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_log_format.h"
21 #include "xfs_shared.h"
22 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_error.h"
27 #include "xfs_alloc.h"
28 #include "xfs_extent_busy.h"
29 #include "xfs_discard.h"
30 #include "xfs_trans.h"
31 #include "xfs_trans_priv.h"
33 #include "xfs_log_priv.h"
36 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
37 * recover, so we don't allow failure here. Also, we allocate in a context that
38 * we don't want to be issuing transactions from, so we need to tell the
39 * allocation code this as well.
41 * We don't reserve any space for the ticket - we are going to steal whatever
42 * space we require from transactions as they commit. To ensure we reserve all
43 * the space required, we need to set the current reservation of the ticket to
44 * zero so that we know to steal the initial transaction overhead from the
45 * first transaction commit.
47 static struct xlog_ticket *
48 xlog_cil_ticket_alloc(
51 struct xlog_ticket *tic;
53 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
55 tic->t_trans_type = XFS_TRANS_CHECKPOINT;
58 * set the current reservation to zero so we know to steal the basic
59 * transaction overhead reservation from the first transaction commit.
66 * After the first stage of log recovery is done, we know where the head and
67 * tail of the log are. We need this log initialisation done before we can
68 * initialise the first CIL checkpoint context.
70 * Here we allocate a log ticket to track space usage during a CIL push. This
71 * ticket is passed to xlog_write() directly so that we don't slowly leak log
72 * space by failing to account for space used by log headers and additional
73 * region headers for split regions.
76 xlog_cil_init_post_recovery(
79 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
80 log->l_cilp->xc_ctx->sequence = 1;
84 * Prepare the log item for insertion into the CIL. Calculate the difference in
85 * log space and vectors it will consume, and if it is a new item pin it as
91 struct xfs_log_vec *lv,
92 struct xfs_log_vec *old_lv,
96 /* Account for the new LV being passed in */
97 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
98 *diff_len += lv->lv_bytes;
99 *diff_iovecs += lv->lv_niovecs;
103 * If there is no old LV, this is the first time we've seen the item in
104 * this CIL context and so we need to pin it. If we are replacing the
105 * old_lv, then remove the space it accounts for and free it.
108 lv->lv_item->li_ops->iop_pin(lv->lv_item);
109 else if (old_lv != lv) {
110 ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
112 *diff_len -= old_lv->lv_bytes;
113 *diff_iovecs -= old_lv->lv_niovecs;
117 /* attach new log vector to log item */
118 lv->lv_item->li_lv = lv;
121 * If this is the first time the item is being committed to the
122 * CIL, store the sequence number on the log item so we can
123 * tell in future commits whether this is the first checkpoint
124 * the item is being committed into.
126 if (!lv->lv_item->li_seq)
127 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
131 * Format log item into a flat buffers
133 * For delayed logging, we need to hold a formatted buffer containing all the
134 * changes on the log item. This enables us to relog the item in memory and
135 * write it out asynchronously without needing to relock the object that was
136 * modified at the time it gets written into the iclog.
138 * This function builds a vector for the changes in each log item in the
139 * transaction. It then works out the length of the buffer needed for each log
140 * item, allocates them and formats the vector for the item into the buffer.
141 * The buffer is then attached to the log item are then inserted into the
142 * Committed Item List for tracking until the next checkpoint is written out.
144 * We don't set up region headers during this process; we simply copy the
145 * regions into the flat buffer. We can do this because we still have to do a
146 * formatting step to write the regions into the iclog buffer. Writing the
147 * ophdrs during the iclog write means that we can support splitting large
148 * regions across iclog boundares without needing a change in the format of the
149 * item/region encapsulation.
151 * Hence what we need to do now is change the rewrite the vector array to point
152 * to the copied region inside the buffer we just allocated. This allows us to
153 * format the regions into the iclog as though they are being formatted
154 * directly out of the objects themselves.
157 xlog_cil_insert_format_items(
159 struct xfs_trans *tp,
163 struct xfs_log_item_desc *lidp;
166 /* Bail out if we didn't find a log item. */
167 if (list_empty(&tp->t_items)) {
172 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
173 struct xfs_log_item *lip = lidp->lid_item;
174 struct xfs_log_vec *lv;
175 struct xfs_log_vec *old_lv;
179 bool ordered = false;
181 /* Skip items which aren't dirty in this transaction. */
182 if (!(lidp->lid_flags & XFS_LID_DIRTY))
185 /* get number of vecs and size of data to be stored */
186 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
188 /* Skip items that do not have any vectors for writing */
193 * Ordered items need to be tracked but we do not wish to write
194 * them. We need a logvec to track the object, but we do not
195 * need an iovec or buffer to be allocated for copying data.
197 if (niovecs == XFS_LOG_VEC_ORDERED) {
204 * We 64-bit align the length of each iovec so that the start
205 * of the next one is naturally aligned. We'll need to
206 * account for that slack space here. Then round nbytes up
207 * to 64-bit alignment so that the initial buffer alignment is
208 * easy to calculate and verify.
210 nbytes += niovecs * sizeof(uint64_t);
211 nbytes = round_up(nbytes, sizeof(uint64_t));
213 /* grab the old item if it exists for reservation accounting */
217 * The data buffer needs to start 64-bit aligned, so round up
218 * that space to ensure we can align it appropriately and not
219 * overrun the buffer.
222 round_up((sizeof(struct xfs_log_vec) +
223 niovecs * sizeof(struct xfs_log_iovec)),
226 /* compare to existing item size */
227 if (lip->li_lv && buf_size <= lip->li_lv->lv_size) {
228 /* same or smaller, optimise common overwrite case */
236 * set the item up as though it is a new insertion so
237 * that the space reservation accounting is correct.
239 *diff_iovecs -= lv->lv_niovecs;
240 *diff_len -= lv->lv_bytes;
242 /* allocate new data chunk */
243 lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS);
245 lv->lv_size = buf_size;
247 /* track as an ordered logvec */
248 ASSERT(lip->li_lv == NULL);
249 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
252 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
255 /* Ensure the lv is set up according to ->iop_size */
256 lv->lv_niovecs = niovecs;
258 /* The allocated data region lies beyond the iovec region */
261 lv->lv_buf = (char *)lv + buf_size - nbytes;
262 ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
264 lip->li_ops->iop_format(lip, lv);
266 ASSERT(lv->lv_buf_len <= nbytes);
267 xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
272 * Insert the log items into the CIL and calculate the difference in space
273 * consumed by the item. Add the space to the checkpoint ticket and calculate
274 * if the change requires additional log metadata. If it does, take that space
275 * as well. Remove the amount of space we added to the checkpoint ticket from
276 * the current transaction ticket so that the accounting works out correctly.
279 xlog_cil_insert_items(
281 struct xfs_trans *tp)
283 struct xfs_cil *cil = log->l_cilp;
284 struct xfs_cil_ctx *ctx = cil->xc_ctx;
285 struct xfs_log_item_desc *lidp;
293 * We can do this safely because the context can't checkpoint until we
294 * are done so it doesn't matter exactly how we update the CIL.
296 xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
299 * Now (re-)position everything modified at the tail of the CIL.
300 * We do this here so we only need to take the CIL lock once during
301 * the transaction commit.
303 spin_lock(&cil->xc_cil_lock);
304 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
305 struct xfs_log_item *lip = lidp->lid_item;
307 /* Skip items which aren't dirty in this transaction. */
308 if (!(lidp->lid_flags & XFS_LID_DIRTY))
311 list_move_tail(&lip->li_cil, &cil->xc_cil);
314 /* account for space used by new iovec headers */
315 len += diff_iovecs * sizeof(xlog_op_header_t);
316 ctx->nvecs += diff_iovecs;
318 /* attach the transaction to the CIL if it has any busy extents */
319 if (!list_empty(&tp->t_busy))
320 list_splice_init(&tp->t_busy, &ctx->busy_extents);
323 * Now transfer enough transaction reservation to the context ticket
324 * for the checkpoint. The context ticket is special - the unit
325 * reservation has to grow as well as the current reservation as we
326 * steal from tickets so we can correctly determine the space used
327 * during the transaction commit.
329 if (ctx->ticket->t_curr_res == 0) {
330 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
331 tp->t_ticket->t_curr_res -= ctx->ticket->t_unit_res;
334 /* do we need space for more log record headers? */
335 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
336 if (len > 0 && (ctx->space_used / iclog_space !=
337 (ctx->space_used + len) / iclog_space)) {
340 hdrs = (len + iclog_space - 1) / iclog_space;
341 /* need to take into account split region headers, too */
342 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
343 ctx->ticket->t_unit_res += hdrs;
344 ctx->ticket->t_curr_res += hdrs;
345 tp->t_ticket->t_curr_res -= hdrs;
346 ASSERT(tp->t_ticket->t_curr_res >= len);
348 tp->t_ticket->t_curr_res -= len;
349 ctx->space_used += len;
351 spin_unlock(&cil->xc_cil_lock);
355 xlog_cil_free_logvec(
356 struct xfs_log_vec *log_vector)
358 struct xfs_log_vec *lv;
360 for (lv = log_vector; lv; ) {
361 struct xfs_log_vec *next = lv->lv_next;
368 * Mark all items committed and clear busy extents. We free the log vector
369 * chains in a separate pass so that we unpin the log items as quickly as
377 struct xfs_cil_ctx *ctx = args;
378 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
380 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
381 ctx->start_lsn, abort);
383 xfs_extent_busy_sort(&ctx->busy_extents);
384 xfs_extent_busy_clear(mp, &ctx->busy_extents,
385 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
388 * If we are aborting the commit, wake up anyone waiting on the
389 * committing list. If we don't, then a shutdown we can leave processes
390 * waiting in xlog_cil_force_lsn() waiting on a sequence commit that
391 * will never happen because we aborted it.
393 spin_lock(&ctx->cil->xc_push_lock);
395 wake_up_all(&ctx->cil->xc_commit_wait);
396 list_del(&ctx->committing);
397 spin_unlock(&ctx->cil->xc_push_lock);
399 xlog_cil_free_logvec(ctx->lv_chain);
401 if (!list_empty(&ctx->busy_extents)) {
402 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
404 xfs_discard_extents(mp, &ctx->busy_extents);
405 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
412 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
413 * is a background flush and so we can chose to ignore it. Otherwise, if the
414 * current sequence is the same as @push_seq we need to do a flush. If
415 * @push_seq is less than the current sequence, then it has already been
416 * flushed and we don't need to do anything - the caller will wait for it to
417 * complete if necessary.
419 * @push_seq is a value rather than a flag because that allows us to do an
420 * unlocked check of the sequence number for a match. Hence we can allows log
421 * forces to run racily and not issue pushes for the same sequence twice. If we
422 * get a race between multiple pushes for the same sequence they will block on
423 * the first one and then abort, hence avoiding needless pushes.
429 struct xfs_cil *cil = log->l_cilp;
430 struct xfs_log_vec *lv;
431 struct xfs_cil_ctx *ctx;
432 struct xfs_cil_ctx *new_ctx;
433 struct xlog_in_core *commit_iclog;
434 struct xlog_ticket *tic;
437 struct xfs_trans_header thdr;
438 struct xfs_log_iovec lhdr;
439 struct xfs_log_vec lvhdr = { NULL };
440 xfs_lsn_t commit_lsn;
446 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
447 new_ctx->ticket = xlog_cil_ticket_alloc(log);
449 down_write(&cil->xc_ctx_lock);
452 spin_lock(&cil->xc_push_lock);
453 push_seq = cil->xc_push_seq;
454 ASSERT(push_seq <= ctx->sequence);
457 * Check if we've anything to push. If there is nothing, then we don't
458 * move on to a new sequence number and so we have to be able to push
459 * this sequence again later.
461 if (list_empty(&cil->xc_cil)) {
462 cil->xc_push_seq = 0;
463 spin_unlock(&cil->xc_push_lock);
466 spin_unlock(&cil->xc_push_lock);
469 /* check for a previously pushed seqeunce */
470 if (push_seq < cil->xc_ctx->sequence)
474 * pull all the log vectors off the items in the CIL, and
475 * remove the items from the CIL. We don't need the CIL lock
476 * here because it's only needed on the transaction commit
477 * side which is currently locked out by the flush lock.
481 while (!list_empty(&cil->xc_cil)) {
482 struct xfs_log_item *item;
484 item = list_first_entry(&cil->xc_cil,
485 struct xfs_log_item, li_cil);
486 list_del_init(&item->li_cil);
488 ctx->lv_chain = item->li_lv;
490 lv->lv_next = item->li_lv;
493 num_iovecs += lv->lv_niovecs;
497 * initialise the new context and attach it to the CIL. Then attach
498 * the current context to the CIL committing lsit so it can be found
499 * during log forces to extract the commit lsn of the sequence that
500 * needs to be forced.
502 INIT_LIST_HEAD(&new_ctx->committing);
503 INIT_LIST_HEAD(&new_ctx->busy_extents);
504 new_ctx->sequence = ctx->sequence + 1;
506 cil->xc_ctx = new_ctx;
509 * The switch is now done, so we can drop the context lock and move out
510 * of a shared context. We can't just go straight to the commit record,
511 * though - we need to synchronise with previous and future commits so
512 * that the commit records are correctly ordered in the log to ensure
513 * that we process items during log IO completion in the correct order.
515 * For example, if we get an EFI in one checkpoint and the EFD in the
516 * next (e.g. due to log forces), we do not want the checkpoint with
517 * the EFD to be committed before the checkpoint with the EFI. Hence
518 * we must strictly order the commit records of the checkpoints so
519 * that: a) the checkpoint callbacks are attached to the iclogs in the
520 * correct order; and b) the checkpoints are replayed in correct order
523 * Hence we need to add this context to the committing context list so
524 * that higher sequences will wait for us to write out a commit record
527 * xfs_log_force_lsn requires us to mirror the new sequence into the cil
528 * structure atomically with the addition of this sequence to the
529 * committing list. This also ensures that we can do unlocked checks
530 * against the current sequence in log forces without risking
531 * deferencing a freed context pointer.
533 spin_lock(&cil->xc_push_lock);
534 cil->xc_current_sequence = new_ctx->sequence;
535 list_add(&ctx->committing, &cil->xc_committing);
536 spin_unlock(&cil->xc_push_lock);
537 up_write(&cil->xc_ctx_lock);
540 * Build a checkpoint transaction header and write it to the log to
541 * begin the transaction. We need to account for the space used by the
542 * transaction header here as it is not accounted for in xlog_write().
544 * The LSN we need to pass to the log items on transaction commit is
545 * the LSN reported by the first log vector write. If we use the commit
546 * record lsn then we can move the tail beyond the grant write head.
549 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
550 thdr.th_type = XFS_TRANS_CHECKPOINT;
551 thdr.th_tid = tic->t_tid;
552 thdr.th_num_items = num_iovecs;
554 lhdr.i_len = sizeof(xfs_trans_header_t);
555 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
556 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
558 lvhdr.lv_niovecs = 1;
559 lvhdr.lv_iovecp = &lhdr;
560 lvhdr.lv_next = ctx->lv_chain;
562 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
564 goto out_abort_free_ticket;
567 * now that we've written the checkpoint into the log, strictly
568 * order the commit records so replay will get them in the right order.
571 spin_lock(&cil->xc_push_lock);
572 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
574 * Avoid getting stuck in this loop because we were woken by the
575 * shutdown, but then went back to sleep once already in the
578 if (XLOG_FORCED_SHUTDOWN(log)) {
579 spin_unlock(&cil->xc_push_lock);
580 goto out_abort_free_ticket;
584 * Higher sequences will wait for this one so skip them.
585 * Don't wait for our own sequence, either.
587 if (new_ctx->sequence >= ctx->sequence)
589 if (!new_ctx->commit_lsn) {
591 * It is still being pushed! Wait for the push to
592 * complete, then start again from the beginning.
594 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
598 spin_unlock(&cil->xc_push_lock);
600 /* xfs_log_done always frees the ticket on error. */
601 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
602 if (commit_lsn == -1)
605 /* attach all the transactions w/ busy extents to iclog */
606 ctx->log_cb.cb_func = xlog_cil_committed;
607 ctx->log_cb.cb_arg = ctx;
608 error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
613 * now the checkpoint commit is complete and we've attached the
614 * callbacks to the iclog we can assign the commit LSN to the context
615 * and wake up anyone who is waiting for the commit to complete.
617 spin_lock(&cil->xc_push_lock);
618 ctx->commit_lsn = commit_lsn;
619 wake_up_all(&cil->xc_commit_wait);
620 spin_unlock(&cil->xc_push_lock);
622 /* release the hounds! */
623 return xfs_log_release_iclog(log->l_mp, commit_iclog);
626 up_write(&cil->xc_ctx_lock);
627 xfs_log_ticket_put(new_ctx->ticket);
631 out_abort_free_ticket:
632 xfs_log_ticket_put(tic);
634 xlog_cil_committed(ctx, XFS_LI_ABORTED);
640 struct work_struct *work)
642 struct xfs_cil *cil = container_of(work, struct xfs_cil,
644 xlog_cil_push(cil->xc_log);
648 * We need to push CIL every so often so we don't cache more than we can fit in
649 * the log. The limit really is that a checkpoint can't be more than half the
650 * log (the current checkpoint is not allowed to overwrite the previous
651 * checkpoint), but commit latency and memory usage limit this to a smaller
655 xlog_cil_push_background(
658 struct xfs_cil *cil = log->l_cilp;
661 * The cil won't be empty because we are called while holding the
662 * context lock so whatever we added to the CIL will still be there
664 ASSERT(!list_empty(&cil->xc_cil));
667 * don't do a background push if we haven't used up all the
668 * space available yet.
670 if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
673 spin_lock(&cil->xc_push_lock);
674 if (cil->xc_push_seq < cil->xc_current_sequence) {
675 cil->xc_push_seq = cil->xc_current_sequence;
676 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
678 spin_unlock(&cil->xc_push_lock);
683 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
684 * number that is passed. When it returns, the work will be queued for
685 * @push_seq, but it won't be completed. The caller is expected to do any
686 * waiting for push_seq to complete if it is required.
693 struct xfs_cil *cil = log->l_cilp;
698 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
700 /* start on any pending background push to minimise wait time on it */
701 flush_work(&cil->xc_push_work);
704 * If the CIL is empty or we've already pushed the sequence then
705 * there's no work we need to do.
707 spin_lock(&cil->xc_push_lock);
708 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
709 spin_unlock(&cil->xc_push_lock);
713 cil->xc_push_seq = push_seq;
714 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
715 spin_unlock(&cil->xc_push_lock);
722 struct xfs_cil *cil = log->l_cilp;
725 spin_lock(&cil->xc_push_lock);
726 if (list_empty(&cil->xc_cil))
728 spin_unlock(&cil->xc_push_lock);
733 * Commit a transaction with the given vector to the Committed Item List.
735 * To do this, we need to format the item, pin it in memory if required and
736 * account for the space used by the transaction. Once we have done that we
737 * need to release the unused reservation for the transaction, attach the
738 * transaction to the checkpoint context so we carry the busy extents through
739 * to checkpoint completion, and then unlock all the items in the transaction.
741 * Called with the context lock already held in read mode to lock out
742 * background commit, returns without it held once background commits are
747 struct xfs_mount *mp,
748 struct xfs_trans *tp,
749 xfs_lsn_t *commit_lsn,
752 struct xlog *log = mp->m_log;
753 struct xfs_cil *cil = log->l_cilp;
756 if (flags & XFS_TRANS_RELEASE_LOG_RES)
757 log_flags = XFS_LOG_REL_PERM_RESERV;
759 /* lock out background commit */
760 down_read(&cil->xc_ctx_lock);
762 xlog_cil_insert_items(log, tp);
764 /* check we didn't blow the reservation */
765 if (tp->t_ticket->t_curr_res < 0)
766 xlog_print_tic_res(mp, tp->t_ticket);
768 tp->t_commit_lsn = cil->xc_ctx->sequence;
770 *commit_lsn = tp->t_commit_lsn;
772 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
773 xfs_trans_unreserve_and_mod_sb(tp);
776 * Once all the items of the transaction have been copied to the CIL,
777 * the items can be unlocked and freed.
779 * This needs to be done before we drop the CIL context lock because we
780 * have to update state in the log items and unlock them before they go
781 * to disk. If we don't, then the CIL checkpoint can race with us and
782 * we can run checkpoint completion before we've updated and unlocked
783 * the log items. This affects (at least) processing of stale buffers,
786 xfs_trans_free_items(tp, tp->t_commit_lsn, 0);
788 xlog_cil_push_background(log);
790 up_read(&cil->xc_ctx_lock);
794 * Conditionally push the CIL based on the sequence passed in.
796 * We only need to push if we haven't already pushed the sequence
797 * number given. Hence the only time we will trigger a push here is
798 * if the push sequence is the same as the current context.
800 * We return the current commit lsn to allow the callers to determine if a
801 * iclog flush is necessary following this call.
808 struct xfs_cil *cil = log->l_cilp;
809 struct xfs_cil_ctx *ctx;
810 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
812 ASSERT(sequence <= cil->xc_current_sequence);
815 * check to see if we need to force out the current context.
816 * xlog_cil_push() handles racing pushes for the same sequence,
817 * so no need to deal with it here.
820 xlog_cil_push_now(log, sequence);
823 * See if we can find a previous sequence still committing.
824 * We need to wait for all previous sequence commits to complete
825 * before allowing the force of push_seq to go ahead. Hence block
826 * on commits for those as well.
828 spin_lock(&cil->xc_push_lock);
829 list_for_each_entry(ctx, &cil->xc_committing, committing) {
831 * Avoid getting stuck in this loop because we were woken by the
832 * shutdown, but then went back to sleep once already in the
835 if (XLOG_FORCED_SHUTDOWN(log))
837 if (ctx->sequence > sequence)
839 if (!ctx->commit_lsn) {
841 * It is still being pushed! Wait for the push to
842 * complete, then start again from the beginning.
844 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
847 if (ctx->sequence != sequence)
850 commit_lsn = ctx->commit_lsn;
854 * The call to xlog_cil_push_now() executes the push in the background.
855 * Hence by the time we have got here it our sequence may not have been
856 * pushed yet. This is true if the current sequence still matches the
857 * push sequence after the above wait loop and the CIL still contains
860 * When the push occurs, it will empty the CIL and atomically increment
861 * the currect sequence past the push sequence and move it into the
862 * committing list. Of course, if the CIL is clean at the time of the
863 * push, it won't have pushed the CIL at all, so in that case we should
864 * try the push for this sequence again from the start just in case.
866 if (sequence == cil->xc_current_sequence &&
867 !list_empty(&cil->xc_cil)) {
868 spin_unlock(&cil->xc_push_lock);
872 spin_unlock(&cil->xc_push_lock);
876 * We detected a shutdown in progress. We need to trigger the log force
877 * to pass through it's iclog state machine error handling, even though
878 * we are already in a shutdown state. Hence we can't return
879 * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
880 * LSN is already stable), so we return a zero LSN instead.
883 spin_unlock(&cil->xc_push_lock);
888 * Check if the current log item was first committed in this sequence.
889 * We can't rely on just the log item being in the CIL, we have to check
890 * the recorded commit sequence number.
892 * Note: for this to be used in a non-racy manner, it has to be called with
893 * CIL flushing locked out. As a result, it should only be used during the
894 * transaction commit process when deciding what to format into the item.
897 xfs_log_item_in_current_chkpt(
898 struct xfs_log_item *lip)
900 struct xfs_cil_ctx *ctx;
902 if (list_empty(&lip->li_cil))
905 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
908 * li_seq is written on the first commit of a log item to record the
909 * first checkpoint it is written to. Hence if it is different to the
910 * current sequence, we're in a new checkpoint.
912 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
918 * Perform initial CIL structure initialisation.
925 struct xfs_cil_ctx *ctx;
927 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
931 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
937 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
938 INIT_LIST_HEAD(&cil->xc_cil);
939 INIT_LIST_HEAD(&cil->xc_committing);
940 spin_lock_init(&cil->xc_cil_lock);
941 spin_lock_init(&cil->xc_push_lock);
942 init_rwsem(&cil->xc_ctx_lock);
943 init_waitqueue_head(&cil->xc_commit_wait);
945 INIT_LIST_HEAD(&ctx->committing);
946 INIT_LIST_HEAD(&ctx->busy_extents);
950 cil->xc_current_sequence = ctx->sequence;
961 if (log->l_cilp->xc_ctx) {
962 if (log->l_cilp->xc_ctx->ticket)
963 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
964 kmem_free(log->l_cilp->xc_ctx);
967 ASSERT(list_empty(&log->l_cilp->xc_cil));
968 kmem_free(log->l_cilp);