linux/fs/xfs/xfs_extfree_item.c

494 lines
13 KiB
C
Raw Normal View History

/*
* Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
* 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 as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it would 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 the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_types.h"
#include "xfs_log.h"
#include "xfs_trans.h"
#include "xfs_buf_item.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_mount.h"
#include "xfs_trans_priv.h"
#include "xfs_extfree_item.h"
kmem_zone_t *xfs_efi_zone;
kmem_zone_t *xfs_efd_zone;
static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
{
return container_of(lip, struct xfs_efi_log_item, efi_item);
}
void
xfs_efi_item_free(
struct xfs_efi_log_item *efip)
{
if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
kmem_free(efip);
else
kmem_zone_free(xfs_efi_zone, efip);
}
/*
* Freeing the efi requires that we remove it from the AIL if it has already
* been placed there. However, the EFI may not yet have been placed in the AIL
* when called by xfs_efi_release() from EFD processing due to the ordering of
* committed vs unpin operations in bulk insert operations. Hence the
* test_and_clear_bit(XFS_EFI_COMMITTED) to ensure only the last caller frees
* the EFI.
*/
STATIC void
__xfs_efi_release(
struct xfs_efi_log_item *efip)
{
struct xfs_ail *ailp = efip->efi_item.li_ailp;
if (!test_and_clear_bit(XFS_EFI_COMMITTED, &efip->efi_flags)) {
spin_lock(&ailp->xa_lock);
/* xfs_trans_ail_delete() drops the AIL lock. */
xfs_trans_ail_delete(ailp, &efip->efi_item,
SHUTDOWN_LOG_IO_ERROR);
xfs_efi_item_free(efip);
}
}
/*
* This returns the number of iovecs needed to log the given efi item.
* We only need 1 iovec for an efi item. It just logs the efi_log_format
* structure.
*/
STATIC uint
xfs_efi_item_size(
struct xfs_log_item *lip)
{
return 1;
}
/*
* This is called to fill in the vector of log iovecs for the
* given efi log item. We use only 1 iovec, and we point that
* at the efi_log_format structure embedded in the efi item.
* It is at this point that we assert that all of the extent
* slots in the efi item have been filled.
*/
STATIC void
xfs_efi_item_format(
struct xfs_log_item *lip,
struct xfs_log_iovec *log_vector)
{
struct xfs_efi_log_item *efip = EFI_ITEM(lip);
uint size;
ASSERT(atomic_read(&efip->efi_next_extent) ==
efip->efi_format.efi_nextents);
efip->efi_format.efi_type = XFS_LI_EFI;
size = sizeof(xfs_efi_log_format_t);
size += (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
efip->efi_format.efi_size = 1;
log_vector->i_addr = &efip->efi_format;
log_vector->i_len = size;
log_vector->i_type = XLOG_REG_TYPE_EFI_FORMAT;
ASSERT(size >= sizeof(xfs_efi_log_format_t));
}
/*
* Pinning has no meaning for an efi item, so just return.
*/
STATIC void
xfs_efi_item_pin(
struct xfs_log_item *lip)
{
}
/*
* While EFIs cannot really be pinned, the unpin operation is the last place at
* which the EFI is manipulated during a transaction. If we are being asked to
* remove the EFI it's because the transaction has been cancelled and by
* definition that means the EFI cannot be in the AIL so remove it from the
* transaction and free it. Otherwise coordinate with xfs_efi_release() (via
* XFS_EFI_COMMITTED) to determine who gets to free the EFI.
*/
STATIC void
xfs_efi_item_unpin(
struct xfs_log_item *lip,
int remove)
{
struct xfs_efi_log_item *efip = EFI_ITEM(lip);
if (remove) {
ASSERT(!(lip->li_flags & XFS_LI_IN_AIL));
xfs: fix efi item leak on forced shutdown After test 139, kmemleak shows: unreferenced object 0xffff880078b405d8 (size 400): comm "xfs_io", pid 4904, jiffies 4294909383 (age 1186.728s) hex dump (first 32 bytes): 60 c1 17 79 00 88 ff ff 60 c1 17 79 00 88 ff ff `..y....`..y.... 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ backtrace: [<ffffffff81afb04d>] kmemleak_alloc+0x2d/0x60 [<ffffffff8115c6cf>] kmem_cache_alloc+0x13f/0x2b0 [<ffffffff814aaa97>] kmem_zone_alloc+0x77/0xf0 [<ffffffff814aab2e>] kmem_zone_zalloc+0x1e/0x50 [<ffffffff8147cd6b>] xfs_efi_init+0x4b/0xb0 [<ffffffff814a4ee8>] xfs_trans_get_efi+0x58/0x90 [<ffffffff81455fab>] xfs_bmap_finish+0x8b/0x1d0 [<ffffffff814851b4>] xfs_itruncate_finish+0x2c4/0x5d0 [<ffffffff814a970f>] xfs_setattr+0x8df/0xa70 [<ffffffff814b5c7b>] xfs_vn_setattr+0x1b/0x20 [<ffffffff8117dc00>] notify_change+0x170/0x2e0 [<ffffffff81163bf6>] do_truncate+0x66/0xa0 [<ffffffff81163d0b>] sys_ftruncate+0xdb/0xe0 [<ffffffff8103a002>] system_call_fastpath+0x16/0x1b [<ffffffffffffffff>] 0xffffffffffffffff The cause of the leak is that the "remove" parameter of IOP_UNPIN() is never set when a CIL push is aborted. This means that the EFI item is never freed if it was in the push being cancelled. The problem is specific to delayed logging, but has uncovered a couple of problems with the handling of IOP_UNPIN(remove). Firstly, we cannot safely call xfs_trans_del_item() from IOP_UNPIN() in the CIL commit failure path or the iclog write failure path because for delayed loging we have no transaction context. Hence we must only call xfs_trans_del_item() if the log item being unpinned has an active log item descriptor. Secondly, xfs_trans_uncommit() does not handle log item descriptor freeing during the traversal of log items on a transaction. It can reference a freed log item descriptor when unpinning an EFI item. Hence it needs to use a safe list traversal method to allow items to be removed from the transaction during IOP_UNPIN(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Alex Elder <aelder@sgi.com>
2011-01-27 01:13:35 +00:00
if (lip->li_desc)
xfs_trans_del_item(lip);
xfs_efi_item_free(efip);
return;
}
__xfs_efi_release(efip);
}
/*
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 05:58:39 +00:00
* Efi items have no locking or pushing. However, since EFIs are pulled from
* the AIL when their corresponding EFDs are committed to disk, their situation
* is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller
* will eventually flush the log. This should help in getting the EFI out of
* the AIL.
*/
STATIC uint
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 05:58:39 +00:00
xfs_efi_item_push(
struct xfs_log_item *lip,
struct list_head *buffer_list)
{
return XFS_ITEM_PINNED;
}
STATIC void
xfs_efi_item_unlock(
struct xfs_log_item *lip)
{
if (lip->li_flags & XFS_LI_ABORTED)
xfs_efi_item_free(EFI_ITEM(lip));
}
/*
* The EFI is logged only once and cannot be moved in the log, so simply return
* the lsn at which it's been logged. For bulk transaction committed
* processing, the EFI may be processed but not yet unpinned prior to the EFD
* being processed. Set the XFS_EFI_COMMITTED flag so this case can be detected
* when processing the EFD.
*/
STATIC xfs_lsn_t
xfs_efi_item_committed(
struct xfs_log_item *lip,
xfs_lsn_t lsn)
{
struct xfs_efi_log_item *efip = EFI_ITEM(lip);
set_bit(XFS_EFI_COMMITTED, &efip->efi_flags);
return lsn;
}
/*
* The EFI dependency tracking op doesn't do squat. It can't because
* it doesn't know where the free extent is coming from. The dependency
* tracking has to be handled by the "enclosing" metadata object. For
* example, for inodes, the inode is locked throughout the extent freeing
* so the dependency should be recorded there.
*/
STATIC void
xfs_efi_item_committing(
struct xfs_log_item *lip,
xfs_lsn_t lsn)
{
}
/*
* This is the ops vector shared by all efi log items.
*/
static const struct xfs_item_ops xfs_efi_item_ops = {
.iop_size = xfs_efi_item_size,
.iop_format = xfs_efi_item_format,
.iop_pin = xfs_efi_item_pin,
.iop_unpin = xfs_efi_item_unpin,
.iop_unlock = xfs_efi_item_unlock,
.iop_committed = xfs_efi_item_committed,
.iop_push = xfs_efi_item_push,
.iop_committing = xfs_efi_item_committing
};
/*
* Allocate and initialize an efi item with the given number of extents.
*/
struct xfs_efi_log_item *
xfs_efi_init(
struct xfs_mount *mp,
uint nextents)
{
struct xfs_efi_log_item *efip;
uint size;
ASSERT(nextents > 0);
if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
size = (uint)(sizeof(xfs_efi_log_item_t) +
((nextents - 1) * sizeof(xfs_extent_t)));
efip = kmem_zalloc(size, KM_SLEEP);
} else {
efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
}
xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
efip->efi_format.efi_nextents = nextents;
efip->efi_format.efi_id = (__psint_t)(void*)efip;
atomic_set(&efip->efi_next_extent, 0);
return efip;
}
/*
* Copy an EFI format buffer from the given buf, and into the destination
* EFI format structure.
* The given buffer can be in 32 bit or 64 bit form (which has different padding),
* one of which will be the native format for this kernel.
* It will handle the conversion of formats if necessary.
*/
int
xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
{
xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
uint i;
uint len = sizeof(xfs_efi_log_format_t) +
(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
uint len32 = sizeof(xfs_efi_log_format_32_t) +
(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
uint len64 = sizeof(xfs_efi_log_format_64_t) +
(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
if (buf->i_len == len) {
memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
return 0;
} else if (buf->i_len == len32) {
xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type;
dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size;
dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id;
for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
dst_efi_fmt->efi_extents[i].ext_start =
src_efi_fmt_32->efi_extents[i].ext_start;
dst_efi_fmt->efi_extents[i].ext_len =
src_efi_fmt_32->efi_extents[i].ext_len;
}
return 0;
} else if (buf->i_len == len64) {
xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type;
dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size;
dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id;
for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
dst_efi_fmt->efi_extents[i].ext_start =
src_efi_fmt_64->efi_extents[i].ext_start;
dst_efi_fmt->efi_extents[i].ext_len =
src_efi_fmt_64->efi_extents[i].ext_len;
}
return 0;
}
return EFSCORRUPTED;
}
/*
* This is called by the efd item code below to release references to the given
* efi item. Each efd calls this with the number of extents that it has
* logged, and when the sum of these reaches the total number of extents logged
* by this efi item we can free the efi item.
*/
void
xfs_efi_release(xfs_efi_log_item_t *efip,
uint nextents)
{
ASSERT(atomic_read(&efip->efi_next_extent) >= nextents);
if (atomic_sub_and_test(nextents, &efip->efi_next_extent))
__xfs_efi_release(efip);
}
static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
{
return container_of(lip, struct xfs_efd_log_item, efd_item);
}
STATIC void
xfs_efd_item_free(struct xfs_efd_log_item *efdp)
{
if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
kmem_free(efdp);
else
kmem_zone_free(xfs_efd_zone, efdp);
}
/*
* This returns the number of iovecs needed to log the given efd item.
* We only need 1 iovec for an efd item. It just logs the efd_log_format
* structure.
*/
STATIC uint
xfs_efd_item_size(
struct xfs_log_item *lip)
{
return 1;
}
/*
* This is called to fill in the vector of log iovecs for the
* given efd log item. We use only 1 iovec, and we point that
* at the efd_log_format structure embedded in the efd item.
* It is at this point that we assert that all of the extent
* slots in the efd item have been filled.
*/
STATIC void
xfs_efd_item_format(
struct xfs_log_item *lip,
struct xfs_log_iovec *log_vector)
{
struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
uint size;
ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
efdp->efd_format.efd_type = XFS_LI_EFD;
size = sizeof(xfs_efd_log_format_t);
size += (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
efdp->efd_format.efd_size = 1;
log_vector->i_addr = &efdp->efd_format;
log_vector->i_len = size;
log_vector->i_type = XLOG_REG_TYPE_EFD_FORMAT;
ASSERT(size >= sizeof(xfs_efd_log_format_t));
}
/*
* Pinning has no meaning for an efd item, so just return.
*/
STATIC void
xfs_efd_item_pin(
struct xfs_log_item *lip)
{
}
/*
* Since pinning has no meaning for an efd item, unpinning does
* not either.
*/
STATIC void
xfs_efd_item_unpin(
struct xfs_log_item *lip,
int remove)
{
}
/*
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 05:58:39 +00:00
* There isn't much you can do to push on an efd item. It is simply stuck
* waiting for the log to be flushed to disk.
*/
STATIC uint
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 05:58:39 +00:00
xfs_efd_item_push(
struct xfs_log_item *lip,
struct list_head *buffer_list)
{
xfs: on-stack delayed write buffer lists Queue delwri buffers on a local on-stack list instead of a per-buftarg one, and write back the buffers per-process instead of by waking up xfsbufd. This is now easily doable given that we have very few places left that write delwri buffers: - log recovery: Only done at mount time, and already forcing out the buffers synchronously using xfs_flush_buftarg - quotacheck: Same story. - dquot reclaim: Writes out dirty dquots on the LRU under memory pressure. We might want to look into doing more of this via xfsaild, but it's already more optimal than the synchronous inode reclaim that writes each buffer synchronously. - xfsaild: This is the main beneficiary of the change. By keeping a local list of buffers to write we reduce latency of writing out buffers, and more importably we can remove all the delwri list promotions which were hitting the buffer cache hard under sustained metadata loads. The implementation is very straight forward - xfs_buf_delwri_queue now gets a new list_head pointer that it adds the delwri buffers to, and all callers need to eventually submit the list using xfs_buf_delwi_submit or xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are skipped in xfs_buf_delwri_queue, assuming they already are on another delwri list. The biggest change to pass down the buffer list was done to the AIL pushing. Now that we operate on buffers the trylock, push and pushbuf log item methods are merged into a single push routine, which tries to lock the item, and if possible add the buffer that needs writeback to the buffer list. This leads to much simpler code than the previous split but requires the individual IOP_PUSH instances to unlock and reacquire the AIL around calls to blocking routines. Given that xfsailds now also handle writing out buffers, the conditions for log forcing and the sleep times needed some small changes. The most important one is that we consider an AIL busy as long we still have buffers to push, and the other one is that we do increment the pushed LSN for buffers that are under flushing at this moment, but still count them towards the stuck items for restart purposes. Without this we could hammer on stuck items without ever forcing the log and not make progress under heavy random delete workloads on fast flash storage devices. [ Dave Chinner: - rebase on previous patches. - improved comments for XBF_DELWRI_Q handling - fix XBF_ASYNC handling in queue submission (test 106 failure) - rename delwri submit function buffer list parameters for clarity - xfs_efd_item_push() should return XFS_ITEM_PINNED ] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 05:58:39 +00:00
return XFS_ITEM_PINNED;
}
STATIC void
xfs_efd_item_unlock(
struct xfs_log_item *lip)
{
if (lip->li_flags & XFS_LI_ABORTED)
xfs_efd_item_free(EFD_ITEM(lip));
}
/*
* When the efd item is committed to disk, all we need to do
* is delete our reference to our partner efi item and then
* free ourselves. Since we're freeing ourselves we must
* return -1 to keep the transaction code from further referencing
* this item.
*/
STATIC xfs_lsn_t
xfs_efd_item_committed(
struct xfs_log_item *lip,
xfs_lsn_t lsn)
{
struct xfs_efd_log_item *efdp = EFD_ITEM(lip);
/*
* If we got a log I/O error, it's always the case that the LR with the
* EFI got unpinned and freed before the EFD got aborted.
*/
if (!(lip->li_flags & XFS_LI_ABORTED))
xfs_efi_release(efdp->efd_efip, efdp->efd_format.efd_nextents);
xfs_efd_item_free(efdp);
return (xfs_lsn_t)-1;
}
/*
* The EFD dependency tracking op doesn't do squat. It can't because
* it doesn't know where the free extent is coming from. The dependency
* tracking has to be handled by the "enclosing" metadata object. For
* example, for inodes, the inode is locked throughout the extent freeing
* so the dependency should be recorded there.
*/
STATIC void
xfs_efd_item_committing(
struct xfs_log_item *lip,
xfs_lsn_t lsn)
{
}
/*
* This is the ops vector shared by all efd log items.
*/
static const struct xfs_item_ops xfs_efd_item_ops = {
.iop_size = xfs_efd_item_size,
.iop_format = xfs_efd_item_format,
.iop_pin = xfs_efd_item_pin,
.iop_unpin = xfs_efd_item_unpin,
.iop_unlock = xfs_efd_item_unlock,
.iop_committed = xfs_efd_item_committed,
.iop_push = xfs_efd_item_push,
.iop_committing = xfs_efd_item_committing
};
/*
* Allocate and initialize an efd item with the given number of extents.
*/
struct xfs_efd_log_item *
xfs_efd_init(
struct xfs_mount *mp,
struct xfs_efi_log_item *efip,
uint nextents)
{
struct xfs_efd_log_item *efdp;
uint size;
ASSERT(nextents > 0);
if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
size = (uint)(sizeof(xfs_efd_log_item_t) +
((nextents - 1) * sizeof(xfs_extent_t)));
efdp = kmem_zalloc(size, KM_SLEEP);
} else {
efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
}
xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
efdp->efd_efip = efip;
efdp->efd_format.efd_nextents = nextents;
efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
return efdp;
}