linux/drivers/md/dm-io.c

526 lines
13 KiB
C
Raw Normal View History

/*
* Copyright (C) 2003 Sistina Software
* Copyright (C) 2006 Red Hat GmbH
*
* This file is released under the GPL.
*/
#include "dm.h"
#include <linux/device-mapper.h>
#include <linux/bio.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/dm-io.h>
#define DM_MSG_PREFIX "io"
#define DM_IO_MAX_REGIONS BITS_PER_LONG
#define MIN_IOS 16
#define MIN_BIOS 16
struct dm_io_client {
mempool_t *pool;
struct bio_set *bios;
};
/*
* Aligning 'struct io' reduces the number of bits required to store
* its address. Refer to store_io_and_region_in_bio() below.
*/
struct io {
unsigned long error_bits;
atomic_t count;
struct task_struct *sleeper;
struct dm_io_client *client;
io_notify_fn callback;
void *context;
void *vma_invalidate_address;
unsigned long vma_invalidate_size;
} __attribute__((aligned(DM_IO_MAX_REGIONS)));
static struct kmem_cache *_dm_io_cache;
/*
* Create a client with mempool and bioset.
*/
struct dm_io_client *dm_io_client_create(void)
{
struct dm_io_client *client;
client = kmalloc(sizeof(*client), GFP_KERNEL);
if (!client)
return ERR_PTR(-ENOMEM);
client->pool = mempool_create_slab_pool(MIN_IOS, _dm_io_cache);
if (!client->pool)
goto bad;
client->bios = bioset_create(MIN_BIOS, 0);
if (!client->bios)
goto bad;
return client;
bad:
if (client->pool)
mempool_destroy(client->pool);
kfree(client);
return ERR_PTR(-ENOMEM);
}
EXPORT_SYMBOL(dm_io_client_create);
void dm_io_client_destroy(struct dm_io_client *client)
{
mempool_destroy(client->pool);
bioset_free(client->bios);
kfree(client);
}
EXPORT_SYMBOL(dm_io_client_destroy);
/*-----------------------------------------------------------------
* We need to keep track of which region a bio is doing io for.
* To avoid a memory allocation to store just 5 or 6 bits, we
* ensure the 'struct io' pointer is aligned so enough low bits are
* always zero and then combine it with the region number directly in
* bi_private.
*---------------------------------------------------------------*/
static void store_io_and_region_in_bio(struct bio *bio, struct io *io,
unsigned region)
{
if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) {
DMCRIT("Unaligned struct io pointer %p", io);
BUG();
}
bio->bi_private = (void *)((unsigned long)io | region);
}
static void retrieve_io_and_region_from_bio(struct bio *bio, struct io **io,
unsigned *region)
{
unsigned long val = (unsigned long)bio->bi_private;
*io = (void *)(val & -(unsigned long)DM_IO_MAX_REGIONS);
*region = val & (DM_IO_MAX_REGIONS - 1);
}
/*-----------------------------------------------------------------
* We need an io object to keep track of the number of bios that
* have been dispatched for a particular io.
*---------------------------------------------------------------*/
static void dec_count(struct io *io, unsigned int region, int error)
{
dm: implement REQ_FLUSH/FUA support for bio-based dm This patch converts bio-based dm to support REQ_FLUSH/FUA instead of now deprecated REQ_HARDBARRIER. * -EOPNOTSUPP handling logic dropped. * Preflush is handled as before but postflush is dropped and replaced with passing down REQ_FUA to member request_queues. This replaces one array wide cache flush w/ member specific FUA writes. * __split_and_process_bio() now calls __clone_and_map_flush() directly for flushes and guarantees all FLUSH bio's going to targets are zero ` length. * It's now guaranteed that all FLUSH bio's which are passed onto dm targets are zero length. bio_empty_barrier() tests are replaced with REQ_FLUSH tests. * Empty WRITE_BARRIERs are replaced with WRITE_FLUSHes. * Dropped unlikely() around REQ_FLUSH tests. Flushes are not unlikely enough to be marked with unlikely(). * Block layer now filters out REQ_FLUSH/FUA bio's if the request_queue doesn't support cache flushing. Advertise REQ_FLUSH | REQ_FUA capability. * Request based dm isn't converted yet. dm_init_request_based_queue() resets flush support to 0 for now. To avoid disturbing request based dm code, dm->flush_error is added for bio based dm while requested based dm continues to use dm->barrier_error. Lightly tested linear, stripe, raid1, snap and crypt targets. Please proceed with caution as I'm not familiar with the code base. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: dm-devel@redhat.com Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2010-09-03 11:56:19 +02:00
if (error)
set_bit(region, &io->error_bits);
if (atomic_dec_and_test(&io->count)) {
if (io->vma_invalidate_size)
invalidate_kernel_vmap_range(io->vma_invalidate_address,
io->vma_invalidate_size);
if (io->sleeper)
wake_up_process(io->sleeper);
else {
unsigned long r = io->error_bits;
io_notify_fn fn = io->callback;
void *context = io->context;
mempool_free(io, io->client->pool);
fn(r, context);
}
}
}
static void endio(struct bio *bio, int error)
{
struct io *io;
unsigned region;
if (error && bio_data_dir(bio) == READ)
zero_fill_bio(bio);
/*
* The bio destructor in bio_put() may use the io object.
*/
retrieve_io_and_region_from_bio(bio, &io, &region);
bio_put(bio);
dec_count(io, region, error);
}
/*-----------------------------------------------------------------
* These little objects provide an abstraction for getting a new
* destination page for io.
*---------------------------------------------------------------*/
struct dpages {
void (*get_page)(struct dpages *dp,
struct page **p, unsigned long *len, unsigned *offset);
void (*next_page)(struct dpages *dp);
unsigned context_u;
void *context_ptr;
void *vma_invalidate_address;
unsigned long vma_invalidate_size;
};
/*
* Functions for getting the pages from a list.
*/
static void list_get_page(struct dpages *dp,
struct page **p, unsigned long *len, unsigned *offset)
{
unsigned o = dp->context_u;
struct page_list *pl = (struct page_list *) dp->context_ptr;
*p = pl->page;
*len = PAGE_SIZE - o;
*offset = o;
}
static void list_next_page(struct dpages *dp)
{
struct page_list *pl = (struct page_list *) dp->context_ptr;
dp->context_ptr = pl->next;
dp->context_u = 0;
}
static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset)
{
dp->get_page = list_get_page;
dp->next_page = list_next_page;
dp->context_u = offset;
dp->context_ptr = pl;
}
/*
* Functions for getting the pages from a bvec.
*/
static void bvec_get_page(struct dpages *dp,
struct page **p, unsigned long *len, unsigned *offset)
{
struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
*p = bvec->bv_page;
*len = bvec->bv_len;
*offset = bvec->bv_offset;
}
static void bvec_next_page(struct dpages *dp)
{
struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
dp->context_ptr = bvec + 1;
}
static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec)
{
dp->get_page = bvec_get_page;
dp->next_page = bvec_next_page;
dp->context_ptr = bvec;
}
/*
* Functions for getting the pages from a VMA.
*/
static void vm_get_page(struct dpages *dp,
struct page **p, unsigned long *len, unsigned *offset)
{
*p = vmalloc_to_page(dp->context_ptr);
*offset = dp->context_u;
*len = PAGE_SIZE - dp->context_u;
}
static void vm_next_page(struct dpages *dp)
{
dp->context_ptr += PAGE_SIZE - dp->context_u;
dp->context_u = 0;
}
static void vm_dp_init(struct dpages *dp, void *data)
{
dp->get_page = vm_get_page;
dp->next_page = vm_next_page;
dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
dp->context_ptr = data;
}
/*
* Functions for getting the pages from kernel memory.
*/
static void km_get_page(struct dpages *dp, struct page **p, unsigned long *len,
unsigned *offset)
{
*p = virt_to_page(dp->context_ptr);
*offset = dp->context_u;
*len = PAGE_SIZE - dp->context_u;
}
static void km_next_page(struct dpages *dp)
{
dp->context_ptr += PAGE_SIZE - dp->context_u;
dp->context_u = 0;
}
static void km_dp_init(struct dpages *dp, void *data)
{
dp->get_page = km_get_page;
dp->next_page = km_next_page;
dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
dp->context_ptr = data;
}
/*-----------------------------------------------------------------
* IO routines that accept a list of pages.
*---------------------------------------------------------------*/
static void do_region(int rw, unsigned region, struct dm_io_region *where,
struct dpages *dp, struct io *io)
{
struct bio *bio;
struct page *page;
unsigned long len;
unsigned offset;
unsigned num_bvecs;
sector_t remaining = where->count;
struct request_queue *q = bdev_get_queue(where->bdev);
unsigned short logical_block_size = queue_logical_block_size(q);
sector_t num_sectors;
/*
dm: implement REQ_FLUSH/FUA support for bio-based dm This patch converts bio-based dm to support REQ_FLUSH/FUA instead of now deprecated REQ_HARDBARRIER. * -EOPNOTSUPP handling logic dropped. * Preflush is handled as before but postflush is dropped and replaced with passing down REQ_FUA to member request_queues. This replaces one array wide cache flush w/ member specific FUA writes. * __split_and_process_bio() now calls __clone_and_map_flush() directly for flushes and guarantees all FLUSH bio's going to targets are zero ` length. * It's now guaranteed that all FLUSH bio's which are passed onto dm targets are zero length. bio_empty_barrier() tests are replaced with REQ_FLUSH tests. * Empty WRITE_BARRIERs are replaced with WRITE_FLUSHes. * Dropped unlikely() around REQ_FLUSH tests. Flushes are not unlikely enough to be marked with unlikely(). * Block layer now filters out REQ_FLUSH/FUA bio's if the request_queue doesn't support cache flushing. Advertise REQ_FLUSH | REQ_FUA capability. * Request based dm isn't converted yet. dm_init_request_based_queue() resets flush support to 0 for now. To avoid disturbing request based dm code, dm->flush_error is added for bio based dm while requested based dm continues to use dm->barrier_error. Lightly tested linear, stripe, raid1, snap and crypt targets. Please proceed with caution as I'm not familiar with the code base. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: dm-devel@redhat.com Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2010-09-03 11:56:19 +02:00
* where->count may be zero if rw holds a flush and we need to
* send a zero-sized flush.
*/
do {
/*
* Allocate a suitably sized-bio.
*/
if ((rw & REQ_DISCARD) || (rw & REQ_WRITE_SAME))
num_bvecs = 1;
else
num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev),
dm_sector_div_up(remaining, (PAGE_SIZE >> SECTOR_SHIFT)));
bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios);
bio->bi_sector = where->sector + (where->count - remaining);
bio->bi_bdev = where->bdev;
bio->bi_end_io = endio;
store_io_and_region_in_bio(bio, io, region);
if (rw & REQ_DISCARD) {
num_sectors = min_t(sector_t, q->limits.max_discard_sectors, remaining);
bio->bi_size = num_sectors << SECTOR_SHIFT;
remaining -= num_sectors;
} else if (rw & REQ_WRITE_SAME) {
/*
* WRITE SAME only uses a single page.
*/
dp->get_page(dp, &page, &len, &offset);
bio_add_page(bio, page, logical_block_size, offset);
num_sectors = min_t(sector_t, q->limits.max_write_same_sectors, remaining);
bio->bi_size = num_sectors << SECTOR_SHIFT;
offset = 0;
remaining -= num_sectors;
dp->next_page(dp);
} else while (remaining) {
/*
* Try and add as many pages as possible.
*/
dp->get_page(dp, &page, &len, &offset);
len = min(len, to_bytes(remaining));
if (!bio_add_page(bio, page, len, offset))
break;
offset = 0;
remaining -= to_sector(len);
dp->next_page(dp);
}
atomic_inc(&io->count);
submit_bio(rw, bio);
} while (remaining);
}
static void dispatch_io(int rw, unsigned int num_regions,
struct dm_io_region *where, struct dpages *dp,
struct io *io, int sync)
{
int i;
struct dpages old_pages = *dp;
BUG_ON(num_regions > DM_IO_MAX_REGIONS);
if (sync)
rw |= REQ_SYNC;
/*
* For multiple regions we need to be careful to rewind
* the dp object for each call to do_region.
*/
for (i = 0; i < num_regions; i++) {
*dp = old_pages;
dm: implement REQ_FLUSH/FUA support for bio-based dm This patch converts bio-based dm to support REQ_FLUSH/FUA instead of now deprecated REQ_HARDBARRIER. * -EOPNOTSUPP handling logic dropped. * Preflush is handled as before but postflush is dropped and replaced with passing down REQ_FUA to member request_queues. This replaces one array wide cache flush w/ member specific FUA writes. * __split_and_process_bio() now calls __clone_and_map_flush() directly for flushes and guarantees all FLUSH bio's going to targets are zero ` length. * It's now guaranteed that all FLUSH bio's which are passed onto dm targets are zero length. bio_empty_barrier() tests are replaced with REQ_FLUSH tests. * Empty WRITE_BARRIERs are replaced with WRITE_FLUSHes. * Dropped unlikely() around REQ_FLUSH tests. Flushes are not unlikely enough to be marked with unlikely(). * Block layer now filters out REQ_FLUSH/FUA bio's if the request_queue doesn't support cache flushing. Advertise REQ_FLUSH | REQ_FUA capability. * Request based dm isn't converted yet. dm_init_request_based_queue() resets flush support to 0 for now. To avoid disturbing request based dm code, dm->flush_error is added for bio based dm while requested based dm continues to use dm->barrier_error. Lightly tested linear, stripe, raid1, snap and crypt targets. Please proceed with caution as I'm not familiar with the code base. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: dm-devel@redhat.com Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2010-09-03 11:56:19 +02:00
if (where[i].count || (rw & REQ_FLUSH))
do_region(rw, i, where + i, dp, io);
}
/*
* Drop the extra reference that we were holding to avoid
* the io being completed too early.
*/
dec_count(io, 0, 0);
}
static int sync_io(struct dm_io_client *client, unsigned int num_regions,
struct dm_io_region *where, int rw, struct dpages *dp,
unsigned long *error_bits)
{
/*
* gcc <= 4.3 can't do the alignment for stack variables, so we must
* align it on our own.
* volatile prevents the optimizer from removing or reusing
* "io_" field from the stack frame (allowed in ANSI C).
*/
volatile char io_[sizeof(struct io) + __alignof__(struct io) - 1];
struct io *io = (struct io *)PTR_ALIGN(&io_, __alignof__(struct io));
dm: unplug queues in threads Remove an avoidable 3ms delay on some dm-raid1 and kcopyd I/O. It is specified that any submitted bio without BIO_RW_SYNC flag may plug the queue (i.e. block the requests from being dispatched to the physical device). The queue is unplugged when the caller calls blk_unplug() function. Usually, the sequence is that someone calls submit_bh to submit IO on a buffer. The IO plugs the queue and waits (to be possibly joined with other adjacent bios). Then, when the caller calls wait_on_buffer(), it unplugs the queue and submits the IOs to the disk. This was happenning: When doing O_SYNC writes, function fsync_buffers_list() submits a list of bios to dm_raid1, the bios are added to dm_raid1 write queue and kmirrord is woken up. fsync_buffers_list() calls wait_on_buffer(). That unplugs the queue, but there are no bios on the device queue as they are still in the dm_raid1 queue. wait_on_buffer() starts waiting until the IO is finished. kmirrord is scheduled, kmirrord takes bios and submits them to the devices. The submitted bio plugs the harddisk queue but there is no one to unplug it. (The process that called wait_on_buffer() is already sleeping.) So there is a 3ms timeout, after which the queues on the harddisks are unplugged and requests are processed. This 3ms timeout meant that in certain workloads (e.g. O_SYNC, 8kb writes), dm-raid1 is 10 times slower than md raid1. Every time we submit something asynchronously via dm_io, we must unplug the queue actually to send the request to the device. This patch adds an unplug call to kmirrord - while processing requests, it keeps the queue plugged (so that adjacent bios can be merged); when it finishes processing all the bios, it unplugs the queue to submit the bios. It also fixes kcopyd which has the same potential problem. All kcopyd requests are submitted with BIO_RW_SYNC. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Acked-by: Jens Axboe <jens.axboe@oracle.com>
2008-04-24 23:10:47 +02:00
if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
WARN_ON(1);
return -EIO;
}
io->error_bits = 0;
atomic_set(&io->count, 1); /* see dispatch_io() */
io->sleeper = current;
io->client = client;
io->vma_invalidate_address = dp->vma_invalidate_address;
io->vma_invalidate_size = dp->vma_invalidate_size;
dispatch_io(rw, num_regions, where, dp, io, 1);
while (1) {
set_current_state(TASK_UNINTERRUPTIBLE);
if (!atomic_read(&io->count))
break;
io_schedule();
}
set_current_state(TASK_RUNNING);
if (error_bits)
*error_bits = io->error_bits;
return io->error_bits ? -EIO : 0;
}
static int async_io(struct dm_io_client *client, unsigned int num_regions,
struct dm_io_region *where, int rw, struct dpages *dp,
io_notify_fn fn, void *context)
{
struct io *io;
dm: unplug queues in threads Remove an avoidable 3ms delay on some dm-raid1 and kcopyd I/O. It is specified that any submitted bio without BIO_RW_SYNC flag may plug the queue (i.e. block the requests from being dispatched to the physical device). The queue is unplugged when the caller calls blk_unplug() function. Usually, the sequence is that someone calls submit_bh to submit IO on a buffer. The IO plugs the queue and waits (to be possibly joined with other adjacent bios). Then, when the caller calls wait_on_buffer(), it unplugs the queue and submits the IOs to the disk. This was happenning: When doing O_SYNC writes, function fsync_buffers_list() submits a list of bios to dm_raid1, the bios are added to dm_raid1 write queue and kmirrord is woken up. fsync_buffers_list() calls wait_on_buffer(). That unplugs the queue, but there are no bios on the device queue as they are still in the dm_raid1 queue. wait_on_buffer() starts waiting until the IO is finished. kmirrord is scheduled, kmirrord takes bios and submits them to the devices. The submitted bio plugs the harddisk queue but there is no one to unplug it. (The process that called wait_on_buffer() is already sleeping.) So there is a 3ms timeout, after which the queues on the harddisks are unplugged and requests are processed. This 3ms timeout meant that in certain workloads (e.g. O_SYNC, 8kb writes), dm-raid1 is 10 times slower than md raid1. Every time we submit something asynchronously via dm_io, we must unplug the queue actually to send the request to the device. This patch adds an unplug call to kmirrord - while processing requests, it keeps the queue plugged (so that adjacent bios can be merged); when it finishes processing all the bios, it unplugs the queue to submit the bios. It also fixes kcopyd which has the same potential problem. All kcopyd requests are submitted with BIO_RW_SYNC. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Acked-by: Jens Axboe <jens.axboe@oracle.com>
2008-04-24 23:10:47 +02:00
if (num_regions > 1 && (rw & RW_MASK) != WRITE) {
WARN_ON(1);
fn(1, context);
return -EIO;
}
io = mempool_alloc(client->pool, GFP_NOIO);
io->error_bits = 0;
atomic_set(&io->count, 1); /* see dispatch_io() */
io->sleeper = NULL;
io->client = client;
io->callback = fn;
io->context = context;
io->vma_invalidate_address = dp->vma_invalidate_address;
io->vma_invalidate_size = dp->vma_invalidate_size;
dispatch_io(rw, num_regions, where, dp, io, 0);
return 0;
}
static int dp_init(struct dm_io_request *io_req, struct dpages *dp,
unsigned long size)
{
/* Set up dpages based on memory type */
dp->vma_invalidate_address = NULL;
dp->vma_invalidate_size = 0;
switch (io_req->mem.type) {
case DM_IO_PAGE_LIST:
list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset);
break;
case DM_IO_BVEC:
bvec_dp_init(dp, io_req->mem.ptr.bvec);
break;
case DM_IO_VMA:
flush_kernel_vmap_range(io_req->mem.ptr.vma, size);
if ((io_req->bi_rw & RW_MASK) == READ) {
dp->vma_invalidate_address = io_req->mem.ptr.vma;
dp->vma_invalidate_size = size;
}
vm_dp_init(dp, io_req->mem.ptr.vma);
break;
case DM_IO_KMEM:
km_dp_init(dp, io_req->mem.ptr.addr);
break;
default:
return -EINVAL;
}
return 0;
}
/*
dm: unplug queues in threads Remove an avoidable 3ms delay on some dm-raid1 and kcopyd I/O. It is specified that any submitted bio without BIO_RW_SYNC flag may plug the queue (i.e. block the requests from being dispatched to the physical device). The queue is unplugged when the caller calls blk_unplug() function. Usually, the sequence is that someone calls submit_bh to submit IO on a buffer. The IO plugs the queue and waits (to be possibly joined with other adjacent bios). Then, when the caller calls wait_on_buffer(), it unplugs the queue and submits the IOs to the disk. This was happenning: When doing O_SYNC writes, function fsync_buffers_list() submits a list of bios to dm_raid1, the bios are added to dm_raid1 write queue and kmirrord is woken up. fsync_buffers_list() calls wait_on_buffer(). That unplugs the queue, but there are no bios on the device queue as they are still in the dm_raid1 queue. wait_on_buffer() starts waiting until the IO is finished. kmirrord is scheduled, kmirrord takes bios and submits them to the devices. The submitted bio plugs the harddisk queue but there is no one to unplug it. (The process that called wait_on_buffer() is already sleeping.) So there is a 3ms timeout, after which the queues on the harddisks are unplugged and requests are processed. This 3ms timeout meant that in certain workloads (e.g. O_SYNC, 8kb writes), dm-raid1 is 10 times slower than md raid1. Every time we submit something asynchronously via dm_io, we must unplug the queue actually to send the request to the device. This patch adds an unplug call to kmirrord - while processing requests, it keeps the queue plugged (so that adjacent bios can be merged); when it finishes processing all the bios, it unplugs the queue to submit the bios. It also fixes kcopyd which has the same potential problem. All kcopyd requests are submitted with BIO_RW_SYNC. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Acked-by: Jens Axboe <jens.axboe@oracle.com>
2008-04-24 23:10:47 +02:00
* New collapsed (a)synchronous interface.
*
* If the IO is asynchronous (i.e. it has notify.fn), you must either unplug
* the queue with blk_unplug() some time later or set REQ_SYNC in
io_req->bi_rw. If you fail to do one of these, the IO will be submitted to
dm: unplug queues in threads Remove an avoidable 3ms delay on some dm-raid1 and kcopyd I/O. It is specified that any submitted bio without BIO_RW_SYNC flag may plug the queue (i.e. block the requests from being dispatched to the physical device). The queue is unplugged when the caller calls blk_unplug() function. Usually, the sequence is that someone calls submit_bh to submit IO on a buffer. The IO plugs the queue and waits (to be possibly joined with other adjacent bios). Then, when the caller calls wait_on_buffer(), it unplugs the queue and submits the IOs to the disk. This was happenning: When doing O_SYNC writes, function fsync_buffers_list() submits a list of bios to dm_raid1, the bios are added to dm_raid1 write queue and kmirrord is woken up. fsync_buffers_list() calls wait_on_buffer(). That unplugs the queue, but there are no bios on the device queue as they are still in the dm_raid1 queue. wait_on_buffer() starts waiting until the IO is finished. kmirrord is scheduled, kmirrord takes bios and submits them to the devices. The submitted bio plugs the harddisk queue but there is no one to unplug it. (The process that called wait_on_buffer() is already sleeping.) So there is a 3ms timeout, after which the queues on the harddisks are unplugged and requests are processed. This 3ms timeout meant that in certain workloads (e.g. O_SYNC, 8kb writes), dm-raid1 is 10 times slower than md raid1. Every time we submit something asynchronously via dm_io, we must unplug the queue actually to send the request to the device. This patch adds an unplug call to kmirrord - while processing requests, it keeps the queue plugged (so that adjacent bios can be merged); when it finishes processing all the bios, it unplugs the queue to submit the bios. It also fixes kcopyd which has the same potential problem. All kcopyd requests are submitted with BIO_RW_SYNC. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Acked-by: Jens Axboe <jens.axboe@oracle.com>
2008-04-24 23:10:47 +02:00
* the disk after q->unplug_delay, which defaults to 3ms in blk-settings.c.
*/
int dm_io(struct dm_io_request *io_req, unsigned num_regions,
struct dm_io_region *where, unsigned long *sync_error_bits)
{
int r;
struct dpages dp;
r = dp_init(io_req, &dp, (unsigned long)where->count << SECTOR_SHIFT);
if (r)
return r;
if (!io_req->notify.fn)
return sync_io(io_req->client, num_regions, where,
io_req->bi_rw, &dp, sync_error_bits);
return async_io(io_req->client, num_regions, where, io_req->bi_rw,
&dp, io_req->notify.fn, io_req->notify.context);
}
EXPORT_SYMBOL(dm_io);
int __init dm_io_init(void)
{
_dm_io_cache = KMEM_CACHE(io, 0);
if (!_dm_io_cache)
return -ENOMEM;
return 0;
}
void dm_io_exit(void)
{
kmem_cache_destroy(_dm_io_cache);
_dm_io_cache = NULL;
}