linux/fs/xfs/xfs_log.c

3877 lines
104 KiB
C
Raw Normal View History

/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it would be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_types.h"
#include "xfs_log.h"
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_mount.h"
#include "xfs_error.h"
#include "xfs_log_priv.h"
#include "xfs_buf_item.h"
#include "xfs_bmap_btree.h"
#include "xfs_alloc_btree.h"
#include "xfs_ialloc_btree.h"
#include "xfs_log_recover.h"
#include "xfs_trans_priv.h"
#include "xfs_dinode.h"
#include "xfs_inode.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>
2009-12-14 23:14:59 +00:00
#include "xfs_trace.h"
#include "xfs_fsops.h"
2012-11-12 11:54:24 +00:00
#include "xfs_cksum.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 int
xlog_sync(
struct xlog *log,
struct xlog_in_core *iclog);
STATIC void
xlog_dealloc_log(
struct xlog *log);
/* local state machine functions */
STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
STATIC void
xlog_state_do_callback(
struct xlog *log,
int aborted,
struct xlog_in_core *iclog);
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 int
xlog_state_release_iclog(
struct xlog *log,
struct xlog_in_core *iclog);
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);
#if defined(DEBUG)
STATIC void
xlog_verify_dest_ptr(
struct xlog *log,
char *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,
boolean_t syncing);
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,d)
#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;
list_for_each_entry(tic, &head->waiters, t_queue) {
need_bytes = xlog_ticket_reservation(log, head, tic);
if (*free_bytes < need_bytes)
return false;
*free_bytes -= need_bytes;
trace_xfs_log_grant_wake_up(log, tic);
wake_up_process(tic->t_task);
}
return true;
}
STATIC int
xlog_grant_head_wait(
struct xlog *log,
struct xlog_grant_head *head,
struct xlog_ticket *tic,
int need_bytes)
{
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(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 XFS_ERROR(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 XFS_ERROR(EIO);
XFS_STATS_INC(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 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,
uint t_type)
{
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 XFS_ERROR(EIO);
XFS_STATS_INC(xs_try_logspace);
ASSERT(*ticp == NULL);
tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent,
KM_SLEEP | KM_MAYFAIL);
if (!tic)
return XFS_ERROR(ENOMEM);
tic->t_trans_type = t_type;
*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 02:01:00 +00:00
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,
uint flags)
{
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;
if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
flags |= XFS_LOG_REL_PERM_RESERV;
}
}
if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
(flags & XFS_LOG_REL_PERM_RESERV)) {
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>
2009-12-14 23:14:59 +00:00
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);
xfs_log_ticket_put(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>
2009-12-14 23:14:59 +00:00
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;
}
return lsn;
}
/*
* Attaches a new iclog I/O completion callback routine during
* transaction commit. If the log is in error state, a non-zero
* return code is handed back and the caller is responsible for
* executing the callback at an appropriate time.
*/
int
xfs_log_notify(
struct xfs_mount *mp,
struct xlog_in_core *iclog,
xfs_log_callback_t *cb)
{
int abortflg;
spin_lock(&iclog->ic_callback_lock);
abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
if (!abortflg) {
ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
(iclog->ic_state == XLOG_STATE_WANT_SYNC));
cb->cb_next = NULL;
*(iclog->ic_callback_tail) = cb;
iclog->ic_callback_tail = &(cb->cb_next);
}
spin_unlock(&iclog->ic_callback_lock);
return abortflg;
}
int
xfs_log_release_iclog(
struct xfs_mount *mp,
struct xlog_in_core *iclog)
{
if (xlog_state_release_iclog(mp->m_log, iclog)) {
xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
return EIO;
}
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 01:13:32 +00:00
xfs_log_mount(
xfs_mount_t *mp,
xfs_buftarg_t *log_target,
xfs_daddr_t blk_offset,
int num_bblks)
{
[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 01:13:32 +00:00
int error;
if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
xfs_notice(mp, "Mounting Filesystem");
else {
xfs_notice(mp,
"Mounting filesystem in no-recovery mode. Filesystem will be inconsistent.");
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;
}
[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 01:13:32 +00:00
/*
* 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 01:13:32 +00:00
}
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 01:13:32 +00:00
/*
* 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 01:13:32 +00:00
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);
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 04:37:18 +00:00
/*
* 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 01:13:32 +00:00
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(xfs_mount_t *mp)
{
int error = 0;
if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
error = xlog_recover_finish(mp->m_log);
if (!error)
xfs_log_work_queue(mp);
} else {
ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
}
return error;
}
/*
* 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.
*/
/*
* 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 "nUmount" is a bit foo.
* As far as I know, there weren't any dependencies on the old behaviour.
*/
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
xlog_ticket_t *tic = NULL;
xfs_lsn_t lsn;
int error;
/*
* Don't write out unmount record on read-only mounts.
* Or, if we are doing a forced umount (typically because of IO errors).
*/
if (mp->m_flags & XFS_MOUNT_RDONLY)
return 0;
error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
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))) {
error = xfs_log_reserve(mp, 600, 1, &tic,
XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
if (!error) {
/* the data section must be 32 bit size aligned */
struct {
__uint16_t magic;
__uint16_t pad1;
__uint32_t pad2; /* may as well make it 64 bits */
} 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,
};
/* 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, XLOG_UNMOUNT_TRANS);
/*
* At this point, we're umounting anyway,
* so there's no point in transitioning log state
* to IOERROR. Just continue...
*/
}
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);
spin_unlock(&log->l_icloglock);
error = xlog_state_release_iclog(log, iclog);
spin_lock(&log->l_icloglock);
if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)) {
if (!XLOG_FORCED_SHUTDOWN(log)) {
xlog_wait(&iclog->ic_force_wait,
&log->l_icloglock);
} else {
spin_unlock(&log->l_icloglock);
}
} else {
spin_unlock(&log->l_icloglock);
}
if (tic) {
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>
2009-12-14 23:14:59 +00:00
trace_xfs_log_umount_write(log, tic);
xlog_ungrant_log_space(log, tic);
xfs_log_ticket_put(tic);
}
} 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);
spin_unlock(&log->l_icloglock);
error = xlog_state_release_iclog(log, iclog);
spin_lock(&log->l_icloglock);
if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
|| iclog->ic_state == XLOG_STATE_DIRTY
|| iclog->ic_state == XLOG_STATE_IOERROR) ) {
xlog_wait(&iclog->ic_force_wait,
&log->l_icloglock);
} else {
spin_unlock(&log->l_icloglock);
}
}
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 01:13:32 +00:00
xfs_trans_ail_destroy(mp);
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 04:37:18 +00:00
item->li_lv = NULL;
INIT_LIST_HEAD(&item->li_ail);
INIT_LIST_HEAD(&item->li_cil);
}
/*
* 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);
}
}
/*
* Determine if we have a transaction that has gone to disk
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 05:06:44 +00:00
* that needs to be covered. To begin the transition to the idle state
* firstly the log needs to be idle (no AIL and nothing in the iclogs).
* 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.
*
* Because this is called as part of the sync process, we should also indicate
* that dummy transactions should be issued in anything but the covered or
* idle states. This ensures that the log tail is accurately reflected in
* the log at the end of the sync, hence if a crash occurrs avoids replay
* of transactions where the metadata is already on disk.
*/
int
xfs_log_need_covered(xfs_mount_t *mp)
{
int needed = 0;
struct xlog *log = mp->m_log;
[XFS] Lazy Superblock Counters When we have a couple of hundred transactions on the fly at once, they all typically modify the on disk superblock in some way. create/unclink/mkdir/rmdir modify inode counts, allocation/freeing modify free block counts. When these counts are modified in a transaction, they must eventually lock the superblock buffer and apply the mods. The buffer then remains locked until the transaction is committed into the incore log buffer. The result of this is that with enough transactions on the fly the incore superblock buffer becomes a bottleneck. The result of contention on the incore superblock buffer is that transaction rates fall - the more pressure that is put on the superblock buffer, the slower things go. The key to removing the contention is to not require the superblock fields in question to be locked. We do that by not marking the superblock dirty in the transaction. IOWs, we modify the incore superblock but do not modify the cached superblock buffer. In short, we do not log superblock modifications to critical fields in the superblock on every transaction. In fact we only do it just before we write the superblock to disk every sync period or just before unmount. This creates an interesting problem - if we don't log or write out the fields in every transaction, then how do the values get recovered after a crash? the answer is simple - we keep enough duplicate, logged information in other structures that we can reconstruct the correct count after log recovery has been performed. It is the AGF and AGI structures that contain the duplicate information; after recovery, we walk every AGI and AGF and sum their individual counters to get the correct value, and we do a transaction into the log to correct them. An optimisation of this is that if we have a clean unmount record, we know the value in the superblock is correct, so we can avoid the summation walk under normal conditions and so mount/recovery times do not change under normal operation. One wrinkle that was discovered during development was that the blocks used in the freespace btrees are never accounted for in the AGF counters. This was once a valid optimisation to make; when the filesystem is full, the free space btrees are empty and consume no space. Hence when it matters, the "accounting" is correct. But that means the when we do the AGF summations, we would not have a correct count and xfs_check would complain. Hence a new counter was added to track the number of blocks used by the free space btrees. This is an *on-disk format change*. As a result of this, lazy superblock counters are a mkfs option and at the moment on linux there is no way to convert an old filesystem. This is possible - xfs_db can be used to twiddle the right bits and then xfs_repair will do the format conversion for you. Similarly, you can convert backwards as well. At some point we'll add functionality to xfs_admin to do the bit twiddling easily.... SGI-PV: 964999 SGI-Modid: xfs-linux-melb:xfs-kern:28652a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-05-24 05:26:31 +00:00
if (!xfs_fs_writable(mp))
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 05:06:44 +00:00
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:
if (!xfs_ail_min_lsn(log->l_ailp) &&
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 05:06:44 +00:00
xlog_iclogs_empty(log)) {
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;
}
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 05:06:44 +00:00
/* FALLTHRU */
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 05:06:44 +00:00
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->xa_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);
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->xa_lock);
tail_lsn = xlog_assign_tail_lsn_locked(mp);
spin_unlock(&mp->m_ail->xa_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\n"
" tail_cycle = %d, tail_bytes = %d\n"
" GH cycle = %d, GH bytes = %d",
tail_cycle, tail_bytes, head_cycle, head_bytes);
ASSERT(0);
free_bytes = log->l_logsize;
}
return free_bytes;
}
/*
* Log function which is called when an io completes.
*
* The log manager needs its own routine, in order to control what
* happens with the buffer after the write completes.
*/
void
xlog_iodone(xfs_buf_t *bp)
{
struct xlog_in_core *iclog = bp->b_fspriv;
struct xlog *l = iclog->ic_log;
int aborted = 0;
/*
* Race to shutdown the filesystem if we see an error.
*/
if (XFS_TEST_ERROR((xfs_buf_geterror(bp)), l->l_mp,
XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
xfs_buf_ioerror_alert(bp, __func__);
xfs_buf_stale(bp);
xfs_force_shutdown(l->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 = XFS_LI_ABORTED;
} else if (iclog->ic_state & XLOG_STATE_IOERROR) {
aborted = XFS_LI_ABORTED;
}
/* log I/O is always issued ASYNC */
ASSERT(XFS_BUF_ISASYNC(bp));
xlog_state_done_syncing(iclog, aborted);
/*
* do not reference the buffer (bp) here as we could race
* with it being freed after writing the unmount record to the
* log.
*/
}
/*
* 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)
{
int size;
int xhdrs;
if (mp->m_logbufs <= 0)
log->l_iclog_bufs = XLOG_MAX_ICLOGS;
else
log->l_iclog_bufs = mp->m_logbufs;
/*
* Buffer size passed in from mount system call.
*/
if (mp->m_logbsize > 0) {
size = log->l_iclog_size = mp->m_logbsize;
log->l_iclog_size_log = 0;
while (size != 1) {
log->l_iclog_size_log++;
size >>= 1;
}
if (xfs_sb_version_haslogv2(&mp->m_sb)) {
/* # headers = size / 32k
* one header holds cycles from 32k of data
*/
xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
xhdrs++;
log->l_iclog_hsize = xhdrs << BBSHIFT;
log->l_iclog_heads = xhdrs;
} else {
ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
log->l_iclog_hsize = BBSIZE;
log->l_iclog_heads = 1;
}
goto done;
}
/* All machines use 32kB buffers by default. */
log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
/* the default log size is 16k or 32k which is one header sector */
log->l_iclog_hsize = BBSIZE;
log->l_iclog_heads = 1;
done:
/* are we being asked to make the sizes selected above visible? */
if (mp->m_logbufs == 0)
mp->m_logbufs = log->l_iclog_bufs;
if (mp->m_logbsize == 0)
mp->m_logbsize = log->l_iclog_size;
} /* xlog_get_iclog_buffer_size */
void
xfs_log_work_queue(
struct xfs_mount *mp)
{
queue_delayed_work(mp->m_log_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.
*/
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))
xfs_fs_log_dummy(mp);
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;
xfs_buf_t *bp;
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);
error = ENOMEM;
bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0);
if (!bp)
goto out_free_log;
bp->b_iodone = xlog_iodone;
ASSERT(xfs_buf_islocked(bp));
log->l_xbuf = bp;
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++) {
*iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
if (!*iclogp)
goto out_free_iclog;
iclog = *iclogp;
iclog->ic_prev = prev_iclog;
prev_iclog = iclog;
bp = xfs_buf_get_uncached(mp->m_logdev_targp,
BTOBB(log->l_iclog_size), 0);
if (!bp)
goto out_free_iclog;
bp->b_iodone = xlog_iodone;
iclog->ic_bp = bp;
iclog->ic_data = bp->b_addr;
#ifdef DEBUG
log->l_iclog_bak[i] = (xfs_caddr_t)&(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 = BBTOB(bp->b_length) - 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);
iclog->ic_callback_tail = &(iclog->ic_callback);
iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
ASSERT(xfs_buf_islocked(iclog->ic_bp));
init_waitqueue_head(&iclog->ic_force_wait);
init_waitqueue_head(&iclog->ic_write_wait);
iclogp = &iclog->ic_next;
}
*iclogp = log->l_iclog; /* complete ring */
log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
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 04:37:18 +00:00
error = xlog_cil_init(log);
if (error)
goto out_free_iclog;
return log;
out_free_iclog:
for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
prev_iclog = iclog->ic_next;
if (iclog->ic_bp)
xfs_buf_free(iclog->ic_bp);
kmem_free(iclog);
}
spinlock_destroy(&log->l_icloglock);
xfs_buf_free(log->l_xbuf);
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 11:54:24 +00:00
/*
* 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;
xfs_caddr_t dp;
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 11:54:24 +00:00
xlog_cksum(
struct xlog *log,
struct xlog_rec_header *rhead,
char *dp,
int size)
{
__uint32_t crc;
/* first generate the crc for the record header ... */
crc = xfs_start_cksum((char *)rhead,
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;
for (i = 1; i < log->l_iclog_heads; i++) {
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);
}
/*
* The bdstrat callback function for log bufs. This gives us a central
* place to trap bufs in case we get hit by a log I/O error and need to
* shutdown. Actually, in practice, even when we didn't get a log error,
* we transition the iclogs to IOERROR state *after* flushing all existing
* iclogs to disk. This is because we don't want anymore new transactions to be
* started or completed afterwards.
*/
STATIC int
xlog_bdstrat(
struct xfs_buf *bp)
{
struct xlog_in_core *iclog = bp->b_fspriv;
if (iclog->ic_state & XLOG_STATE_IOERROR) {
xfs_buf_ioerror(bp, EIO);
xfs_buf_stale(bp);
xfs_buf_ioend(bp, 0);
/*
* 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.
*/
return 0;
}
xfs_buf_iorequest(bp);
return 0;
}
/*
* 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
* log_sync (and then only bwrite()) know about the fact that the log may
* not start with block zero on a given device. The log block start offset
* is added immediately before calling bwrite().
*/
STATIC int
xlog_sync(
struct xlog *log,
struct xlog_in_core *iclog)
{
xfs_buf_t *bp;
int i;
uint count; /* byte count of bwrite */
uint count_init; /* initial count before roundup */
int roundoff; /* roundoff to BB or stripe */
int split = 0; /* split write into two regions */
int error;
int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
2012-11-12 11:54:24 +00:00
int size;
XFS_STATS_INC(xs_log_writes);
ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
/* Add for LR header */
count_init = log->l_iclog_hsize + iclog->ic_offset;
/* Round out the log write size */
if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
/* we have a v2 stripe unit to use */
count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
} else {
count = BBTOB(BTOBB(count_init));
}
roundoff = count - count_init;
ASSERT(roundoff >= 0);
ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
roundoff < log->l_mp->m_sb.sb_logsunit)
||
(log->l_mp->m_sb.sb_logsunit <= 1 &&
roundoff < BBTOB(1)));
/* 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 11:54:24 +00:00
size = iclog->ic_offset;
if (v2)
size += roundoff;
iclog->ic_header.h_len = cpu_to_be32(size);
bp = iclog->ic_bp;
XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
/* Do we need to split this write into 2 parts? */
if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
2012-11-12 11:54:24 +00:00
char *dptr;
split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
2012-11-12 11:54:24 +00:00
iclog->ic_bwritecnt = 2;
/*
* Bump the cycle numbers at the start of each block in the
* part of the iclog that ends up in the buffer that gets
* written to the start of the log.
*
* Watch out for the header magic number case, though.
*/
dptr = (char *)&iclog->ic_header + count;
for (i = 0; i < split; i += BBSIZE) {
__uint32_t cycle = be32_to_cpu(*(__be32 *)dptr);
if (++cycle == XLOG_HEADER_MAGIC_NUM)
cycle++;
*(__be32 *)dptr = cpu_to_be32(cycle);
dptr += BBSIZE;
}
} else {
iclog->ic_bwritecnt = 1;
}
2012-11-12 11:54:24 +00:00
/* calculcate the checksum */
iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
iclog->ic_datap, size);
bp->b_io_length = BTOBB(count);
bp->b_fspriv = iclog;
XFS_BUF_ZEROFLAGS(bp);
XFS_BUF_ASYNC(bp);
bp->b_flags |= XBF_SYNCIO;
if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
bp->b_flags |= XBF_FUA;
/*
* 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_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
else
bp->b_flags |= XBF_FLUSH;
}
ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
xlog_verify_iclog(log, iclog, count, B_TRUE);
/* account for log which doesn't start at block #0 */
XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
/*
* Don't call xfs_bwrite here. We do log-syncs even when the filesystem
* is shutting down.
*/
XFS_BUF_WRITE(bp);
error = xlog_bdstrat(bp);
if (error) {
xfs_buf_ioerror_alert(bp, "xlog_sync");
return error;
}
if (split) {
bp = iclog->ic_log->l_xbuf;
XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
xfs_buf_associate_memory(bp,
(char *)&iclog->ic_header + count, split);
bp->b_fspriv = iclog;
XFS_BUF_ZEROFLAGS(bp);
XFS_BUF_ASYNC(bp);
bp->b_flags |= XBF_SYNCIO;
if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
bp->b_flags |= XBF_FUA;
ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
/* account for internal log which doesn't start at block #0 */
XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
XFS_BUF_WRITE(bp);
error = xlog_bdstrat(bp);
if (error) {
xfs_buf_ioerror_alert(bp, "xlog_sync (split)");
return error;
}
}
return 0;
} /* xlog_sync */
/*
* 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 04:37:18 +00:00
xlog_cil_destroy(log);
/*
* always need to ensure that the extra buffer does not point to memory
* owned by another log buffer before we free it.
*/
xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
xfs_buf_free(log->l_xbuf);
iclog = log->l_iclog;
for (i=0; i<log->l_iclog_bufs; i++) {
xfs_buf_free(iclog->ic_bp);
next_iclog = iclog->ic_next;
kmem_free(iclog);
iclog = next_iclog;
}
spinlock_destroy(&log->l_icloglock);
log->l_mp->m_log = NULL;
kmem_free(log);
} /* xlog_dealloc_log */
/*
* Update counters atomically now that memcpy is done.
*/
/* ARGSUSED */
static inline void
xlog_state_finish_copy(
struct xlog *log,
struct xlog_in_core *iclog,
int record_cnt,
int copy_bytes)
{
spin_lock(&log->l_icloglock);
be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
iclog->ic_offset += copy_bytes;
spin_unlock(&log->l_icloglock);
} /* xlog_state_finish_copy */
/*
* 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 04:37:18 +00:00
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 */
static char *res_type_str[XLOG_REG_TYPE_MAX] = {
"bformat",
"bchunk",
"efi_format",
"efd_format",
"iformat",
"icore",
"iext",
"ibroot",
"ilocal",
"iattr_ext",
"iattr_broot",
"iattr_local",
"qformat",
"dquot",
"quotaoff",
"LR header",
"unmount",
"commit",
"trans header"
};
static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
"SETATTR_NOT_SIZE",
"SETATTR_SIZE",
"INACTIVE",
"CREATE",
"CREATE_TRUNC",
"TRUNCATE_FILE",
"REMOVE",
"LINK",
"RENAME",
"MKDIR",
"RMDIR",
"SYMLINK",
"SET_DMATTRS",
"GROWFS",
"STRAT_WRITE",
"DIOSTRAT",
"WRITE_SYNC",
"WRITEID",
"ADDAFORK",
"ATTRINVAL",
"ATRUNCATE",
"ATTR_SET",
"ATTR_RM",
"ATTR_FLAG",
"CLEAR_AGI_BUCKET",
"QM_SBCHANGE",
"DUMMY1",
"DUMMY2",
"QM_QUOTAOFF",
"QM_DQALLOC",
"QM_SETQLIM",
"QM_DQCLUSTER",
"QM_QINOCREATE",
"QM_QUOTAOFF_END",
"SB_UNIT",
"FSYNC_TS",
"GROWFSRT_ALLOC",
"GROWFSRT_ZERO",
"GROWFSRT_FREE",
"SWAPEXT"
};
xfs_warn(mp,
"xlog_write: reservation summary:\n"
" trans type = %s (%u)\n"
" unit res = %d bytes\n"
" current res = %d bytes\n"
" total reg = %u bytes (o/flow = %u bytes)\n"
" ophdrs = %u (ophdr space = %u bytes)\n"
" ophdr + reg = %u bytes\n"
" num regions = %u\n",
((ticket->t_trans_type <= 0 ||
ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
"bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
ticket->t_trans_type,
ticket->t_unit_res,
ticket->t_curr_res,
ticket->t_res_arr_sum, ticket->t_res_o_flow,
ticket->t_res_num_ophdrs, ophdr_spc,
ticket->t_res_arr_sum +
ticket->t_res_o_flow + ophdr_spc,
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\n", i,
((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
"bad-rtype" : res_type_str[r_type-1]),
ticket->t_res_arr[i].r_len);
}
xfs_alert_tag(mp, XFS_PTAG_LOGRES,
"xlog_write: reservation ran out. Need to up reservation");
xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
}
/*
* 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) {
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 0x%p",
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
* 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)
{
if (*partial_copy) {
/*
* This iclog has already been marked WANT_SYNC by
* xlog_state_get_iclog_space.
*/
xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
*record_cnt = 0;
*data_cnt = 0;
return xlog_state_release_iclog(log, 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. */
xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
*record_cnt = 0;
*data_cnt = 0;
spin_lock(&log->l_icloglock);
xlog_state_want_sync(log, iclog);
spin_unlock(&log->l_icloglock);
if (!commit_iclog)
return xlog_state_release_iclog(log, iclog);
ASSERT(flags & XLOG_COMMIT_TRANS);
*commit_iclog = iclog;
}
return 0;
}
/*
* 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 04:37:18 +00:00
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;
*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 04:37:18 +00:00
/*
* 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 04:37:18 +00:00
if (ticket->t_curr_res < 0)
xlog_print_tic_res(log->l_mp, ticket);
index = 0;
lv = log_vector;
vecp = lv->lv_iovecp;
while (lv && 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().
*/
while (lv && index < lv->lv_niovecs) {
struct xfs_log_iovec *reg = &vecp[index];
struct xlog_op_header *ophdr;
int start_rec_copy;
int copy_len;
int copy_off;
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 XFS_ERROR(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 */
ASSERT(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) {
lv = lv->lv_next;
index = 0;
if (lv)
vecp = lv->lv_iovecp;
}
if (record_cnt == 0) {
if (!lv)
return 0;
break;
}
}
}
ASSERT(len == 0);
xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
if (!commit_iclog)
return xlog_state_release_iclog(log, iclog);
ASSERT(flags & XLOG_COMMIT_TRANS);
*commit_iclog = iclog;
return 0;
}
/*****************************************************************************
*
* State Machine functions
*
*****************************************************************************
*/
/* Clean iclogs starting from the head. This ordering 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.
*
* State Change: DIRTY -> ACTIVE
*/
STATIC void
xlog_state_clean_log(
struct xlog *log)
{
xlog_in_core_t *iclog;
int changed = 0;
iclog = log->l_iclog;
do {
if (iclog->ic_state == XLOG_STATE_DIRTY) {
iclog->ic_state = XLOG_STATE_ACTIVE;
iclog->ic_offset = 0;
ASSERT(iclog->ic_callback == NULL);
/*
* 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);
/* log is locked when we are called */
/*
* 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);
}
}
} /* xlog_state_clean_log */
STATIC xfs_lsn_t
xlog_get_lowest_lsn(
struct xlog *log)
{
xlog_in_core_t *lsn_log;
xfs_lsn_t lowest_lsn, lsn;
lsn_log = log->l_iclog;
lowest_lsn = 0;
do {
if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
if ((lsn && !lowest_lsn) ||
(XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
lowest_lsn = lsn;
}
}
lsn_log = lsn_log->ic_next;
} while (lsn_log != log->l_iclog);
return lowest_lsn;
}
STATIC void
xlog_state_do_callback(
struct xlog *log,
int aborted,
struct xlog_in_core *ciclog)
{
xlog_in_core_t *iclog;
xlog_in_core_t *first_iclog; /* used to know when we've
* processed all iclogs once */
xfs_log_callback_t *cb, *cb_next;
int flushcnt = 0;
xfs_lsn_t lowest_lsn;
int ioerrors; /* counter: iclogs with errors */
int loopdidcallbacks; /* flag: inner loop did callbacks*/
int funcdidcallbacks; /* flag: function did callbacks */
int repeats; /* for issuing console warnings if
* looping too many times */
int wake = 0;
spin_lock(&log->l_icloglock);
first_iclog = iclog = log->l_iclog;
ioerrors = 0;
funcdidcallbacks = 0;
repeats = 0;
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;
loopdidcallbacks = 0;
repeats++;
do {
/* skip all iclogs in the ACTIVE & DIRTY states */
if (iclog->ic_state &
(XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
iclog = iclog->ic_next;
continue;
}
/*
* 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.
*/
if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
/*
* Can only perform callbacks in order. Since
* this iclog is not in the DONE_SYNC/
* DO_CALLBACK state, we skip the rest and
* just try to clean up. If we set our iclog
* to DO_CALLBACK, we will not process it when
* we retry since a previous iclog is in the
* CALLBACK and the state cannot change since
* we are holding the l_icloglock.
*/
if (!(iclog->ic_state &
(XLOG_STATE_DONE_SYNC |
XLOG_STATE_DO_CALLBACK))) {
if (ciclog && (ciclog->ic_state ==
XLOG_STATE_DONE_SYNC)) {
ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
}
break;
}
/*
* We now have an iclog that is in either the
* DO_CALLBACK or DONE_SYNC states. The other
* states (WANT_SYNC, SYNCING, or CALLBACK were
* caught by the above if and are going to
* clean (i.e. we aren't doing their callbacks)
* see the above if.
*/
/*
* We will do one more check here to see if we
* have chased our tail around.
*/
lowest_lsn = xlog_get_lowest_lsn(log);
if (lowest_lsn &&
XFS_LSN_CMP(lowest_lsn,
be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
iclog = iclog->ic_next;
continue; /* Leave this iclog for
* another thread */
}
iclog->ic_state = XLOG_STATE_CALLBACK;
/*
xfs: only update the last_sync_lsn when a transaction completes The log write code stamps each iclog with the current tail LSN in the iclog header so that recovery knows where to find the tail of thelog once it has found the head. Normally this is taken from the first item on the AIL - the log item that corresponds to the oldest active item in the log. The problem is that when the AIL is empty, the tail lsn is dervied from the the l_last_sync_lsn, which is the LSN of the last iclog to be written to the log. In most cases this doesn't happen, because the AIL is rarely empty on an active filesystem. However, when it does, it opens up an interesting case when the transaction being committed to the iclog spans multiple iclogs. That is, the first iclog is stamped with the l_last_sync_lsn, and IO is issued. Then the next iclog is setup, the changes copied into the iclog (takes some time), and then the l_last_sync_lsn is stamped into the header and IO is issued. This is still the same transaction, so the tail lsn of both iclogs must be the same for log recovery to find the entire transaction to be able to replay it. The problem arises in that the iclog buffer IO completion updates the l_last_sync_lsn with it's own LSN. Therefore, If the first iclog completes it's IO before the second iclog is filled and has the tail lsn stamped in it, it will stamp the LSN of the first iclog into it's tail lsn field. If the system fails at this point, log recovery will not see a complete transaction, so the transaction will no be replayed. The fix is simple - the l_last_sync_lsn is updated when a iclog buffer IO completes, and this is incorrect. The l_last_sync_lsn shoul dbe updated when a transaction is completed by a iclog buffer IO. That is, only iclog buffers that have transaction commit callbacks attached to them should update the l_last_sync_lsn. This means that the last_sync_lsn will only move forward when a commit record it written, not in the middle of a large transaction that is rolling through multiple iclog buffers. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-10-08 10:56:12 +00:00
* 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 th etail 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.
*/
ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
xfs: only update the last_sync_lsn when a transaction completes The log write code stamps each iclog with the current tail LSN in the iclog header so that recovery knows where to find the tail of thelog once it has found the head. Normally this is taken from the first item on the AIL - the log item that corresponds to the oldest active item in the log. The problem is that when the AIL is empty, the tail lsn is dervied from the the l_last_sync_lsn, which is the LSN of the last iclog to be written to the log. In most cases this doesn't happen, because the AIL is rarely empty on an active filesystem. However, when it does, it opens up an interesting case when the transaction being committed to the iclog spans multiple iclogs. That is, the first iclog is stamped with the l_last_sync_lsn, and IO is issued. Then the next iclog is setup, the changes copied into the iclog (takes some time), and then the l_last_sync_lsn is stamped into the header and IO is issued. This is still the same transaction, so the tail lsn of both iclogs must be the same for log recovery to find the entire transaction to be able to replay it. The problem arises in that the iclog buffer IO completion updates the l_last_sync_lsn with it's own LSN. Therefore, If the first iclog completes it's IO before the second iclog is filled and has the tail lsn stamped in it, it will stamp the LSN of the first iclog into it's tail lsn field. If the system fails at this point, log recovery will not see a complete transaction, so the transaction will no be replayed. The fix is simple - the l_last_sync_lsn is updated when a iclog buffer IO completes, and this is incorrect. The l_last_sync_lsn shoul dbe updated when a transaction is completed by a iclog buffer IO. That is, only iclog buffers that have transaction commit callbacks attached to them should update the l_last_sync_lsn. This means that the last_sync_lsn will only move forward when a commit record it written, not in the middle of a large transaction that is rolling through multiple iclog buffers. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ben Myers <bpm@sgi.com>
2012-10-08 10:56:12 +00:00
if (iclog->ic_callback)
atomic64_set(&log->l_last_sync_lsn,
be64_to_cpu(iclog->ic_header.h_lsn));
} else
ioerrors++;
spin_unlock(&log->l_icloglock);
/*
* Keep processing entries in the 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.
*/
spin_lock(&iclog->ic_callback_lock);
cb = iclog->ic_callback;
while (cb) {
iclog->ic_callback_tail = &(iclog->ic_callback);
iclog->ic_callback = NULL;
spin_unlock(&iclog->ic_callback_lock);
/* perform callbacks in the order given */
for (; cb; cb = cb_next) {
cb_next = cb->cb_next;
cb->cb_func(cb->cb_arg, aborted);
}
spin_lock(&iclog->ic_callback_lock);
cb = iclog->ic_callback;
}
loopdidcallbacks++;
funcdidcallbacks++;
spin_lock(&log->l_icloglock);
ASSERT(iclog->ic_callback == NULL);
spin_unlock(&iclog->ic_callback_lock);
if (!(iclog->ic_state & XLOG_STATE_IOERROR))
iclog->ic_state = XLOG_STATE_DIRTY;
/*
* Transition from DIRTY to ACTIVE if applicable.
* NOP if STATE_IOERROR.
*/
xlog_state_clean_log(log);
/* wake up threads waiting in xfs_log_force() */
wake_up_all(&iclog->ic_force_wait);
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 (!ioerrors && loopdidcallbacks);
/*
* make one last gasp attempt to see if iclogs are being left in
* limbo..
*/
#ifdef DEBUG
if (funcdidcallbacks) {
first_iclog = iclog = log->l_iclog;
do {
ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
/*
* Terminate the loop if iclogs are found in states
* which will cause other threads to clean up iclogs.
*
* SYNCING - i/o completion will go through logs
* DONE_SYNC - interrupt thread should be waiting for
* l_icloglock
* IOERROR - give up hope all ye who enter here
*/
if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
iclog->ic_state == XLOG_STATE_SYNCING ||
iclog->ic_state == XLOG_STATE_DONE_SYNC ||
iclog->ic_state == XLOG_STATE_IOERROR )
break;
iclog = iclog->ic_next;
} while (first_iclog != iclog);
}
#endif
if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
wake = 1;
spin_unlock(&log->l_icloglock);
if (wake)
wake_up_all(&log->l_flush_wait);
}
/*
* 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(
xlog_in_core_t *iclog,
int aborted)
{
struct xlog *log = iclog->ic_log;
spin_lock(&log->l_icloglock);
ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
iclog->ic_state == XLOG_STATE_IOERROR);
ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
/*
* 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_IOERROR) {
if (--iclog->ic_bwritecnt == 1) {
spin_unlock(&log->l_icloglock);
return;
}
iclog->ic_state = XLOG_STATE_DONE_SYNC;
}
/*
* 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, iclog); /* 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;
int error;
restart:
spin_lock(&log->l_icloglock);
if (XLOG_FORCED_SHUTDOWN(log)) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
iclog = log->l_iclog;
if (iclog->ic_state != XLOG_STATE_ACTIVE) {
XFS_STATS_INC(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)) {
xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
/*
* If I'm 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)) {
/* we are the only one */
spin_unlock(&log->l_icloglock);
error = xlog_state_release_iclog(log, iclog);
if (error)
return error;
} else {
spin_unlock(&log->l_icloglock);
}
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>
2009-12-14 23:14:59 +00:00
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>
2009-12-14 23:14:59 +00:00
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>
2009-12-14 23:14:59 +00:00
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>
2009-12-14 23:14:59 +00:00
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>
2009-12-14 23:14:59 +00:00
trace_xfs_log_ungrant_exit(log, ticket);
xfs_log_space_wake(log->l_mp);
}
/*
* Flush iclog to disk if this is the last reference to the given iclog and
* the WANT_SYNC bit is set.
*
* When this function is entered, the iclog is not necessarily in the
* WANT_SYNC state. It may be sitting around waiting to get filled.
*
*
*/
STATIC int
xlog_state_release_iclog(
struct xlog *log,
struct xlog_in_core *iclog)
{
int sync = 0; /* do we sync? */
if (iclog->ic_state & XLOG_STATE_IOERROR)
return XFS_ERROR(EIO);
ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
return 0;
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_WANT_SYNC);
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);
sync++;
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 */
}
spin_unlock(&log->l_icloglock);
/*
* We let the log lock go, so it's possible that we hit a log I/O
* error or some other SHUTDOWN condition that marks the iclog
* as XLOG_STATE_IOERROR before the bwrite. However, we know that
* this iclog has consistent data, so we ignore IOERROR
* flags after this point.
*/
if (sync)
return xlog_sync(log, iclog);
return 0;
} /* xlog_state_release_iclog */
/*
* 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) {
log->l_curr_cycle++;
if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
log->l_curr_cycle++;
log->l_curr_block -= log->l_logBBsize;
ASSERT(log->l_curr_block >= 0);
}
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,
int *log_flushed)
{
struct xlog *log = mp->m_log;
struct xlog_in_core *iclog;
xfs_lsn_t lsn;
XFS_STATS_INC(xs_log_force);
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 04:37:18 +00:00
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
/* If the head iclog is not active nor dirty, we just attach
* ourselves to the head and go to sleep.
*/
if (iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY) {
/*
* 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.
*/
if (iclog->ic_state == XLOG_STATE_DIRTY ||
(atomic_read(&iclog->ic_refcnt) == 0
&& iclog->ic_offset == 0)) {
iclog = iclog->ic_prev;
if (iclog->ic_state == XLOG_STATE_ACTIVE ||
iclog->ic_state == XLOG_STATE_DIRTY)
goto no_sleep;
else
goto maybe_sleep;
} else {
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);
spin_unlock(&log->l_icloglock);
if (xlog_state_release_iclog(log, iclog))
return XFS_ERROR(EIO);
if (log_flushed)
*log_flushed = 1;
spin_lock(&log->l_icloglock);
if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
iclog->ic_state != XLOG_STATE_DIRTY)
goto maybe_sleep;
else
goto no_sleep;
} 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);
goto maybe_sleep;
}
}
}
/* By the time we come around again, the iclog could've been filled
* which would give it another lsn. If we have a new lsn, just
* return because the relevant data has been flushed.
*/
maybe_sleep:
if (flags & XFS_LOG_SYNC) {
/*
* We must check if we're shutting down here, before
* we wait, while we're holding the l_icloglock.
* Then we check again after waking up, in case our
* sleep was disturbed by a bad news.
*/
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
XFS_STATS_INC(xs_log_force_sleep);
xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
/*
* No need to grab the log lock here since we're
* only deciding whether or not to return EIO
* and the memory read should be atomic.
*/
if (iclog->ic_state & XLOG_STATE_IOERROR)
return XFS_ERROR(EIO);
if (log_flushed)
*log_flushed = 1;
} else {
no_sleep:
spin_unlock(&log->l_icloglock);
}
return 0;
}
/*
* Wrapper for _xfs_log_force(), to be used when caller doesn't care
* about errors or whether the log was flushed or not. This is the normal
* interface to use when trying to unpin items or move the log forward.
*/
void
xfs_log_force(
xfs_mount_t *mp,
uint flags)
{
int error;
trace_xfs_log_force(mp, 0);
error = _xfs_log_force(mp, flags, NULL);
if (error)
xfs_warn(mp, "%s: error %d returned.", __func__, error);
}
/*
* 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 signal variable. All callers
* to force a given lsn to disk will wait on a the sv 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
* sv.
*/
int
_xfs_log_force_lsn(
struct xfs_mount *mp,
xfs_lsn_t lsn,
uint flags,
int *log_flushed)
{
struct xlog *log = mp->m_log;
struct xlog_in_core *iclog;
int already_slept = 0;
ASSERT(lsn != 0);
XFS_STATS_INC(xs_log_force);
lsn = xlog_cil_force_lsn(log, lsn);
if (lsn == NULLCOMMITLSN)
return 0;
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 04:37:18 +00:00
try_again:
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
do {
if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
iclog = iclog->ic_next;
continue;
}
if (iclog->ic_state == XLOG_STATE_DIRTY) {
spin_unlock(&log->l_icloglock);
return 0;
}
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 | XLOG_STATE_SYNCING))) {
ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
XFS_STATS_INC(xs_log_force_sleep);
xlog_wait(&iclog->ic_prev->ic_write_wait,
&log->l_icloglock);
if (log_flushed)
*log_flushed = 1;
already_slept = 1;
goto try_again;
}
atomic_inc(&iclog->ic_refcnt);
xlog_state_switch_iclogs(log, iclog, 0);
spin_unlock(&log->l_icloglock);
if (xlog_state_release_iclog(log, iclog))
return XFS_ERROR(EIO);
if (log_flushed)
*log_flushed = 1;
spin_lock(&log->l_icloglock);
}
if ((flags & XFS_LOG_SYNC) && /* sleep */
!(iclog->ic_state &
(XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
/*
* Don't wait on completion if we know that we've
* gotten a log write error.
*/
if (iclog->ic_state & XLOG_STATE_IOERROR) {
spin_unlock(&log->l_icloglock);
return XFS_ERROR(EIO);
}
XFS_STATS_INC(xs_log_force_sleep);
xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
/*
* No need to grab the log lock here since we're
* only deciding whether or not to return EIO
* and the memory read should be atomic.
*/
if (iclog->ic_state & XLOG_STATE_IOERROR)
return XFS_ERROR(EIO);
if (log_flushed)
*log_flushed = 1;
} else { /* just return */
spin_unlock(&log->l_icloglock);
}
return 0;
} while (iclog != log->l_iclog);
spin_unlock(&log->l_icloglock);
return 0;
}
/*
* Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
* about errors or whether the log was flushed or not. This is the normal
* interface to use when trying to unpin items or move the log forward.
*/
void
xfs_log_force_lsn(
xfs_mount_t *mp,
xfs_lsn_t lsn,
uint flags)
{
int error;
trace_xfs_log_force(mp, lsn);
error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
if (error)
xfs_warn(mp, "%s: error %d returned.", __func__, error);
}
/*
* 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|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_zone_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;
}
/*
* Allocate and initialise a new log ticket.
*/
struct xlog_ticket *
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 00:21:11 +00:00
xlog_ticket_alloc(
struct xlog *log,
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 00:21:11 +00:00
int unit_bytes,
int cnt,
char client,
bool permanent,
xfs_km_flags_t alloc_flags)
{
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 00:21:11 +00:00
struct xlog_ticket *tic;
uint num_headers;
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 00:21:11 +00:00
int iclog_space;
tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
if (!tic)
return NULL;
/*
* 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 00:21:11 +00:00
/*
* 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(&log->l_mp->m_sb) &&
log->l_mp->m_sb.sb_logsunit > 1) {
/* log su roundoff */
unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
} else {
/* BB roundoff */
unit_bytes += 2*BBSIZE;
}
atomic_set(&tic->t_ref, 1);
tic->t_task = current;
INIT_LIST_HEAD(&tic->t_queue);
tic->t_unit_res = unit_bytes;
tic->t_curr_res = unit_bytes;
tic->t_cnt = cnt;
tic->t_ocnt = cnt;
tic->t_tid = random32();
tic->t_clientid = client;
tic->t_flags = XLOG_TIC_INITED;
tic->t_trans_type = 0;
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.
*/
void
xlog_verify_dest_ptr(
struct xlog *log,
char *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,
boolean_t syncing)
{
xlog_op_header_t *ophead;
xlog_in_core_t *icptr;
xlog_in_core_2_t *xhdr;
xfs_caddr_t ptr;
xfs_caddr_t base_ptr;
__psint_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++) {
if (icptr == NULL)
xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
icptr = icptr->ic_next;
}
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__);
ptr = (xfs_caddr_t) &iclog->ic_header;
for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + 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);
ptr = iclog->ic_datap;
base_ptr = ptr;
ophead = (xlog_op_header_t *)ptr;
xhdr = iclog->ic_data;
for (i = 0; i < len; i++) {
ophead = (xlog_op_header_t *)ptr;
/* clientid is only 1 byte */
field_offset = (__psint_t)
((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
if (syncing == B_FALSE || (field_offset & 0x1ff)) {
clientid = ophead->oh_clientid;
} else {
idx = BTOBBT((xfs_caddr_t)&(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 0x%p offset 0x%lx",
__func__, clientid, ophead,
(unsigned long)field_offset);
/* check length */
field_offset = (__psint_t)
((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
if (syncing == B_FALSE || (field_offset & 0x1ff)) {
op_len = be32_to_cpu(ophead->oh_len);
} else {
idx = BTOBBT((__psint_t)&ophead->oh_len -
(__psint_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. the filesystem gets marked 'SHUTDOWN' for all interested
* parties to find out, 'atomically'.
* b. those who're sleeping on log reservations, pinned objects and
* other resources get woken up, and be told the bad news.
* c. nothing new gets queued up after (a) and (b) are done.
* d. if !logerror, flush the iclogs to disk, then seal them off
* for business.
*
* Note: for delayed logging the !logerror case needs to flush the regions
* held in memory out to the iclogs before flushing them to disk. This needs
* to be done before the log is marked as shutdown, otherwise the flush to the
* iclogs 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)
XFS_BUF_DONE(mp->m_sb_bp);
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;
}
retval = 0;
/*
* Flush the in memory commit item list before marking the log as
* being shut down. We need to do it in this order to ensure all the
* completed transactions are flushed to disk with the xfs_log_force()
* call below.
*/
if (!logerror)
xfs: Reduce log force overhead for delayed logging Delayed logging adds some serialisation to the log force process to ensure that it does not deference a bad commit context structure when determining if a CIL push is necessary or not. It does this by grabing the CIL context lock exclusively, then dropping it before pushing the CIL if necessary. This causes serialisation of all log forces and pushes regardless of whether a force is necessary or not. As a result fsync heavy workloads (like dbench) can be significantly slower with delayed logging than without. To avoid this penalty, copy the current sequence from the context to the CIL structure when they are swapped. This allows us to do unlocked checks on the current sequence without having to worry about dereferencing context structures that may have already been freed. Hence we can remove the CIL context locking in the forcing code and only call into the push code if the current context matches the sequence we need to force. By passing the sequence into the push code, we can check the sequence again once we have the CIL lock held exclusive and abort if the sequence has already been pushed. This avoids a lock round-trip and unnecessary CIL pushes when we have racing push calls. The result is that the regression in dbench performance goes away - this change improves dbench performance on a ramdisk from ~2100MB/s to ~2500MB/s. This compares favourably to not using delayed logging which retuns ~2500MB/s for the same workload. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-08-24 01:40:03 +00:00
xlog_cil_force(log);
/*
* 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)
XFS_BUF_DONE(mp->m_sb_bp);
/*
* This flag is sort of redundant because of the mount flag, but
* it's good to maintain the separation between the log and the rest
* of XFS.
*/
log->l_flags |= XLOG_IO_ERROR;
/*
* If we hit a log error, we want to mark all the iclogs IOERROR
* while we're still holding the loglock.
*/
if (logerror)
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);
if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
ASSERT(!logerror);
/*
* Force the incore logs to disk before shutting the
* log down completely.
*/
_xfs_log_force(mp, XFS_LOG_SYNC, NULL);
spin_lock(&log->l_icloglock);
retval = xlog_state_ioerror(log);
spin_unlock(&log->l_icloglock);
}
/*
* Wake up everybody waiting on xfs_log_force.
* Callback all log item committed functions as if the
* log writes were completed.
*/
xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
#ifdef XFSERRORDEBUG
{
xlog_in_core_t *iclog;
spin_lock(&log->l_icloglock);
iclog = log->l_iclog;
do {
ASSERT(iclog->ic_callback == 0);
iclog = iclog->ic_next;
} while (iclog != log->l_iclog);
spin_unlock(&log->l_icloglock);
}
#endif
/* 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;
}