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_types.h"
22 #include "xfs_trans.h"
23 #include "xfs_trans_priv.h"
24 #include "xfs_log_priv.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_alloc.h"
30 #include "xfs_extent_busy.h"
31 #include "xfs_discard.h"
34 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
35 * recover, so we don't allow failure here. Also, we allocate in a context that
36 * we don't want to be issuing transactions from, so we need to tell the
37 * allocation code this as well.
39 * We don't reserve any space for the ticket - we are going to steal whatever
40 * space we require from transactions as they commit. To ensure we reserve all
41 * the space required, we need to set the current reservation of the ticket to
42 * zero so that we know to steal the initial transaction overhead from the
43 * first transaction commit.
45 static struct xlog_ticket *
46 xlog_cil_ticket_alloc(
49 struct xlog_ticket *tic;
51 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
53 tic->t_trans_type = XFS_TRANS_CHECKPOINT;
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;
79 log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle,
84 xlog_cil_lv_item_format(
85 struct xfs_log_item *lip,
86 struct xfs_log_vec *lv)
91 /* format new vectors into array */
92 lip->li_ops->iop_format(lip, lv->lv_iovecp);
94 /* copy data into existing array */
96 for (index = 0; index < lv->lv_niovecs; index++) {
97 struct xfs_log_iovec *vec = &lv->lv_iovecp[index];
99 memcpy(ptr, vec->i_addr, vec->i_len);
105 * some size calculations for log vectors over-estimate, so the caller
106 * doesn't know the amount of space actually used by the item. Return
107 * the byte count to the caller so they can check and store it
110 return ptr - lv->lv_buf;
114 * Format log item into a flat buffers
116 * For delayed logging, we need to hold a formatted buffer containing all the
117 * changes on the log item. This enables us to relog the item in memory and
118 * write it out asynchronously without needing to relock the object that was
119 * modified at the time it gets written into the iclog.
121 * This function builds a vector for the changes in each log item in the
122 * transaction. It then works out the length of the buffer needed for each log
123 * item, allocates them and formats the vector for the item into the buffer.
124 * The buffer is then attached to the log item are then inserted into the
125 * Committed Item List for tracking until the next checkpoint is written out.
127 * We don't set up region headers during this process; we simply copy the
128 * regions into the flat buffer. We can do this because we still have to do a
129 * formatting step to write the regions into the iclog buffer. Writing the
130 * ophdrs during the iclog write means that we can support splitting large
131 * regions across iclog boundares without needing a change in the format of the
132 * item/region encapsulation.
134 * Hence what we need to do now is change the rewrite the vector array to point
135 * to the copied region inside the buffer we just allocated. This allows us to
136 * format the regions into the iclog as though they are being formatted
137 * directly out of the objects themselves.
139 static struct xfs_log_vec *
140 xlog_cil_prepare_log_vecs(
141 struct xfs_trans *tp)
143 struct xfs_log_item_desc *lidp;
144 struct xfs_log_vec *prev_lv = NULL;
145 struct xfs_log_vec *ret_lv = NULL;
148 /* Bail out if we didn't find a log item. */
149 if (list_empty(&tp->t_items)) {
154 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
155 struct xfs_log_item *lip = lidp->lid_item;
156 struct xfs_log_vec *lv;
160 bool ordered = false;
162 /* Skip items which aren't dirty in this transaction. */
163 if (!(lidp->lid_flags & XFS_LID_DIRTY))
166 /* get number of vecs and size of data to be stored */
167 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
169 /* Skip items that do not have any vectors for writing */
174 * Ordered items need to be tracked but we do not wish to write
175 * them. We need a logvec to track the object, but we do not
176 * need an iovec or buffer to be allocated for copying data.
178 if (niovecs == XFS_LOG_VEC_ORDERED) {
184 /* calc buffer size */
185 buf_size = sizeof(struct xfs_log_vec) + nbytes +
186 niovecs * sizeof(struct xfs_log_iovec);
188 /* compare to existing item size */
189 if (lip->li_lv && buf_size <= lip->li_lv->lv_size) {
190 /* same or smaller, optimise common overwrite case */
197 /* Ensure the lv is set up according to ->iop_size */
198 lv->lv_niovecs = niovecs;
199 lv->lv_buf = (char *)lv + buf_size - nbytes;
200 lv->lv_buf_len = xlog_cil_lv_item_format(lip, lv);
204 /* allocate new data chunk */
205 lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS);
207 lv->lv_size = buf_size;
208 lv->lv_niovecs = niovecs;
210 /* track as an ordered logvec */
211 ASSERT(lip->li_lv == NULL);
212 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
216 /* The allocated iovec region lies beyond the log vector. */
217 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
219 /* The allocated data region lies beyond the iovec region */
220 lv->lv_buf = (char *)lv + buf_size - nbytes;
222 lv->lv_buf_len = xlog_cil_lv_item_format(lip, lv);
224 ASSERT(lv->lv_buf_len <= nbytes);
228 prev_lv->lv_next = lv;
236 * Prepare the log item for insertion into the CIL. Calculate the difference in
237 * log space and vectors it will consume, and if it is a new item pin it as
241 xfs_cil_prepare_item(
243 struct xfs_log_vec *lv,
247 struct xfs_log_vec *old = lv->lv_item->li_lv;
250 /* new lv, must pin the log item */
251 ASSERT(!lv->lv_item->li_lv);
253 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
254 *len += lv->lv_buf_len;
255 *diff_iovecs += lv->lv_niovecs;
257 lv->lv_item->li_ops->iop_pin(lv->lv_item);
259 } else if (old != lv) {
260 /* existing lv on log item, space used is a delta */
261 ASSERT((old->lv_buf && old->lv_buf_len && old->lv_niovecs) ||
262 old->lv_buf_len == XFS_LOG_VEC_ORDERED);
265 * If the new item is ordered, keep the old one that is already
266 * tracking dirty or ordered regions
268 if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) {
274 *len += lv->lv_buf_len - old->lv_buf_len;
275 *diff_iovecs += lv->lv_niovecs - old->lv_niovecs;
279 /* XXX: can't account for len/diff_iovecs yet */
282 /* attach new log vector to log item */
283 lv->lv_item->li_lv = lv;
286 * If this is the first time the item is being committed to the
287 * CIL, store the sequence number on the log item so we can
288 * tell in future commits whether this is the first checkpoint
289 * the item is being committed into.
291 if (!lv->lv_item->li_seq)
292 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
296 * Insert the log items into the CIL and calculate the difference in space
297 * consumed by the item. Add the space to the checkpoint ticket and calculate
298 * if the change requires additional log metadata. If it does, take that space
299 * as well. Remove the amount of space we added to the checkpoint ticket from
300 * the current transaction ticket so that the accounting works out correctly.
303 xlog_cil_insert_items(
305 struct xfs_log_vec *log_vector,
306 struct xlog_ticket *ticket)
308 struct xfs_cil *cil = log->l_cilp;
309 struct xfs_cil_ctx *ctx = cil->xc_ctx;
310 struct xfs_log_vec *lv;
318 * Do all the accounting aggregation and switching of log vectors
319 * around in a separate loop to the insertion of items into the CIL.
320 * Then we can do a separate loop to update the CIL within a single
321 * lock/unlock pair. This reduces the number of round trips on the CIL
322 * lock from O(nr_logvectors) to O(1) and greatly reduces the overall
323 * hold time for the transaction commit.
325 * If this is the first time the item is being placed into the CIL in
326 * this context, pin it so it can't be written to disk until the CIL is
327 * flushed to the iclog and the iclog written to disk.
329 * We can do this safely because the context can't checkpoint until we
330 * are done so it doesn't matter exactly how we update the CIL.
332 spin_lock(&cil->xc_cil_lock);
333 for (lv = log_vector; lv; ) {
334 struct xfs_log_vec *next = lv->lv_next;
336 ASSERT(lv->lv_item->li_lv || list_empty(&lv->lv_item->li_cil));
340 * xfs_cil_prepare_item() may free the lv, so move the item on
343 list_move_tail(&lv->lv_item->li_cil, &cil->xc_cil);
344 xfs_cil_prepare_item(log, lv, &len, &diff_iovecs);
348 /* account for space used by new iovec headers */
349 len += diff_iovecs * sizeof(xlog_op_header_t);
350 ctx->nvecs += diff_iovecs;
353 * Now transfer enough transaction reservation to the context ticket
354 * for the checkpoint. The context ticket is special - the unit
355 * reservation has to grow as well as the current reservation as we
356 * steal from tickets so we can correctly determine the space used
357 * during the transaction commit.
359 if (ctx->ticket->t_curr_res == 0) {
360 /* first commit in checkpoint, steal the header reservation */
361 ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len);
362 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
363 ticket->t_curr_res -= ctx->ticket->t_unit_res;
366 /* do we need space for more log record headers? */
367 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
368 if (len > 0 && (ctx->space_used / iclog_space !=
369 (ctx->space_used + len) / iclog_space)) {
372 hdrs = (len + iclog_space - 1) / iclog_space;
373 /* need to take into account split region headers, too */
374 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
375 ctx->ticket->t_unit_res += hdrs;
376 ctx->ticket->t_curr_res += hdrs;
377 ticket->t_curr_res -= hdrs;
378 ASSERT(ticket->t_curr_res >= len);
380 ticket->t_curr_res -= len;
381 ctx->space_used += len;
383 spin_unlock(&cil->xc_cil_lock);
387 xlog_cil_free_logvec(
388 struct xfs_log_vec *log_vector)
390 struct xfs_log_vec *lv;
392 for (lv = log_vector; lv; ) {
393 struct xfs_log_vec *next = lv->lv_next;
400 * Mark all items committed and clear busy extents. We free the log vector
401 * chains in a separate pass so that we unpin the log items as quickly as
409 struct xfs_cil_ctx *ctx = args;
410 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
412 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
413 ctx->start_lsn, abort);
415 xfs_extent_busy_sort(&ctx->busy_extents);
416 xfs_extent_busy_clear(mp, &ctx->busy_extents,
417 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
419 spin_lock(&ctx->cil->xc_cil_lock);
420 list_del(&ctx->committing);
421 spin_unlock(&ctx->cil->xc_cil_lock);
423 xlog_cil_free_logvec(ctx->lv_chain);
425 if (!list_empty(&ctx->busy_extents)) {
426 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
428 xfs_discard_extents(mp, &ctx->busy_extents);
429 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
436 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
437 * is a background flush and so we can chose to ignore it. Otherwise, if the
438 * current sequence is the same as @push_seq we need to do a flush. If
439 * @push_seq is less than the current sequence, then it has already been
440 * flushed and we don't need to do anything - the caller will wait for it to
441 * complete if necessary.
443 * @push_seq is a value rather than a flag because that allows us to do an
444 * unlocked check of the sequence number for a match. Hence we can allows log
445 * forces to run racily and not issue pushes for the same sequence twice. If we
446 * get a race between multiple pushes for the same sequence they will block on
447 * the first one and then abort, hence avoiding needless pushes.
453 struct xfs_cil *cil = log->l_cilp;
454 struct xfs_log_vec *lv;
455 struct xfs_cil_ctx *ctx;
456 struct xfs_cil_ctx *new_ctx;
457 struct xlog_in_core *commit_iclog;
458 struct xlog_ticket *tic;
461 struct xfs_trans_header thdr;
462 struct xfs_log_iovec lhdr;
463 struct xfs_log_vec lvhdr = { NULL };
464 xfs_lsn_t commit_lsn;
470 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
471 new_ctx->ticket = xlog_cil_ticket_alloc(log);
473 down_write(&cil->xc_ctx_lock);
476 spin_lock(&cil->xc_cil_lock);
477 push_seq = cil->xc_push_seq;
478 ASSERT(push_seq <= ctx->sequence);
481 * Check if we've anything to push. If there is nothing, then we don't
482 * move on to a new sequence number and so we have to be able to push
483 * this sequence again later.
485 if (list_empty(&cil->xc_cil)) {
486 cil->xc_push_seq = 0;
487 spin_unlock(&cil->xc_cil_lock);
490 spin_unlock(&cil->xc_cil_lock);
493 /* check for a previously pushed seqeunce */
494 if (push_seq < cil->xc_ctx->sequence)
498 * pull all the log vectors off the items in the CIL, and
499 * remove the items from the CIL. We don't need the CIL lock
500 * here because it's only needed on the transaction commit
501 * side which is currently locked out by the flush lock.
505 while (!list_empty(&cil->xc_cil)) {
506 struct xfs_log_item *item;
508 item = list_first_entry(&cil->xc_cil,
509 struct xfs_log_item, li_cil);
510 list_del_init(&item->li_cil);
512 ctx->lv_chain = item->li_lv;
514 lv->lv_next = item->li_lv;
517 num_iovecs += lv->lv_niovecs;
521 * initialise the new context and attach it to the CIL. Then attach
522 * the current context to the CIL committing lsit so it can be found
523 * during log forces to extract the commit lsn of the sequence that
524 * needs to be forced.
526 INIT_LIST_HEAD(&new_ctx->committing);
527 INIT_LIST_HEAD(&new_ctx->busy_extents);
528 new_ctx->sequence = ctx->sequence + 1;
530 cil->xc_ctx = new_ctx;
533 * mirror the new sequence into the cil structure so that we can do
534 * unlocked checks against the current sequence in log forces without
535 * risking deferencing a freed context pointer.
537 cil->xc_current_sequence = new_ctx->sequence;
540 * The switch is now done, so we can drop the context lock and move out
541 * of a shared context. We can't just go straight to the commit record,
542 * though - we need to synchronise with previous and future commits so
543 * that the commit records are correctly ordered in the log to ensure
544 * that we process items during log IO completion in the correct order.
546 * For example, if we get an EFI in one checkpoint and the EFD in the
547 * next (e.g. due to log forces), we do not want the checkpoint with
548 * the EFD to be committed before the checkpoint with the EFI. Hence
549 * we must strictly order the commit records of the checkpoints so
550 * that: a) the checkpoint callbacks are attached to the iclogs in the
551 * correct order; and b) the checkpoints are replayed in correct order
554 * Hence we need to add this context to the committing context list so
555 * that higher sequences will wait for us to write out a commit record
558 spin_lock(&cil->xc_cil_lock);
559 list_add(&ctx->committing, &cil->xc_committing);
560 spin_unlock(&cil->xc_cil_lock);
561 up_write(&cil->xc_ctx_lock);
564 * Build a checkpoint transaction header and write it to the log to
565 * begin the transaction. We need to account for the space used by the
566 * transaction header here as it is not accounted for in xlog_write().
568 * The LSN we need to pass to the log items on transaction commit is
569 * the LSN reported by the first log vector write. If we use the commit
570 * record lsn then we can move the tail beyond the grant write head.
573 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
574 thdr.th_type = XFS_TRANS_CHECKPOINT;
575 thdr.th_tid = tic->t_tid;
576 thdr.th_num_items = num_iovecs;
578 lhdr.i_len = sizeof(xfs_trans_header_t);
579 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
580 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
582 lvhdr.lv_niovecs = 1;
583 lvhdr.lv_iovecp = &lhdr;
584 lvhdr.lv_next = ctx->lv_chain;
586 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
588 goto out_abort_free_ticket;
591 * now that we've written the checkpoint into the log, strictly
592 * order the commit records so replay will get them in the right order.
595 spin_lock(&cil->xc_cil_lock);
596 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
598 * Higher sequences will wait for this one so skip them.
599 * Don't wait for own own sequence, either.
601 if (new_ctx->sequence >= ctx->sequence)
603 if (!new_ctx->commit_lsn) {
605 * It is still being pushed! Wait for the push to
606 * complete, then start again from the beginning.
608 xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock);
612 spin_unlock(&cil->xc_cil_lock);
614 /* xfs_log_done always frees the ticket on error. */
615 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
616 if (commit_lsn == -1)
619 /* attach all the transactions w/ busy extents to iclog */
620 ctx->log_cb.cb_func = xlog_cil_committed;
621 ctx->log_cb.cb_arg = ctx;
622 error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
627 * now the checkpoint commit is complete and we've attached the
628 * callbacks to the iclog we can assign the commit LSN to the context
629 * and wake up anyone who is waiting for the commit to complete.
631 spin_lock(&cil->xc_cil_lock);
632 ctx->commit_lsn = commit_lsn;
633 wake_up_all(&cil->xc_commit_wait);
634 spin_unlock(&cil->xc_cil_lock);
636 /* release the hounds! */
637 return xfs_log_release_iclog(log->l_mp, commit_iclog);
640 up_write(&cil->xc_ctx_lock);
641 xfs_log_ticket_put(new_ctx->ticket);
645 out_abort_free_ticket:
646 xfs_log_ticket_put(tic);
648 xlog_cil_committed(ctx, XFS_LI_ABORTED);
649 return XFS_ERROR(EIO);
654 struct work_struct *work)
656 struct xfs_cil *cil = container_of(work, struct xfs_cil,
658 xlog_cil_push(cil->xc_log);
662 * We need to push CIL every so often so we don't cache more than we can fit in
663 * the log. The limit really is that a checkpoint can't be more than half the
664 * log (the current checkpoint is not allowed to overwrite the previous
665 * checkpoint), but commit latency and memory usage limit this to a smaller
669 xlog_cil_push_background(
672 struct xfs_cil *cil = log->l_cilp;
675 * The cil won't be empty because we are called while holding the
676 * context lock so whatever we added to the CIL will still be there
678 ASSERT(!list_empty(&cil->xc_cil));
681 * don't do a background push if we haven't used up all the
682 * space available yet.
684 if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
687 spin_lock(&cil->xc_cil_lock);
688 if (cil->xc_push_seq < cil->xc_current_sequence) {
689 cil->xc_push_seq = cil->xc_current_sequence;
690 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
692 spin_unlock(&cil->xc_cil_lock);
697 xlog_cil_push_foreground(
701 struct xfs_cil *cil = log->l_cilp;
706 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
708 /* start on any pending background push to minimise wait time on it */
709 flush_work(&cil->xc_push_work);
712 * If the CIL is empty or we've already pushed the sequence then
713 * there's no work we need to do.
715 spin_lock(&cil->xc_cil_lock);
716 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
717 spin_unlock(&cil->xc_cil_lock);
721 cil->xc_push_seq = push_seq;
722 spin_unlock(&cil->xc_cil_lock);
724 /* do the push now */
729 * Commit a transaction with the given vector to the Committed Item List.
731 * To do this, we need to format the item, pin it in memory if required and
732 * account for the space used by the transaction. Once we have done that we
733 * need to release the unused reservation for the transaction, attach the
734 * transaction to the checkpoint context so we carry the busy extents through
735 * to checkpoint completion, and then unlock all the items in the transaction.
737 * Called with the context lock already held in read mode to lock out
738 * background commit, returns without it held once background commits are
743 struct xfs_mount *mp,
744 struct xfs_trans *tp,
745 xfs_lsn_t *commit_lsn,
748 struct xlog *log = mp->m_log;
750 struct xfs_log_vec *log_vector;
752 if (flags & XFS_TRANS_RELEASE_LOG_RES)
753 log_flags = XFS_LOG_REL_PERM_RESERV;
755 /* lock out background commit */
756 down_read(&log->l_cilp->xc_ctx_lock);
758 log_vector = xlog_cil_prepare_log_vecs(tp);
763 *commit_lsn = log->l_cilp->xc_ctx->sequence;
765 /* xlog_cil_insert_items() destroys log_vector list */
766 xlog_cil_insert_items(log, log_vector, tp->t_ticket);
768 /* check we didn't blow the reservation */
769 if (tp->t_ticket->t_curr_res < 0)
770 xlog_print_tic_res(log->l_mp, tp->t_ticket);
772 /* attach the transaction to the CIL if it has any busy extents */
773 if (!list_empty(&tp->t_busy)) {
774 spin_lock(&log->l_cilp->xc_cil_lock);
775 list_splice_init(&tp->t_busy,
776 &log->l_cilp->xc_ctx->busy_extents);
777 spin_unlock(&log->l_cilp->xc_cil_lock);
780 tp->t_commit_lsn = *commit_lsn;
781 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
782 xfs_trans_unreserve_and_mod_sb(tp);
785 * Once all the items of the transaction have been copied to the CIL,
786 * the items can be unlocked and freed.
788 * This needs to be done before we drop the CIL context lock because we
789 * have to update state in the log items and unlock them before they go
790 * to disk. If we don't, then the CIL checkpoint can race with us and
791 * we can run checkpoint completion before we've updated and unlocked
792 * the log items. This affects (at least) processing of stale buffers,
795 xfs_trans_free_items(tp, *commit_lsn, 0);
797 xlog_cil_push_background(log);
799 up_read(&log->l_cilp->xc_ctx_lock);
804 * Conditionally push the CIL based on the sequence passed in.
806 * We only need to push if we haven't already pushed the sequence
807 * number given. Hence the only time we will trigger a push here is
808 * if the push sequence is the same as the current context.
810 * We return the current commit lsn to allow the callers to determine if a
811 * iclog flush is necessary following this call.
818 struct xfs_cil *cil = log->l_cilp;
819 struct xfs_cil_ctx *ctx;
820 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
822 ASSERT(sequence <= cil->xc_current_sequence);
825 * check to see if we need to force out the current context.
826 * xlog_cil_push() handles racing pushes for the same sequence,
827 * so no need to deal with it here.
829 xlog_cil_push_foreground(log, sequence);
832 * See if we can find a previous sequence still committing.
833 * We need to wait for all previous sequence commits to complete
834 * before allowing the force of push_seq to go ahead. Hence block
835 * on commits for those as well.
838 spin_lock(&cil->xc_cil_lock);
839 list_for_each_entry(ctx, &cil->xc_committing, committing) {
840 if (ctx->sequence > sequence)
842 if (!ctx->commit_lsn) {
844 * It is still being pushed! Wait for the push to
845 * complete, then start again from the beginning.
847 xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock);
850 if (ctx->sequence != sequence)
853 commit_lsn = ctx->commit_lsn;
855 spin_unlock(&cil->xc_cil_lock);
860 * Check if the current log item was first committed in this sequence.
861 * We can't rely on just the log item being in the CIL, we have to check
862 * the recorded commit sequence number.
864 * Note: for this to be used in a non-racy manner, it has to be called with
865 * CIL flushing locked out. As a result, it should only be used during the
866 * transaction commit process when deciding what to format into the item.
869 xfs_log_item_in_current_chkpt(
870 struct xfs_log_item *lip)
872 struct xfs_cil_ctx *ctx;
874 if (list_empty(&lip->li_cil))
877 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
880 * li_seq is written on the first commit of a log item to record the
881 * first checkpoint it is written to. Hence if it is different to the
882 * current sequence, we're in a new checkpoint.
884 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
890 * Perform initial CIL structure initialisation.
897 struct xfs_cil_ctx *ctx;
899 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
903 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
909 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
910 INIT_LIST_HEAD(&cil->xc_cil);
911 INIT_LIST_HEAD(&cil->xc_committing);
912 spin_lock_init(&cil->xc_cil_lock);
913 init_rwsem(&cil->xc_ctx_lock);
914 init_waitqueue_head(&cil->xc_commit_wait);
916 INIT_LIST_HEAD(&ctx->committing);
917 INIT_LIST_HEAD(&ctx->busy_extents);
921 cil->xc_current_sequence = ctx->sequence;
932 if (log->l_cilp->xc_ctx) {
933 if (log->l_cilp->xc_ctx->ticket)
934 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
935 kmem_free(log->l_cilp->xc_ctx);
938 ASSERT(list_empty(&log->l_cilp->xc_cil));
939 kmem_free(log->l_cilp);