2 * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write the Free Software Foundation,
15 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_shared.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_error.h"
26 #include "xfs_alloc.h"
27 #include "xfs_extent_busy.h"
28 #include "xfs_discard.h"
29 #include "xfs_trans.h"
30 #include "xfs_trans_priv.h"
32 #include "xfs_log_priv.h"
35 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
36 * recover, so we don't allow failure here. Also, we allocate in a context that
37 * we don't want to be issuing transactions from, so we need to tell the
38 * allocation code this as well.
40 * We don't reserve any space for the ticket - we are going to steal whatever
41 * space we require from transactions as they commit. To ensure we reserve all
42 * the space required, we need to set the current reservation of the ticket to
43 * zero so that we know to steal the initial transaction overhead from the
44 * first transaction commit.
46 static struct xlog_ticket *
47 xlog_cil_ticket_alloc(
50 struct xlog_ticket *tic;
52 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
56 * set the current reservation to zero so we know to steal the basic
57 * transaction overhead reservation from the first transaction commit.
64 * After the first stage of log recovery is done, we know where the head and
65 * tail of the log are. We need this log initialisation done before we can
66 * initialise the first CIL checkpoint context.
68 * Here we allocate a log ticket to track space usage during a CIL push. This
69 * ticket is passed to xlog_write() directly so that we don't slowly leak log
70 * space by failing to account for space used by log headers and additional
71 * region headers for split regions.
74 xlog_cil_init_post_recovery(
77 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
78 log->l_cilp->xc_ctx->sequence = 1;
82 * Prepare the log item for insertion into the CIL. Calculate the difference in
83 * log space and vectors it will consume, and if it is a new item pin it as
89 struct xfs_log_vec *lv,
90 struct xfs_log_vec *old_lv,
94 /* Account for the new LV being passed in */
95 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
96 *diff_len += lv->lv_bytes;
97 *diff_iovecs += lv->lv_niovecs;
101 * If there is no old LV, this is the first time we've seen the item in
102 * this CIL context and so we need to pin it. If we are replacing the
103 * old_lv, then remove the space it accounts for and free it.
106 lv->lv_item->li_ops->iop_pin(lv->lv_item);
107 else if (old_lv != lv) {
108 ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
110 *diff_len -= old_lv->lv_bytes;
111 *diff_iovecs -= old_lv->lv_niovecs;
115 /* attach new log vector to log item */
116 lv->lv_item->li_lv = lv;
119 * If this is the first time the item is being committed to the
120 * CIL, store the sequence number on the log item so we can
121 * tell in future commits whether this is the first checkpoint
122 * the item is being committed into.
124 if (!lv->lv_item->li_seq)
125 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
129 * Format log item into a flat buffers
131 * For delayed logging, we need to hold a formatted buffer containing all the
132 * changes on the log item. This enables us to relog the item in memory and
133 * write it out asynchronously without needing to relock the object that was
134 * modified at the time it gets written into the iclog.
136 * This function builds a vector for the changes in each log item in the
137 * transaction. It then works out the length of the buffer needed for each log
138 * item, allocates them and formats the vector for the item into the buffer.
139 * The buffer is then attached to the log item are then inserted into the
140 * Committed Item List for tracking until the next checkpoint is written out.
142 * We don't set up region headers during this process; we simply copy the
143 * regions into the flat buffer. We can do this because we still have to do a
144 * formatting step to write the regions into the iclog buffer. Writing the
145 * ophdrs during the iclog write means that we can support splitting large
146 * regions across iclog boundares without needing a change in the format of the
147 * item/region encapsulation.
149 * Hence what we need to do now is change the rewrite the vector array to point
150 * to the copied region inside the buffer we just allocated. This allows us to
151 * format the regions into the iclog as though they are being formatted
152 * directly out of the objects themselves.
155 xlog_cil_insert_format_items(
157 struct xfs_trans *tp,
161 struct xfs_log_item_desc *lidp;
164 /* Bail out if we didn't find a log item. */
165 if (list_empty(&tp->t_items)) {
170 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
171 struct xfs_log_item *lip = lidp->lid_item;
172 struct xfs_log_vec *lv;
173 struct xfs_log_vec *old_lv;
177 bool ordered = false;
179 /* Skip items which aren't dirty in this transaction. */
180 if (!(lidp->lid_flags & XFS_LID_DIRTY))
183 /* get number of vecs and size of data to be stored */
184 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
186 /* Skip items that do not have any vectors for writing */
191 * Ordered items need to be tracked but we do not wish to write
192 * them. We need a logvec to track the object, but we do not
193 * need an iovec or buffer to be allocated for copying data.
195 if (niovecs == XFS_LOG_VEC_ORDERED) {
202 * We 64-bit align the length of each iovec so that the start
203 * of the next one is naturally aligned. We'll need to
204 * account for that slack space here. Then round nbytes up
205 * to 64-bit alignment so that the initial buffer alignment is
206 * easy to calculate and verify.
208 nbytes += niovecs * sizeof(uint64_t);
209 nbytes = round_up(nbytes, sizeof(uint64_t));
211 /* grab the old item if it exists for reservation accounting */
215 * The data buffer needs to start 64-bit aligned, so round up
216 * that space to ensure we can align it appropriately and not
217 * overrun the buffer.
220 round_up((sizeof(struct xfs_log_vec) +
221 niovecs * sizeof(struct xfs_log_iovec)),
224 /* compare to existing item size */
225 if (lip->li_lv && buf_size <= lip->li_lv->lv_size) {
226 /* same or smaller, optimise common overwrite case */
234 * set the item up as though it is a new insertion so
235 * that the space reservation accounting is correct.
237 *diff_iovecs -= lv->lv_niovecs;
238 *diff_len -= lv->lv_bytes;
240 /* allocate new data chunk */
241 lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS);
243 lv->lv_size = buf_size;
245 /* track as an ordered logvec */
246 ASSERT(lip->li_lv == NULL);
247 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
250 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
253 /* Ensure the lv is set up according to ->iop_size */
254 lv->lv_niovecs = niovecs;
256 /* The allocated data region lies beyond the iovec region */
259 lv->lv_buf = (char *)lv + buf_size - nbytes;
260 ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
262 lip->li_ops->iop_format(lip, lv);
264 ASSERT(lv->lv_buf_len <= nbytes);
265 xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
270 * Insert the log items into the CIL and calculate the difference in space
271 * consumed by the item. Add the space to the checkpoint ticket and calculate
272 * if the change requires additional log metadata. If it does, take that space
273 * as well. Remove the amount of space we added to the checkpoint ticket from
274 * the current transaction ticket so that the accounting works out correctly.
277 xlog_cil_insert_items(
279 struct xfs_trans *tp)
281 struct xfs_cil *cil = log->l_cilp;
282 struct xfs_cil_ctx *ctx = cil->xc_ctx;
283 struct xfs_log_item_desc *lidp;
291 * We can do this safely because the context can't checkpoint until we
292 * are done so it doesn't matter exactly how we update the CIL.
294 xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
297 * Now (re-)position everything modified at the tail of the CIL.
298 * We do this here so we only need to take the CIL lock once during
299 * the transaction commit.
301 spin_lock(&cil->xc_cil_lock);
302 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
303 struct xfs_log_item *lip = lidp->lid_item;
305 /* Skip items which aren't dirty in this transaction. */
306 if (!(lidp->lid_flags & XFS_LID_DIRTY))
310 * Only move the item if it isn't already at the tail. This is
311 * to prevent a transient list_empty() state when reinserting
312 * an item that is already the only item in the CIL.
314 if (!list_is_last(&lip->li_cil, &cil->xc_cil))
315 list_move_tail(&lip->li_cil, &cil->xc_cil);
318 /* account for space used by new iovec headers */
319 len += diff_iovecs * sizeof(xlog_op_header_t);
320 ctx->nvecs += diff_iovecs;
322 /* attach the transaction to the CIL if it has any busy extents */
323 if (!list_empty(&tp->t_busy))
324 list_splice_init(&tp->t_busy, &ctx->busy_extents);
327 * Now transfer enough transaction reservation to the context ticket
328 * for the checkpoint. The context ticket is special - the unit
329 * reservation has to grow as well as the current reservation as we
330 * steal from tickets so we can correctly determine the space used
331 * during the transaction commit.
333 if (ctx->ticket->t_curr_res == 0) {
334 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
335 tp->t_ticket->t_curr_res -= ctx->ticket->t_unit_res;
338 /* do we need space for more log record headers? */
339 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
340 if (len > 0 && (ctx->space_used / iclog_space !=
341 (ctx->space_used + len) / iclog_space)) {
344 hdrs = (len + iclog_space - 1) / iclog_space;
345 /* need to take into account split region headers, too */
346 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
347 ctx->ticket->t_unit_res += hdrs;
348 ctx->ticket->t_curr_res += hdrs;
349 tp->t_ticket->t_curr_res -= hdrs;
350 ASSERT(tp->t_ticket->t_curr_res >= len);
352 tp->t_ticket->t_curr_res -= len;
353 ctx->space_used += len;
355 spin_unlock(&cil->xc_cil_lock);
359 xlog_cil_free_logvec(
360 struct xfs_log_vec *log_vector)
362 struct xfs_log_vec *lv;
364 for (lv = log_vector; lv; ) {
365 struct xfs_log_vec *next = lv->lv_next;
372 * Mark all items committed and clear busy extents. We free the log vector
373 * chains in a separate pass so that we unpin the log items as quickly as
381 struct xfs_cil_ctx *ctx = args;
382 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
384 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
385 ctx->start_lsn, abort);
387 xfs_extent_busy_sort(&ctx->busy_extents);
388 xfs_extent_busy_clear(mp, &ctx->busy_extents,
389 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
392 * If we are aborting the commit, wake up anyone waiting on the
393 * committing list. If we don't, then a shutdown we can leave processes
394 * waiting in xlog_cil_force_lsn() waiting on a sequence commit that
395 * will never happen because we aborted it.
397 spin_lock(&ctx->cil->xc_push_lock);
399 wake_up_all(&ctx->cil->xc_commit_wait);
400 list_del(&ctx->committing);
401 spin_unlock(&ctx->cil->xc_push_lock);
403 xlog_cil_free_logvec(ctx->lv_chain);
405 if (!list_empty(&ctx->busy_extents)) {
406 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
408 xfs_discard_extents(mp, &ctx->busy_extents);
409 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
416 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
417 * is a background flush and so we can chose to ignore it. Otherwise, if the
418 * current sequence is the same as @push_seq we need to do a flush. If
419 * @push_seq is less than the current sequence, then it has already been
420 * flushed and we don't need to do anything - the caller will wait for it to
421 * complete if necessary.
423 * @push_seq is a value rather than a flag because that allows us to do an
424 * unlocked check of the sequence number for a match. Hence we can allows log
425 * forces to run racily and not issue pushes for the same sequence twice. If we
426 * get a race between multiple pushes for the same sequence they will block on
427 * the first one and then abort, hence avoiding needless pushes.
433 struct xfs_cil *cil = log->l_cilp;
434 struct xfs_log_vec *lv;
435 struct xfs_cil_ctx *ctx;
436 struct xfs_cil_ctx *new_ctx;
437 struct xlog_in_core *commit_iclog;
438 struct xlog_ticket *tic;
441 struct xfs_trans_header thdr;
442 struct xfs_log_iovec lhdr;
443 struct xfs_log_vec lvhdr = { NULL };
444 xfs_lsn_t commit_lsn;
450 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
451 new_ctx->ticket = xlog_cil_ticket_alloc(log);
453 down_write(&cil->xc_ctx_lock);
456 spin_lock(&cil->xc_push_lock);
457 push_seq = cil->xc_push_seq;
458 ASSERT(push_seq <= ctx->sequence);
461 * Check if we've anything to push. If there is nothing, then we don't
462 * move on to a new sequence number and so we have to be able to push
463 * this sequence again later.
465 if (list_empty(&cil->xc_cil)) {
466 cil->xc_push_seq = 0;
467 spin_unlock(&cil->xc_push_lock);
472 /* check for a previously pushed seqeunce */
473 if (push_seq < cil->xc_ctx->sequence) {
474 spin_unlock(&cil->xc_push_lock);
479 * We are now going to push this context, so add it to the committing
480 * list before we do anything else. This ensures that anyone waiting on
481 * this push can easily detect the difference between a "push in
482 * progress" and "CIL is empty, nothing to do".
484 * IOWs, a wait loop can now check for:
485 * the current sequence not being found on the committing list;
487 * an unchanged sequence number
488 * to detect a push that had nothing to do and therefore does not need
489 * waiting on. If the CIL is not empty, we get put on the committing
490 * list before emptying the CIL and bumping the sequence number. Hence
491 * an empty CIL and an unchanged sequence number means we jumped out
492 * above after doing nothing.
494 * Hence the waiter will either find the commit sequence on the
495 * committing list or the sequence number will be unchanged and the CIL
496 * still dirty. In that latter case, the push has not yet started, and
497 * so the waiter will have to continue trying to check the CIL
498 * committing list until it is found. In extreme cases of delay, the
499 * sequence may fully commit between the attempts the wait makes to wait
500 * on the commit sequence.
502 list_add(&ctx->committing, &cil->xc_committing);
503 spin_unlock(&cil->xc_push_lock);
506 * pull all the log vectors off the items in the CIL, and
507 * remove the items from the CIL. We don't need the CIL lock
508 * here because it's only needed on the transaction commit
509 * side which is currently locked out by the flush lock.
513 while (!list_empty(&cil->xc_cil)) {
514 struct xfs_log_item *item;
516 item = list_first_entry(&cil->xc_cil,
517 struct xfs_log_item, li_cil);
518 list_del_init(&item->li_cil);
520 ctx->lv_chain = item->li_lv;
522 lv->lv_next = item->li_lv;
525 num_iovecs += lv->lv_niovecs;
529 * initialise the new context and attach it to the CIL. Then attach
530 * the current context to the CIL committing lsit so it can be found
531 * during log forces to extract the commit lsn of the sequence that
532 * needs to be forced.
534 INIT_LIST_HEAD(&new_ctx->committing);
535 INIT_LIST_HEAD(&new_ctx->busy_extents);
536 new_ctx->sequence = ctx->sequence + 1;
538 cil->xc_ctx = new_ctx;
541 * The switch is now done, so we can drop the context lock and move out
542 * of a shared context. We can't just go straight to the commit record,
543 * though - we need to synchronise with previous and future commits so
544 * that the commit records are correctly ordered in the log to ensure
545 * that we process items during log IO completion in the correct order.
547 * For example, if we get an EFI in one checkpoint and the EFD in the
548 * next (e.g. due to log forces), we do not want the checkpoint with
549 * the EFD to be committed before the checkpoint with the EFI. Hence
550 * we must strictly order the commit records of the checkpoints so
551 * that: a) the checkpoint callbacks are attached to the iclogs in the
552 * correct order; and b) the checkpoints are replayed in correct order
555 * Hence we need to add this context to the committing context list so
556 * that higher sequences will wait for us to write out a commit record
559 * xfs_log_force_lsn requires us to mirror the new sequence into the cil
560 * structure atomically with the addition of this sequence to the
561 * committing list. This also ensures that we can do unlocked checks
562 * against the current sequence in log forces without risking
563 * deferencing a freed context pointer.
565 spin_lock(&cil->xc_push_lock);
566 cil->xc_current_sequence = new_ctx->sequence;
567 spin_unlock(&cil->xc_push_lock);
568 up_write(&cil->xc_ctx_lock);
571 * Build a checkpoint transaction header and write it to the log to
572 * begin the transaction. We need to account for the space used by the
573 * transaction header here as it is not accounted for in xlog_write().
575 * The LSN we need to pass to the log items on transaction commit is
576 * the LSN reported by the first log vector write. If we use the commit
577 * record lsn then we can move the tail beyond the grant write head.
580 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
581 thdr.th_type = XFS_TRANS_CHECKPOINT;
582 thdr.th_tid = tic->t_tid;
583 thdr.th_num_items = num_iovecs;
585 lhdr.i_len = sizeof(xfs_trans_header_t);
586 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
587 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
589 lvhdr.lv_niovecs = 1;
590 lvhdr.lv_iovecp = &lhdr;
591 lvhdr.lv_next = ctx->lv_chain;
593 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
595 goto out_abort_free_ticket;
598 * now that we've written the checkpoint into the log, strictly
599 * order the commit records so replay will get them in the right order.
602 spin_lock(&cil->xc_push_lock);
603 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
605 * Avoid getting stuck in this loop because we were woken by the
606 * shutdown, but then went back to sleep once already in the
609 if (XLOG_FORCED_SHUTDOWN(log)) {
610 spin_unlock(&cil->xc_push_lock);
611 goto out_abort_free_ticket;
615 * Higher sequences will wait for this one so skip them.
616 * Don't wait for our own sequence, either.
618 if (new_ctx->sequence >= ctx->sequence)
620 if (!new_ctx->commit_lsn) {
622 * It is still being pushed! Wait for the push to
623 * complete, then start again from the beginning.
625 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
629 spin_unlock(&cil->xc_push_lock);
631 /* xfs_log_done always frees the ticket on error. */
632 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, false);
633 if (commit_lsn == -1)
636 /* attach all the transactions w/ busy extents to iclog */
637 ctx->log_cb.cb_func = xlog_cil_committed;
638 ctx->log_cb.cb_arg = ctx;
639 error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
644 * now the checkpoint commit is complete and we've attached the
645 * callbacks to the iclog we can assign the commit LSN to the context
646 * and wake up anyone who is waiting for the commit to complete.
648 spin_lock(&cil->xc_push_lock);
649 ctx->commit_lsn = commit_lsn;
650 wake_up_all(&cil->xc_commit_wait);
651 spin_unlock(&cil->xc_push_lock);
653 /* release the hounds! */
654 return xfs_log_release_iclog(log->l_mp, commit_iclog);
657 up_write(&cil->xc_ctx_lock);
658 xfs_log_ticket_put(new_ctx->ticket);
662 out_abort_free_ticket:
663 xfs_log_ticket_put(tic);
665 xlog_cil_committed(ctx, XFS_LI_ABORTED);
671 struct work_struct *work)
673 struct xfs_cil *cil = container_of(work, struct xfs_cil,
675 xlog_cil_push(cil->xc_log);
679 * We need to push CIL every so often so we don't cache more than we can fit in
680 * the log. The limit really is that a checkpoint can't be more than half the
681 * log (the current checkpoint is not allowed to overwrite the previous
682 * checkpoint), but commit latency and memory usage limit this to a smaller
686 xlog_cil_push_background(
689 struct xfs_cil *cil = log->l_cilp;
692 * The cil won't be empty because we are called while holding the
693 * context lock so whatever we added to the CIL will still be there
695 ASSERT(!list_empty(&cil->xc_cil));
698 * don't do a background push if we haven't used up all the
699 * space available yet.
701 if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
704 spin_lock(&cil->xc_push_lock);
705 if (cil->xc_push_seq < cil->xc_current_sequence) {
706 cil->xc_push_seq = cil->xc_current_sequence;
707 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
709 spin_unlock(&cil->xc_push_lock);
714 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
715 * number that is passed. When it returns, the work will be queued for
716 * @push_seq, but it won't be completed. The caller is expected to do any
717 * waiting for push_seq to complete if it is required.
724 struct xfs_cil *cil = log->l_cilp;
729 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
731 /* start on any pending background push to minimise wait time on it */
732 flush_work(&cil->xc_push_work);
735 * If the CIL is empty or we've already pushed the sequence then
736 * there's no work we need to do.
738 spin_lock(&cil->xc_push_lock);
739 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
740 spin_unlock(&cil->xc_push_lock);
744 cil->xc_push_seq = push_seq;
745 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
746 spin_unlock(&cil->xc_push_lock);
753 struct xfs_cil *cil = log->l_cilp;
756 spin_lock(&cil->xc_push_lock);
757 if (list_empty(&cil->xc_cil))
759 spin_unlock(&cil->xc_push_lock);
764 * Commit a transaction with the given vector to the Committed Item List.
766 * To do this, we need to format the item, pin it in memory if required and
767 * account for the space used by the transaction. Once we have done that we
768 * need to release the unused reservation for the transaction, attach the
769 * transaction to the checkpoint context so we carry the busy extents through
770 * to checkpoint completion, and then unlock all the items in the transaction.
772 * Called with the context lock already held in read mode to lock out
773 * background commit, returns without it held once background commits are
778 struct xfs_mount *mp,
779 struct xfs_trans *tp,
780 xfs_lsn_t *commit_lsn,
783 struct xlog *log = mp->m_log;
784 struct xfs_cil *cil = log->l_cilp;
786 /* lock out background commit */
787 down_read(&cil->xc_ctx_lock);
789 xlog_cil_insert_items(log, tp);
791 /* check we didn't blow the reservation */
792 if (tp->t_ticket->t_curr_res < 0)
793 xlog_print_tic_res(mp, tp->t_ticket);
795 tp->t_commit_lsn = cil->xc_ctx->sequence;
797 *commit_lsn = tp->t_commit_lsn;
799 xfs_log_done(mp, tp->t_ticket, NULL, regrant);
800 xfs_trans_unreserve_and_mod_sb(tp);
803 * Once all the items of the transaction have been copied to the CIL,
804 * the items can be unlocked and freed.
806 * This needs to be done before we drop the CIL context lock because we
807 * have to update state in the log items and unlock them before they go
808 * to disk. If we don't, then the CIL checkpoint can race with us and
809 * we can run checkpoint completion before we've updated and unlocked
810 * the log items. This affects (at least) processing of stale buffers,
813 xfs_trans_free_items(tp, tp->t_commit_lsn, false);
815 xlog_cil_push_background(log);
817 up_read(&cil->xc_ctx_lock);
821 * Conditionally push the CIL based on the sequence passed in.
823 * We only need to push if we haven't already pushed the sequence
824 * number given. Hence the only time we will trigger a push here is
825 * if the push sequence is the same as the current context.
827 * We return the current commit lsn to allow the callers to determine if a
828 * iclog flush is necessary following this call.
835 struct xfs_cil *cil = log->l_cilp;
836 struct xfs_cil_ctx *ctx;
837 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
839 ASSERT(sequence <= cil->xc_current_sequence);
842 * check to see if we need to force out the current context.
843 * xlog_cil_push() handles racing pushes for the same sequence,
844 * so no need to deal with it here.
847 xlog_cil_push_now(log, sequence);
850 * See if we can find a previous sequence still committing.
851 * We need to wait for all previous sequence commits to complete
852 * before allowing the force of push_seq to go ahead. Hence block
853 * on commits for those as well.
855 spin_lock(&cil->xc_push_lock);
856 list_for_each_entry(ctx, &cil->xc_committing, committing) {
858 * Avoid getting stuck in this loop because we were woken by the
859 * shutdown, but then went back to sleep once already in the
862 if (XLOG_FORCED_SHUTDOWN(log))
864 if (ctx->sequence > sequence)
866 if (!ctx->commit_lsn) {
868 * It is still being pushed! Wait for the push to
869 * complete, then start again from the beginning.
871 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
874 if (ctx->sequence != sequence)
877 commit_lsn = ctx->commit_lsn;
881 * The call to xlog_cil_push_now() executes the push in the background.
882 * Hence by the time we have got here it our sequence may not have been
883 * pushed yet. This is true if the current sequence still matches the
884 * push sequence after the above wait loop and the CIL still contains
885 * dirty objects. This is guaranteed by the push code first adding the
886 * context to the committing list before emptying the CIL.
888 * Hence if we don't find the context in the committing list and the
889 * current sequence number is unchanged then the CIL contents are
890 * significant. If the CIL is empty, if means there was nothing to push
891 * and that means there is nothing to wait for. If the CIL is not empty,
892 * it means we haven't yet started the push, because if it had started
893 * we would have found the context on the committing list.
895 if (sequence == cil->xc_current_sequence &&
896 !list_empty(&cil->xc_cil)) {
897 spin_unlock(&cil->xc_push_lock);
901 spin_unlock(&cil->xc_push_lock);
905 * We detected a shutdown in progress. We need to trigger the log force
906 * to pass through it's iclog state machine error handling, even though
907 * we are already in a shutdown state. Hence we can't return
908 * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
909 * LSN is already stable), so we return a zero LSN instead.
912 spin_unlock(&cil->xc_push_lock);
917 * Check if the current log item was first committed in this sequence.
918 * We can't rely on just the log item being in the CIL, we have to check
919 * the recorded commit sequence number.
921 * Note: for this to be used in a non-racy manner, it has to be called with
922 * CIL flushing locked out. As a result, it should only be used during the
923 * transaction commit process when deciding what to format into the item.
926 xfs_log_item_in_current_chkpt(
927 struct xfs_log_item *lip)
929 struct xfs_cil_ctx *ctx;
931 if (list_empty(&lip->li_cil))
934 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
937 * li_seq is written on the first commit of a log item to record the
938 * first checkpoint it is written to. Hence if it is different to the
939 * current sequence, we're in a new checkpoint.
941 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
947 * Perform initial CIL structure initialisation.
954 struct xfs_cil_ctx *ctx;
956 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
960 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
966 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
967 INIT_LIST_HEAD(&cil->xc_cil);
968 INIT_LIST_HEAD(&cil->xc_committing);
969 spin_lock_init(&cil->xc_cil_lock);
970 spin_lock_init(&cil->xc_push_lock);
971 init_rwsem(&cil->xc_ctx_lock);
972 init_waitqueue_head(&cil->xc_commit_wait);
974 INIT_LIST_HEAD(&ctx->committing);
975 INIT_LIST_HEAD(&ctx->busy_extents);
979 cil->xc_current_sequence = ctx->sequence;
990 if (log->l_cilp->xc_ctx) {
991 if (log->l_cilp->xc_ctx->ticket)
992 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
993 kmem_free(log->l_cilp->xc_ctx);
996 ASSERT(list_empty(&log->l_cilp->xc_cil));
997 kmem_free(log->l_cilp);