linux_dsm_epyc7002/fs/xfs/xfs_log.c

4015 lines
111 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_errortag.h"
#include "xfs_error.h"
#include "xfs_trans.h"
#include "xfs_trans_priv.h"
#include "xfs_log.h"
#include "xfs_log_priv.h"
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
#include "xfs_trace.h"
#include "xfs_sysfs.h"
#include "xfs_sb.h"
#include "xfs_health.h"
kmem_zone_t *xfs_log_ticket_zone;
/* Local miscellaneous function prototypes */
STATIC int
xlog_commit_record(
struct xlog *log,
struct xlog_ticket *ticket,
struct xlog_in_core **iclog,
xfs_lsn_t *commitlsnp);
STATIC struct xlog *
xlog_alloc_log(
struct xfs_mount *mp,
struct xfs_buftarg *log_target,
xfs_daddr_t blk_offset,
int num_bblks);
STATIC int
xlog_space_left(
struct xlog *log,
atomic64_t *head);
STATIC void
xlog_dealloc_log(
struct xlog *log);
/* local state machine functions */
STATIC void xlog_state_done_syncing(
struct xlog_in_core *iclog,
bool aborted);
STATIC int
xlog_state_get_iclog_space(
struct xlog *log,
int len,
struct xlog_in_core **iclog,
struct xlog_ticket *ticket,
int *continued_write,
int *logoffsetp);
STATIC void
xlog_state_switch_iclogs(
struct xlog *log,
struct xlog_in_core *iclog,
int eventual_size);
STATIC void
xlog_state_want_sync(
struct xlog *log,
struct xlog_in_core *iclog);
STATIC void
xlog_grant_push_ail(
struct xlog *log,
int need_bytes);
STATIC void
xlog_regrant_reserve_log_space(
struct xlog *log,
struct xlog_ticket *ticket);
STATIC void
xlog_ungrant_log_space(
struct xlog *log,
struct xlog_ticket *ticket);
STATIC void
xlog_sync(
struct xlog *log,
struct xlog_in_core *iclog);
#if defined(DEBUG)
STATIC void
xlog_verify_dest_ptr(
struct xlog *log,
void *ptr);
STATIC void
xlog_verify_grant_tail(
struct xlog *log);
STATIC void
xlog_verify_iclog(
struct xlog *log,
struct xlog_in_core *iclog,
int count);
STATIC void
xlog_verify_tail_lsn(
struct xlog *log,
struct xlog_in_core *iclog,
xfs_lsn_t tail_lsn);
#else
#define xlog_verify_dest_ptr(a,b)
#define xlog_verify_grant_tail(a)
#define xlog_verify_iclog(a,b,c)
#define xlog_verify_tail_lsn(a,b,c)
#endif
STATIC int
xlog_iclogs_empty(
struct xlog *log);
static void
xlog_grant_sub_space(
struct xlog *log,
atomic64_t *head,
int bytes)
{
int64_t head_val = atomic64_read(head);
int64_t new, old;
do {
int cycle, space;
xlog_crack_grant_head_val(head_val, &cycle, &space);
space -= bytes;
if (space < 0) {
space += log->l_logsize;
cycle--;
}
old = head_val;
new = xlog_assign_grant_head_val(cycle, space);
head_val = atomic64_cmpxchg(head, old, new);
} while (head_val != old);
}
static void
xlog_grant_add_space(
struct xlog *log,
atomic64_t *head,
int bytes)
{
int64_t head_val = atomic64_read(head);
int64_t new, old;
do {
int tmp;
int cycle, space;
xlog_crack_grant_head_val(head_val, &cycle, &space);
tmp = log->l_logsize - space;
if (tmp > bytes)
space += bytes;
else {
space = bytes - tmp;
cycle++;
}
old = head_val;
new = xlog_assign_grant_head_val(cycle, space);
head_val = atomic64_cmpxchg(head, old, new);
} while (head_val != old);
}
STATIC void
xlog_grant_head_init(
struct xlog_grant_head *head)
{
xlog_assign_grant_head(&head->grant, 1, 0);
INIT_LIST_HEAD(&head->waiters);
spin_lock_init(&head->lock);
}
STATIC void
xlog_grant_head_wake_all(
struct xlog_grant_head *head)
{
struct xlog_ticket *tic;
spin_lock(&head->lock);
list_for_each_entry(tic, &head->waiters, t_queue)
wake_up_process(tic->t_task);
spin_unlock(&head->lock);
}
static inline int
xlog_ticket_reservation(
struct xlog *log,
struct xlog_grant_head *head,
struct xlog_ticket *tic)
{
if (head == &log->l_write_head) {
ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
return tic->t_unit_res;
} else {
if (tic->t_flags & XLOG_TIC_PERM_RESERV)
return tic->t_unit_res * tic->t_cnt;
else
return tic->t_unit_res;
}
}
STATIC bool
xlog_grant_head_wake(
struct xlog *log,
struct xlog_grant_head *head,
int *free_bytes)
{
struct xlog_ticket *tic;
int need_bytes;
xfs: push the AIL in xlog_grant_head_wake In the situation where the log is full and the CIL has not recently flushed, the AIL push threshold is throttled back to the where the last write of the head of the log was completed. This is stored in log->l_last_sync_lsn. Hence if the CIL holds > 25% of the log space pinned by flushes and/or aggregation in progress, we can get the situation where the head of the log lags a long way behind the reservation grant head. When this happens, the AIL push target is trimmed back from where the reservation grant head wants to push the log tail to, back to where the head of the log currently is. This means the push target doesn't reach far enough into the log to actually move the tail before the transaction reservation goes to sleep. When the CIL push completes, it moves the log head forward such that the AIL push target can now be moved, but that has no mechanism for puhsing the log tail. Further, if the next tail movement of the log is not large enough wake the waiter (i.e. still not enough space for it to have a reservation granted), we don't wake anything up, and hence we do not update the AIL push target to take into account the head of the log moving and allowing the push target to be moved forwards. To avoid this particular condition, if we fail to wake the first waiter on the grant head because we don't have enough space, push on the AIL again. This will pick up any movement of the log head and allow the push target to move forward due to completion of CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 07:32:48 +07:00
bool woken_task = false;
list_for_each_entry(tic, &head->waiters, t_queue) {
xfs: push the AIL in xlog_grant_head_wake In the situation where the log is full and the CIL has not recently flushed, the AIL push threshold is throttled back to the where the last write of the head of the log was completed. This is stored in log->l_last_sync_lsn. Hence if the CIL holds > 25% of the log space pinned by flushes and/or aggregation in progress, we can get the situation where the head of the log lags a long way behind the reservation grant head. When this happens, the AIL push target is trimmed back from where the reservation grant head wants to push the log tail to, back to where the head of the log currently is. This means the push target doesn't reach far enough into the log to actually move the tail before the transaction reservation goes to sleep. When the CIL push completes, it moves the log head forward such that the AIL push target can now be moved, but that has no mechanism for puhsing the log tail. Further, if the next tail movement of the log is not large enough wake the waiter (i.e. still not enough space for it to have a reservation granted), we don't wake anything up, and hence we do not update the AIL push target to take into account the head of the log moving and allowing the push target to be moved forwards. To avoid this particular condition, if we fail to wake the first waiter on the grant head because we don't have enough space, push on the AIL again. This will pick up any movement of the log head and allow the push target to move forward due to completion of CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 07:32:48 +07:00
/*
* There is a chance that the size of the CIL checkpoints in
* progress at the last AIL push target calculation resulted in
* limiting the target to the log head (l_last_sync_lsn) at the
* time. This may not reflect where the log head is now as the
* CIL checkpoints may have completed.
*
* Hence when we are woken here, it may be that the head of the
* log that has moved rather than the tail. As the tail didn't
* move, there still won't be space available for the
* reservation we require. However, if the AIL has already
* pushed to the target defined by the old log head location, we
* will hang here waiting for something else to update the AIL
* push target.
*
* Therefore, if there isn't space to wake the first waiter on
* the grant head, we need to push the AIL again to ensure the
* target reflects both the current log tail and log head
* position before we wait for the tail to move again.
*/
need_bytes = xlog_ticket_reservation(log, head, tic);
xfs: push the AIL in xlog_grant_head_wake In the situation where the log is full and the CIL has not recently flushed, the AIL push threshold is throttled back to the where the last write of the head of the log was completed. This is stored in log->l_last_sync_lsn. Hence if the CIL holds > 25% of the log space pinned by flushes and/or aggregation in progress, we can get the situation where the head of the log lags a long way behind the reservation grant head. When this happens, the AIL push target is trimmed back from where the reservation grant head wants to push the log tail to, back to where the head of the log currently is. This means the push target doesn't reach far enough into the log to actually move the tail before the transaction reservation goes to sleep. When the CIL push completes, it moves the log head forward such that the AIL push target can now be moved, but that has no mechanism for puhsing the log tail. Further, if the next tail movement of the log is not large enough wake the waiter (i.e. still not enough space for it to have a reservation granted), we don't wake anything up, and hence we do not update the AIL push target to take into account the head of the log moving and allowing the push target to be moved forwards. To avoid this particular condition, if we fail to wake the first waiter on the grant head because we don't have enough space, push on the AIL again. This will pick up any movement of the log head and allow the push target to move forward due to completion of CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 07:32:48 +07:00
if (*free_bytes < need_bytes) {
if (!woken_task)
xlog_grant_push_ail(log, need_bytes);
return false;
xfs: push the AIL in xlog_grant_head_wake In the situation where the log is full and the CIL has not recently flushed, the AIL push threshold is throttled back to the where the last write of the head of the log was completed. This is stored in log->l_last_sync_lsn. Hence if the CIL holds > 25% of the log space pinned by flushes and/or aggregation in progress, we can get the situation where the head of the log lags a long way behind the reservation grant head. When this happens, the AIL push target is trimmed back from where the reservation grant head wants to push the log tail to, back to where the head of the log currently is. This means the push target doesn't reach far enough into the log to actually move the tail before the transaction reservation goes to sleep. When the CIL push completes, it moves the log head forward such that the AIL push target can now be moved, but that has no mechanism for puhsing the log tail. Further, if the next tail movement of the log is not large enough wake the waiter (i.e. still not enough space for it to have a reservation granted), we don't wake anything up, and hence we do not update the AIL push target to take into account the head of the log moving and allowing the push target to be moved forwards. To avoid this particular condition, if we fail to wake the first waiter on the grant head because we don't have enough space, push on the AIL again. This will pick up any movement of the log head and allow the push target to move forward due to completion of CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 07:32:48 +07:00
}
*free_bytes -= need_bytes;
trace_xfs_log_grant_wake_up(log, tic);
wake_up_process(tic->t_task);
xfs: push the AIL in xlog_grant_head_wake In the situation where the log is full and the CIL has not recently flushed, the AIL push threshold is throttled back to the where the last write of the head of the log was completed. This is stored in log->l_last_sync_lsn. Hence if the CIL holds > 25% of the log space pinned by flushes and/or aggregation in progress, we can get the situation where the head of the log lags a long way behind the reservation grant head. When this happens, the AIL push target is trimmed back from where the reservation grant head wants to push the log tail to, back to where the head of the log currently is. This means the push target doesn't reach far enough into the log to actually move the tail before the transaction reservation goes to sleep. When the CIL push completes, it moves the log head forward such that the AIL push target can now be moved, but that has no mechanism for puhsing the log tail. Further, if the next tail movement of the log is not large enough wake the waiter (i.e. still not enough space for it to have a reservation granted), we don't wake anything up, and hence we do not update the AIL push target to take into account the head of the log moving and allowing the push target to be moved forwards. To avoid this particular condition, if we fail to wake the first waiter on the grant head because we don't have enough space, push on the AIL again. This will pick up any movement of the log head and allow the push target to move forward due to completion of CIL pushing. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-06 07:32:48 +07:00
woken_task = true;
}
return true;
}
STATIC int
xlog_grant_head_wait(
struct xlog *log,
struct xlog_grant_head *head,
struct xlog_ticket *tic,
int need_bytes) __releases(&head->lock)
__acquires(&head->lock)
{
list_add_tail(&tic->t_queue, &head->waiters);
do {
if (XLOG_FORCED_SHUTDOWN(log))
goto shutdown;
xlog_grant_push_ail(log, need_bytes);
__set_current_state(TASK_UNINTERRUPTIBLE);
spin_unlock(&head->lock);
XFS_STATS_INC(log->l_mp, xs_sleep_logspace);
trace_xfs_log_grant_sleep(log, tic);
schedule();
trace_xfs_log_grant_wake(log, tic);
spin_lock(&head->lock);
if (XLOG_FORCED_SHUTDOWN(log))
goto shutdown;
} while (xlog_space_left(log, &head->grant) < need_bytes);
list_del_init(&tic->t_queue);
return 0;
shutdown:
list_del_init(&tic->t_queue);
return -EIO;
}
/*
* Atomically get the log space required for a log ticket.
*
* Once a ticket gets put onto head->waiters, it will only return after the
* needed reservation is satisfied.
*
* This function is structured so that it has a lock free fast path. This is
* necessary because every new transaction reservation will come through this
* path. Hence any lock will be globally hot if we take it unconditionally on
* every pass.
*
* As tickets are only ever moved on and off head->waiters under head->lock, we
* only need to take that lock if we are going to add the ticket to the queue
* and sleep. We can avoid taking the lock if the ticket was never added to
* head->waiters because the t_queue list head will be empty and we hold the
* only reference to it so it can safely be checked unlocked.
*/
STATIC int
xlog_grant_head_check(
struct xlog *log,
struct xlog_grant_head *head,
struct xlog_ticket *tic,
int *need_bytes)
{
int free_bytes;
int error = 0;
ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
/*
* If there are other waiters on the queue then give them a chance at
* logspace before us. Wake up the first waiters, if we do not wake
* up all the waiters then go to sleep waiting for more free space,
* otherwise try to get some space for this transaction.
*/
*need_bytes = xlog_ticket_reservation(log, head, tic);
free_bytes = xlog_space_left(log, &head->grant);
if (!list_empty_careful(&head->waiters)) {
spin_lock(&head->lock);
if (!xlog_grant_head_wake(log, head, &free_bytes) ||
free_bytes < *need_bytes) {
error = xlog_grant_head_wait(log, head, tic,
*need_bytes);
}
spin_unlock(&head->lock);
} else if (free_bytes < *need_bytes) {
spin_lock(&head->lock);
error = xlog_grant_head_wait(log, head, tic, *need_bytes);
spin_unlock(&head->lock);
}
return error;
}
static void
xlog_tic_reset_res(xlog_ticket_t *tic)
{
tic->t_res_num = 0;
tic->t_res_arr_sum = 0;
tic->t_res_num_ophdrs = 0;
}
static void
xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
{
if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
/* add to overflow and start again */
tic->t_res_o_flow += tic->t_res_arr_sum;
tic->t_res_num = 0;
tic->t_res_arr_sum = 0;
}
tic->t_res_arr[tic->t_res_num].r_len = len;
tic->t_res_arr[tic->t_res_num].r_type = type;
tic->t_res_arr_sum += len;
tic->t_res_num++;
}
/*
* Replenish the byte reservation required by moving the grant write head.
*/
int
xfs_log_regrant(
struct xfs_mount *mp,
struct xlog_ticket *tic)
{
struct xlog *log = mp->m_log;
int need_bytes;
int error = 0;
if (XLOG_FORCED_SHUTDOWN(log))
return -EIO;
XFS_STATS_INC(mp, xs_try_logspace);
/*
* This is a new transaction on the ticket, so we need to change the
* transaction ID so that the next transaction has a different TID in
* the log. Just add one to the existing tid so that we can see chains
* of rolling transactions in the log easily.
*/
tic->t_tid++;
xlog_grant_push_ail(log, tic->t_unit_res);
tic->t_curr_res = tic->t_unit_res;
xlog_tic_reset_res(tic);
if (tic->t_cnt > 0)
return 0;
trace_xfs_log_regrant(log, tic);
error = xlog_grant_head_check(log, &log->l_write_head, tic,
&need_bytes);
if (error)
goto out_error;
xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
trace_xfs_log_regrant_exit(log, tic);
xlog_verify_grant_tail(log);
return 0;
out_error:
/*
* If we are failing, make sure the ticket doesn't have any current
* reservations. We don't want to add this back when the ticket/
* transaction gets cancelled.
*/
tic->t_curr_res = 0;
tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
return error;
}
/*
* Reserve log space and return a ticket corresponding to the reservation.
*
* Each reservation is going to reserve extra space for a log record header.
* When writes happen to the on-disk log, we don't subtract the length of the
* log record header from any reservation. By wasting space in each
* reservation, we prevent over allocation problems.
*/
int
xfs_log_reserve(
struct xfs_mount *mp,
int unit_bytes,
int cnt,
struct xlog_ticket **ticp,
uint8_t client,
bool permanent)
{
struct xlog *log = mp->m_log;
struct xlog_ticket *tic;
int need_bytes;
int error = 0;
ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
if (XLOG_FORCED_SHUTDOWN(log))
return -EIO;
XFS_STATS_INC(mp, xs_try_logspace);
ASSERT(*ticp == NULL);
tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, 0);
*ticp = tic;
xfs: fix direct IO nested transaction deadlock. The direct IO path can do a nested transaction reservation when writing past the EOF. The first transaction is the append transaction for setting the filesize at IO completion, but we can also need a transaction for allocation of blocks. If the log is low on space due to reservations and small log, the append transaction can be granted after wating for space as the only active transaction in the system. This then attempts a reservation for an allocation, which there isn't space in the log for, and the reservation sleeps. The result is that there is nothing left in the system to wake up all the processes waiting for log space to come free. The stack trace that shows this deadlock is relatively innocuous: xlog_grant_head_wait xlog_grant_head_check xfs_log_reserve xfs_trans_reserve xfs_iomap_write_direct __xfs_get_blocks xfs_get_blocks_direct do_blockdev_direct_IO __blockdev_direct_IO xfs_vm_direct_IO generic_file_direct_write xfs_file_dio_aio_writ xfs_file_aio_write do_sync_write vfs_write This was discovered on a filesystem with a log of only 10MB, and a log stripe unit of 256k whih increased the base reservations by 512k. Hence a allocation transaction requires 1.2MB of log space to be available instead of only 260k, and so greatly increased the chance that there wouldn't be enough log space available for the nested transaction to succeed. The key to reproducing it is this mkfs command: mkfs.xfs -f -d agcount=16,su=256k,sw=12 -l su=256k,size=2560b $SCRATCH_DEV The test case was a 1000 fsstress processes running with random freeze and unfreezes every few seconds. Thanks to Eryu Guan (eguan@redhat.com) for writing the test that found this on a system with a somewhat unique default configuration.... cc: <stable@vger.kernel.org> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Andrew Dahl <adahl@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-11-28 09:01:00 +07:00
xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt
: tic->t_unit_res);
trace_xfs_log_reserve(log, tic);
error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
&need_bytes);
if (error)
goto out_error;
xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
trace_xfs_log_reserve_exit(log, tic);
xlog_verify_grant_tail(log);
return 0;
out_error:
/*
* If we are failing, make sure the ticket doesn't have any current
* reservations. We don't want to add this back when the ticket/
* transaction gets cancelled.
*/
tic->t_curr_res = 0;
tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
return error;
}
/*
* NOTES:
*
* 1. currblock field gets updated at startup and after in-core logs
* marked as with WANT_SYNC.
*/
/*
* This routine is called when a user of a log manager ticket is done with
* the reservation. If the ticket was ever used, then a commit record for
* the associated transaction is written out as a log operation header with
* no data. The flag XLOG_TIC_INITED is set when the first write occurs with
* a given ticket. If the ticket was one with a permanent reservation, then
* a few operations are done differently. Permanent reservation tickets by
* default don't release the reservation. They just commit the current
* transaction with the belief that the reservation is still needed. A flag
* must be passed in before permanent reservations are actually released.
* When these type of tickets are not released, they need to be set into
* the inited state again. By doing this, a start record will be written
* out when the next write occurs.
*/
xfs_lsn_t
xfs_log_done(
struct xfs_mount *mp,
struct xlog_ticket *ticket,
struct xlog_in_core **iclog,
bool regrant)
{
struct xlog *log = mp->m_log;
xfs_lsn_t lsn = 0;
if (XLOG_FORCED_SHUTDOWN(log) ||
/*
* If nothing was ever written, don't write out commit record.
* If we get an error, just continue and give back the log ticket.
*/
(((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
(xlog_commit_record(log, ticket, iclog, &lsn)))) {
lsn = (xfs_lsn_t) -1;
regrant = false;
}
if (!regrant) {
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_done_nonperm(log, ticket);
/*
* Release ticket if not permanent reservation or a specific
* request has been made to release a permanent reservation.
*/
xlog_ungrant_log_space(log, ticket);
} else {
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_done_perm(log, ticket);
xlog_regrant_reserve_log_space(log, ticket);
/* If this ticket was a permanent reservation and we aren't
* trying to release it, reset the inited flags; so next time
* we write, a start record will be written out.
*/
ticket->t_flags |= XLOG_TIC_INITED;
}
xfs_log_ticket_put(ticket);
return lsn;
}
static bool
__xlog_state_release_iclog(
struct xlog *log,
struct xlog_in_core *iclog)
{
lockdep_assert_held(&log->l_icloglock);
if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
/* update tail before writing to iclog */
xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
iclog->ic_state = XLOG_STATE_SYNCING;
iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
xlog_verify_tail_lsn(log, iclog, tail_lsn);
/* cycle incremented when incrementing curr_block */
return true;
}
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
return false;
}
/*
* Flush iclog to disk if this is the last reference to the given iclog and the
* it is in the WANT_SYNC state.
*/
static int
xlog_state_release_iclog(
struct xlog *log,
struct xlog_in_core *iclog)
{
lockdep_assert_held(&log->l_icloglock);
if (iclog->ic_state == XLOG_STATE_IOERROR)
return -EIO;
if (atomic_dec_and_test(&iclog->ic_refcnt) &&
__xlog_state_release_iclog(log, iclog)) {
spin_unlock(&log->l_icloglock);
xlog_sync(log, iclog);
spin_lock(&log->l_icloglock);
}
return 0;
}
int
xfs_log_release_iclog(
struct xfs_mount *mp,
struct xlog_in_core *iclog)
{
struct xlog *log = mp->m_log;
bool sync;
if (iclog->ic_state == XLOG_STATE_IOERROR) {
xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
return -EIO;
}
if (atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) {
sync = __xlog_state_release_iclog(log, iclog);
spin_unlock(&log->l_icloglock);
if (sync)
xlog_sync(log, iclog);
}
return 0;
}
/*
* Mount a log filesystem
*
* mp - ubiquitous xfs mount point structure
* log_target - buftarg of on-disk log device
* blk_offset - Start block # where block size is 512 bytes (BBSIZE)
* num_bblocks - Number of BBSIZE blocks in on-disk log
*
* Return error or zero.
*/
int
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
xfs_log_mount(
xfs_mount_t *mp,
xfs_buftarg_t *log_target,
xfs_daddr_t blk_offset,
int num_bblks)
{
bool fatal = xfs_sb_version_hascrc(&mp->m_sb);
int error = 0;
int min_logfsbs;
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
xfs_notice(mp, "Mounting V%d Filesystem",
XFS_SB_VERSION_NUM(&mp->m_sb));
} else {
xfs_notice(mp,
"Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.",
XFS_SB_VERSION_NUM(&mp->m_sb));
ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
}
mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
if (IS_ERR(mp->m_log)) {
error = PTR_ERR(mp->m_log);
goto out;
}
/*
* Validate the given log space and drop a critical message via syslog
* if the log size is too small that would lead to some unexpected
* situations in transaction log space reservation stage.
*
* Note: we can't just reject the mount if the validation fails. This
* would mean that people would have to downgrade their kernel just to
* remedy the situation as there is no way to grow the log (short of
* black magic surgery with xfs_db).
*
* We can, however, reject mounts for CRC format filesystems, as the
* mkfs binary being used to make the filesystem should never create a
* filesystem with a log that is too small.
*/
min_logfsbs = xfs_log_calc_minimum_size(mp);
if (mp->m_sb.sb_logblocks < min_logfsbs) {
xfs_warn(mp,
"Log size %d blocks too small, minimum size is %d blocks",
mp->m_sb.sb_logblocks, min_logfsbs);
error = -EINVAL;
} else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) {
xfs_warn(mp,
"Log size %d blocks too large, maximum size is %lld blocks",
mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS);
error = -EINVAL;
} else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) {
xfs_warn(mp,
"log size %lld bytes too large, maximum size is %lld bytes",
XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks),
XFS_MAX_LOG_BYTES);
error = -EINVAL;
} else if (mp->m_sb.sb_logsunit > 1 &&
mp->m_sb.sb_logsunit % mp->m_sb.sb_blocksize) {
xfs_warn(mp,
"log stripe unit %u bytes must be a multiple of block size",
mp->m_sb.sb_logsunit);
error = -EINVAL;
fatal = true;
}
if (error) {
/*
* Log check errors are always fatal on v5; or whenever bad
* metadata leads to a crash.
*/
if (fatal) {
xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");
ASSERT(0);
goto out_free_log;
}
xfs_crit(mp, "Log size out of supported range.");
xfs_crit(mp,
"Continuing onwards, but if log hangs are experienced then please report this message in the bug report.");
}
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
/*
* Initialize the AIL now we have a log.
*/
error = xfs_trans_ail_init(mp);
if (error) {
xfs_warn(mp, "AIL initialisation failed: error %d", error);
goto out_free_log;
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
}
mp->m_log->l_ailp = mp->m_ail;
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
/*
* skip log recovery on a norecovery mount. pretend it all
* just worked.
*/
if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
if (readonly)
mp->m_flags &= ~XFS_MOUNT_RDONLY;
error = xlog_recover(mp->m_log);
if (readonly)
mp->m_flags |= XFS_MOUNT_RDONLY;
if (error) {
xfs_warn(mp, "log mount/recovery failed: error %d",
error);
xlog_recover_cancel(mp->m_log);
goto out_destroy_ail;
}
}
error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj,
"log");
if (error)
goto out_destroy_ail;
/* Normal transactions can now occur */
mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
/*
* Now the log has been fully initialised and we know were our
* space grant counters are, we can initialise the permanent ticket
* needed for delayed logging to work.
*/
xlog_cil_init_post_recovery(mp->m_log);
return 0;
out_destroy_ail:
xfs_trans_ail_destroy(mp);
out_free_log:
xlog_dealloc_log(mp->m_log);
out:
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
return error;
}
/*
* Finish the recovery of the file system. This is separate from the
* xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read
* in the root and real-time bitmap inodes between calling xfs_log_mount() and
* here.
*
* If we finish recovery successfully, start the background log work. If we are
* not doing recovery, then we have a RO filesystem and we don't need to start
* it.
*/
int
xfs_log_mount_finish(
struct xfs_mount *mp)
{
int error = 0;
bool readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
bool recovered = mp->m_log->l_flags & XLOG_RECOVERY_NEEDED;
if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
return 0;
} else if (readonly) {
/* Allow unlinked processing to proceed */
mp->m_flags &= ~XFS_MOUNT_RDONLY;
}
/*
* During the second phase of log recovery, we need iget and
* iput to behave like they do for an active filesystem.
* xfs_fs_drop_inode needs to be able to prevent the deletion
* of inodes before we're done replaying log items on those
* inodes. Turn it off immediately after recovery finishes
* so that we don't leak the quota inodes if subsequent mount
* activities fail.
*
* We let all inodes involved in redo item processing end up on
* the LRU instead of being evicted immediately so that if we do
* something to an unlinked inode, the irele won't cause
* premature truncation and freeing of the inode, which results
* in log recovery failure. We have to evict the unreferenced
Rename superblock flags (MS_xyz -> SB_xyz) This is a pure automated search-and-replace of the internal kernel superblock flags. The s_flags are now called SB_*, with the names and the values for the moment mirroring the MS_* flags that they're equivalent to. Note how the MS_xyz flags are the ones passed to the mount system call, while the SB_xyz flags are what we then use in sb->s_flags. The script to do this was: # places to look in; re security/*: it generally should *not* be # touched (that stuff parses mount(2) arguments directly), but # there are two places where we really deal with superblock flags. FILES="drivers/mtd drivers/staging/lustre fs ipc mm \ include/linux/fs.h include/uapi/linux/bfs_fs.h \ security/apparmor/apparmorfs.c security/apparmor/include/lib.h" # the list of MS_... constants SYMS="RDONLY NOSUID NODEV NOEXEC SYNCHRONOUS REMOUNT MANDLOCK \ DIRSYNC NOATIME NODIRATIME BIND MOVE REC VERBOSE SILENT \ POSIXACL UNBINDABLE PRIVATE SLAVE SHARED RELATIME KERNMOUNT \ I_VERSION STRICTATIME LAZYTIME SUBMOUNT NOREMOTELOCK NOSEC BORN \ ACTIVE NOUSER" SED_PROG= for i in $SYMS; do SED_PROG="$SED_PROG -e s/MS_$i/SB_$i/g"; done # we want files that contain at least one of MS_..., # with fs/namespace.c and fs/pnode.c excluded. L=$(for i in $SYMS; do git grep -w -l MS_$i $FILES; done| sort|uniq|grep -v '^fs/namespace.c'|grep -v '^fs/pnode.c') for f in $L; do sed -i $f $SED_PROG; done Requested-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-11-28 04:05:09 +07:00
* lru inodes after clearing SB_ACTIVE because we don't
* otherwise clean up the lru if there's a subsequent failure in
* xfs_mountfs, which leads to us leaking the inodes if nothing
* else (e.g. quotacheck) references the inodes before the
* mount failure occurs.
*/
Rename superblock flags (MS_xyz -> SB_xyz) This is a pure automated search-and-replace of the internal kernel superblock flags. The s_flags are now called SB_*, with the names and the values for the moment mirroring the MS_* flags that they're equivalent to. Note how the MS_xyz flags are the ones passed to the mount system call, while the SB_xyz flags are what we then use in sb->s_flags. The script to do this was: # places to look in; re security/*: it generally should *not* be # touched (that stuff parses mount(2) arguments directly), but # there are two places where we really deal with superblock flags. FILES="drivers/mtd drivers/staging/lustre fs ipc mm \ include/linux/fs.h include/uapi/linux/bfs_fs.h \ security/apparmor/apparmorfs.c security/apparmor/include/lib.h" # the list of MS_... constants SYMS="RDONLY NOSUID NODEV NOEXEC SYNCHRONOUS REMOUNT MANDLOCK \ DIRSYNC NOATIME NODIRATIME BIND MOVE REC VERBOSE SILENT \ POSIXACL UNBINDABLE PRIVATE SLAVE SHARED RELATIME KERNMOUNT \ I_VERSION STRICTATIME LAZYTIME SUBMOUNT NOREMOTELOCK NOSEC BORN \ ACTIVE NOUSER" SED_PROG= for i in $SYMS; do SED_PROG="$SED_PROG -e s/MS_$i/SB_$i/g"; done # we want files that contain at least one of MS_..., # with fs/namespace.c and fs/pnode.c excluded. L=$(for i in $SYMS; do git grep -w -l MS_$i $FILES; done| sort|uniq|grep -v '^fs/namespace.c'|grep -v '^fs/pnode.c') for f in $L; do sed -i $f $SED_PROG; done Requested-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-11-28 04:05:09 +07:00
mp->m_super->s_flags |= SB_ACTIVE;
error = xlog_recover_finish(mp->m_log);
if (!error)
xfs_log_work_queue(mp);
Rename superblock flags (MS_xyz -> SB_xyz) This is a pure automated search-and-replace of the internal kernel superblock flags. The s_flags are now called SB_*, with the names and the values for the moment mirroring the MS_* flags that they're equivalent to. Note how the MS_xyz flags are the ones passed to the mount system call, while the SB_xyz flags are what we then use in sb->s_flags. The script to do this was: # places to look in; re security/*: it generally should *not* be # touched (that stuff parses mount(2) arguments directly), but # there are two places where we really deal with superblock flags. FILES="drivers/mtd drivers/staging/lustre fs ipc mm \ include/linux/fs.h include/uapi/linux/bfs_fs.h \ security/apparmor/apparmorfs.c security/apparmor/include/lib.h" # the list of MS_... constants SYMS="RDONLY NOSUID NODEV NOEXEC SYNCHRONOUS REMOUNT MANDLOCK \ DIRSYNC NOATIME NODIRATIME BIND MOVE REC VERBOSE SILENT \ POSIXACL UNBINDABLE PRIVATE SLAVE SHARED RELATIME KERNMOUNT \ I_VERSION STRICTATIME LAZYTIME SUBMOUNT NOREMOTELOCK NOSEC BORN \ ACTIVE NOUSER" SED_PROG= for i in $SYMS; do SED_PROG="$SED_PROG -e s/MS_$i/SB_$i/g"; done # we want files that contain at least one of MS_..., # with fs/namespace.c and fs/pnode.c excluded. L=$(for i in $SYMS; do git grep -w -l MS_$i $FILES; done| sort|uniq|grep -v '^fs/namespace.c'|grep -v '^fs/pnode.c') for f in $L; do sed -i $f $SED_PROG; done Requested-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-11-28 04:05:09 +07:00
mp->m_super->s_flags &= ~SB_ACTIVE;
evict_inodes(mp->m_super);
/*
* Drain the buffer LRU after log recovery. This is required for v4
* filesystems to avoid leaving around buffers with NULL verifier ops,
* but we do it unconditionally to make sure we're always in a clean
* cache state after mount.
*
* Don't push in the error case because the AIL may have pending intents
* that aren't removed until recovery is cancelled.
*/
if (!error && recovered) {
xfs_log_force(mp, XFS_LOG_SYNC);
xfs_ail_push_all_sync(mp->m_ail);
}
xfs_wait_buftarg(mp->m_ddev_targp);
if (readonly)
mp->m_flags |= XFS_MOUNT_RDONLY;
return error;
}
/*
* The mount has failed. Cancel the recovery if it hasn't completed and destroy
* the log.
*/
void
xfs_log_mount_cancel(
struct xfs_mount *mp)
{
xlog_recover_cancel(mp->m_log);
xfs_log_unmount(mp);
}
/*
* Final log writes as part of unmount.
*
* Mark the filesystem clean as unmount happens. Note that during relocation
* this routine needs to be executed as part of source-bag while the
* deallocation must not be done until source-end.
*/
/* Actually write the unmount record to disk. */
static void
xfs_log_write_unmount_record(
struct xfs_mount *mp)
{
/* the data section must be 32 bit size aligned */
struct xfs_unmount_log_format magic = {
.magic = XLOG_UNMOUNT_TYPE,
};
struct xfs_log_iovec reg = {
.i_addr = &magic,
.i_len = sizeof(magic),
.i_type = XLOG_REG_TYPE_UNMOUNT,
};
struct xfs_log_vec vec = {
.lv_niovecs = 1,
.lv_iovecp = &reg,
};
struct xlog *log = mp->m_log;
struct xlog_in_core *iclog;
struct xlog_ticket *tic = NULL;
xfs_lsn_t lsn;
uint flags = XLOG_UNMOUNT_TRANS;
int error;
error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0);
if (error)
goto out_err;
/*
* If we think the summary counters are bad, clear the unmount header
* flag in the unmount record so that the summary counters will be
* recalculated during log recovery at next mount. Refer to
* xlog_check_unmount_rec for more details.
*/
if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp,
XFS_ERRTAG_FORCE_SUMMARY_RECALC)) {
xfs_alert(mp, "%s: will fix summary counters at next mount",
__func__);
flags &= ~XLOG_UNMOUNT_TRANS;
}
/* remove inited flag, and account for space used */
tic->t_flags = 0;
tic->t_curr_res -= sizeof(magic);
error = xlog_write(log, &vec, tic, &lsn, NULL, flags);
/*
* At this point, we're umounting anyway, so there's no point in
* transitioning log state to IOERROR. Just continue...
*/
out_err:
if (error)
xfs_alert(mp, "%s: unmount record failed", __func__);
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
atomic_inc(&iclog->ic_refcnt);
xlog_state_want_sync(log, iclog);
error = xlog_state_release_iclog(log, iclog);
switch (iclog->ic_state) {
default:
if (!XLOG_FORCED_SHUTDOWN(log)) {
xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
break;
}
/* fall through */
case XLOG_STATE_ACTIVE:
case XLOG_STATE_DIRTY:
spin_unlock(&log->l_icloglock);
break;
}
if (tic) {
trace_xfs_log_umount_write(log, tic);
xlog_ungrant_log_space(log, tic);
xfs_log_ticket_put(tic);
}
}
/*
* Unmount record used to have a string "Unmount filesystem--" in the
* data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
* We just write the magic number now since that particular field isn't
* currently architecture converted and "Unmount" is a bit foo.
* As far as I know, there weren't any dependencies on the old behaviour.
*/
static int
xfs_log_unmount_write(xfs_mount_t *mp)
{
struct xlog *log = mp->m_log;
xlog_in_core_t *iclog;
#ifdef DEBUG
xlog_in_core_t *first_iclog;
#endif
int error;
/*
* Don't write out unmount record on norecovery mounts or ro devices.
* Or, if we are doing a forced umount (typically because of IO errors).
*/
if (mp->m_flags & XFS_MOUNT_NORECOVERY ||
xfs_readonly_buftarg(log->l_targ)) {
ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
return 0;
}
error = xfs_log_force(mp, XFS_LOG_SYNC);
ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
#ifdef DEBUG
first_iclog = iclog = log->l_iclog;
do {
if (iclog->ic_state != XLOG_STATE_IOERROR) {
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
ASSERT(iclog->ic_offset == 0);
}
iclog = iclog->ic_next;
} while (iclog != first_iclog);
#endif
if (! (XLOG_FORCED_SHUTDOWN(log))) {
xfs_log_write_unmount_record(mp);
} else {
/*
* We're already in forced_shutdown mode, couldn't
* even attempt to write out the unmount transaction.
*
* Go through the motions of sync'ing and releasing
* the iclog, even though no I/O will actually happen,
* we need to wait for other log I/Os that may already
* be in progress. Do this as a separate section of
* code so we'll know if we ever get stuck here that
* we're in this odd situation of trying to unmount
* a file system that went into forced_shutdown as
* the result of an unmount..
*/
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
atomic_inc(&iclog->ic_refcnt);
xlog_state_want_sync(log, iclog);
error = xlog_state_release_iclog(log, iclog);
switch (iclog->ic_state) {
case XLOG_STATE_ACTIVE:
case XLOG_STATE_DIRTY:
case XLOG_STATE_IOERROR:
spin_unlock(&log->l_icloglock);
break;
default:
xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
break;
}
}
return error;
} /* xfs_log_unmount_write */
/*
* Empty the log for unmount/freeze.
*
* To do this, we first need to shut down the background log work so it is not
* trying to cover the log as we clean up. We then need to unpin all objects in
* the log so we can then flush them out. Once they have completed their IO and
* run the callbacks removing themselves from the AIL, we can write the unmount
* record.
*/
void
xfs_log_quiesce(
struct xfs_mount *mp)
{
cancel_delayed_work_sync(&mp->m_log->l_work);
xfs_log_force(mp, XFS_LOG_SYNC);
/*
* The superblock buffer is uncached and while xfs_ail_push_all_sync()
* will push it, xfs_wait_buftarg() will not wait for it. Further,
* xfs_buf_iowait() cannot be used because it was pushed with the
* XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for
* the IO to complete.
*/
xfs_ail_push_all_sync(mp->m_ail);
xfs_wait_buftarg(mp->m_ddev_targp);
xfs_buf_lock(mp->m_sb_bp);
xfs_buf_unlock(mp->m_sb_bp);
xfs_log_unmount_write(mp);
}
/*
* Shut down and release the AIL and Log.
*
* During unmount, we need to ensure we flush all the dirty metadata objects
* from the AIL so that the log is empty before we write the unmount record to
* the log. Once this is done, we can tear down the AIL and the log.
*/
void
xfs_log_unmount(
struct xfs_mount *mp)
{
xfs_log_quiesce(mp);
[XFS] Move AIL pushing into it's own thread When many hundreds to thousands of threads all try to do simultaneous transactions and the log is in a tail-pushing situation (i.e. full), we can get multiple threads walking the AIL list and contending on the AIL lock. The AIL push is, in effect, a simple I/O dispatch algorithm complicated by the ordering constraints placed on it by the transaction subsystem. It really does not need multiple threads to push on it - even when only a single CPU is pushing the AIL, it can push the I/O out far faster that pretty much any disk subsystem can handle. So, to avoid contention problems stemming from multiple list walkers, move the list walk off into another thread and simply provide a "target" to push to. When a thread requires a push, it sets the target and wakes the push thread, then goes to sleep waiting for the required amount of space to become available in the log. This mechanism should also be a lot fairer under heavy load as the waiters will queue in arrival order, rather than queuing in "who completed a push first" order. Also, by moving the pushing to a separate thread we can do more effectively overload detection and prevention as we can keep context from loop iteration to loop iteration. That is, we can push only part of the list each loop and not have to loop back to the start of the list every time we run. This should also help by reducing the number of items we try to lock and/or push items that we cannot move. Note that this patch is not intended to solve the inefficiencies in the AIL structure and the associated issues with extremely large list contents. That needs to be addresses separately; parallel access would cause problems to any new structure as well, so I'm only aiming to isolate the structure from unbounded parallelism here. SGI-PV: 972759 SGI-Modid: xfs-linux-melb:xfs-kern:30371a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
2008-02-05 08:13:32 +07:00
xfs_trans_ail_destroy(mp);
xfs_sysfs_del(&mp->m_log->l_kobj);
xlog_dealloc_log(mp->m_log);
}
void
xfs_log_item_init(
struct xfs_mount *mp,
struct xfs_log_item *item,
int type,
const struct xfs_item_ops *ops)
{
item->li_mountp = mp;
item->li_ailp = mp->m_ail;
item->li_type = type;
item->li_ops = ops;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
item->li_lv = NULL;
INIT_LIST_HEAD(&item->li_ail);
INIT_LIST_HEAD(&item->li_cil);
INIT_LIST_HEAD(&item->li_bio_list);
INIT_LIST_HEAD(&item->li_trans);
}
/*
* Wake up processes waiting for log space after we have moved the log tail.
*/
void
xfs_log_space_wake(
struct xfs_mount *mp)
{
struct xlog *log = mp->m_log;
int free_bytes;
if (XLOG_FORCED_SHUTDOWN(log))
return;
if (!list_empty_careful(&log->l_write_head.waiters)) {
ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
spin_lock(&log->l_write_head.lock);
free_bytes = xlog_space_left(log, &log->l_write_head.grant);
xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
spin_unlock(&log->l_write_head.lock);
}
if (!list_empty_careful(&log->l_reserve_head.waiters)) {
ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
spin_lock(&log->l_reserve_head.lock);
free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
spin_unlock(&log->l_reserve_head.lock);
}
}
/*
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-15 05:17:49 +07:00
* Determine if we have a transaction that has gone to disk that needs to be
* covered. To begin the transition to the idle state firstly the log needs to
* be idle. That means the CIL, the AIL and the iclogs needs to be empty before
* we start attempting to cover the log.
xfs: ensure that sync updates the log tail correctly Updates to the VFS layer removed an extra ->sync_fs call into the filesystem during the sync process (from the quota code). Unfortunately the sync code was unknowingly relying on this call to make sure metadata buffers were flushed via a xfs_buftarg_flush() call to move the tail of the log forward in memory before the final transactions of the sync process were issued. As a result, the old code would write a very recent log tail value to the log by the end of the sync process, and so a subsequent crash would leave nothing for log recovery to do. Hence in qa test 182, log recovery only replayed a small handle for inode fsync transactions in this case. However, with the removal of the extra ->sync_fs call, the log tail was now not moved forward with the inode fsync transactions near the end of the sync procese the first (and only) buftarg flush occurred after these transactions went to disk. The result is that log recovery now sees a large number of transactions for metadata that is already on disk. This usually isn't a problem, but when the transactions include inode chunk allocation, the inode create transactions and all subsequent changes are replayed as we cannt rely on what is on disk is valid. As a result, if the inode was written and contains unlogged changes, the unlogged changes are lost, thereby violating sync semantics. The fix is to always issue a transaction after the buftarg flush occurs is the log iѕ not idle or covered. This results in a dummy transaction being written that contains the up-to-date log tail value, which will be very recent. Indeed, it will be at least as recent as the old code would have left on disk, so log recovery will behave exactly as it used to in this situation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-04-13 12:06:44 +07:00
*
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-15 05:17:49 +07:00
* Only if we are then in a state where covering is needed, the caller is
* informed that dummy transactions are required to move the log into the idle
* state.
*
* If there are any items in the AIl or CIL, then we do not want to attempt to
* cover the log as we may be in a situation where there isn't log space
* available to run a dummy transaction and this can lead to deadlocks when the
* tail of the log is pinned by an item that is modified in the CIL. Hence
* there's no point in running a dummy transaction at this point because we
* can't start trying to idle the log until both the CIL and AIL are empty.
*/
static int
xfs_log_need_covered(xfs_mount_t *mp)
{
struct xlog *log = mp->m_log;
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-15 05:17:49 +07:00
int needed = 0;
if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
return 0;
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-15 05:17:49 +07:00
if (!xlog_cil_empty(log))
return 0;
spin_lock(&log->l_icloglock);
xfs: ensure that sync updates the log tail correctly Updates to the VFS layer removed an extra ->sync_fs call into the filesystem during the sync process (from the quota code). Unfortunately the sync code was unknowingly relying on this call to make sure metadata buffers were flushed via a xfs_buftarg_flush() call to move the tail of the log forward in memory before the final transactions of the sync process were issued. As a result, the old code would write a very recent log tail value to the log by the end of the sync process, and so a subsequent crash would leave nothing for log recovery to do. Hence in qa test 182, log recovery only replayed a small handle for inode fsync transactions in this case. However, with the removal of the extra ->sync_fs call, the log tail was now not moved forward with the inode fsync transactions near the end of the sync procese the first (and only) buftarg flush occurred after these transactions went to disk. The result is that log recovery now sees a large number of transactions for metadata that is already on disk. This usually isn't a problem, but when the transactions include inode chunk allocation, the inode create transactions and all subsequent changes are replayed as we cannt rely on what is on disk is valid. As a result, if the inode was written and contains unlogged changes, the unlogged changes are lost, thereby violating sync semantics. The fix is to always issue a transaction after the buftarg flush occurs is the log iѕ not idle or covered. This results in a dummy transaction being written that contains the up-to-date log tail value, which will be very recent. Indeed, it will be at least as recent as the old code would have left on disk, so log recovery will behave exactly as it used to in this situation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-04-13 12:06:44 +07:00
switch (log->l_covered_state) {
case XLOG_STATE_COVER_DONE:
case XLOG_STATE_COVER_DONE2:
case XLOG_STATE_COVER_IDLE:
break;
case XLOG_STATE_COVER_NEED:
case XLOG_STATE_COVER_NEED2:
xfs: prevent deadlock trying to cover an active log Recent analysis of a deadlocked XFS filesystem from a kernel crash dump indicated that the filesystem was stuck waiting for log space. The short story of the hang on the RHEL6 kernel is this: - the tail of the log is pinned by an inode - the inode has been pushed by the xfsaild - the inode has been flushed to it's backing buffer and is currently flush locked and hence waiting for backing buffer IO to complete and remove it from the AIL - the backing buffer is marked for write - it is on the delayed write queue - the inode buffer has been modified directly and logged recently due to unlinked inode list modification - the backing buffer is pinned in memory as it is in the active CIL context. - the xfsbufd won't start buffer writeback because it is pinned - xfssyncd won't force the log because it sees the log as needing to be covered and hence wants to issue a dummy transaction to move the log covering state machine along. Hence there is no trigger to force the CIL to the log and hence unpin the inode buffer and therefore complete the inode IO, remove it from the AIL and hence move the tail of the log along, allowing transactions to start again. Mainline kernels also have the same deadlock, though the signature is slightly different - the inode buffer never reaches the delayed write lists because xfs_buf_item_push() sees that it is pinned and hence never adds it to the delayed write list that the xfsaild flushes. There are two possible solutions here. The first is to simply force the log before trying to cover the log and so ensure that the CIL is emptied before we try to reserve space for the dummy transaction in the xfs_log_worker(). While this might work most of the time, it is still racy and is no guarantee that we don't get stuck in xfs_trans_reserve waiting for log space to come free. Hence it's not the best way to solve the problem. The second solution is to modify xfs_log_need_covered() to be aware of the CIL. We only should be attempting to cover the log if there is no current activity in the log - covering the log is the process of ensuring that the head and tail in the log on disk are identical (i.e. the log is clean and at idle). Hence, by definition, if there are items in the CIL then the log is not at idle and so we don't need to attempt to cover it. When we don't need to cover the log because it is active or idle, we issue a log force from xfs_log_worker() - if the log is idle, then this does nothing. However, if the log is active due to there being items in the CIL, it will force the items in the CIL to the log and unpin them. In the case of the above deadlock scenario, instead of xfs_log_worker() getting stuck in xfs_trans_reserve() attempting to cover the log, it will instead force the log, thereby unpinning the inode buffer, allowing IO to be issued and complete and hence removing the inode that was pinning the tail of the log from the AIL. At that point, everything will start moving along again. i.e. the xfs_log_worker turns back into a watchdog that can alleviate deadlocks based around pinned items that prevent the tail of the log from being moved... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Eric Sandeen <sandeen@redhat.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-15 05:17:49 +07:00
if (xfs_ail_min_lsn(log->l_ailp))
break;
if (!xlog_iclogs_empty(log))
break;
needed = 1;
if (log->l_covered_state == XLOG_STATE_COVER_NEED)
log->l_covered_state = XLOG_STATE_COVER_DONE;
else
log->l_covered_state = XLOG_STATE_COVER_DONE2;
break;
xfs: ensure that sync updates the log tail correctly Updates to the VFS layer removed an extra ->sync_fs call into the filesystem during the sync process (from the quota code). Unfortunately the sync code was unknowingly relying on this call to make sure metadata buffers were flushed via a xfs_buftarg_flush() call to move the tail of the log forward in memory before the final transactions of the sync process were issued. As a result, the old code would write a very recent log tail value to the log by the end of the sync process, and so a subsequent crash would leave nothing for log recovery to do. Hence in qa test 182, log recovery only replayed a small handle for inode fsync transactions in this case. However, with the removal of the extra ->sync_fs call, the log tail was now not moved forward with the inode fsync transactions near the end of the sync procese the first (and only) buftarg flush occurred after these transactions went to disk. The result is that log recovery now sees a large number of transactions for metadata that is already on disk. This usually isn't a problem, but when the transactions include inode chunk allocation, the inode create transactions and all subsequent changes are replayed as we cannt rely on what is on disk is valid. As a result, if the inode was written and contains unlogged changes, the unlogged changes are lost, thereby violating sync semantics. The fix is to always issue a transaction after the buftarg flush occurs is the log iѕ not idle or covered. This results in a dummy transaction being written that contains the up-to-date log tail value, which will be very recent. Indeed, it will be at least as recent as the old code would have left on disk, so log recovery will behave exactly as it used to in this situation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-04-13 12:06:44 +07:00
default:
needed = 1;
xfs: ensure that sync updates the log tail correctly Updates to the VFS layer removed an extra ->sync_fs call into the filesystem during the sync process (from the quota code). Unfortunately the sync code was unknowingly relying on this call to make sure metadata buffers were flushed via a xfs_buftarg_flush() call to move the tail of the log forward in memory before the final transactions of the sync process were issued. As a result, the old code would write a very recent log tail value to the log by the end of the sync process, and so a subsequent crash would leave nothing for log recovery to do. Hence in qa test 182, log recovery only replayed a small handle for inode fsync transactions in this case. However, with the removal of the extra ->sync_fs call, the log tail was now not moved forward with the inode fsync transactions near the end of the sync procese the first (and only) buftarg flush occurred after these transactions went to disk. The result is that log recovery now sees a large number of transactions for metadata that is already on disk. This usually isn't a problem, but when the transactions include inode chunk allocation, the inode create transactions and all subsequent changes are replayed as we cannt rely on what is on disk is valid. As a result, if the inode was written and contains unlogged changes, the unlogged changes are lost, thereby violating sync semantics. The fix is to always issue a transaction after the buftarg flush occurs is the log iѕ not idle or covered. This results in a dummy transaction being written that contains the up-to-date log tail value, which will be very recent. Indeed, it will be at least as recent as the old code would have left on disk, so log recovery will behave exactly as it used to in this situation. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-04-13 12:06:44 +07:00
break;
}
spin_unlock(&log->l_icloglock);
return needed;
}
/*
* We may be holding the log iclog lock upon entering this routine.
*/
xfs_lsn_t
xlog_assign_tail_lsn_locked(
struct xfs_mount *mp)
{
struct xlog *log = mp->m_log;
struct xfs_log_item *lip;
xfs_lsn_t tail_lsn;
assert_spin_locked(&mp->m_ail->ail_lock);
/*
* To make sure we always have a valid LSN for the log tail we keep
* track of the last LSN which was committed in log->l_last_sync_lsn,
* and use that when the AIL was empty.
*/
lip = xfs_ail_min(mp->m_ail);
if (lip)
tail_lsn = lip->li_lsn;
else
tail_lsn = atomic64_read(&log->l_last_sync_lsn);
trace_xfs_log_assign_tail_lsn(log, tail_lsn);
atomic64_set(&log->l_tail_lsn, tail_lsn);
return tail_lsn;
}
xfs_lsn_t
xlog_assign_tail_lsn(
struct xfs_mount *mp)
{
xfs_lsn_t tail_lsn;
spin_lock(&mp->m_ail->ail_lock);
tail_lsn = xlog_assign_tail_lsn_locked(mp);
spin_unlock(&mp->m_ail->ail_lock);
return tail_lsn;
}
/*
* Return the space in the log between the tail and the head. The head
* is passed in the cycle/bytes formal parms. In the special case where
* the reserve head has wrapped passed the tail, this calculation is no
* longer valid. In this case, just return 0 which means there is no space
* in the log. This works for all places where this function is called
* with the reserve head. Of course, if the write head were to ever
* wrap the tail, we should blow up. Rather than catch this case here,
* we depend on other ASSERTions in other parts of the code. XXXmiken
*
* This code also handles the case where the reservation head is behind
* the tail. The details of this case are described below, but the end
* result is that we return the size of the log as the amount of space left.
*/
STATIC int
xlog_space_left(
struct xlog *log,
atomic64_t *head)
{
int free_bytes;
int tail_bytes;
int tail_cycle;
int head_cycle;
int head_bytes;
xlog_crack_grant_head(head, &head_cycle, &head_bytes);
xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
tail_bytes = BBTOB(tail_bytes);
if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
free_bytes = log->l_logsize - (head_bytes - tail_bytes);
else if (tail_cycle + 1 < head_cycle)
return 0;
else if (tail_cycle < head_cycle) {
ASSERT(tail_cycle == (head_cycle - 1));
free_bytes = tail_bytes - head_bytes;
} else {
/*
* The reservation head is behind the tail.
* In this case we just want to return the size of the
* log as the amount of space left.
*/
xfs_alert(log->l_mp, "xlog_space_left: head behind tail");
xfs_alert(log->l_mp,
" tail_cycle = %d, tail_bytes = %d",
tail_cycle, tail_bytes);
xfs_alert(log->l_mp,
" GH cycle = %d, GH bytes = %d",
head_cycle, head_bytes);
ASSERT(0);
free_bytes = log->l_logsize;
}
return free_bytes;
}
static void
xlog_ioend_work(
struct work_struct *work)
{
struct xlog_in_core *iclog =
container_of(work, struct xlog_in_core, ic_end_io_work);
struct xlog *log = iclog->ic_log;
bool aborted = false;
int error;
error = blk_status_to_errno(iclog->ic_bio.bi_status);
#ifdef DEBUG
/* treat writes with injected CRC errors as failed */
if (iclog->ic_fail_crc)
error = -EIO;
#endif
/*
* Race to shutdown the filesystem if we see an error.
*/
if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) {
xfs_alert(log->l_mp, "log I/O error %d", error);
xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR);
/*
* This flag will be propagated to the trans-committed
* callback routines to let them know that the log-commit
* didn't succeed.
*/
aborted = true;
} else if (iclog->ic_state == XLOG_STATE_IOERROR) {
aborted = true;
}
xlog_state_done_syncing(iclog, aborted);
bio_uninit(&iclog->ic_bio);
xfs: unmount does not wait for shutdown during unmount And interesting situation can occur if a log IO error occurs during the unmount of a filesystem. The cases reported have the same signature - the update of the superblock counters fails due to a log write IO error: XFS (dm-16): xfs_do_force_shutdown(0x2) called from line 1170 of file fs/xfs/xfs_log.c. Return address = 0xffffffffa08a44a1 XFS (dm-16): Log I/O Error Detected. Shutting down filesystem XFS (dm-16): Unable to update superblock counters. Freespace may not be correct on next mount. XFS (dm-16): xfs_log_force: error 5 returned. XFS (¿-¿¿¿): Please umount the filesystem and rectify the problem(s) It can be seen that the last line of output contains a corrupt device name - this is because the log and xfs_mount structures have already been freed by the time this message is printed. A kernel oops closely follows. The issue is that the shutdown is occurring in a separate IO completion thread to the unmount. Once the shutdown processing has started and all the iclogs are marked with XLOG_STATE_IOERROR, the log shutdown code wakes anyone waiting on a log force so they can process the shutdown error. This wakes up the unmount code that is doing a synchronous transaction to update the superblock counters. The unmount path now sees all the iclogs are marked with XLOG_STATE_IOERROR and so never waits on them again, knowing that if it does, there will not be a wakeup trigger for it and we will hang the unmount if we do. Hence the unmount runs through all the remaining code and frees all the filesystem structures while the xlog_iodone() is still processing the shutdown. When the log shutdown processing completes, xfs_do_force_shutdown() emits the "Please umount the filesystem and rectify the problem(s)" message, and xlog_iodone() then aborts all the objects attached to the iclog. An iclog that has already been freed.... The real issue here is that there is no serialisation point between the log IO and the unmount. We have serialisations points for log writes, log forces, reservations, etc, but we don't actually have any code that wakes for log IO to fully complete. We do that for all other types of object, so why not iclogbufs? Well, it turns out that we can easily do this. We've got xfs_buf handles, and that's what everyone else uses for IO serialisation. i.e. bp->b_sema. So, lets hold iclogbufs locked over IO, and only release the lock in xlog_iodone() when we are finished with the buffer. That way before we tear down the iclog, we can lock and unlock the buffer to ensure IO completion has finished completely before we tear it down. Signed-off-by: Dave Chinner <dchinner@redhat.com> Tested-by: Mike Snitzer <snitzer@redhat.com> Tested-by: Bob Mastors <bob.mastors@solidfire.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-04-17 05:15:26 +07:00
/*
* Drop the lock to signal that we are done. Nothing references the
* iclog after this, so an unmount waiting on this lock can now tear it
* down safely. As such, it is unsafe to reference the iclog after the
* unlock as we could race with it being freed.
*/
up(&iclog->ic_sema);
}
/*
* Return size of each in-core log record buffer.
*
* All machines get 8 x 32kB buffers by default, unless tuned otherwise.
*
* If the filesystem blocksize is too large, we may need to choose a
* larger size since the directory code currently logs entire blocks.
*/
STATIC void
xlog_get_iclog_buffer_size(
struct xfs_mount *mp,
struct xlog *log)
{
if (mp->m_logbufs <= 0)
mp->m_logbufs = XLOG_MAX_ICLOGS;
if (mp->m_logbsize <= 0)
mp->m_logbsize = XLOG_BIG_RECORD_BSIZE;
log->l_iclog_bufs = mp->m_logbufs;
log->l_iclog_size = mp->m_logbsize;
/*
* # headers = size / 32k - one header holds cycles from 32k of data.
*/
log->l_iclog_heads =
DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE);
log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT;
}
void
xfs_log_work_queue(
struct xfs_mount *mp)
{
queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work,
msecs_to_jiffies(xfs_syncd_centisecs * 10));
}
/*
* Every sync period we need to unpin all items in the AIL and push them to
* disk. If there is nothing dirty, then we might need to cover the log to
* indicate that the filesystem is idle.
*/
static void
xfs_log_worker(
struct work_struct *work)
{
struct xlog *log = container_of(to_delayed_work(work),
struct xlog, l_work);
struct xfs_mount *mp = log->l_mp;
/* dgc: errors ignored - not fatal and nowhere to report them */
if (xfs_log_need_covered(mp)) {
/*
* Dump a transaction into the log that contains no real change.
* This is needed to stamp the current tail LSN into the log
* during the covering operation.
*
* We cannot use an inode here for this - that will push dirty
* state back up into the VFS and then periodic inode flushing
* will prevent log covering from making progress. Hence we
* synchronously log the superblock instead to ensure the
* superblock is immediately unpinned and can be written back.
*/
xfs_sync_sb(mp, true);
} else
xfs_log_force(mp, 0);
/* start pushing all the metadata that is currently dirty */
xfs_ail_push_all(mp->m_ail);
/* queue us up again */
xfs_log_work_queue(mp);
}
/*
* This routine initializes some of the log structure for a given mount point.
* Its primary purpose is to fill in enough, so recovery can occur. However,
* some other stuff may be filled in too.
*/
STATIC struct xlog *
xlog_alloc_log(
struct xfs_mount *mp,
struct xfs_buftarg *log_target,
xfs_daddr_t blk_offset,
int num_bblks)
{
struct xlog *log;
xlog_rec_header_t *head;
xlog_in_core_t **iclogp;
xlog_in_core_t *iclog, *prev_iclog=NULL;
int i;
int error = -ENOMEM;
uint log2_size = 0;
log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
if (!log) {
xfs_warn(mp, "Log allocation failed: No memory!");
goto out;
}
log->l_mp = mp;
log->l_targ = log_target;
log->l_logsize = BBTOB(num_bblks);
log->l_logBBstart = blk_offset;
log->l_logBBsize = num_bblks;
log->l_covered_state = XLOG_STATE_COVER_IDLE;
log->l_flags |= XLOG_ACTIVE_RECOVERY;
INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);
log->l_prev_block = -1;
/* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
xlog_grant_head_init(&log->l_reserve_head);
xlog_grant_head_init(&log->l_write_head);
error = -EFSCORRUPTED;
if (xfs_sb_version_hassector(&mp->m_sb)) {
log2_size = mp->m_sb.sb_logsectlog;
if (log2_size < BBSHIFT) {
xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
log2_size, BBSHIFT);
goto out_free_log;
}
log2_size -= BBSHIFT;
if (log2_size > mp->m_sectbb_log) {
xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
log2_size, mp->m_sectbb_log);
goto out_free_log;
}
/* for larger sector sizes, must have v2 or external log */
if (log2_size && log->l_logBBstart > 0 &&
!xfs_sb_version_haslogv2(&mp->m_sb)) {
xfs_warn(mp,
"log sector size (0x%x) invalid for configuration.",
log2_size);
goto out_free_log;
}
}
log->l_sectBBsize = 1 << log2_size;
xlog_get_iclog_buffer_size(mp, log);
spin_lock_init(&log->l_icloglock);
init_waitqueue_head(&log->l_flush_wait);
iclogp = &log->l_iclog;
/*
* The amount of memory to allocate for the iclog structure is
* rather funky due to the way the structure is defined. It is
* done this way so that we can use different sizes for machines
* with different amounts of memory. See the definition of
* xlog_in_core_t in xfs_log_priv.h for details.
*/
ASSERT(log->l_iclog_size >= 4096);
for (i = 0; i < log->l_iclog_bufs; i++) {
int align_mask = xfs_buftarg_dma_alignment(mp->m_logdev_targp);
size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) *
sizeof(struct bio_vec);
iclog = kmem_zalloc(sizeof(*iclog) + bvec_size, KM_MAYFAIL);
if (!iclog)
goto out_free_iclog;
*iclogp = iclog;
iclog->ic_prev = prev_iclog;
prev_iclog = iclog;
iclog->ic_data = kmem_alloc_io(log->l_iclog_size, align_mask,
KM_MAYFAIL | KM_ZERO);
if (!iclog->ic_data)
goto out_free_iclog;
#ifdef DEBUG
log->l_iclog_bak[i] = &iclog->ic_header;
#endif
head = &iclog->ic_header;
memset(head, 0, sizeof(xlog_rec_header_t));
head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
head->h_version = cpu_to_be32(
xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
head->h_size = cpu_to_be32(log->l_iclog_size);
/* new fields */
head->h_fmt = cpu_to_be32(XLOG_FMT);
memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize;
iclog->ic_state = XLOG_STATE_ACTIVE;
iclog->ic_log = log;
atomic_set(&iclog->ic_refcnt, 0);
spin_lock_init(&iclog->ic_callback_lock);
INIT_LIST_HEAD(&iclog->ic_callbacks);
iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
init_waitqueue_head(&iclog->ic_force_wait);
init_waitqueue_head(&iclog->ic_write_wait);
INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work);
sema_init(&iclog->ic_sema, 1);
iclogp = &iclog->ic_next;
}
*iclogp = log->l_iclog; /* complete ring */
log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s",
WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_HIGHPRI, 0,
mp->m_super->s_id);
if (!log->l_ioend_workqueue)
goto out_free_iclog;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
error = xlog_cil_init(log);
if (error)
goto out_destroy_workqueue;
return log;
out_destroy_workqueue:
destroy_workqueue(log->l_ioend_workqueue);
out_free_iclog:
for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
prev_iclog = iclog->ic_next;
kmem_free(iclog->ic_data);
kmem_free(iclog);
if (prev_iclog == log->l_iclog)
break;
}
out_free_log:
kmem_free(log);
out:
return ERR_PTR(error);
} /* xlog_alloc_log */
/*
* Write out the commit record of a transaction associated with the given
* ticket. Return the lsn of the commit record.
*/
STATIC int
xlog_commit_record(
struct xlog *log,
struct xlog_ticket *ticket,
struct xlog_in_core **iclog,
xfs_lsn_t *commitlsnp)
{
struct xfs_mount *mp = log->l_mp;
int error;
struct xfs_log_iovec reg = {
.i_addr = NULL,
.i_len = 0,
.i_type = XLOG_REG_TYPE_COMMIT,
};
struct xfs_log_vec vec = {
.lv_niovecs = 1,
.lv_iovecp = &reg,
};
ASSERT_ALWAYS(iclog);
error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
XLOG_COMMIT_TRANS);
if (error)
xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
return error;
}
/*
* Push on the buffer cache code if we ever use more than 75% of the on-disk
* log space. This code pushes on the lsn which would supposedly free up
* the 25% which we want to leave free. We may need to adopt a policy which
* pushes on an lsn which is further along in the log once we reach the high
* water mark. In this manner, we would be creating a low water mark.
*/
STATIC void
xlog_grant_push_ail(
struct xlog *log,
int need_bytes)
{
xfs_lsn_t threshold_lsn = 0;
xfs_lsn_t last_sync_lsn;
int free_blocks;
int free_bytes;
int threshold_block;
int threshold_cycle;
int free_threshold;
ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
free_blocks = BTOBBT(free_bytes);
/*
* Set the threshold for the minimum number of free blocks in the
* log to the maximum of what the caller needs, one quarter of the
* log, and 256 blocks.
*/
free_threshold = BTOBB(need_bytes);
free_threshold = max(free_threshold, (log->l_logBBsize >> 2));
free_threshold = max(free_threshold, 256);
if (free_blocks >= free_threshold)
return;
xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
&threshold_block);
threshold_block += free_threshold;
if (threshold_block >= log->l_logBBsize) {
threshold_block -= log->l_logBBsize;
threshold_cycle += 1;
}
threshold_lsn = xlog_assign_lsn(threshold_cycle,
threshold_block);
/*
* Don't pass in an lsn greater than the lsn of the last
* log record known to be on disk. Use a snapshot of the last sync lsn
* so that it doesn't change between the compare and the set.
*/
last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
threshold_lsn = last_sync_lsn;
/*
* Get the transaction layer to kick the dirty buffers out to
* disk asynchronously. No point in trying to do this if
* the filesystem is shutting down.
*/
if (!XLOG_FORCED_SHUTDOWN(log))
xfs_ail_push(log->l_ailp, threshold_lsn);
}
2012-11-12 18:54:24 +07:00
/*
* Stamp cycle number in every block
*/
STATIC void
xlog_pack_data(
struct xlog *log,
struct xlog_in_core *iclog,
int roundoff)
{
int i, j, k;
int size = iclog->ic_offset + roundoff;
__be32 cycle_lsn;
char *dp;
2012-11-12 18:54:24 +07:00
cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
dp = iclog->ic_datap;
for (i = 0; i < BTOBB(size); i++) {
if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))
break;
iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
*(__be32 *)dp = cycle_lsn;
dp += BBSIZE;
}
if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
xlog_in_core_2_t *xhdr = iclog->ic_data;
for ( ; i < BTOBB(size); i++) {
j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
*(__be32 *)dp = cycle_lsn;
dp += BBSIZE;
}
for (i = 1; i < log->l_iclog_heads; i++)
xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
}
}
/*
* Calculate the checksum for a log buffer.
*
* This is a little more complicated than it should be because the various
* headers and the actual data are non-contiguous.
*/
__le32
2012-11-12 18:54:24 +07:00
xlog_cksum(
struct xlog *log,
struct xlog_rec_header *rhead,
char *dp,
int size)
{
uint32_t crc;
2012-11-12 18:54:24 +07:00
/* first generate the crc for the record header ... */
crc = xfs_start_cksum_update((char *)rhead,
2012-11-12 18:54:24 +07:00
sizeof(struct xlog_rec_header),
offsetof(struct xlog_rec_header, h_crc));
/* ... then for additional cycle data for v2 logs ... */
if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;
int i;
xfs: checksum log record ext headers based on record size The first 4 bytes of every basic block in the physical log is stamped with the current lsn. To support this mechanism, the log record header (first block of each new log record) contains space for the original first byte of each log record block before it is replaced with the lsn. The log record header has space for 32k worth of blocks. The version 2 log adds new extended record headers for each additional 32k worth of blocks beyond what is supported by the record header. The log record checksum incorporates the log record header, the extended headers and the record payload. xlog_cksum() checksums the extended headers based on log->l_iclog_heads, which specifies the number of extended headers in a log record based on the log buffer size mount option. The log buffer size is variable, however, and thus means the checksum can be calculated differently based on how a filesystem is mounted. This is problematic if a filesystem crashes and recovery occurs on a subsequent mount using a different log buffer size. For example, crash an active filesystem that is mounted with the default (32k) logbsize, attempt remount/recovery using '-o logbsize=64k' and the mount fails on or warns about log checksum failures. To avoid this problem, update xlog_cksum() to calculate the checksum based on the size of the log buffer according to the log record. The size is already included in the h_size field of the log record header and thus is available at log recovery time. Extended log record headers are also only written when the log record is large enough to require them. This makes checksum calculation of log records consistent with the extended record header mechanism as well as how on-disk records are checksummed with various log buffer size mount options. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 06:59:50 +07:00
int xheads;
2012-11-12 18:54:24 +07:00
xfs: checksum log record ext headers based on record size The first 4 bytes of every basic block in the physical log is stamped with the current lsn. To support this mechanism, the log record header (first block of each new log record) contains space for the original first byte of each log record block before it is replaced with the lsn. The log record header has space for 32k worth of blocks. The version 2 log adds new extended record headers for each additional 32k worth of blocks beyond what is supported by the record header. The log record checksum incorporates the log record header, the extended headers and the record payload. xlog_cksum() checksums the extended headers based on log->l_iclog_heads, which specifies the number of extended headers in a log record based on the log buffer size mount option. The log buffer size is variable, however, and thus means the checksum can be calculated differently based on how a filesystem is mounted. This is problematic if a filesystem crashes and recovery occurs on a subsequent mount using a different log buffer size. For example, crash an active filesystem that is mounted with the default (32k) logbsize, attempt remount/recovery using '-o logbsize=64k' and the mount fails on or warns about log checksum failures. To avoid this problem, update xlog_cksum() to calculate the checksum based on the size of the log buffer according to the log record. The size is already included in the h_size field of the log record header and thus is available at log recovery time. Extended log record headers are also only written when the log record is large enough to require them. This makes checksum calculation of log records consistent with the extended record header mechanism as well as how on-disk records are checksummed with various log buffer size mount options. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 06:59:50 +07:00
xheads = size / XLOG_HEADER_CYCLE_SIZE;
if (size % XLOG_HEADER_CYCLE_SIZE)
xheads++;
2012-11-12 18:54:24 +07:00
xfs: checksum log record ext headers based on record size The first 4 bytes of every basic block in the physical log is stamped with the current lsn. To support this mechanism, the log record header (first block of each new log record) contains space for the original first byte of each log record block before it is replaced with the lsn. The log record header has space for 32k worth of blocks. The version 2 log adds new extended record headers for each additional 32k worth of blocks beyond what is supported by the record header. The log record checksum incorporates the log record header, the extended headers and the record payload. xlog_cksum() checksums the extended headers based on log->l_iclog_heads, which specifies the number of extended headers in a log record based on the log buffer size mount option. The log buffer size is variable, however, and thus means the checksum can be calculated differently based on how a filesystem is mounted. This is problematic if a filesystem crashes and recovery occurs on a subsequent mount using a different log buffer size. For example, crash an active filesystem that is mounted with the default (32k) logbsize, attempt remount/recovery using '-o logbsize=64k' and the mount fails on or warns about log checksum failures. To avoid this problem, update xlog_cksum() to calculate the checksum based on the size of the log buffer according to the log record. The size is already included in the h_size field of the log record header and thus is available at log recovery time. Extended log record headers are also only written when the log record is large enough to require them. This makes checksum calculation of log records consistent with the extended record header mechanism as well as how on-disk records are checksummed with various log buffer size mount options. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 06:59:50 +07:00
for (i = 1; i < xheads; i++) {
2012-11-12 18:54:24 +07:00
crc = crc32c(crc, &xhdr[i].hic_xheader,
sizeof(struct xlog_rec_ext_header));
}
}
/* ... and finally for the payload */
crc = crc32c(crc, dp, size);
return xfs_end_cksum(crc);
}
static void
xlog_bio_end_io(
struct bio *bio)
{
struct xlog_in_core *iclog = bio->bi_private;
queue_work(iclog->ic_log->l_ioend_workqueue,
&iclog->ic_end_io_work);
}
static void
xlog_map_iclog_data(
struct bio *bio,
void *data,
size_t count)
{
do {
struct page *page = kmem_to_page(data);
unsigned int off = offset_in_page(data);
size_t len = min_t(size_t, count, PAGE_SIZE - off);
WARN_ON_ONCE(bio_add_page(bio, page, len, off) != len);
data += len;
count -= len;
} while (count);
}
STATIC void
xlog_write_iclog(
struct xlog *log,
struct xlog_in_core *iclog,
uint64_t bno,
unsigned int count,
bool need_flush)
{
ASSERT(bno < log->l_logBBsize);
/*
* We lock the iclogbufs here so that we can serialise against I/O
* completion during unmount. We might be processing a shutdown
* triggered during unmount, and that can occur asynchronously to the
* unmount thread, and hence we need to ensure that completes before
* tearing down the iclogbufs. Hence we need to hold the buffer lock
* across the log IO to archieve that.
*/
down(&iclog->ic_sema);
if (unlikely(iclog->ic_state == XLOG_STATE_IOERROR)) {
/*
* It would seem logical to return EIO here, but we rely on
* the log state machine to propagate I/O errors instead of
* doing it here. We kick of the state machine and unlock
* the buffer manually, the code needs to be kept in sync
* with the I/O completion path.
*/
xlog_state_done_syncing(iclog, true);
up(&iclog->ic_sema);
return;
}
bio_init(&iclog->ic_bio, iclog->ic_bvec, howmany(count, PAGE_SIZE));
bio_set_dev(&iclog->ic_bio, log->l_targ->bt_bdev);
iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno;
iclog->ic_bio.bi_end_io = xlog_bio_end_io;
iclog->ic_bio.bi_private = iclog;
iclog->ic_bio.bi_opf = REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_FUA;
if (need_flush)
iclog->ic_bio.bi_opf |= REQ_PREFLUSH;
xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count);
if (is_vmalloc_addr(iclog->ic_data))
flush_kernel_vmap_range(iclog->ic_data, count);
/*
* If this log buffer would straddle the end of the log we will have
* to split it up into two bios, so that we can continue at the start.
*/
if (bno + BTOBB(count) > log->l_logBBsize) {
struct bio *split;
split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno,
GFP_NOIO, &fs_bio_set);
bio_chain(split, &iclog->ic_bio);
submit_bio(split);
/* restart at logical offset zero for the remainder */
iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart;
}
submit_bio(&iclog->ic_bio);
}
/*
* We need to bump cycle number for the part of the iclog that is
* written to the start of the log. Watch out for the header magic
* number case, though.
*/
static void
xlog_split_iclog(
struct xlog *log,
void *data,
uint64_t bno,
unsigned int count)
{
unsigned int split_offset = BBTOB(log->l_logBBsize - bno);
unsigned int i;
for (i = split_offset; i < count; i += BBSIZE) {
uint32_t cycle = get_unaligned_be32(data + i);
if (++cycle == XLOG_HEADER_MAGIC_NUM)
cycle++;
put_unaligned_be32(cycle, data + i);
}
}
static int
xlog_calc_iclog_size(
struct xlog *log,
struct xlog_in_core *iclog,
uint32_t *roundoff)
{
uint32_t count_init, count;
bool use_lsunit;
use_lsunit = xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
log->l_mp->m_sb.sb_logsunit > 1;
/* Add for LR header */
count_init = log->l_iclog_hsize + iclog->ic_offset;
/* Round out the log write size */
if (use_lsunit) {
/* we have a v2 stripe unit to use */
count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
} else {
count = BBTOB(BTOBB(count_init));
}
ASSERT(count >= count_init);
*roundoff = count - count_init;
if (use_lsunit)
ASSERT(*roundoff < log->l_mp->m_sb.sb_logsunit);
else
ASSERT(*roundoff < BBTOB(1));
return count;
}
/*
* Flush out the in-core log (iclog) to the on-disk log in an asynchronous
* fashion. Previously, we should have moved the current iclog
* ptr in the log to point to the next available iclog. This allows further
* write to continue while this code syncs out an iclog ready to go.
* Before an in-core log can be written out, the data section must be scanned
* to save away the 1st word of each BBSIZE block into the header. We replace
* it with the current cycle count. Each BBSIZE block is tagged with the
* cycle count because there in an implicit assumption that drives will
* guarantee that entire 512 byte blocks get written at once. In other words,
* we can't have part of a 512 byte block written and part not written. By
* tagging each block, we will know which blocks are valid when recovering
* after an unclean shutdown.
*
* This routine is single threaded on the iclog. No other thread can be in
* this routine with the same iclog. Changing contents of iclog can there-
* fore be done without grabbing the state machine lock. Updating the global
* log will require grabbing the lock though.
*
* The entire log manager uses a logical block numbering scheme. Only
* xlog_write_iclog knows about the fact that the log may not start with
* block zero on a given device.
*/
STATIC void
xlog_sync(
struct xlog *log,
struct xlog_in_core *iclog)
{
unsigned int count; /* byte count of bwrite */
unsigned int roundoff; /* roundoff to BB or stripe */
uint64_t bno;
unsigned int size;
bool need_flush = true, split = false;
ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
count = xlog_calc_iclog_size(log, iclog, &roundoff);
/* move grant heads by roundoff in sync */
xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
/* put cycle number in every block */
xlog_pack_data(log, iclog, roundoff);
/* real byte length */
2012-11-12 18:54:24 +07:00
size = iclog->ic_offset;
if (xfs_sb_version_haslogv2(&log->l_mp->m_sb))
2012-11-12 18:54:24 +07:00
size += roundoff;
iclog->ic_header.h_len = cpu_to_be32(size);
XFS_STATS_INC(log->l_mp, xs_log_writes);
XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count));
bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn));
/* Do we need to split this write into 2 parts? */
if (bno + BTOBB(count) > log->l_logBBsize) {
xlog_split_iclog(log, &iclog->ic_header, bno, count);
split = true;
}
2012-11-12 18:54:24 +07:00
/* calculcate the checksum */
iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
iclog->ic_datap, size);
/*
* Intentionally corrupt the log record CRC based on the error injection
* frequency, if defined. This facilitates testing log recovery in the
* event of torn writes. Hence, set the IOABORT state to abort the log
* write on I/O completion and shutdown the fs. The subsequent mount
* detects the bad CRC and attempts to recover.
*/
#ifdef DEBUG
if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) {
iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA);
iclog->ic_fail_crc = true;
xfs_warn(log->l_mp,
"Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.",
be64_to_cpu(iclog->ic_header.h_lsn));
}
#endif
2012-11-12 18:54:24 +07:00
/*
* Flush the data device before flushing the log to make sure all meta
* data written back from the AIL actually made it to disk before
* stamping the new log tail LSN into the log buffer. For an external
* log we need to issue the flush explicitly, and unfortunately
* synchronously here; for an internal log we can simply use the block
* layer state machine for preflushes.
*/
if (log->l_targ != log->l_mp->m_ddev_targp || split) {
xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
need_flush = false;
}
xlog_verify_iclog(log, iclog, count);
xlog_write_iclog(log, iclog, bno, count, need_flush);
}
/*
* Deallocate a log structure
*/
STATIC void
xlog_dealloc_log(
struct xlog *log)
{
xlog_in_core_t *iclog, *next_iclog;
int i;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
xlog_cil_destroy(log);
/*
xfs: unmount does not wait for shutdown during unmount And interesting situation can occur if a log IO error occurs during the unmount of a filesystem. The cases reported have the same signature - the update of the superblock counters fails due to a log write IO error: XFS (dm-16): xfs_do_force_shutdown(0x2) called from line 1170 of file fs/xfs/xfs_log.c. Return address = 0xffffffffa08a44a1 XFS (dm-16): Log I/O Error Detected. Shutting down filesystem XFS (dm-16): Unable to update superblock counters. Freespace may not be correct on next mount. XFS (dm-16): xfs_log_force: error 5 returned. XFS (¿-¿¿¿): Please umount the filesystem and rectify the problem(s) It can be seen that the last line of output contains a corrupt device name - this is because the log and xfs_mount structures have already been freed by the time this message is printed. A kernel oops closely follows. The issue is that the shutdown is occurring in a separate IO completion thread to the unmount. Once the shutdown processing has started and all the iclogs are marked with XLOG_STATE_IOERROR, the log shutdown code wakes anyone waiting on a log force so they can process the shutdown error. This wakes up the unmount code that is doing a synchronous transaction to update the superblock counters. The unmount path now sees all the iclogs are marked with XLOG_STATE_IOERROR and so never waits on them again, knowing that if it does, there will not be a wakeup trigger for it and we will hang the unmount if we do. Hence the unmount runs through all the remaining code and frees all the filesystem structures while the xlog_iodone() is still processing the shutdown. When the log shutdown processing completes, xfs_do_force_shutdown() emits the "Please umount the filesystem and rectify the problem(s)" message, and xlog_iodone() then aborts all the objects attached to the iclog. An iclog that has already been freed.... The real issue here is that there is no serialisation point between the log IO and the unmount. We have serialisations points for log writes, log forces, reservations, etc, but we don't actually have any code that wakes for log IO to fully complete. We do that for all other types of object, so why not iclogbufs? Well, it turns out that we can easily do this. We've got xfs_buf handles, and that's what everyone else uses for IO serialisation. i.e. bp->b_sema. So, lets hold iclogbufs locked over IO, and only release the lock in xlog_iodone() when we are finished with the buffer. That way before we tear down the iclog, we can lock and unlock the buffer to ensure IO completion has finished completely before we tear it down. Signed-off-by: Dave Chinner <dchinner@redhat.com> Tested-by: Mike Snitzer <snitzer@redhat.com> Tested-by: Bob Mastors <bob.mastors@solidfire.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-04-17 05:15:26 +07:00
* Cycle all the iclogbuf locks to make sure all log IO completion
* is done before we tear down these buffers.
*/
iclog = log->l_iclog;
for (i = 0; i < log->l_iclog_bufs; i++) {
down(&iclog->ic_sema);
up(&iclog->ic_sema);
xfs: unmount does not wait for shutdown during unmount And interesting situation can occur if a log IO error occurs during the unmount of a filesystem. The cases reported have the same signature - the update of the superblock counters fails due to a log write IO error: XFS (dm-16): xfs_do_force_shutdown(0x2) called from line 1170 of file fs/xfs/xfs_log.c. Return address = 0xffffffffa08a44a1 XFS (dm-16): Log I/O Error Detected. Shutting down filesystem XFS (dm-16): Unable to update superblock counters. Freespace may not be correct on next mount. XFS (dm-16): xfs_log_force: error 5 returned. XFS (¿-¿¿¿): Please umount the filesystem and rectify the problem(s) It can be seen that the last line of output contains a corrupt device name - this is because the log and xfs_mount structures have already been freed by the time this message is printed. A kernel oops closely follows. The issue is that the shutdown is occurring in a separate IO completion thread to the unmount. Once the shutdown processing has started and all the iclogs are marked with XLOG_STATE_IOERROR, the log shutdown code wakes anyone waiting on a log force so they can process the shutdown error. This wakes up the unmount code that is doing a synchronous transaction to update the superblock counters. The unmount path now sees all the iclogs are marked with XLOG_STATE_IOERROR and so never waits on them again, knowing that if it does, there will not be a wakeup trigger for it and we will hang the unmount if we do. Hence the unmount runs through all the remaining code and frees all the filesystem structures while the xlog_iodone() is still processing the shutdown. When the log shutdown processing completes, xfs_do_force_shutdown() emits the "Please umount the filesystem and rectify the problem(s)" message, and xlog_iodone() then aborts all the objects attached to the iclog. An iclog that has already been freed.... The real issue here is that there is no serialisation point between the log IO and the unmount. We have serialisations points for log writes, log forces, reservations, etc, but we don't actually have any code that wakes for log IO to fully complete. We do that for all other types of object, so why not iclogbufs? Well, it turns out that we can easily do this. We've got xfs_buf handles, and that's what everyone else uses for IO serialisation. i.e. bp->b_sema. So, lets hold iclogbufs locked over IO, and only release the lock in xlog_iodone() when we are finished with the buffer. That way before we tear down the iclog, we can lock and unlock the buffer to ensure IO completion has finished completely before we tear it down. Signed-off-by: Dave Chinner <dchinner@redhat.com> Tested-by: Mike Snitzer <snitzer@redhat.com> Tested-by: Bob Mastors <bob.mastors@solidfire.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-04-17 05:15:26 +07:00
iclog = iclog->ic_next;
}
iclog = log->l_iclog;
xfs: unmount does not wait for shutdown during unmount And interesting situation can occur if a log IO error occurs during the unmount of a filesystem. The cases reported have the same signature - the update of the superblock counters fails due to a log write IO error: XFS (dm-16): xfs_do_force_shutdown(0x2) called from line 1170 of file fs/xfs/xfs_log.c. Return address = 0xffffffffa08a44a1 XFS (dm-16): Log I/O Error Detected. Shutting down filesystem XFS (dm-16): Unable to update superblock counters. Freespace may not be correct on next mount. XFS (dm-16): xfs_log_force: error 5 returned. XFS (¿-¿¿¿): Please umount the filesystem and rectify the problem(s) It can be seen that the last line of output contains a corrupt device name - this is because the log and xfs_mount structures have already been freed by the time this message is printed. A kernel oops closely follows. The issue is that the shutdown is occurring in a separate IO completion thread to the unmount. Once the shutdown processing has started and all the iclogs are marked with XLOG_STATE_IOERROR, the log shutdown code wakes anyone waiting on a log force so they can process the shutdown error. This wakes up the unmount code that is doing a synchronous transaction to update the superblock counters. The unmount path now sees all the iclogs are marked with XLOG_STATE_IOERROR and so never waits on them again, knowing that if it does, there will not be a wakeup trigger for it and we will hang the unmount if we do. Hence the unmount runs through all the remaining code and frees all the filesystem structures while the xlog_iodone() is still processing the shutdown. When the log shutdown processing completes, xfs_do_force_shutdown() emits the "Please umount the filesystem and rectify the problem(s)" message, and xlog_iodone() then aborts all the objects attached to the iclog. An iclog that has already been freed.... The real issue here is that there is no serialisation point between the log IO and the unmount. We have serialisations points for log writes, log forces, reservations, etc, but we don't actually have any code that wakes for log IO to fully complete. We do that for all other types of object, so why not iclogbufs? Well, it turns out that we can easily do this. We've got xfs_buf handles, and that's what everyone else uses for IO serialisation. i.e. bp->b_sema. So, lets hold iclogbufs locked over IO, and only release the lock in xlog_iodone() when we are finished with the buffer. That way before we tear down the iclog, we can lock and unlock the buffer to ensure IO completion has finished completely before we tear it down. Signed-off-by: Dave Chinner <dchinner@redhat.com> Tested-by: Mike Snitzer <snitzer@redhat.com> Tested-by: Bob Mastors <bob.mastors@solidfire.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-04-17 05:15:26 +07:00
for (i = 0; i < log->l_iclog_bufs; i++) {
next_iclog = iclog->ic_next;
kmem_free(iclog->ic_data);
kmem_free(iclog);
iclog = next_iclog;
}
log->l_mp->m_log = NULL;
destroy_workqueue(log->l_ioend_workqueue);
kmem_free(log);
} /* xlog_dealloc_log */
/*
* Update counters atomically now that memcpy is done.
*/
static inline void
xlog_state_finish_copy(
struct xlog *log,
struct xlog_in_core *iclog,
int record_cnt,
int copy_bytes)
{
lockdep_assert_held(&log->l_icloglock);
be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
iclog->ic_offset += copy_bytes;
}
/*
* print out info relating to regions written which consume
* the reservation
*/
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
void
xlog_print_tic_res(
struct xfs_mount *mp,
struct xlog_ticket *ticket)
{
uint i;
uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
/* match with XLOG_REG_TYPE_* in xfs_log.h */
#define REG_TYPE_STR(type, str) [XLOG_REG_TYPE_##type] = str
static char *res_type_str[] = {
REG_TYPE_STR(BFORMAT, "bformat"),
REG_TYPE_STR(BCHUNK, "bchunk"),
REG_TYPE_STR(EFI_FORMAT, "efi_format"),
REG_TYPE_STR(EFD_FORMAT, "efd_format"),
REG_TYPE_STR(IFORMAT, "iformat"),
REG_TYPE_STR(ICORE, "icore"),
REG_TYPE_STR(IEXT, "iext"),
REG_TYPE_STR(IBROOT, "ibroot"),
REG_TYPE_STR(ILOCAL, "ilocal"),
REG_TYPE_STR(IATTR_EXT, "iattr_ext"),
REG_TYPE_STR(IATTR_BROOT, "iattr_broot"),
REG_TYPE_STR(IATTR_LOCAL, "iattr_local"),
REG_TYPE_STR(QFORMAT, "qformat"),
REG_TYPE_STR(DQUOT, "dquot"),
REG_TYPE_STR(QUOTAOFF, "quotaoff"),
REG_TYPE_STR(LRHEADER, "LR header"),
REG_TYPE_STR(UNMOUNT, "unmount"),
REG_TYPE_STR(COMMIT, "commit"),
REG_TYPE_STR(TRANSHDR, "trans header"),
REG_TYPE_STR(ICREATE, "inode create"),
REG_TYPE_STR(RUI_FORMAT, "rui_format"),
REG_TYPE_STR(RUD_FORMAT, "rud_format"),
REG_TYPE_STR(CUI_FORMAT, "cui_format"),
REG_TYPE_STR(CUD_FORMAT, "cud_format"),
REG_TYPE_STR(BUI_FORMAT, "bui_format"),
REG_TYPE_STR(BUD_FORMAT, "bud_format"),
};
BUILD_BUG_ON(ARRAY_SIZE(res_type_str) != XLOG_REG_TYPE_MAX + 1);
#undef REG_TYPE_STR
xfs_warn(mp, "ticket reservation summary:");
xfs_warn(mp, " unit res = %d bytes",
ticket->t_unit_res);
xfs_warn(mp, " current res = %d bytes",
ticket->t_curr_res);
xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)",
ticket->t_res_arr_sum, ticket->t_res_o_flow);
xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)",
ticket->t_res_num_ophdrs, ophdr_spc);
xfs_warn(mp, " ophdr + reg = %u bytes",
ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc);
xfs_warn(mp, " num regions = %u",
ticket->t_res_num);
for (i = 0; i < ticket->t_res_num; i++) {
uint r_type = ticket->t_res_arr[i].r_type;
xfs_warn(mp, "region[%u]: %s - %u bytes", i,
((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
"bad-rtype" : res_type_str[r_type]),
ticket->t_res_arr[i].r_len);
}
}
/*
* Print a summary of the transaction.
*/
void
xlog_print_trans(
struct xfs_trans *tp)
{
struct xfs_mount *mp = tp->t_mountp;
struct xfs_log_item *lip;
/* dump core transaction and ticket info */
xfs_warn(mp, "transaction summary:");
xfs_warn(mp, " log res = %d", tp->t_log_res);
xfs_warn(mp, " log count = %d", tp->t_log_count);
xfs_warn(mp, " flags = 0x%x", tp->t_flags);
xlog_print_tic_res(mp, tp->t_ticket);
/* dump each log item */
list_for_each_entry(lip, &tp->t_items, li_trans) {
struct xfs_log_vec *lv = lip->li_lv;
struct xfs_log_iovec *vec;
int i;
xfs_warn(mp, "log item: ");
xfs_warn(mp, " type = 0x%x", lip->li_type);
xfs_warn(mp, " flags = 0x%lx", lip->li_flags);
if (!lv)
continue;
xfs_warn(mp, " niovecs = %d", lv->lv_niovecs);
xfs_warn(mp, " size = %d", lv->lv_size);
xfs_warn(mp, " bytes = %d", lv->lv_bytes);
xfs_warn(mp, " buf len = %d", lv->lv_buf_len);
/* dump each iovec for the log item */
vec = lv->lv_iovecp;
for (i = 0; i < lv->lv_niovecs; i++) {
int dumplen = min(vec->i_len, 32);
xfs_warn(mp, " iovec[%d]", i);
xfs_warn(mp, " type = 0x%x", vec->i_type);
xfs_warn(mp, " len = %d", vec->i_len);
xfs_warn(mp, " first %d bytes of iovec[%d]:", dumplen, i);
xfs_hex_dump(vec->i_addr, dumplen);
vec++;
}
}
}
/*
* Calculate the potential space needed by the log vector. Each region gets
* its own xlog_op_header_t and may need to be double word aligned.
*/
static int
xlog_write_calc_vec_length(
struct xlog_ticket *ticket,
struct xfs_log_vec *log_vector)
{
struct xfs_log_vec *lv;
int headers = 0;
int len = 0;
int i;
/* acct for start rec of xact */
if (ticket->t_flags & XLOG_TIC_INITED)
headers++;
for (lv = log_vector; lv; lv = lv->lv_next) {
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 13:04:51 +07:00
/* we don't write ordered log vectors */
if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED)
continue;
headers += lv->lv_niovecs;
for (i = 0; i < lv->lv_niovecs; i++) {
struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
len += vecp->i_len;
xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
}
}
ticket->t_res_num_ophdrs += headers;
len += headers * sizeof(struct xlog_op_header);
return len;
}
/*
* If first write for transaction, insert start record We can't be trying to
* commit if we are inited. We can't have any "partial_copy" if we are inited.
*/
static int
xlog_write_start_rec(
struct xlog_op_header *ophdr,
struct xlog_ticket *ticket)
{
if (!(ticket->t_flags & XLOG_TIC_INITED))
return 0;
ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
ophdr->oh_clientid = ticket->t_clientid;
ophdr->oh_len = 0;
ophdr->oh_flags = XLOG_START_TRANS;
ophdr->oh_res2 = 0;
ticket->t_flags &= ~XLOG_TIC_INITED;
return sizeof(struct xlog_op_header);
}
static xlog_op_header_t *
xlog_write_setup_ophdr(
struct xlog *log,
struct xlog_op_header *ophdr,
struct xlog_ticket *ticket,
uint flags)
{
ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
ophdr->oh_clientid = ticket->t_clientid;
ophdr->oh_res2 = 0;
/* are we copying a commit or unmount record? */
ophdr->oh_flags = flags;
/*
* We've seen logs corrupted with bad transaction client ids. This
* makes sure that XFS doesn't generate them on. Turn this into an EIO
* and shut down the filesystem.
*/
switch (ophdr->oh_clientid) {
case XFS_TRANSACTION:
case XFS_VOLUME:
case XFS_LOG:
break;
default:
xfs_warn(log->l_mp,
"Bad XFS transaction clientid 0x%x in ticket "PTR_FMT,
ophdr->oh_clientid, ticket);
return NULL;
}
return ophdr;
}
/*
* Set up the parameters of the region copy into the log. This has
* to handle region write split across multiple log buffers - this
* state is kept external to this function so that this code can
* be written in an obvious, self documenting manner.
*/
static int
xlog_write_setup_copy(
struct xlog_ticket *ticket,
struct xlog_op_header *ophdr,
int space_available,
int space_required,
int *copy_off,
int *copy_len,
int *last_was_partial_copy,
int *bytes_consumed)
{
int still_to_copy;
still_to_copy = space_required - *bytes_consumed;
*copy_off = *bytes_consumed;
if (still_to_copy <= space_available) {
/* write of region completes here */
*copy_len = still_to_copy;
ophdr->oh_len = cpu_to_be32(*copy_len);
if (*last_was_partial_copy)
ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
*last_was_partial_copy = 0;
*bytes_consumed = 0;
return 0;
}
/* partial write of region, needs extra log op header reservation */
*copy_len = space_available;
ophdr->oh_len = cpu_to_be32(*copy_len);
ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
if (*last_was_partial_copy)
ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
*bytes_consumed += *copy_len;
(*last_was_partial_copy)++;
/* account for new log op header */
ticket->t_curr_res -= sizeof(struct xlog_op_header);
ticket->t_res_num_ophdrs++;
return sizeof(struct xlog_op_header);
}
static int
xlog_write_copy_finish(
struct xlog *log,
struct xlog_in_core *iclog,
uint flags,
int *record_cnt,
int *data_cnt,
int *partial_copy,
int *partial_copy_len,
int log_offset,
struct xlog_in_core **commit_iclog)
{
int error;
if (*partial_copy) {
/*
* This iclog has already been marked WANT_SYNC by
* xlog_state_get_iclog_space.
*/
spin_lock(&log->l_icloglock);
xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
*record_cnt = 0;
*data_cnt = 0;
goto release_iclog;
}
*partial_copy = 0;
*partial_copy_len = 0;
if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
/* no more space in this iclog - push it. */
spin_lock(&log->l_icloglock);
xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
*record_cnt = 0;
*data_cnt = 0;
xlog_state_want_sync(log, iclog);
if (!commit_iclog)
goto release_iclog;
spin_unlock(&log->l_icloglock);
ASSERT(flags & XLOG_COMMIT_TRANS);
*commit_iclog = iclog;
}
return 0;
release_iclog:
error = xlog_state_release_iclog(log, iclog);
spin_unlock(&log->l_icloglock);
return error;
}
/*
* Write some region out to in-core log
*
* This will be called when writing externally provided regions or when
* writing out a commit record for a given transaction.
*
* General algorithm:
* 1. Find total length of this write. This may include adding to the
* lengths passed in.
* 2. Check whether we violate the tickets reservation.
* 3. While writing to this iclog
* A. Reserve as much space in this iclog as can get
* B. If this is first write, save away start lsn
* C. While writing this region:
* 1. If first write of transaction, write start record
* 2. Write log operation header (header per region)
* 3. Find out if we can fit entire region into this iclog
* 4. Potentially, verify destination memcpy ptr
* 5. Memcpy (partial) region
* 6. If partial copy, release iclog; otherwise, continue
* copying more regions into current iclog
* 4. Mark want sync bit (in simulation mode)
* 5. Release iclog for potential flush to on-disk log.
*
* ERRORS:
* 1. Panic if reservation is overrun. This should never happen since
* reservation amounts are generated internal to the filesystem.
* NOTES:
* 1. Tickets are single threaded data structures.
* 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
* syncing routine. When a single log_write region needs to span
* multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
* on all log operation writes which don't contain the end of the
* region. The XLOG_END_TRANS bit is used for the in-core log
* operation which contains the end of the continued log_write region.
* 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
* we don't really know exactly how much space will be used. As a result,
* we don't update ic_offset until the end when we know exactly how many
* bytes have been written out.
*/
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
int
xlog_write(
struct xlog *log,
struct xfs_log_vec *log_vector,
struct xlog_ticket *ticket,
xfs_lsn_t *start_lsn,
struct xlog_in_core **commit_iclog,
uint flags)
{
struct xlog_in_core *iclog = NULL;
struct xfs_log_iovec *vecp;
struct xfs_log_vec *lv;
int len;
int index;
int partial_copy = 0;
int partial_copy_len = 0;
int contwr = 0;
int record_cnt = 0;
int data_cnt = 0;
int error = 0;
*start_lsn = 0;
len = xlog_write_calc_vec_length(ticket, log_vector);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
/*
* Region headers and bytes are already accounted for.
* We only need to take into account start records and
* split regions in this function.
*/
if (ticket->t_flags & XLOG_TIC_INITED)
ticket->t_curr_res -= sizeof(xlog_op_header_t);
/*
* Commit record headers need to be accounted for. These
* come in as separate writes so are easy to detect.
*/
if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
ticket->t_curr_res -= sizeof(xlog_op_header_t);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
if (ticket->t_curr_res < 0) {
xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
"ctx ticket reservation ran out. Need to up reservation");
xlog_print_tic_res(log->l_mp, ticket);
xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR);
}
index = 0;
lv = log_vector;
vecp = lv->lv_iovecp;
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 13:04:51 +07:00
while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
void *ptr;
int log_offset;
error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
&contwr, &log_offset);
if (error)
return error;
ASSERT(log_offset <= iclog->ic_size - 1);
ptr = iclog->ic_datap + log_offset;
/* start_lsn is the first lsn written to. That's all we need. */
if (!*start_lsn)
*start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
/*
* This loop writes out as many regions as can fit in the amount
* of space which was allocated by xlog_state_get_iclog_space().
*/
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 13:04:51 +07:00
while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
struct xfs_log_iovec *reg;
struct xlog_op_header *ophdr;
int start_rec_copy;
int copy_len;
int copy_off;
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 13:04:51 +07:00
bool ordered = false;
/* ordered log vectors have no regions to write */
if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) {
ASSERT(lv->lv_niovecs == 0);
ordered = true;
goto next_lv;
}
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 13:04:51 +07:00
reg = &vecp[index];
ASSERT(reg->i_len % sizeof(int32_t) == 0);
ASSERT((unsigned long)ptr % sizeof(int32_t) == 0);
start_rec_copy = xlog_write_start_rec(ptr, ticket);
if (start_rec_copy) {
record_cnt++;
xlog_write_adv_cnt(&ptr, &len, &log_offset,
start_rec_copy);
}
ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
if (!ophdr)
return -EIO;
xlog_write_adv_cnt(&ptr, &len, &log_offset,
sizeof(struct xlog_op_header));
len += xlog_write_setup_copy(ticket, ophdr,
iclog->ic_size-log_offset,
reg->i_len,
&copy_off, &copy_len,
&partial_copy,
&partial_copy_len);
xlog_verify_dest_ptr(log, ptr);
/*
* Copy region.
*
* Unmount records just log an opheader, so can have
* empty payloads with no data region to copy. Hence we
* only copy the payload if the vector says it has data
* to copy.
*/
ASSERT(copy_len >= 0);
if (copy_len > 0) {
memcpy(ptr, reg->i_addr + copy_off, copy_len);
xlog_write_adv_cnt(&ptr, &len, &log_offset,
copy_len);
}
copy_len += start_rec_copy + sizeof(xlog_op_header_t);
record_cnt++;
data_cnt += contwr ? copy_len : 0;
error = xlog_write_copy_finish(log, iclog, flags,
&record_cnt, &data_cnt,
&partial_copy,
&partial_copy_len,
log_offset,
commit_iclog);
if (error)
return error;
/*
* if we had a partial copy, we need to get more iclog
* space but we don't want to increment the region
* index because there is still more is this region to
* write.
*
* If we completed writing this region, and we flushed
* the iclog (indicated by resetting of the record
* count), then we also need to get more log space. If
* this was the last record, though, we are done and
* can just return.
*/
if (partial_copy)
break;
if (++index == lv->lv_niovecs) {
xfs: Introduce ordered log vector support And "ordered log vector" is a log vector that is used for tracking a log item through the CIL and into the AIL as part of the log checkpointing. These ordered log vectors are special in that they are not written to to journal in any way, and are not accounted to the checkpoint being written. The reason for this behaviour is to allow operations to attach items to transactions and have them follow the normal transactional lifecycle without actually having to write them to the journal. This allows logging of items that track high level logical changes and writing them to the log, while the physical items being modified pass through into the AIL and pin the tail of the log (and therefore the logical item in the log) until all the modified items are physically written to disk. IOWs, it allows us to write metadata without physically logging every individual change but still maintain the full transactional integrity guarantees we currently have w.r.t. crash recovery. This change modifies some of the CIL item insertion loops, as ordered log vectors introduce some new constraints as they don't track any data. One advantage of this change is that it combines two log vector chain walks into a single pass, so there is less overhead in the transaction commit pass as well. It also kills some unused code in the log vector walk loop when committing the CIL. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-06-27 13:04:51 +07:00
next_lv:
lv = lv->lv_next;
index = 0;
if (lv)
vecp = lv->lv_iovecp;
}
if (record_cnt == 0 && !ordered) {
if (!lv)
return 0;
break;
}
}
}
ASSERT(len == 0);
spin_lock(&log->l_icloglock);
xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
if (commit_iclog) {
ASSERT(flags & XLOG_COMMIT_TRANS);
*commit_iclog = iclog;
} else {
error = xlog_state_release_iclog(log, iclog);
}
spin_unlock(&log->l_icloglock);
return error;
}
/*****************************************************************************
*
* State Machine functions
*
*****************************************************************************
*/
/*
* An iclog has just finished IO completion processing, so we need to update
* the iclog state and propagate that up into the overall log state. Hence we
* prepare the iclog for cleaning, and then clean all the pending dirty iclogs
* starting from the head, and then wake up any threads that are waiting for the
* iclog to be marked clean.
*
* The ordering of marking iclogs ACTIVE must be maintained, so an iclog
* doesn't become ACTIVE beyond one that is SYNCING. This is also required to
* maintain the notion that we use a ordered wait queue to hold off would be
* writers to the log when every iclog is trying to sync to disk.
*
* Caller must hold the icloglock before calling us.
*
* State Change: !IOERROR -> DIRTY -> ACTIVE
*/
STATIC void
xlog_state_clean_iclog(
struct xlog *log,
struct xlog_in_core *dirty_iclog)
{
struct xlog_in_core *iclog;
int changed = 0;
/* Prepare the completed iclog. */
if (dirty_iclog->ic_state != XLOG_STATE_IOERROR)
dirty_iclog->ic_state = XLOG_STATE_DIRTY;
/* Walk all the iclogs to update the ordered active state. */
iclog = log->l_iclog;
do {
if (iclog->ic_state == XLOG_STATE_DIRTY) {
iclog->ic_state = XLOG_STATE_ACTIVE;
iclog->ic_offset = 0;
ASSERT(list_empty_careful(&iclog->ic_callbacks));
/*
* If the number of ops in this iclog indicate it just
* contains the dummy transaction, we can
* change state into IDLE (the second time around).
* Otherwise we should change the state into
* NEED a dummy.
* We don't need to cover the dummy.
*/
if (!changed &&
(be32_to_cpu(iclog->ic_header.h_num_logops) ==
XLOG_COVER_OPS)) {
changed = 1;
} else {
/*
* We have two dirty iclogs so start over
* This could also be num of ops indicates
* this is not the dummy going out.
*/
changed = 2;
}
iclog->ic_header.h_num_logops = 0;
memset(iclog->ic_header.h_cycle_data, 0,
sizeof(iclog->ic_header.h_cycle_data));
iclog->ic_header.h_lsn = 0;
} else if (iclog->ic_state == XLOG_STATE_ACTIVE)
/* do nothing */;
else
break; /* stop cleaning */
iclog = iclog->ic_next;
} while (iclog != log->l_iclog);
/*
* Wake up threads waiting in xfs_log_force() for the dirty iclog
* to be cleaned.
*/
wake_up_all(&dirty_iclog->ic_force_wait);
/*
* Change state for the dummy log recording.
* We usually go to NEED. But we go to NEED2 if the changed indicates
* we are done writing the dummy record.
* If we are done with the second dummy recored (DONE2), then
* we go to IDLE.
*/
if (changed) {
switch (log->l_covered_state) {
case XLOG_STATE_COVER_IDLE:
case XLOG_STATE_COVER_NEED:
case XLOG_STATE_COVER_NEED2:
log->l_covered_state = XLOG_STATE_COVER_NEED;
break;
case XLOG_STATE_COVER_DONE:
if (changed == 1)
log->l_covered_state = XLOG_STATE_COVER_NEED2;
else
log->l_covered_state = XLOG_STATE_COVER_NEED;
break;
case XLOG_STATE_COVER_DONE2:
if (changed == 1)
log->l_covered_state = XLOG_STATE_COVER_IDLE;
else
log->l_covered_state = XLOG_STATE_COVER_NEED;
break;
default:
ASSERT(0);
}
}
}
STATIC xfs_lsn_t
xlog_get_lowest_lsn(
struct xlog *log)
{
struct xlog_in_core *iclog = log->l_iclog;
xfs_lsn_t lowest_lsn = 0, lsn;
do {
if (iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)
continue;
lsn = be64_to_cpu(iclog->ic_header.h_lsn);
if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0)
lowest_lsn = lsn;
} while ((iclog = iclog->ic_next) != log->l_iclog);
return lowest_lsn;
}
/*
* Completion of a iclog IO does not imply that a transaction has completed, as
* transactions can be large enough to span many iclogs. We cannot change the
* tail of the log half way through a transaction as this may be the only
* transaction in the log and moving the tail to point to the middle of it
* will prevent recovery from finding the start of the transaction. Hence we
* should only update the last_sync_lsn if this iclog contains transaction
* completion callbacks on it.
*
* We have to do this before we drop the icloglock to ensure we are the only one
* that can update it.
*
* If we are moving the last_sync_lsn forwards, we also need to ensure we kick
* the reservation grant head pushing. This is due to the fact that the push
* target is bound by the current last_sync_lsn value. Hence if we have a large
* amount of log space bound up in this committing transaction then the
* last_sync_lsn value may be the limiting factor preventing tail pushing from
* freeing space in the log. Hence once we've updated the last_sync_lsn we
* should push the AIL to ensure the push target (and hence the grant head) is
* no longer bound by the old log head location and can move forwards and make
* progress again.
*/
static void
xlog_state_set_callback(
struct xlog *log,
struct xlog_in_core *iclog,
xfs_lsn_t header_lsn)
{
iclog->ic_state = XLOG_STATE_CALLBACK;
ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
header_lsn) <= 0);
if (list_empty_careful(&iclog->ic_callbacks))
return;
atomic64_set(&log->l_last_sync_lsn, header_lsn);
xlog_grant_push_ail(log, 0);
}
/*
* Return true if we need to stop processing, false to continue to the next
* iclog. The caller will need to run callbacks if the iclog is returned in the
* XLOG_STATE_CALLBACK state.
*/
static bool
xlog_state_iodone_process_iclog(
struct xlog *log,
struct xlog_in_core *iclog,
bool *ioerror)
{
xfs_lsn_t lowest_lsn;
xfs_lsn_t header_lsn;
switch (iclog->ic_state) {
case XLOG_STATE_ACTIVE:
case XLOG_STATE_DIRTY:
/*
* Skip all iclogs in the ACTIVE & DIRTY states:
*/
return false;
case XLOG_STATE_IOERROR:
/*
* Between marking a filesystem SHUTDOWN and stopping the log,
* we do flush all iclogs to disk (if there wasn't a log I/O
* error). So, we do want things to go smoothly in case of just
* a SHUTDOWN w/o a LOG_IO_ERROR.
*/
*ioerror = true;
return false;
case XLOG_STATE_DONE_SYNC:
/*
* Now that we have an iclog that is in the DONE_SYNC state, do
* one more check here to see if we have chased our tail around.
* If this is not the lowest lsn iclog, then we will leave it
* for another completion to process.
*/
header_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
lowest_lsn = xlog_get_lowest_lsn(log);
if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0)
return false;
xlog_state_set_callback(log, iclog, header_lsn);
return false;
default:
/*
* Can only perform callbacks in order. Since this iclog is not
* in the DONE_SYNC state, we skip the rest and just try to
* clean up.
*/
return true;
}
}
/*
* Keep processing entries in the iclog callback list until we come around and
* it is empty. We need to atomically see that the list is empty and change the
* state to DIRTY so that we don't miss any more callbacks being added.
*
* This function is called with the icloglock held and returns with it held. We
* drop it while running callbacks, however, as holding it over thousands of
* callbacks is unnecessary and causes excessive contention if we do.
*/
static void
xlog_state_do_iclog_callbacks(
struct xlog *log,
struct xlog_in_core *iclog,
bool aborted)
__releases(&log->l_icloglock)
__acquires(&log->l_icloglock)
{
spin_unlock(&log->l_icloglock);
spin_lock(&iclog->ic_callback_lock);
while (!list_empty(&iclog->ic_callbacks)) {
LIST_HEAD(tmp);
list_splice_init(&iclog->ic_callbacks, &tmp);
spin_unlock(&iclog->ic_callback_lock);
xlog_cil_process_committed(&tmp, aborted);
spin_lock(&iclog->ic_callback_lock);
}
/*
* Pick up the icloglock while still holding the callback lock so we
* serialise against anyone trying to add more callbacks to this iclog
* now we've finished processing.
*/
spin_lock(&log->l_icloglock);
spin_unlock(&iclog->ic_callback_lock);
}
STATIC void
xlog_state_do_callback(
struct xlog *log,
bool aborted)
{
struct xlog_in_core *iclog;
struct xlog_in_core *first_iclog;
bool cycled_icloglock;
bool ioerror;
int flushcnt = 0;
int repeats = 0;
spin_lock(&log->l_icloglock);
do {
/*
* Scan all iclogs starting with the one pointed to by the
* log. Reset this starting point each time the log is
* unlocked (during callbacks).
*
* Keep looping through iclogs until one full pass is made
* without running any callbacks.
*/
first_iclog = log->l_iclog;
iclog = log->l_iclog;
cycled_icloglock = false;
ioerror = false;
repeats++;
do {
if (xlog_state_iodone_process_iclog(log, iclog,
&ioerror))
break;
if (iclog->ic_state != XLOG_STATE_CALLBACK &&
iclog->ic_state != XLOG_STATE_IOERROR) {
iclog = iclog->ic_next;
continue;
}
/*
* Running callbacks will drop the icloglock which means
* we'll have to run at least one more complete loop.
*/
cycled_icloglock = true;
xlog_state_do_iclog_callbacks(log, iclog, aborted);
xlog_state_clean_iclog(log, iclog);
iclog = iclog->ic_next;
} while (first_iclog != iclog);
if (repeats > 5000) {
flushcnt += repeats;
repeats = 0;
xfs_warn(log->l_mp,
"%s: possible infinite loop (%d iterations)",
__func__, flushcnt);
}
} while (!ioerror && cycled_icloglock);
if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE ||
log->l_iclog->ic_state == XLOG_STATE_IOERROR)
wake_up_all(&log->l_flush_wait);
spin_unlock(&log->l_icloglock);
}
/*
* Finish transitioning this iclog to the dirty state.
*
* Make sure that we completely execute this routine only when this is
* the last call to the iclog. There is a good chance that iclog flushes,
* when we reach the end of the physical log, get turned into 2 separate
* calls to bwrite. Hence, one iclog flush could generate two calls to this
* routine. By using the reference count bwritecnt, we guarantee that only
* the second completion goes through.
*
* Callbacks could take time, so they are done outside the scope of the
* global state machine log lock.
*/
STATIC void
xlog_state_done_syncing(
struct xlog_in_core *iclog,
bool aborted)
{
struct xlog *log = iclog->ic_log;
spin_lock(&log->l_icloglock);
ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
/*
* If we got an error, either on the first buffer, or in the case of
* split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
* and none should ever be attempted to be written to disk
* again.
*/
if (iclog->ic_state == XLOG_STATE_SYNCING)
iclog->ic_state = XLOG_STATE_DONE_SYNC;
else
ASSERT(iclog->ic_state == XLOG_STATE_IOERROR);
/*
* Someone could be sleeping prior to writing out the next
* iclog buffer, we wake them all, one will get to do the
* I/O, the others get to wait for the result.
*/
wake_up_all(&iclog->ic_write_wait);
spin_unlock(&log->l_icloglock);
xlog_state_do_callback(log, aborted); /* also cleans log */
} /* xlog_state_done_syncing */
/*
* If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
* sleep. We wait on the flush queue on the head iclog as that should be
* the first iclog to complete flushing. Hence if all iclogs are syncing,
* we will wait here and all new writes will sleep until a sync completes.
*
* The in-core logs are used in a circular fashion. They are not used
* out-of-order even when an iclog past the head is free.
*
* return:
* * log_offset where xlog_write() can start writing into the in-core
* log's data space.
* * in-core log pointer to which xlog_write() should write.
* * boolean indicating this is a continued write to an in-core log.
* If this is the last write, then the in-core log's offset field
* needs to be incremented, depending on the amount of data which
* is copied.
*/
STATIC int
xlog_state_get_iclog_space(
struct xlog *log,
int len,
struct xlog_in_core **iclogp,
struct xlog_ticket *ticket,
int *continued_write,
int *logoffsetp)
{
int log_offset;
xlog_rec_header_t *head;
xlog_in_core_t *iclog;
restart:
spin_lock(&log->l_icloglock);
if (XLOG_FORCED_SHUTDOWN(log)) {
spin_unlock(&log->l_icloglock);
return -EIO;
}
iclog = log->l_iclog;
if (iclog->ic_state != XLOG_STATE_ACTIVE) {
XFS_STATS_INC(log->l_mp, xs_log_noiclogs);
/* Wait for log writes to have flushed */
xlog_wait(&log->l_flush_wait, &log->l_icloglock);
goto restart;
}
head = &iclog->ic_header;
atomic_inc(&iclog->ic_refcnt); /* prevents sync */
log_offset = iclog->ic_offset;
/* On the 1st write to an iclog, figure out lsn. This works
* if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
* committing to. If the offset is set, that's how many blocks
* must be written.
*/
if (log_offset == 0) {
ticket->t_curr_res -= log->l_iclog_hsize;
xlog_tic_add_region(ticket,
log->l_iclog_hsize,
XLOG_REG_TYPE_LRHEADER);
head->h_cycle = cpu_to_be32(log->l_curr_cycle);
head->h_lsn = cpu_to_be64(
xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
ASSERT(log->l_curr_block >= 0);
}
/* If there is enough room to write everything, then do it. Otherwise,
* claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
* bit is on, so this will get flushed out. Don't update ic_offset
* until you know exactly how many bytes get copied. Therefore, wait
* until later to update ic_offset.
*
* xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
* can fit into remaining data section.
*/
if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
int error = 0;
xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
/*
* If we are the only one writing to this iclog, sync it to
* disk. We need to do an atomic compare and decrement here to
* avoid racing with concurrent atomic_dec_and_lock() calls in
* xlog_state_release_iclog() when there is more than one
* reference to the iclog.
*/
if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1))
error = xlog_state_release_iclog(log, iclog);
spin_unlock(&log->l_icloglock);
if (error)
return error;
goto restart;
}
/* Do we have enough room to write the full amount in the remainder
* of this iclog? Or must we continue a write on the next iclog and
* mark this iclog as completely taken? In the case where we switch
* iclogs (to mark it taken), this particular iclog will release/sync
* to disk in xlog_write().
*/
if (len <= iclog->ic_size - iclog->ic_offset) {
*continued_write = 0;
iclog->ic_offset += len;
} else {
*continued_write = 1;
xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
}
*iclogp = iclog;
ASSERT(iclog->ic_offset <= iclog->ic_size);
spin_unlock(&log->l_icloglock);
*logoffsetp = log_offset;
return 0;
} /* xlog_state_get_iclog_space */
/* The first cnt-1 times through here we don't need to
* move the grant write head because the permanent
* reservation has reserved cnt times the unit amount.
* Release part of current permanent unit reservation and
* reset current reservation to be one units worth. Also
* move grant reservation head forward.
*/
STATIC void
xlog_regrant_reserve_log_space(
struct xlog *log,
struct xlog_ticket *ticket)
{
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_regrant_reserve_enter(log, ticket);
if (ticket->t_cnt > 0)
ticket->t_cnt--;
xlog_grant_sub_space(log, &log->l_reserve_head.grant,
ticket->t_curr_res);
xlog_grant_sub_space(log, &log->l_write_head.grant,
ticket->t_curr_res);
ticket->t_curr_res = ticket->t_unit_res;
xlog_tic_reset_res(ticket);
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_regrant_reserve_sub(log, ticket);
/* just return if we still have some of the pre-reserved space */
if (ticket->t_cnt > 0)
return;
xlog_grant_add_space(log, &log->l_reserve_head.grant,
ticket->t_unit_res);
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_regrant_reserve_exit(log, ticket);
ticket->t_curr_res = ticket->t_unit_res;
xlog_tic_reset_res(ticket);
} /* xlog_regrant_reserve_log_space */
/*
* Give back the space left from a reservation.
*
* All the information we need to make a correct determination of space left
* is present. For non-permanent reservations, things are quite easy. The
* count should have been decremented to zero. We only need to deal with the
* space remaining in the current reservation part of the ticket. If the
* ticket contains a permanent reservation, there may be left over space which
* needs to be released. A count of N means that N-1 refills of the current
* reservation can be done before we need to ask for more space. The first
* one goes to fill up the first current reservation. Once we run out of
* space, the count will stay at zero and the only space remaining will be
* in the current reservation field.
*/
STATIC void
xlog_ungrant_log_space(
struct xlog *log,
struct xlog_ticket *ticket)
{
int bytes;
if (ticket->t_cnt > 0)
ticket->t_cnt--;
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_ungrant_enter(log, ticket);
trace_xfs_log_ungrant_sub(log, ticket);
/*
* If this is a permanent reservation ticket, we may be able to free
* up more space based on the remaining count.
*/
bytes = ticket->t_curr_res;
if (ticket->t_cnt > 0) {
ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
bytes += ticket->t_unit_res*ticket->t_cnt;
}
xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-15 06:14:59 +07:00
trace_xfs_log_ungrant_exit(log, ticket);
xfs_log_space_wake(log->l_mp);
}
/*
* This routine will mark the current iclog in the ring as WANT_SYNC
* and move the current iclog pointer to the next iclog in the ring.
* When this routine is called from xlog_state_get_iclog_space(), the
* exact size of the iclog has not yet been determined. All we know is
* that every data block. We have run out of space in this log record.
*/
STATIC void
xlog_state_switch_iclogs(
struct xlog *log,
struct xlog_in_core *iclog,
int eventual_size)
{
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
if (!eventual_size)
eventual_size = iclog->ic_offset;
iclog->ic_state = XLOG_STATE_WANT_SYNC;
iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
log->l_prev_block = log->l_curr_block;
log->l_prev_cycle = log->l_curr_cycle;
/* roll log?: ic_offset changed later */
log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
/* Round up to next log-sunit */
if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
log->l_mp->m_sb.sb_logsunit > 1) {
uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
}
if (log->l_curr_block >= log->l_logBBsize) {
xfs: validate metadata LSNs against log on v5 superblocks Since the onset of v5 superblocks, the LSN of the last modification has been included in a variety of on-disk data structures. This LSN is used to provide log recovery ordering guarantees (e.g., to ensure an older log recovery item is not replayed over a newer target data structure). While this works correctly from the point a filesystem is formatted and mounted, userspace tools have some problematic behaviors that defeat this mechanism. For example, xfs_repair historically zeroes out the log unconditionally (regardless of whether corruption is detected). If this occurs, the LSN of the filesystem is reset and the log is now in a problematic state with respect to on-disk metadata structures that might have a larger LSN. Until either the log catches up to the highest previously used metadata LSN or each affected data structure is modified and written out without incident (which resets the metadata LSN), log recovery is susceptible to filesystem corruption. This problem is ultimately addressed and repaired in the associated userspace tools. The kernel is still responsible to detect the problem and notify the user that something is wrong. Check the superblock LSN at mount time and fail the mount if it is invalid. From that point on, trigger verifier failure on any metadata I/O where an invalid LSN is detected. This results in a filesystem shutdown and guarantees that we do not log metadata changes with invalid LSNs on disk. Since this is a known issue with a known recovery path, present a warning to instruct the user how to recover. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 11:59:25 +07:00
/*
* Rewind the current block before the cycle is bumped to make
* sure that the combined LSN never transiently moves forward
* when the log wraps to the next cycle. This is to support the
* unlocked sample of these fields from xlog_valid_lsn(). Most
* other cases should acquire l_icloglock.
*/
log->l_curr_block -= log->l_logBBsize;
ASSERT(log->l_curr_block >= 0);
smp_wmb();
log->l_curr_cycle++;
if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
log->l_curr_cycle++;
}
ASSERT(iclog == log->l_iclog);
log->l_iclog = iclog->ic_next;
} /* xlog_state_switch_iclogs */
/*
* Write out all data in the in-core log as of this exact moment in time.
*
* Data may be written to the in-core log during this call. However,
* we don't guarantee this data will be written out. A change from past
* implementation means this routine will *not* write out zero length LRs.
*
* Basically, we try and perform an intelligent scan of the in-core logs.
* If we determine there is no flushable data, we just return. There is no
* flushable data if:
*
* 1. the current iclog is active and has no data; the previous iclog
* is in the active or dirty state.
* 2. the current iclog is drity, and the previous iclog is in the
* active or dirty state.
*
* We may sleep if:
*
* 1. the current iclog is not in the active nor dirty state.
* 2. the current iclog dirty, and the previous iclog is not in the
* active nor dirty state.
* 3. the current iclog is active, and there is another thread writing
* to this particular iclog.
* 4. a) the current iclog is active and has no other writers
* b) when we return from flushing out this iclog, it is still
* not in the active nor dirty state.
*/
int
xfs_log_force(
struct xfs_mount *mp,
uint flags)
{
struct xlog *log = mp->m_log;
struct xlog_in_core *iclog;
xfs_lsn_t lsn;
XFS_STATS_INC(mp, xs_log_force);
trace_xfs_log_force(mp, 0, _RET_IP_);
xlog_cil_force(log);
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
if (iclog->ic_state == XLOG_STATE_IOERROR)
goto out_error;
if (iclog->ic_state == XLOG_STATE_DIRTY ||
(iclog->ic_state == XLOG_STATE_ACTIVE &&
atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) {
/*
* If the head is dirty or (active and empty), then we need to
* look at the previous iclog.
*
* If the previous iclog is active or dirty we are done. There
* is nothing to sync out. Otherwise, we attach ourselves to the
* previous iclog and go to sleep.
*/
iclog = iclog->ic_prev;
if (iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)
goto out_unlock;
} else if (iclog->ic_state == XLOG_STATE_ACTIVE) {
if (atomic_read(&iclog->ic_refcnt) == 0) {
/*
* We are the only one with access to this iclog.
*
* Flush it out now. There should be a roundoff of zero
* to show that someone has already taken care of the
* roundoff from the previous sync.
*/
atomic_inc(&iclog->ic_refcnt);
lsn = be64_to_cpu(iclog->ic_header.h_lsn);
xlog_state_switch_iclogs(log, iclog, 0);
if (xlog_state_release_iclog(log, iclog))
goto out_error;
if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn ||
iclog->ic_state == XLOG_STATE_DIRTY)
goto out_unlock;
} else {
/*
* Someone else is writing to this iclog.
*
* Use its call to flush out the data. However, the
* other thread may not force out this LR, so we mark
* it WANT_SYNC.
*/
xlog_state_switch_iclogs(log, iclog, 0);
}
} else {
/*
* If the head iclog is not active nor dirty, we just attach
* ourselves to the head and go to sleep if necessary.
*/
;
}
if (!(flags & XFS_LOG_SYNC))
goto out_unlock;
if (iclog->ic_state == XLOG_STATE_IOERROR)
goto out_error;
XFS_STATS_INC(mp, xs_log_force_sleep);
xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
if (iclog->ic_state == XLOG_STATE_IOERROR)
return -EIO;
return 0;
out_unlock:
spin_unlock(&log->l_icloglock);
return 0;
out_error:
spin_unlock(&log->l_icloglock);
return -EIO;
}
static int
__xfs_log_force_lsn(
struct xfs_mount *mp,
xfs_lsn_t lsn,
uint flags,
int *log_flushed,
bool already_slept)
{
struct xlog *log = mp->m_log;
struct xlog_in_core *iclog;
xfs: Introduce delayed logging core code The delayed logging code only changes in-memory structures and as such can be enabled and disabled with a mount option. Add the mount option and emit a warning that this is an experimental feature that should not be used in production yet. We also need infrastructure to track committed items that have not yet been written to the log. This is what the Committed Item List (CIL) is for. The log item also needs to be extended to track the current log vector, the associated memory buffer and it's location in the Commit Item List. Extend the log item and log vector structures to enable this tracking. To maintain the current log format for transactions with delayed logging, we need to introduce a checkpoint transaction and a context for tracking each checkpoint from initiation to transaction completion. This includes adding a log ticket for tracking space log required/used by the context checkpoint. To track all the changes we need an io vector array per log item, rather than a single array for the entire transaction. Using the new log vector structure for this requires two passes - the first to allocate the log vector structures and chain them together, and the second to fill them out. This log vector chain can then be passed to the CIL for formatting, pinning and insertion into the CIL. Formatting of the log vector chain is relatively simple - it's just a loop over the iovecs on each log vector, but it is made slightly more complex because we re-write the iovec after the copy to point back at the memory buffer we just copied into. This code also needs to pin log items. If the log item is not already tracked in this checkpoint context, then it needs to be pinned. Otherwise it is already pinned and we don't need to pin it again. The only other complexity is calculating the amount of new log space the formatting has consumed. This needs to be accounted to the transaction in progress, and the accounting is made more complex becase we need also to steal space from it for log metadata in the checkpoint transaction. Calculate all this at insert time and update all the tickets, counters, etc correctly. Once we've formatted all the log items in the transaction, attach the busy extents to the checkpoint context so the busy extents live until checkpoint completion and can be processed at that point in time. Transactions can then be freed at this point in time. Now we need to issue checkpoints - we are tracking the amount of log space used by the items in the CIL, so we can trigger background checkpoints when the space usage gets to a certain threshold. Otherwise, checkpoints need ot be triggered when a log synchronisation point is reached - a log force event. Because the log write code already handles chained log vectors, writing the transaction is trivial, too. Construct a transaction header, add it to the head of the chain and write it into the log, then issue a commit record write. Then we can release the checkpoint log ticket and attach the context to the log buffer so it can be called during Io completion to complete the checkpoint. We also need to allow for synchronising multiple in-flight checkpoints. This is needed for two things - the first is to ensure that checkpoint commit records appear in the log in the correct sequence order (so they are replayed in the correct order). The second is so that xfs_log_force_lsn() operates correctly and only flushes and/or waits for the specific sequence it was provided with. To do this we need a wait variable and a list tracking the checkpoint commits in progress. We can walk this list and wait for the checkpoints to change state or complete easily, an this provides the necessary synchronisation for correct operation in both cases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2010-05-21 11:37:18 +07:00
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
if (iclog->ic_state == XLOG_STATE_IOERROR)
goto out_error;
while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
iclog = iclog->ic_next;
if (iclog == log->l_iclog)
goto out_unlock;
}
if (iclog->ic_state == XLOG_STATE_DIRTY)
goto out_unlock;
if (iclog->ic_state == XLOG_STATE_ACTIVE) {
/*
* We sleep here if we haven't already slept (e.g. this is the
* first time we've looked at the correct iclog buf) and the
* buffer before us is going to be sync'ed. The reason for this
* is that if we are doing sync transactions here, by waiting
* for the previous I/O to complete, we can allow a few more
* transactions into this iclog before we close it down.
*
* Otherwise, we mark the buffer WANT_SYNC, and bump up the
* refcnt so we can release the log (which drops the ref count).
* The state switch keeps new transaction commits from using
* this buffer. When the current commits finish writing into
* the buffer, the refcount will drop to zero and the buffer
* will go out then.
*/
if (!already_slept &&
(iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC ||
iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) {
XFS_STATS_INC(mp, xs_log_force_sleep);
xlog_wait(&iclog->ic_prev->ic_write_wait,
&log->l_icloglock);
return -EAGAIN;
}
atomic_inc(&iclog->ic_refcnt);
xlog_state_switch_iclogs(log, iclog, 0);
if (xlog_state_release_iclog(log, iclog))
goto out_error;
if (log_flushed)
*log_flushed = 1;
}
if (!(flags & XFS_LOG_SYNC) ||
(iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY))
goto out_unlock;
if (iclog->ic_state == XLOG_STATE_IOERROR)
goto out_error;
XFS_STATS_INC(mp, xs_log_force_sleep);
xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
if (iclog->ic_state == XLOG_STATE_IOERROR)
return -EIO;
return 0;
out_unlock:
spin_unlock(&log->l_icloglock);
return 0;
out_error:
spin_unlock(&log->l_icloglock);
return -EIO;
}
/*
* Force the in-core log to disk for a specific LSN.
*
* Find in-core log with lsn.
* If it is in the DIRTY state, just return.
* If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
* state and go to sleep or return.
* If it is in any other state, go to sleep or return.
*
* Synchronous forces are implemented with a wait queue. All callers trying
* to force a given lsn to disk must wait on the queue attached to the
* specific in-core log. When given in-core log finally completes its write
* to disk, that thread will wake up all threads waiting on the queue.
*/
int
xfs_log_force_lsn(
struct xfs_mount *mp,
xfs_lsn_t lsn,
uint flags,
int *log_flushed)
{
int ret;
ASSERT(lsn != 0);
XFS_STATS_INC(mp, xs_log_force);
trace_xfs_log_force(mp, lsn, _RET_IP_);
lsn = xlog_cil_force_lsn(mp->m_log, lsn);
if (lsn == NULLCOMMITLSN)
return 0;
ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, false);
if (ret == -EAGAIN)
ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, true);
return ret;
}
/*
* Called when we want to mark the current iclog as being ready to sync to
* disk.
*/
STATIC void
xlog_state_want_sync(
struct xlog *log,
struct xlog_in_core *iclog)
{
assert_spin_locked(&log->l_icloglock);
if (iclog->ic_state == XLOG_STATE_ACTIVE) {
xlog_state_switch_iclogs(log, iclog, 0);
} else {
ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC ||
iclog->ic_state == XLOG_STATE_IOERROR);
}
}
/*****************************************************************************
*
* TICKET functions
*
*****************************************************************************
*/
/*
* Free a used ticket when its refcount falls to zero.
*/
void
xfs_log_ticket_put(
xlog_ticket_t *ticket)
{
ASSERT(atomic_read(&ticket->t_ref) > 0);
if (atomic_dec_and_test(&ticket->t_ref))
kmem_cache_free(xfs_log_ticket_zone, ticket);
}
xlog_ticket_t *
xfs_log_ticket_get(
xlog_ticket_t *ticket)
{
ASSERT(atomic_read(&ticket->t_ref) > 0);
atomic_inc(&ticket->t_ref);
return ticket;
}
/*
* Figure out the total log space unit (in bytes) that would be
* required for a log ticket.
*/
int
xfs_log_calc_unit_res(
struct xfs_mount *mp,
int unit_bytes)
{
struct xlog *log = mp->m_log;
int iclog_space;
uint num_headers;
/*
* Permanent reservations have up to 'cnt'-1 active log operations
* in the log. A unit in this case is the amount of space for one
* of these log operations. Normal reservations have a cnt of 1
* and their unit amount is the total amount of space required.
*
* The following lines of code account for non-transaction data
* which occupy space in the on-disk log.
*
* Normal form of a transaction is:
* <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
* and then there are LR hdrs, split-recs and roundoff at end of syncs.
*
* We need to account for all the leadup data and trailer data
* around the transaction data.
* And then we need to account for the worst case in terms of using
* more space.
* The worst case will happen if:
* - the placement of the transaction happens to be such that the
* roundoff is at its maximum
* - the transaction data is synced before the commit record is synced
* i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
* Therefore the commit record is in its own Log Record.
* This can happen as the commit record is called with its
* own region to xlog_write().
* This then means that in the worst case, roundoff can happen for
* the commit-rec as well.
* The commit-rec is smaller than padding in this scenario and so it is
* not added separately.
*/
/* for trans header */
unit_bytes += sizeof(xlog_op_header_t);
unit_bytes += sizeof(xfs_trans_header_t);
/* for start-rec */
unit_bytes += sizeof(xlog_op_header_t);
xfs: log ticket reservation underestimates the number of iclogs When allocation a ticket for a transaction, the ticket is initialised with the worst case log space usage based on the number of bytes the transaction may consume. Part of this calculation is the number of log headers required for the iclog space used up by the transaction. This calculation makes an undocumented assumption that if the transaction uses the log header space reservation on an iclog, then it consumes either the entire iclog or it completes. That is - the transaction that is first in an iclog is the transaction that the log header reservation is accounted to. If the transaction is larger than the iclog, then it will use the entire iclog itself. Document this assumption. Further, the current calculation uses the rule that we can fit iclog_size bytes of transaction data into an iclog. This is in correct - the amount of space available in an iclog for transaction data is the size of the iclog minus the space used for log record headers. This means that the calculation is out by 512 bytes per 32k of log space the transaction can consume. This is rarely an issue because maximally sized transactions are extremely uncommon, and for 4k block size filesystems maximal transaction reservations are about 400kb. Hence the error in this case is less than the size of an iclog, so that makes it even harder to hit. However, anyone using larger directory blocks (16k directory blocks push the maximum transaction size to approx. 900k on a 4k block size filesystem) or larger block size (e.g. 64k blocks push transactions to the 3-4MB size) could see the error grow to more than an iclog and at this point the transaction is guaranteed to get a reservation underrun and shutdown the filesystem. Fix this by adjusting the calculation to calculate the correct number of iclogs required and account for them all up front. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-03-23 07:21:11 +07:00
/*
* for LR headers - the space for data in an iclog is the size minus
* the space used for the headers. If we use the iclog size, then we
* undercalculate the number of headers required.
*
* Furthermore - the addition of op headers for split-recs might
* increase the space required enough to require more log and op
* headers, so take that into account too.
*
* IMPORTANT: This reservation makes the assumption that if this
* transaction is the first in an iclog and hence has the LR headers
* accounted to it, then the remaining space in the iclog is
* exclusively for this transaction. i.e. if the transaction is larger
* than the iclog, it will be the only thing in that iclog.
* Fundamentally, this means we must pass the entire log vector to
* xlog_write to guarantee this.
*/
iclog_space = log->l_iclog_size - log->l_iclog_hsize;
num_headers = howmany(unit_bytes, iclog_space);
/* for split-recs - ophdrs added when data split over LRs */
unit_bytes += sizeof(xlog_op_header_t) * num_headers;
/* add extra header reservations if we overrun */
while (!num_headers ||
howmany(unit_bytes, iclog_space) > num_headers) {
unit_bytes += sizeof(xlog_op_header_t);
num_headers++;
}
unit_bytes += log->l_iclog_hsize * num_headers;
/* for commit-rec LR header - note: padding will subsume the ophdr */
unit_bytes += log->l_iclog_hsize;
/* for roundoff padding for transaction data and one for commit record */
if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) {
/* log su roundoff */
unit_bytes += 2 * mp->m_sb.sb_logsunit;
} else {
/* BB roundoff */
unit_bytes += 2 * BBSIZE;
}
return unit_bytes;
}
/*
* Allocate and initialise a new log ticket.
*/
struct xlog_ticket *
xlog_ticket_alloc(
struct xlog *log,
int unit_bytes,
int cnt,
char client,
bool permanent,
xfs_km_flags_t alloc_flags)
{
struct xlog_ticket *tic;
int unit_res;
tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
if (!tic)
return NULL;
unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes);
atomic_set(&tic->t_ref, 1);
tic->t_task = current;
INIT_LIST_HEAD(&tic->t_queue);
tic->t_unit_res = unit_res;
tic->t_curr_res = unit_res;
tic->t_cnt = cnt;
tic->t_ocnt = cnt;
tic->t_tid = prandom_u32();
tic->t_clientid = client;
tic->t_flags = XLOG_TIC_INITED;
if (permanent)
tic->t_flags |= XLOG_TIC_PERM_RESERV;
xlog_tic_reset_res(tic);
return tic;
}
/******************************************************************************
*
* Log debug routines
*
******************************************************************************
*/
#if defined(DEBUG)
/*
* Make sure that the destination ptr is within the valid data region of
* one of the iclogs. This uses backup pointers stored in a different
* part of the log in case we trash the log structure.
*/
STATIC void
xlog_verify_dest_ptr(
struct xlog *log,
void *ptr)
{
int i;
int good_ptr = 0;
for (i = 0; i < log->l_iclog_bufs; i++) {
if (ptr >= log->l_iclog_bak[i] &&
ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
good_ptr++;
}
if (!good_ptr)
xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
}
/*
* Check to make sure the grant write head didn't just over lap the tail. If
* the cycles are the same, we can't be overlapping. Otherwise, make sure that
* the cycles differ by exactly one and check the byte count.
*
* This check is run unlocked, so can give false positives. Rather than assert
* on failures, use a warn-once flag and a panic tag to allow the admin to
* determine if they want to panic the machine when such an error occurs. For
* debug kernels this will have the same effect as using an assert but, unlinke
* an assert, it can be turned off at runtime.
*/
STATIC void
xlog_verify_grant_tail(
struct xlog *log)
{
int tail_cycle, tail_blocks;
int cycle, space;
xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
if (tail_cycle != cycle) {
if (cycle - 1 != tail_cycle &&
!(log->l_flags & XLOG_TAIL_WARN)) {
xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
"%s: cycle - 1 != tail_cycle", __func__);
log->l_flags |= XLOG_TAIL_WARN;
}
if (space > BBTOB(tail_blocks) &&
!(log->l_flags & XLOG_TAIL_WARN)) {
xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
"%s: space > BBTOB(tail_blocks)", __func__);
log->l_flags |= XLOG_TAIL_WARN;
}
}
}
/* check if it will fit */
STATIC void
xlog_verify_tail_lsn(
struct xlog *log,
struct xlog_in_core *iclog,
xfs_lsn_t tail_lsn)
{
int blocks;
if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
blocks =
log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
} else {
ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
if (blocks < BTOBB(iclog->ic_offset) + 1)
xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
}
} /* xlog_verify_tail_lsn */
/*
* Perform a number of checks on the iclog before writing to disk.
*
* 1. Make sure the iclogs are still circular
* 2. Make sure we have a good magic number
* 3. Make sure we don't have magic numbers in the data
* 4. Check fields of each log operation header for:
* A. Valid client identifier
* B. tid ptr value falls in valid ptr space (user space code)
* C. Length in log record header is correct according to the
* individual operation headers within record.
* 5. When a bwrite will occur within 5 blocks of the front of the physical
* log, check the preceding blocks of the physical log to make sure all
* the cycle numbers agree with the current cycle number.
*/
STATIC void
xlog_verify_iclog(
struct xlog *log,
struct xlog_in_core *iclog,
int count)
{
xlog_op_header_t *ophead;
xlog_in_core_t *icptr;
xlog_in_core_2_t *xhdr;
void *base_ptr, *ptr, *p;
ptrdiff_t field_offset;
uint8_t clientid;
int len, i, j, k, op_len;
int idx;
/* check validity of iclog pointers */
spin_lock(&log->l_icloglock);
icptr = log->l_iclog;
for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)
ASSERT(icptr);
if (icptr != log->l_iclog)
xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
spin_unlock(&log->l_icloglock);
/* check log magic numbers */
if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
base_ptr = ptr = &iclog->ic_header;
p = &iclog->ic_header;
for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) {
if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
xfs_emerg(log->l_mp, "%s: unexpected magic num",
__func__);
}
/* check fields */
len = be32_to_cpu(iclog->ic_header.h_num_logops);
base_ptr = ptr = iclog->ic_datap;
ophead = ptr;
xhdr = iclog->ic_data;
for (i = 0; i < len; i++) {
ophead = ptr;
/* clientid is only 1 byte */
p = &ophead->oh_clientid;
field_offset = p - base_ptr;
if (field_offset & 0x1ff) {
clientid = ophead->oh_clientid;
} else {
idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap);
if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
clientid = xlog_get_client_id(
xhdr[j].hic_xheader.xh_cycle_data[k]);
} else {
clientid = xlog_get_client_id(
iclog->ic_header.h_cycle_data[idx]);
}
}
if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
xfs_warn(log->l_mp,
"%s: invalid clientid %d op "PTR_FMT" offset 0x%lx",
__func__, clientid, ophead,
(unsigned long)field_offset);
/* check length */
p = &ophead->oh_len;
field_offset = p - base_ptr;
if (field_offset & 0x1ff) {
op_len = be32_to_cpu(ophead->oh_len);
} else {
idx = BTOBBT((uintptr_t)&ophead->oh_len -
(uintptr_t)iclog->ic_datap);
if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
} else {
op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
}
}
ptr += sizeof(xlog_op_header_t) + op_len;
}
} /* xlog_verify_iclog */
#endif
/*
* Mark all iclogs IOERROR. l_icloglock is held by the caller.
*/
STATIC int
xlog_state_ioerror(
struct xlog *log)
{
xlog_in_core_t *iclog, *ic;
iclog = log->l_iclog;
if (iclog->ic_state != XLOG_STATE_IOERROR) {
/*
* Mark all the incore logs IOERROR.
* From now on, no log flushes will result.
*/
ic = iclog;
do {
ic->ic_state = XLOG_STATE_IOERROR;
ic = ic->ic_next;
} while (ic != iclog);
return 0;
}
/*
* Return non-zero, if state transition has already happened.
*/
return 1;
}
/*
* This is called from xfs_force_shutdown, when we're forcibly
* shutting down the filesystem, typically because of an IO error.
* Our main objectives here are to make sure that:
* a. if !logerror, flush the logs to disk. Anything modified
* after this is ignored.
* b. the filesystem gets marked 'SHUTDOWN' for all interested
* parties to find out, 'atomically'.
* c. those who're sleeping on log reservations, pinned objects and
* other resources get woken up, and be told the bad news.
* d. nothing new gets queued up after (b) and (c) are done.
*
* Note: for the !logerror case we need to flush the regions held in memory out
* to disk first. This needs to be done before the log is marked as shutdown,
* otherwise the iclog writes will fail.
*/
int
xfs_log_force_umount(
struct xfs_mount *mp,
int logerror)
{
struct xlog *log;
int retval;
log = mp->m_log;
/*
* If this happens during log recovery, don't worry about
* locking; the log isn't open for business yet.
*/
if (!log ||
log->l_flags & XLOG_ACTIVE_RECOVERY) {
mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
if (mp->m_sb_bp)
mp->m_sb_bp->b_flags |= XBF_DONE;
return 0;
}
/*
* Somebody could've already done the hard work for us.
* No need to get locks for this.
*/
if (logerror && log->l_iclog->ic_state == XLOG_STATE_IOERROR) {
ASSERT(XLOG_FORCED_SHUTDOWN(log));
return 1;
}
/*
* Flush all the completed transactions to disk before marking the log
* being shut down. We need to do it in this order to ensure that
* completed operations are safely on disk before we shut down, and that
* we don't have to issue any buffer IO after the shutdown flags are set
* to guarantee this.
*/
if (!logerror)
xfs_log_force(mp, XFS_LOG_SYNC);
/*
* mark the filesystem and the as in a shutdown state and wake
* everybody up to tell them the bad news.
*/
spin_lock(&log->l_icloglock);
mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
if (mp->m_sb_bp)
mp->m_sb_bp->b_flags |= XBF_DONE;
/*
* Mark the log and the iclogs with IO error flags to prevent any
* further log IO from being issued or completed.
*/
log->l_flags |= XLOG_IO_ERROR;
retval = xlog_state_ioerror(log);
spin_unlock(&log->l_icloglock);
/*
* We don't want anybody waiting for log reservations after this. That
* means we have to wake up everybody queued up on reserveq as well as
* writeq. In addition, we make sure in xlog_{re}grant_log_space that
* we don't enqueue anything once the SHUTDOWN flag is set, and this
* action is protected by the grant locks.
*/
xlog_grant_head_wake_all(&log->l_reserve_head);
xlog_grant_head_wake_all(&log->l_write_head);
/*
* Wake up everybody waiting on xfs_log_force. Wake the CIL push first
* as if the log writes were completed. The abort handling in the log
* item committed callback functions will do this again under lock to
* avoid races.
*/
spin_lock(&log->l_cilp->xc_push_lock);
wake_up_all(&log->l_cilp->xc_commit_wait);
spin_unlock(&log->l_cilp->xc_push_lock);
xlog_state_do_callback(log, true);
/* return non-zero if log IOERROR transition had already happened */
return retval;
}
STATIC int
xlog_iclogs_empty(
struct xlog *log)
{
xlog_in_core_t *iclog;
iclog = log->l_iclog;
do {
/* endianness does not matter here, zero is zero in
* any language.
*/
if (iclog->ic_header.h_num_logops)
return 0;
iclog = iclog->ic_next;
} while (iclog != log->l_iclog);
return 1;
}
xfs: validate metadata LSNs against log on v5 superblocks Since the onset of v5 superblocks, the LSN of the last modification has been included in a variety of on-disk data structures. This LSN is used to provide log recovery ordering guarantees (e.g., to ensure an older log recovery item is not replayed over a newer target data structure). While this works correctly from the point a filesystem is formatted and mounted, userspace tools have some problematic behaviors that defeat this mechanism. For example, xfs_repair historically zeroes out the log unconditionally (regardless of whether corruption is detected). If this occurs, the LSN of the filesystem is reset and the log is now in a problematic state with respect to on-disk metadata structures that might have a larger LSN. Until either the log catches up to the highest previously used metadata LSN or each affected data structure is modified and written out without incident (which resets the metadata LSN), log recovery is susceptible to filesystem corruption. This problem is ultimately addressed and repaired in the associated userspace tools. The kernel is still responsible to detect the problem and notify the user that something is wrong. Check the superblock LSN at mount time and fail the mount if it is invalid. From that point on, trigger verifier failure on any metadata I/O where an invalid LSN is detected. This results in a filesystem shutdown and guarantees that we do not log metadata changes with invalid LSNs on disk. Since this is a known issue with a known recovery path, present a warning to instruct the user how to recover. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-10-12 11:59:25 +07:00
/*
* Verify that an LSN stamped into a piece of metadata is valid. This is
* intended for use in read verifiers on v5 superblocks.
*/
bool
xfs_log_check_lsn(
struct xfs_mount *mp,
xfs_lsn_t lsn)
{
struct xlog *log = mp->m_log;
bool valid;
/*
* norecovery mode skips mount-time log processing and unconditionally
* resets the in-core LSN. We can't validate in this mode, but
* modifications are not allowed anyways so just return true.
*/
if (mp->m_flags & XFS_MOUNT_NORECOVERY)
return true;
/*
* Some metadata LSNs are initialized to NULL (e.g., the agfl). This is
* handled by recovery and thus safe to ignore here.
*/
if (lsn == NULLCOMMITLSN)
return true;
valid = xlog_valid_lsn(mp->m_log, lsn);
/* warn the user about what's gone wrong before verifier failure */
if (!valid) {
spin_lock(&log->l_icloglock);
xfs_warn(mp,
"Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). "
"Please unmount and run xfs_repair (>= v4.3) to resolve.",
CYCLE_LSN(lsn), BLOCK_LSN(lsn),
log->l_curr_cycle, log->l_curr_block);
spin_unlock(&log->l_icloglock);
}
return valid;
}
bool
xfs_log_in_recovery(
struct xfs_mount *mp)
{
struct xlog *log = mp->m_log;
return log->l_flags & XLOG_ACTIVE_RECOVERY;
}