:00
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
|
|
|
node = &entry->rb_node;
|
2008-01-15 14:40:48 +01:00
|
|
|
rb_erase(node, &tree->tree);
|
2008-07-17 18:53:50 +02:00
|
|
|
tree->last = NULL;
|
|
|
|
set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
|
2008-07-24 17:57:52 +02:00
|
|
|
|
2009-10-08 19:34:05 +02:00
|
|
|
spin_lock(&BTRFS_I(inode)->accounting_lock);
|
2010-03-19 19:07:23 +01:00
|
|
|
WARN_ON(!BTRFS_I(inode)->outstanding_extents);
|
2009-10-08 19:34:05 +02:00
|
|
|
BTRFS_I(inode)->outstanding_extents--;
|
|
|
|
spin_unlock(&BTRFS_I(inode)->accounting_lock);
|
|
|
|
btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
|
|
|
|
inode, 1);
|
|
|
|
|
2010-03-19 19:07:23 +01:00
|
|
|
spin_lock(&root->fs_info->ordered_extent_lock);
|
2008-07-24 17:57:52 +02:00
|
|
|
list_del_init(&entry->root_extent_list);
|
2009-03-31 19:27:11 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* we have no more ordered extents for this inode and
|
|
|
|
* no dirty pages. We can safely remove it from the
|
|
|
|
* list of ordered extents
|
|
|
|
*/
|
|
|
|
if (RB_EMPTY_ROOT(&tree->tree) &&
|
|
|
|
!mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
|
|
|
|
list_del_init(&BTRFS_I(inode)->ordered_operations);
|
|
|
|
}
|
2010-03-19 19:07:23 +01:00
|
|
|
spin_unlock(&root->fs_info->ordered_extent_lock);
|
2008-07-24 17:57:52 +02:00
|
|
|
|
2009-11-12 10:34:21 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* remove an ordered extent from the tree. No references are dropped
|
|
|
|
* but any waiters are woken.
|
|
|
|
*/
|
|
|
|
int btrfs_remove_ordered_extent(struct inode *inode,
|
|
|
|
struct btrfs_ordered_extent *entry)
|
|
|
|
{
|
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2009-11-12 10:34:21 +01:00
|
|
|
ret = __btrfs_remove_ordered_extent(inode, entry);
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
wake_up(&entry->wait);
|
2009-11-12 10:34:21 +01:00
|
|
|
|
|
|
|
return ret;
|
2008-01-15 14:40:48 +01:00
|
|
|
}
|
|
|
|
|
2008-09-29 21:18:18 +02:00
|
|
|
/*
|
|
|
|
* wait for all the ordered extents in a root. This is done when balancing
|
|
|
|
* space between drives.
|
|
|
|
*/
|
2009-11-12 10:36:34 +01:00
|
|
|
int btrfs_wait_ordered_extents(struct btrfs_root *root,
|
|
|
|
int nocow_only, int delay_iput)
|
2008-07-24 17:57:52 +02:00
|
|
|
{
|
|
|
|
struct list_head splice;
|
|
|
|
struct list_head *cur;
|
|
|
|
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);
|
2008-09-26 16:05:38 +02:00
|
|
|
while (!list_empty(&splice)) {
|
2008-07-24 17:57:52 +02:00
|
|
|
cur = splice.next;
|
|
|
|
ordered = list_entry(cur, struct btrfs_ordered_extent,
|
|
|
|
root_extent_list);
|
2008-08-05 19:05:02 +02:00
|
|
|
if (nocow_only &&
|
2008-10-30 19:25:28 +01:00
|
|
|
!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
|
|
|
|
!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
|
2008-09-26 16:05:38 +02:00
|
|
|
list_move(&ordered->root_extent_list,
|
|
|
|
&root->fs_info->ordered_extents);
|
2008-08-05 19:05:02 +02:00
|
|
|
cond_resched_lock(&root->fs_info->ordered_extent_lock);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2008-07-24 17:57:52 +02:00
|
|
|
list_del_init(&ordered->root_extent_list);
|
|
|
|
atomic_inc(&ordered->refs);
|
|
|
|
|
|
|
|
/*
|
2008-09-26 16:05:38 +02:00
|
|
|
* the inode may be getting freed (in sys_unlink path).
|
2008-07-24 17:57:52 +02:00
|
|
|
*/
|
2008-09-26 16:05:38 +02:00
|
|
|
inode = igrab(ordered->inode);
|
|
|
|
|
2008-07-24 17:57:52 +02:00
|
|
|
spin_unlock(&root->fs_info->ordered_extent_lock);
|
|
|
|
|
2008-09-26 16:05:38 +02:00
|
|
|
if (inode) {
|
|
|
|
btrfs_start_ordered_extent(inode, ordered, 1);
|
|
|
|
btrfs_put_ordered_extent(ordered);
|
2009-11-12 10:36:34 +01:00
|
|
|
if (delay_iput)
|
|
|
|
btrfs_add_delayed_iput(inode);
|
|
|
|
else
|
|
|
|
iput(inode);
|
2008-09-26 16:05:38 +02:00
|
|
|
} else {
|
|
|
|
btrfs_put_ordered_extent(ordered);
|
|
|
|
}
|
2008-07-24 17:57:52 +02:00
|
|
|
|
|
|
|
spin_lock(&root->fs_info->ordered_extent_lock);
|
|
|
|
}
|
|
|
|
spin_unlock(&root->fs_info->ordered_extent_lock);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-03-31 19:27:11 +02:00
|
|
|
/*
|
|
|
|
* this is used during transaction commit to write all the inodes
|
|
|
|
* added to the ordered operation list. These files must be fully on
|
|
|
|
* disk before the transaction commits.
|
|
|
|
*
|
|
|
|
* we have two modes here, one is to just start the IO via filemap_flush
|
|
|
|
* and the other is to wait for all the io. When we wait, we have an
|
|
|
|
* extra check to make sure the ordered operation list really is empty
|
|
|
|
* before we return
|
|
|
|
*/
|
|
|
|
int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
|
|
|
|
{
|
|
|
|
struct btrfs_inode *btrfs_inode;
|
|
|
|
struct inode *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);
|
|
|
|
again:
|
|
|
|
list_splice_init(&root->fs_info->ordered_operations, &splice);
|
|
|
|
|
|
|
|
while (!list_empty(&splice)) {
|
|
|
|
btrfs_inode = list_entry(splice.next, struct btrfs_inode,
|
|
|
|
ordered_operations);
|
|
|
|
|
|
|
|
inode = &btrfs_inode->vfs_inode;
|
|
|
|
|
|
|
|
list_del_init(&btrfs_inode->ordered_operations);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* the inode may be getting freed (in sys_unlink path).
|
|
|
|
*/
|
|
|
|
inode = igrab(inode);
|
|
|
|
|
|
|
|
if (!wait && inode) {
|
|
|
|
list_add_tail(&BTRFS_I(inode)->ordered_operations,
|
|
|
|
&root->fs_info->ordered_operations);
|
|
|
|
}
|
|
|
|
spin_unlock(&root->fs_info->ordered_extent_lock);
|
|
|
|
|
|
|
|
if (inode) {
|
|
|
|
if (wait)
|
|
|
|
btrfs_wait_ordered_range(inode, 0, (u64)-1);
|
|
|
|
else
|
|
|
|
filemap_flush(inode->i_mapping);
|
2009-11-12 10:36:34 +01:00
|
|
|
btrfs_add_delayed_iput(inode);
|
2009-03-31 19:27:11 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
cond_resched();
|
|
|
|
spin_lock(&root->fs_info->ordered_extent_lock);
|
|
|
|
}
|
|
|
|
if (wait && !list_empty(&root->fs_info->ordered_operations))
|
|
|
|
goto again;
|
|
|
|
|
|
|
|
spin_unlock(&root->fs_info->ordered_extent_lock);
|
|
|
|
mutex_unlock(&root->fs_info->ordered_operations_mutex);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* Used to start IO or wait for a given ordered extent to finish.
|
|
|
|
*
|
|
|
|
* If wait is one, this effectively waits on page writeback for all the pages
|
|
|
|
* in the extent, and it waits on the io completion code to insert
|
|
|
|
* metadata into the btree corresponding to the extent
|
|
|
|
*/
|
|
|
|
void btrfs_start_ordered_extent(struct inode *inode,
|
|
|
|
struct btrfs_ordered_extent *entry,
|
|
|
|
int wait)
|
2008-07-17 18:53:50 +02:00
|
|
|
{
|
|
|
|
u64 start = entry->file_offset;
|
|
|
|
u64 end = start + entry->len - 1;
|
2008-05-27 16:55:43 +02:00
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* pages in the range can be dirty, clean or writeback. We
|
|
|
|
* start IO on any dirty ones so the wait doesn't stall waiting
|
|
|
|
* for pdflush to find them
|
|
|
|
*/
|
2009-10-01 18:58:30 +02:00
|
|
|
filemap_fdatawrite_range(inode->i_mapping, start, end);
|
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
|
|
|
if (wait) {
|
2008-07-17 18:53:50 +02:00
|
|
|
wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
|
|
|
|
&entry->flags));
|
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
|
|
|
}
|
2008-07-17 18:53:50 +02:00
|
|
|
}
|
2008-01-15 14:40:48 +01:00
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* Used to wait on ordered extents across a large range of bytes.
|
|
|
|
*/
|
2008-10-03 18:30:02 +02:00
|
|
|
int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
|
2008-07-17 18:53:50 +02:00
|
|
|
{
|
|
|
|
u64 end;
|
2008-07-19 02:42:20 +02:00
|
|
|
u64 orig_end;
|
|
|
|
u64 wait_end;
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_extent *ordered;
|
2009-09-02 22:53:46 +02:00
|
|
|
int found;
|
2008-07-19 02:42:20 +02:00
|
|
|
|
|
|
|
if (start + len < start) {
|
2008-07-22 17:18:09 +02:00
|
|
|
orig_end = INT_LIMIT(loff_t);
|
2008-07-19 02:42:20 +02:00
|
|
|
} else {
|
|
|
|
orig_end = start + len - 1;
|
2008-07-22 17:18:09 +02:00
|
|
|
if (orig_end > INT_LIMIT(loff_t))
|
|
|
|
orig_end = INT_LIMIT(loff_t);
|
2008-07-19 02:42:20 +02:00
|
|
|
}
|
2008-07-22 17:18:09 +02:00
|
|
|
wait_end = orig_end;
|
2008-07-21 16:29:44 +02:00
|
|
|
again:
|
2008-07-19 02:42:20 +02:00
|
|
|
/* start IO across the range first to instantiate any delalloc
|
|
|
|
* extents
|
|
|
|
*/
|
2009-10-01 18:58:30 +02:00
|
|
|
filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
|
2008-07-22 17:18:09 +02:00
|
|
|
|
2008-11-07 04:02:51 +01:00
|
|
|
/* The compression code will leave pages locked but return from
|
|
|
|
* writepage without setting the page writeback. Starting again
|
|
|
|
* with WB_SYNC_ALL will end up waiting for the IO to actually start.
|
|
|
|
*/
|
2009-10-01 18:58:30 +02:00
|
|
|
filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
|
2008-11-07 04:02:51 +01:00
|
|
|
|
2009-10-01 18:58:30 +02:00
|
|
|
filemap_fdatawait_range(inode->i_mapping, start, orig_end);
|
2008-07-19 02:42:20 +02:00
|
|
|
|
2008-07-22 17:18:09 +02:00
|
|
|
end = orig_end;
|
2009-09-02 22:53:46 +02:00
|
|
|
found = 0;
|
2009-01-06 03:25:51 +01:00
|
|
|
while (1) {
|
2008-07-17 18:53:50 +02:00
|
|
|
ordered = btrfs_lookup_first_ordered_extent(inode, end);
|
2009-01-06 03:25:51 +01:00
|
|
|
if (!ordered)
|
2008-07-17 18:53:50 +02:00
|
|
|
break;
|
2008-07-19 02:42:20 +02:00
|
|
|
if (ordered->file_offset > orig_end) {
|
2008-07-17 18:53:50 +02:00
|
|
|
btrfs_put_ordered_extent(ordered);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (ordered->file_offset + ordered->len < start) {
|
|
|
|
btrfs_put_ordered_extent(ordered);
|
|
|
|
break;
|
|
|
|
}
|
2009-09-02 22:53:46 +02:00
|
|
|
found++;
|
2008-07-19 02:42:20 +02:00
|
|
|
btrfs_start_ordered_extent(inode, ordered, 1);
|
2008-07-17 18:53:50 +02:00
|
|
|
end = ordered->file_offset;
|
|
|
|
btrfs_put_ordered_extent(ordered);
|
2008-07-19 02:42:20 +02:00
|
|
|
if (end == 0 || end == start)
|
2008-07-17 18:53:50 +02:00
|
|
|
break;
|
|
|
|
end--;
|
|
|
|
}
|
2009-09-02 22:53:46 +02:00
|
|
|
if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
|
|
|
|
EXTENT_DELALLOC, 0, NULL)) {
|
2008-11-07 04:02:51 +01:00
|
|
|
schedule_timeout(1);
|
2008-07-21 16:29:44 +02:00
|
|
|
goto again;
|
|
|
|
}
|
2008-10-03 18:30:02 +02:00
|
|
|
return 0;
|
2008-01-15 14:40:48 +01:00
|
|
|
}
|
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* find an ordered extent corresponding to file_offset. return NULL if
|
|
|
|
* nothing is found, otherwise take a reference on the extent and return it
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
|
|
|
|
u64 file_offset)
|
|
|
|
{
|
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
|
|
|
struct rb_node *node;
|
|
|
|
struct btrfs_ordered_extent *entry = NULL;
|
|
|
|
|
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
node = tree_search(tree, file_offset);
|
|
|
|
if (!node)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
|
|
|
|
if (!offset_in_entry(entry, file_offset))
|
|
|
|
entry = NULL;
|
|
|
|
if (entry)
|
|
|
|
atomic_inc(&entry->refs);
|
|
|
|
out:
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
return entry;
|
|
|
|
}
|
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* lookup and return any extent before 'file_offset'. NULL is returned
|
|
|
|
* if none is found
|
|
|
|
*/
|
2008-07-17 18:53:50 +02:00
|
|
|
struct btrfs_ordered_extent *
|
2009-01-06 03:25:51 +01:00
|
|
|
btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
|
2008-07-17 18:53:50 +02:00
|
|
|
{
|
|
|
|
struct btrfs_ordered_inode_tree *tree;
|
|
|
|
struct rb_node *node;
|
|
|
|
struct btrfs_ordered_extent *entry = NULL;
|
|
|
|
|
|
|
|
tree = &BTRFS_I(inode)->ordered_tree;
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
node = tree_search(tree, file_offset);
|
|
|
|
if (!node)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
|
|
|
|
atomic_inc(&entry->refs);
|
|
|
|
out:
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2008-07-17 18:53:50 +02:00
|
|
|
return entry;
|
2008-04-25 14:51:48 +02:00
|
|
|
}
|
2008-07-17 18:54:05 +02:00
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* After an extent is done, call this to conditionally update the on disk
|
|
|
|
* i_size. i_size is updated to cover any fully written part of the file.
|
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
|
2008-07-17 18:54:05 +02:00
|
|
|
struct btrfs_ordered_extent *ordered)
|
|
|
|
{
|
|
|
|
struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
|
|
|
|
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
|
|
|
|
u64 disk_i_size;
|
|
|
|
u64 new_i_size;
|
|
|
|
u64 i_size_test;
|
2009-11-12 10:34:21 +01:00
|
|
|
u64 i_size = i_size_read(inode);
|
2008-07-17 18:54:05 +02:00
|
|
|
struct rb_node *node;
|
2009-11-12 10:34:21 +01:00
|
|
|
struct rb_node *prev = NULL;
|
2008-07-17 18:54:05 +02:00
|
|
|
struct btrfs_ordered_extent *test;
|
2009-11-12 10:34:21 +01:00
|
|
|
int ret = 1;
|
|
|
|
|
|
|
|
if (ordered)
|
|
|
|
offset = entry_end(ordered);
|
2009-12-28 06:01:58 +01:00
|
|
|
else
|
|
|
|
offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
|
2008-07-17 18:54:05 +02:00
|
|
|
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2008-07-17 18:54:05 +02:00
|
|
|
disk_i_size = BTRFS_I(inode)->disk_i_size;
|
|
|
|
|
2009-11-12 10:34:21 +01:00
|
|
|
/* truncate file */
|
|
|
|
if (disk_i_size > i_size) {
|
|
|
|
BTRFS_I(inode)->disk_i_size = i_size;
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2008-07-17 18:54:05 +02:00
|
|
|
/*
|
|
|
|
* if the disk i_size is already at the inode->i_size, or
|
|
|
|
* this ordered extent is inside the disk i_size, we're done
|
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
if (disk_i_size == i_size || offset <= disk_i_size) {
|
2008-07-17 18:54:05 +02:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* we can't update the disk_isize if there are delalloc bytes
|
|
|
|
* between disk_i_size and this ordered extent
|
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
if (test_range_bit(io_tree, disk_i_size, offset - 1,
|
2009-09-02 21:22:30 +02:00
|
|
|
EXTENT_DELALLOC, 0, NULL)) {
|
2008-07-17 18:54:05 +02:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* walk backward from this ordered extent to disk_i_size.
|
|
|
|
* if we find an ordered extent then we can't update disk i_size
|
|
|
|
* yet
|
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
if (ordered) {
|
|
|
|
node = rb_prev(&ordered->rb_node);
|
|
|
|
} else {
|
|
|
|
prev = tree_search(tree, offset);
|
|
|
|
/*
|
|
|
|
* we insert file extents without involving ordered struct,
|
|
|
|
* so there should be no ordered struct cover this offset
|
|
|
|
*/
|
|
|
|
if (prev) {
|
|
|
|
test = rb_entry(prev, struct btrfs_ordered_extent,
|
|
|
|
rb_node);
|
|
|
|
BUG_ON(offset_in_entry(test, offset));
|
|
|
|
}
|
|
|
|
node = prev;
|
|
|
|
}
|
|
|
|
while (node) {
|
2008-07-17 18:54:05 +02:00
|
|
|
test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
|
|
|
|
if (test->file_offset + test->len <= disk_i_size)
|
|
|
|
break;
|
2009-11-12 10:34:21 +01:00
|
|
|
if (test->file_offset >= i_size)
|
2008-07-17 18:54:05 +02:00
|
|
|
break;
|
|
|
|
if (test->file_offset >= disk_i_size)
|
|
|
|
goto out;
|
2009-11-12 10:34:21 +01:00
|
|
|
node = rb_prev(node);
|
2008-07-17 18:54:05 +02:00
|
|
|
}
|
2009-11-12 10:34:21 +01:00
|
|
|
new_i_size = min_t(u64, offset, i_size);
|
2008-07-17 18:54:05 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* at this point, we know we can safely update i_size to at least
|
|
|
|
* the offset from this ordered extent. But, we need to
|
|
|
|
* walk forward and see if ios from higher up in the file have
|
|
|
|
* finished.
|
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
if (ordered) {
|
|
|
|
node = rb_next(&ordered->rb_node);
|
|
|
|
} else {
|
|
|
|
if (prev)
|
|
|
|
node = rb_next(prev);
|
|
|
|
else
|
|
|
|
node = rb_first(&tree->tree);
|
|
|
|
}
|
2008-07-17 18:54:05 +02:00
|
|
|
i_size_test = 0;
|
|
|
|
if (node) {
|
|
|
|
/*
|
|
|
|
* do we have an area where IO might have finished
|
|
|
|
* between our ordered extent and the next one.
|
|
|
|
*/
|
|
|
|
test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
|
2009-11-12 10:34:21 +01:00
|
|
|
if (test->file_offset > offset)
|
2008-08-05 05:23:47 +02:00
|
|
|
i_size_test = test->file_offset;
|
2008-07-17 18:54:05 +02:00
|
|
|
} else {
|
2009-11-12 10:34:21 +01:00
|
|
|
i_size_test = i_size;
|
2008-07-17 18:54:05 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* i_size_test is the end of a region after this ordered
|
|
|
|
* extent where there are no ordered extents. As long as there
|
|
|
|
* are no delalloc bytes in this area, it is safe to update
|
|
|
|
* disk_i_size to the end of the region.
|
|
|
|
*/
|
2009-11-12 10:34:21 +01:00
|
|
|
if (i_size_test > offset &&
|
|
|
|
!test_range_bit(io_tree, offset, i_size_test - 1,
|
|
|
|
EXTENT_DELALLOC, 0, NULL)) {
|
|
|
|
new_i_size = min_t(u64, i_size_test, i_size);
|
2008-07-17 18:54:05 +02:00
|
|
|
}
|
|
|
|
BTRFS_I(inode)->disk_i_size = new_i_size;
|
2009-11-12 10:34:21 +01:00
|
|
|
ret = 0;
|
2008-07-17 18:54:05 +02:00
|
|
|
out:
|
2009-11-12 10:34:21 +01:00
|
|
|
/*
|
|
|
|
* we need to remove the ordered extent with the tree lock held
|
|
|
|
* so that other people calling this function don't find our fully
|
|
|
|
* processed ordered entry and skip updating the i_size
|
|
|
|
*/
|
|
|
|
if (ordered)
|
|
|
|
__btrfs_remove_ordered_extent(inode, ordered);
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2009-11-12 10:34:21 +01:00
|
|
|
if (ordered)
|
|
|
|
wake_up(&ordered->wait);
|
|
|
|
return ret;
|
2008-07-17 18:54:05 +02:00
|
|
|
}
|
2008-07-17 18:54:15 +02:00
|
|
|
|
2008-07-17 19:53:27 +02:00
|
|
|
/*
|
|
|
|
* search the ordered extents for one corresponding to 'offset' and
|
|
|
|
* try to find a checksum. This is used because we allow pages to
|
|
|
|
* be reclaimed before their checksum is actually put into the btree
|
|
|
|
*/
|
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
|
|
|
int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
|
|
|
|
u32 *sum)
|
2008-07-17 18:54:15 +02:00
|
|
|
{
|
|
|
|
struct btrfs_ordered_sum *ordered_sum;
|
|
|
|
struct btrfs_sector_sum *sector_sums;
|
|
|
|
struct btrfs_ordered_extent *ordered;
|
|
|
|
struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
|
2008-07-18 12:17:13 +02:00
|
|
|
unsigned long num_sectors;
|
|
|
|
unsigned long i;
|
|
|
|
u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
|
2008-07-17 18:54:15 +02:00
|
|
|
int ret = 1;
|
|
|
|
|
|
|
|
ordered = btrfs_lookup_ordered_extent(inode, offset);
|
|
|
|
if (!ordered)
|
|
|
|
return 1;
|
|
|
|
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_lock(&tree->lock);
|
2009-01-21 16:59:08 +01:00
|
|
|
list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
|
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 (disk_bytenr >= ordered_sum->bytenr) {
|
2008-07-18 12:17:13 +02:00
|
|
|
num_sectors = ordered_sum->len / sectorsize;
|
2008-07-23 05:06:42 +02:00
|
|
|
sector_sums = ordered_sum->sums;
|
2008-07-18 12:17:13 +02:00
|
|
|
for (i = 0; i < num_sectors; i++) {
|
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 (sector_sums[i].bytenr == disk_bytenr) {
|
2008-07-18 12:17:13 +02:00
|
|
|
*sum = sector_sums[i].sum;
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
2008-07-17 18:54:15 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
out:
|
2010-02-02 22:48:28 +01:00
|
|
|
spin_unlock(&tree->lock);
|
2008-07-24 15:41:53 +02:00
|
|
|
btrfs_put_ordered_extent(ordered);
|
2008-07-17 18:54:15 +02:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-07-22 17:18:09 +02:00
|
|
|
|
2009-03-31 19:27:11 +02:00
|
|
|
/*
|
|
|
|
* add a given inode to the list of inodes that must be fully on
|
|
|
|
* disk before a transaction commit finishes.
|
|
|
|
*
|
|
|
|
* This basically gives us the ext3 style data=ordered mode, and it is mostly
|
|
|
|
* used to make sure renamed files are fully on disk.
|
|
|
|
*
|
|
|
|
* It is a noop if the inode is already fully on disk.
|
|
|
|
*
|
|
|
|
* If trans is not null, we'll do a friendly check for a transaction that
|
|
|
|
* is already flushing things and force the IO down ourselves.
|
|
|
|
*/
|
|
|
|
int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_root *root,
|
|
|
|
struct inode *inode)
|
|
|
|
{
|
|
|
|
u64 last_mod;
|
|
|
|
|
|
|
|
last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* if this file hasn't been changed since the last transaction
|
|
|
|
* commit, we can safely return without doing anything
|
|
|
|
*/
|
|
|
|
if (last_mod < root->fs_info->last_trans_committed)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* the transaction is already committing. Just start the IO and
|
|
|
|
* don't bother with all of this list nonsense
|
|
|
|
*/
|
|
|
|
if (trans && root->fs_info->running_transaction->blocked) {
|
|
|
|
btrfs_wait_ordered_range(inode, 0, (u64)-1);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_lock(&root->fs_info->ordered_extent_lock);
|
|
|
|
if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
|
|
|
|
list_add_tail(&BTRFS_I(inode)->ordered_operations,
|
|
|
|
&root->fs_info->ordered_operations);
|
|
|
|
}
|
|
|
|
spin_unlock(&root->fs_info->ordered_extent_lock);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|