1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_shared.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_error.h"
14 #include "xfs_alloc.h"
15 #include "xfs_extent_busy.h"
16 #include "xfs_discard.h"
17 #include "xfs_trans.h"
18 #include "xfs_trans_priv.h"
20 #include "xfs_log_priv.h"
21 #include "xfs_trace.h"
23 struct workqueue_struct *xfs_discard_wq;
26 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
27 * recover, so we don't allow failure here. Also, we allocate in a context that
28 * we don't want to be issuing transactions from, so we need to tell the
29 * allocation code this as well.
31 * We don't reserve any space for the ticket - we are going to steal whatever
32 * space we require from transactions as they commit. To ensure we reserve all
33 * the space required, we need to set the current reservation of the ticket to
34 * zero so that we know to steal the initial transaction overhead from the
35 * first transaction commit.
37 static struct xlog_ticket *
38 xlog_cil_ticket_alloc(
41 struct xlog_ticket *tic;
43 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
47 * set the current reservation to zero so we know to steal the basic
48 * transaction overhead reservation from the first transaction commit.
55 * After the first stage of log recovery is done, we know where the head and
56 * tail of the log are. We need this log initialisation done before we can
57 * initialise the first CIL checkpoint context.
59 * Here we allocate a log ticket to track space usage during a CIL push. This
60 * ticket is passed to xlog_write() directly so that we don't slowly leak log
61 * space by failing to account for space used by log headers and additional
62 * region headers for split regions.
65 xlog_cil_init_post_recovery(
68 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
69 log->l_cilp->xc_ctx->sequence = 1;
76 return round_up((sizeof(struct xfs_log_vec) +
77 niovecs * sizeof(struct xfs_log_iovec)),
82 * Allocate or pin log vector buffers for CIL insertion.
84 * The CIL currently uses disposable buffers for copying a snapshot of the
85 * modified items into the log during a push. The biggest problem with this is
86 * the requirement to allocate the disposable buffer during the commit if:
87 * a) does not exist; or
90 * If we do this allocation within xlog_cil_insert_format_items(), it is done
91 * under the xc_ctx_lock, which means that a CIL push cannot occur during
92 * the memory allocation. This means that we have a potential deadlock situation
93 * under low memory conditions when we have lots of dirty metadata pinned in
94 * the CIL and we need a CIL commit to occur to free memory.
96 * To avoid this, we need to move the memory allocation outside the
97 * xc_ctx_lock, but because the log vector buffers are disposable, that opens
98 * up a TOCTOU race condition w.r.t. the CIL committing and removing the log
99 * vector buffers between the check and the formatting of the item into the
100 * log vector buffer within the xc_ctx_lock.
102 * Because the log vector buffer needs to be unchanged during the CIL push
103 * process, we cannot share the buffer between the transaction commit (which
104 * modifies the buffer) and the CIL push context that is writing the changes
105 * into the log. This means skipping preallocation of buffer space is
106 * unreliable, but we most definitely do not want to be allocating and freeing
107 * buffers unnecessarily during commits when overwrites can be done safely.
109 * The simplest solution to this problem is to allocate a shadow buffer when a
110 * log item is committed for the second time, and then to only use this buffer
111 * if necessary. The buffer can remain attached to the log item until such time
112 * it is needed, and this is the buffer that is reallocated to match the size of
113 * the incoming modification. Then during the formatting of the item we can swap
114 * the active buffer with the new one if we can't reuse the existing buffer. We
115 * don't free the old buffer as it may be reused on the next modification if
116 * it's size is right, otherwise we'll free and reallocate it at that point.
118 * This function builds a vector for the changes in each log item in the
119 * transaction. It then works out the length of the buffer needed for each log
120 * item, allocates them and attaches the vector to the log item in preparation
121 * for the formatting step which occurs under the xc_ctx_lock.
123 * While this means the memory footprint goes up, it avoids the repeated
124 * alloc/free pattern that repeated modifications of an item would otherwise
125 * cause, and hence minimises the CPU overhead of such behaviour.
128 xlog_cil_alloc_shadow_bufs(
130 struct xfs_trans *tp)
132 struct xfs_log_item *lip;
134 list_for_each_entry(lip, &tp->t_items, li_trans) {
135 struct xfs_log_vec *lv;
139 bool ordered = false;
141 /* Skip items which aren't dirty in this transaction. */
142 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
145 /* get number of vecs and size of data to be stored */
146 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
149 * Ordered items need to be tracked but we do not wish to write
150 * them. We need a logvec to track the object, but we do not
151 * need an iovec or buffer to be allocated for copying data.
153 if (niovecs == XFS_LOG_VEC_ORDERED) {
160 * We 64-bit align the length of each iovec so that the start
161 * of the next one is naturally aligned. We'll need to
162 * account for that slack space here. Then round nbytes up
163 * to 64-bit alignment so that the initial buffer alignment is
164 * easy to calculate and verify.
166 nbytes += niovecs * sizeof(uint64_t);
167 nbytes = round_up(nbytes, sizeof(uint64_t));
170 * The data buffer needs to start 64-bit aligned, so round up
171 * that space to ensure we can align it appropriately and not
172 * overrun the buffer.
174 buf_size = nbytes + xlog_cil_iovec_space(niovecs);
177 * if we have no shadow buffer, or it is too small, we need to
180 if (!lip->li_lv_shadow ||
181 buf_size > lip->li_lv_shadow->lv_size) {
184 * We free and allocate here as a realloc would copy
185 * unecessary data. We don't use kmem_zalloc() for the
186 * same reason - we don't need to zero the data area in
187 * the buffer, only the log vector header and the iovec
190 kmem_free(lip->li_lv_shadow);
192 lv = kmem_alloc_large(buf_size, KM_SLEEP | KM_NOFS);
193 memset(lv, 0, xlog_cil_iovec_space(niovecs));
196 lv->lv_size = buf_size;
198 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
200 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
201 lip->li_lv_shadow = lv;
203 /* same or smaller, optimise common overwrite case */
204 lv = lip->li_lv_shadow;
206 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
213 /* Ensure the lv is set up according to ->iop_size */
214 lv->lv_niovecs = niovecs;
216 /* The allocated data region lies beyond the iovec region */
217 lv->lv_buf = (char *)lv + xlog_cil_iovec_space(niovecs);
223 * Prepare the log item for insertion into the CIL. Calculate the difference in
224 * log space and vectors it will consume, and if it is a new item pin it as
228 xfs_cil_prepare_item(
230 struct xfs_log_vec *lv,
231 struct xfs_log_vec *old_lv,
235 /* Account for the new LV being passed in */
236 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
237 *diff_len += lv->lv_bytes;
238 *diff_iovecs += lv->lv_niovecs;
242 * If there is no old LV, this is the first time we've seen the item in
243 * this CIL context and so we need to pin it. If we are replacing the
244 * old_lv, then remove the space it accounts for and make it the shadow
245 * buffer for later freeing. In both cases we are now switching to the
246 * shadow buffer, so update the the pointer to it appropriately.
249 lv->lv_item->li_ops->iop_pin(lv->lv_item);
250 lv->lv_item->li_lv_shadow = NULL;
251 } else if (old_lv != lv) {
252 ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
254 *diff_len -= old_lv->lv_bytes;
255 *diff_iovecs -= old_lv->lv_niovecs;
256 lv->lv_item->li_lv_shadow = old_lv;
259 /* attach new log vector to log item */
260 lv->lv_item->li_lv = lv;
263 * If this is the first time the item is being committed to the
264 * CIL, store the sequence number on the log item so we can
265 * tell in future commits whether this is the first checkpoint
266 * the item is being committed into.
268 if (!lv->lv_item->li_seq)
269 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
273 * Format log item into a flat buffers
275 * For delayed logging, we need to hold a formatted buffer containing all the
276 * changes on the log item. This enables us to relog the item in memory and
277 * write it out asynchronously without needing to relock the object that was
278 * modified at the time it gets written into the iclog.
280 * This function takes the prepared log vectors attached to each log item, and
281 * formats the changes into the log vector buffer. The buffer it uses is
282 * dependent on the current state of the vector in the CIL - the shadow lv is
283 * guaranteed to be large enough for the current modification, but we will only
284 * use that if we can't reuse the existing lv. If we can't reuse the existing
285 * lv, then simple swap it out for the shadow lv. We don't free it - that is
286 * done lazily either by th enext modification or the freeing of the log item.
288 * We don't set up region headers during this process; we simply copy the
289 * regions into the flat buffer. We can do this because we still have to do a
290 * formatting step to write the regions into the iclog buffer. Writing the
291 * ophdrs during the iclog write means that we can support splitting large
292 * regions across iclog boundares without needing a change in the format of the
293 * item/region encapsulation.
295 * Hence what we need to do now is change the rewrite the vector array to point
296 * to the copied region inside the buffer we just allocated. This allows us to
297 * format the regions into the iclog as though they are being formatted
298 * directly out of the objects themselves.
301 xlog_cil_insert_format_items(
303 struct xfs_trans *tp,
307 struct xfs_log_item *lip;
310 /* Bail out if we didn't find a log item. */
311 if (list_empty(&tp->t_items)) {
316 list_for_each_entry(lip, &tp->t_items, li_trans) {
317 struct xfs_log_vec *lv;
318 struct xfs_log_vec *old_lv = NULL;
319 struct xfs_log_vec *shadow;
320 bool ordered = false;
322 /* Skip items which aren't dirty in this transaction. */
323 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
327 * The formatting size information is already attached to
328 * the shadow lv on the log item.
330 shadow = lip->li_lv_shadow;
331 if (shadow->lv_buf_len == XFS_LOG_VEC_ORDERED)
334 /* Skip items that do not have any vectors for writing */
335 if (!shadow->lv_niovecs && !ordered)
338 /* compare to existing item size */
340 if (lip->li_lv && shadow->lv_size <= lip->li_lv->lv_size) {
341 /* same or smaller, optimise common overwrite case */
349 * set the item up as though it is a new insertion so
350 * that the space reservation accounting is correct.
352 *diff_iovecs -= lv->lv_niovecs;
353 *diff_len -= lv->lv_bytes;
355 /* Ensure the lv is set up according to ->iop_size */
356 lv->lv_niovecs = shadow->lv_niovecs;
358 /* reset the lv buffer information for new formatting */
361 lv->lv_buf = (char *)lv +
362 xlog_cil_iovec_space(lv->lv_niovecs);
364 /* switch to shadow buffer! */
368 /* track as an ordered logvec */
369 ASSERT(lip->li_lv == NULL);
374 ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
375 lip->li_ops->iop_format(lip, lv);
377 xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
382 * Insert the log items into the CIL and calculate the difference in space
383 * consumed by the item. Add the space to the checkpoint ticket and calculate
384 * if the change requires additional log metadata. If it does, take that space
385 * as well. Remove the amount of space we added to the checkpoint ticket from
386 * the current transaction ticket so that the accounting works out correctly.
389 xlog_cil_insert_items(
391 struct xfs_trans *tp)
393 struct xfs_cil *cil = log->l_cilp;
394 struct xfs_cil_ctx *ctx = cil->xc_ctx;
395 struct xfs_log_item *lip;
399 int iovhdr_res = 0, split_res = 0, ctx_res = 0;
404 * We can do this safely because the context can't checkpoint until we
405 * are done so it doesn't matter exactly how we update the CIL.
407 xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
409 spin_lock(&cil->xc_cil_lock);
411 /* account for space used by new iovec headers */
412 iovhdr_res = diff_iovecs * sizeof(xlog_op_header_t);
414 ctx->nvecs += diff_iovecs;
416 /* attach the transaction to the CIL if it has any busy extents */
417 if (!list_empty(&tp->t_busy))
418 list_splice_init(&tp->t_busy, &ctx->busy_extents);
421 * Now transfer enough transaction reservation to the context ticket
422 * for the checkpoint. The context ticket is special - the unit
423 * reservation has to grow as well as the current reservation as we
424 * steal from tickets so we can correctly determine the space used
425 * during the transaction commit.
427 if (ctx->ticket->t_curr_res == 0) {
428 ctx_res = ctx->ticket->t_unit_res;
429 ctx->ticket->t_curr_res = ctx_res;
430 tp->t_ticket->t_curr_res -= ctx_res;
433 /* do we need space for more log record headers? */
434 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
435 if (len > 0 && (ctx->space_used / iclog_space !=
436 (ctx->space_used + len) / iclog_space)) {
437 split_res = (len + iclog_space - 1) / iclog_space;
438 /* need to take into account split region headers, too */
439 split_res *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
440 ctx->ticket->t_unit_res += split_res;
441 ctx->ticket->t_curr_res += split_res;
442 tp->t_ticket->t_curr_res -= split_res;
443 ASSERT(tp->t_ticket->t_curr_res >= len);
445 tp->t_ticket->t_curr_res -= len;
446 ctx->space_used += len;
449 * If we've overrun the reservation, dump the tx details before we move
450 * the log items. Shutdown is imminent...
452 if (WARN_ON(tp->t_ticket->t_curr_res < 0)) {
453 xfs_warn(log->l_mp, "Transaction log reservation overrun:");
455 " log items: %d bytes (iov hdrs: %d bytes)",
457 xfs_warn(log->l_mp, " split region headers: %d bytes",
459 xfs_warn(log->l_mp, " ctx ticket: %d bytes", ctx_res);
460 xlog_print_trans(tp);
464 * Now (re-)position everything modified at the tail of the CIL.
465 * We do this here so we only need to take the CIL lock once during
466 * the transaction commit.
468 list_for_each_entry(lip, &tp->t_items, li_trans) {
470 /* Skip items which aren't dirty in this transaction. */
471 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
475 * Only move the item if it isn't already at the tail. This is
476 * to prevent a transient list_empty() state when reinserting
477 * an item that is already the only item in the CIL.
479 if (!list_is_last(&lip->li_cil, &cil->xc_cil))
480 list_move_tail(&lip->li_cil, &cil->xc_cil);
483 spin_unlock(&cil->xc_cil_lock);
485 if (tp->t_ticket->t_curr_res < 0)
486 xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR);
490 xlog_cil_free_logvec(
491 struct xfs_log_vec *log_vector)
493 struct xfs_log_vec *lv;
495 for (lv = log_vector; lv; ) {
496 struct xfs_log_vec *next = lv->lv_next;
503 xlog_discard_endio_work(
504 struct work_struct *work)
506 struct xfs_cil_ctx *ctx =
507 container_of(work, struct xfs_cil_ctx, discard_endio_work);
508 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
510 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
515 * Queue up the actual completion to a thread to avoid IRQ-safe locking for
516 * pagb_lock. Note that we need a unbounded workqueue, otherwise we might
517 * get the execution delayed up to 30 seconds for weird reasons.
523 struct xfs_cil_ctx *ctx = bio->bi_private;
525 INIT_WORK(&ctx->discard_endio_work, xlog_discard_endio_work);
526 queue_work(xfs_discard_wq, &ctx->discard_endio_work);
531 xlog_discard_busy_extents(
532 struct xfs_mount *mp,
533 struct xfs_cil_ctx *ctx)
535 struct list_head *list = &ctx->busy_extents;
536 struct xfs_extent_busy *busyp;
537 struct bio *bio = NULL;
538 struct blk_plug plug;
541 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
543 blk_start_plug(&plug);
544 list_for_each_entry(busyp, list, list) {
545 trace_xfs_discard_extent(mp, busyp->agno, busyp->bno,
548 error = __blkdev_issue_discard(mp->m_ddev_targp->bt_bdev,
549 XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno),
550 XFS_FSB_TO_BB(mp, busyp->length),
552 if (error && error != -EOPNOTSUPP) {
554 "discard failed for extent [0x%llx,%u], error %d",
555 (unsigned long long)busyp->bno,
563 bio->bi_private = ctx;
564 bio->bi_end_io = xlog_discard_endio;
567 xlog_discard_endio_work(&ctx->discard_endio_work);
569 blk_finish_plug(&plug);
573 * Mark all items committed and clear busy extents. We free the log vector
574 * chains in a separate pass so that we unpin the log items as quickly as
582 struct xfs_cil_ctx *ctx = args;
583 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
586 * If the I/O failed, we're aborting the commit and already shutdown.
587 * Wake any commit waiters before aborting the log items so we don't
588 * block async log pushers on callbacks. Async log pushers explicitly do
589 * not wait on log force completion because they may be holding locks
590 * required to unpin items.
593 spin_lock(&ctx->cil->xc_push_lock);
594 wake_up_all(&ctx->cil->xc_commit_wait);
595 spin_unlock(&ctx->cil->xc_push_lock);
598 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
599 ctx->start_lsn, abort);
601 xfs_extent_busy_sort(&ctx->busy_extents);
602 xfs_extent_busy_clear(mp, &ctx->busy_extents,
603 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
605 spin_lock(&ctx->cil->xc_push_lock);
606 list_del(&ctx->committing);
607 spin_unlock(&ctx->cil->xc_push_lock);
609 xlog_cil_free_logvec(ctx->lv_chain);
611 if (!list_empty(&ctx->busy_extents))
612 xlog_discard_busy_extents(mp, ctx);
618 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
619 * is a background flush and so we can chose to ignore it. Otherwise, if the
620 * current sequence is the same as @push_seq we need to do a flush. If
621 * @push_seq is less than the current sequence, then it has already been
622 * flushed and we don't need to do anything - the caller will wait for it to
623 * complete if necessary.
625 * @push_seq is a value rather than a flag because that allows us to do an
626 * unlocked check of the sequence number for a match. Hence we can allows log
627 * forces to run racily and not issue pushes for the same sequence twice. If we
628 * get a race between multiple pushes for the same sequence they will block on
629 * the first one and then abort, hence avoiding needless pushes.
635 struct xfs_cil *cil = log->l_cilp;
636 struct xfs_log_vec *lv;
637 struct xfs_cil_ctx *ctx;
638 struct xfs_cil_ctx *new_ctx;
639 struct xlog_in_core *commit_iclog;
640 struct xlog_ticket *tic;
643 struct xfs_trans_header thdr;
644 struct xfs_log_iovec lhdr;
645 struct xfs_log_vec lvhdr = { NULL };
646 xfs_lsn_t commit_lsn;
652 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
653 new_ctx->ticket = xlog_cil_ticket_alloc(log);
655 down_write(&cil->xc_ctx_lock);
658 spin_lock(&cil->xc_push_lock);
659 push_seq = cil->xc_push_seq;
660 ASSERT(push_seq <= ctx->sequence);
663 * Check if we've anything to push. If there is nothing, then we don't
664 * move on to a new sequence number and so we have to be able to push
665 * this sequence again later.
667 if (list_empty(&cil->xc_cil)) {
668 cil->xc_push_seq = 0;
669 spin_unlock(&cil->xc_push_lock);
674 /* check for a previously pushed seqeunce */
675 if (push_seq < cil->xc_ctx->sequence) {
676 spin_unlock(&cil->xc_push_lock);
681 * We are now going to push this context, so add it to the committing
682 * list before we do anything else. This ensures that anyone waiting on
683 * this push can easily detect the difference between a "push in
684 * progress" and "CIL is empty, nothing to do".
686 * IOWs, a wait loop can now check for:
687 * the current sequence not being found on the committing list;
689 * an unchanged sequence number
690 * to detect a push that had nothing to do and therefore does not need
691 * waiting on. If the CIL is not empty, we get put on the committing
692 * list before emptying the CIL and bumping the sequence number. Hence
693 * an empty CIL and an unchanged sequence number means we jumped out
694 * above after doing nothing.
696 * Hence the waiter will either find the commit sequence on the
697 * committing list or the sequence number will be unchanged and the CIL
698 * still dirty. In that latter case, the push has not yet started, and
699 * so the waiter will have to continue trying to check the CIL
700 * committing list until it is found. In extreme cases of delay, the
701 * sequence may fully commit between the attempts the wait makes to wait
702 * on the commit sequence.
704 list_add(&ctx->committing, &cil->xc_committing);
705 spin_unlock(&cil->xc_push_lock);
708 * pull all the log vectors off the items in the CIL, and
709 * remove the items from the CIL. We don't need the CIL lock
710 * here because it's only needed on the transaction commit
711 * side which is currently locked out by the flush lock.
715 while (!list_empty(&cil->xc_cil)) {
716 struct xfs_log_item *item;
718 item = list_first_entry(&cil->xc_cil,
719 struct xfs_log_item, li_cil);
720 list_del_init(&item->li_cil);
722 ctx->lv_chain = item->li_lv;
724 lv->lv_next = item->li_lv;
727 num_iovecs += lv->lv_niovecs;
731 * initialise the new context and attach it to the CIL. Then attach
732 * the current context to the CIL committing lsit so it can be found
733 * during log forces to extract the commit lsn of the sequence that
734 * needs to be forced.
736 INIT_LIST_HEAD(&new_ctx->committing);
737 INIT_LIST_HEAD(&new_ctx->busy_extents);
738 new_ctx->sequence = ctx->sequence + 1;
740 cil->xc_ctx = new_ctx;
743 * The switch is now done, so we can drop the context lock and move out
744 * of a shared context. We can't just go straight to the commit record,
745 * though - we need to synchronise with previous and future commits so
746 * that the commit records are correctly ordered in the log to ensure
747 * that we process items during log IO completion in the correct order.
749 * For example, if we get an EFI in one checkpoint and the EFD in the
750 * next (e.g. due to log forces), we do not want the checkpoint with
751 * the EFD to be committed before the checkpoint with the EFI. Hence
752 * we must strictly order the commit records of the checkpoints so
753 * that: a) the checkpoint callbacks are attached to the iclogs in the
754 * correct order; and b) the checkpoints are replayed in correct order
757 * Hence we need to add this context to the committing context list so
758 * that higher sequences will wait for us to write out a commit record
761 * xfs_log_force_lsn requires us to mirror the new sequence into the cil
762 * structure atomically with the addition of this sequence to the
763 * committing list. This also ensures that we can do unlocked checks
764 * against the current sequence in log forces without risking
765 * deferencing a freed context pointer.
767 spin_lock(&cil->xc_push_lock);
768 cil->xc_current_sequence = new_ctx->sequence;
769 spin_unlock(&cil->xc_push_lock);
770 up_write(&cil->xc_ctx_lock);
773 * Build a checkpoint transaction header and write it to the log to
774 * begin the transaction. We need to account for the space used by the
775 * transaction header here as it is not accounted for in xlog_write().
777 * The LSN we need to pass to the log items on transaction commit is
778 * the LSN reported by the first log vector write. If we use the commit
779 * record lsn then we can move the tail beyond the grant write head.
782 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
783 thdr.th_type = XFS_TRANS_CHECKPOINT;
784 thdr.th_tid = tic->t_tid;
785 thdr.th_num_items = num_iovecs;
787 lhdr.i_len = sizeof(xfs_trans_header_t);
788 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
789 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
791 lvhdr.lv_niovecs = 1;
792 lvhdr.lv_iovecp = &lhdr;
793 lvhdr.lv_next = ctx->lv_chain;
795 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
797 goto out_abort_free_ticket;
800 * now that we've written the checkpoint into the log, strictly
801 * order the commit records so replay will get them in the right order.
804 spin_lock(&cil->xc_push_lock);
805 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
807 * Avoid getting stuck in this loop because we were woken by the
808 * shutdown, but then went back to sleep once already in the
811 if (XLOG_FORCED_SHUTDOWN(log)) {
812 spin_unlock(&cil->xc_push_lock);
813 goto out_abort_free_ticket;
817 * Higher sequences will wait for this one so skip them.
818 * Don't wait for our own sequence, either.
820 if (new_ctx->sequence >= ctx->sequence)
822 if (!new_ctx->commit_lsn) {
824 * It is still being pushed! Wait for the push to
825 * complete, then start again from the beginning.
827 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
831 spin_unlock(&cil->xc_push_lock);
833 /* xfs_log_done always frees the ticket on error. */
834 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, false);
835 if (commit_lsn == -1)
838 /* attach all the transactions w/ busy extents to iclog */
839 ctx->log_cb.cb_func = xlog_cil_committed;
840 ctx->log_cb.cb_arg = ctx;
841 error = xfs_log_notify(commit_iclog, &ctx->log_cb);
846 * now the checkpoint commit is complete and we've attached the
847 * callbacks to the iclog we can assign the commit LSN to the context
848 * and wake up anyone who is waiting for the commit to complete.
850 spin_lock(&cil->xc_push_lock);
851 ctx->commit_lsn = commit_lsn;
852 wake_up_all(&cil->xc_commit_wait);
853 spin_unlock(&cil->xc_push_lock);
855 /* release the hounds! */
856 return xfs_log_release_iclog(log->l_mp, commit_iclog);
859 up_write(&cil->xc_ctx_lock);
860 xfs_log_ticket_put(new_ctx->ticket);
864 out_abort_free_ticket:
865 xfs_log_ticket_put(tic);
867 xlog_cil_committed(ctx, XFS_LI_ABORTED);
873 struct work_struct *work)
875 struct xfs_cil *cil = container_of(work, struct xfs_cil,
877 xlog_cil_push(cil->xc_log);
881 * We need to push CIL every so often so we don't cache more than we can fit in
882 * the log. The limit really is that a checkpoint can't be more than half the
883 * log (the current checkpoint is not allowed to overwrite the previous
884 * checkpoint), but commit latency and memory usage limit this to a smaller
888 xlog_cil_push_background(
891 struct xfs_cil *cil = log->l_cilp;
894 * The cil won't be empty because we are called while holding the
895 * context lock so whatever we added to the CIL will still be there
897 ASSERT(!list_empty(&cil->xc_cil));
900 * don't do a background push if we haven't used up all the
901 * space available yet.
903 if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
906 spin_lock(&cil->xc_push_lock);
907 if (cil->xc_push_seq < cil->xc_current_sequence) {
908 cil->xc_push_seq = cil->xc_current_sequence;
909 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
911 spin_unlock(&cil->xc_push_lock);
916 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
917 * number that is passed. When it returns, the work will be queued for
918 * @push_seq, but it won't be completed. The caller is expected to do any
919 * waiting for push_seq to complete if it is required.
926 struct xfs_cil *cil = log->l_cilp;
931 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
933 /* start on any pending background push to minimise wait time on it */
934 flush_work(&cil->xc_push_work);
937 * If the CIL is empty or we've already pushed the sequence then
938 * there's no work we need to do.
940 spin_lock(&cil->xc_push_lock);
941 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
942 spin_unlock(&cil->xc_push_lock);
946 cil->xc_push_seq = push_seq;
947 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
948 spin_unlock(&cil->xc_push_lock);
955 struct xfs_cil *cil = log->l_cilp;
958 spin_lock(&cil->xc_push_lock);
959 if (list_empty(&cil->xc_cil))
961 spin_unlock(&cil->xc_push_lock);
966 * Commit a transaction with the given vector to the Committed Item List.
968 * To do this, we need to format the item, pin it in memory if required and
969 * account for the space used by the transaction. Once we have done that we
970 * need to release the unused reservation for the transaction, attach the
971 * transaction to the checkpoint context so we carry the busy extents through
972 * to checkpoint completion, and then unlock all the items in the transaction.
974 * Called with the context lock already held in read mode to lock out
975 * background commit, returns without it held once background commits are
980 struct xfs_mount *mp,
981 struct xfs_trans *tp,
982 xfs_lsn_t *commit_lsn,
985 struct xlog *log = mp->m_log;
986 struct xfs_cil *cil = log->l_cilp;
987 xfs_lsn_t xc_commit_lsn;
990 * Do all necessary memory allocation before we lock the CIL.
991 * This ensures the allocation does not deadlock with a CIL
992 * push in memory reclaim (e.g. from kswapd).
994 xlog_cil_alloc_shadow_bufs(log, tp);
996 /* lock out background commit */
997 down_read(&cil->xc_ctx_lock);
999 xlog_cil_insert_items(log, tp);
1001 xc_commit_lsn = cil->xc_ctx->sequence;
1003 *commit_lsn = xc_commit_lsn;
1005 xfs_log_done(mp, tp->t_ticket, NULL, regrant);
1006 tp->t_ticket = NULL;
1007 xfs_trans_unreserve_and_mod_sb(tp);
1010 * Once all the items of the transaction have been copied to the CIL,
1011 * the items can be unlocked and freed.
1013 * This needs to be done before we drop the CIL context lock because we
1014 * have to update state in the log items and unlock them before they go
1015 * to disk. If we don't, then the CIL checkpoint can race with us and
1016 * we can run checkpoint completion before we've updated and unlocked
1017 * the log items. This affects (at least) processing of stale buffers,
1020 xfs_trans_free_items(tp, xc_commit_lsn, false);
1022 xlog_cil_push_background(log);
1024 up_read(&cil->xc_ctx_lock);
1028 * Conditionally push the CIL based on the sequence passed in.
1030 * We only need to push if we haven't already pushed the sequence
1031 * number given. Hence the only time we will trigger a push here is
1032 * if the push sequence is the same as the current context.
1034 * We return the current commit lsn to allow the callers to determine if a
1035 * iclog flush is necessary following this call.
1042 struct xfs_cil *cil = log->l_cilp;
1043 struct xfs_cil_ctx *ctx;
1044 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
1046 ASSERT(sequence <= cil->xc_current_sequence);
1049 * check to see if we need to force out the current context.
1050 * xlog_cil_push() handles racing pushes for the same sequence,
1051 * so no need to deal with it here.
1054 xlog_cil_push_now(log, sequence);
1057 * See if we can find a previous sequence still committing.
1058 * We need to wait for all previous sequence commits to complete
1059 * before allowing the force of push_seq to go ahead. Hence block
1060 * on commits for those as well.
1062 spin_lock(&cil->xc_push_lock);
1063 list_for_each_entry(ctx, &cil->xc_committing, committing) {
1065 * Avoid getting stuck in this loop because we were woken by the
1066 * shutdown, but then went back to sleep once already in the
1069 if (XLOG_FORCED_SHUTDOWN(log))
1071 if (ctx->sequence > sequence)
1073 if (!ctx->commit_lsn) {
1075 * It is still being pushed! Wait for the push to
1076 * complete, then start again from the beginning.
1078 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
1081 if (ctx->sequence != sequence)
1084 commit_lsn = ctx->commit_lsn;
1088 * The call to xlog_cil_push_now() executes the push in the background.
1089 * Hence by the time we have got here it our sequence may not have been
1090 * pushed yet. This is true if the current sequence still matches the
1091 * push sequence after the above wait loop and the CIL still contains
1092 * dirty objects. This is guaranteed by the push code first adding the
1093 * context to the committing list before emptying the CIL.
1095 * Hence if we don't find the context in the committing list and the
1096 * current sequence number is unchanged then the CIL contents are
1097 * significant. If the CIL is empty, if means there was nothing to push
1098 * and that means there is nothing to wait for. If the CIL is not empty,
1099 * it means we haven't yet started the push, because if it had started
1100 * we would have found the context on the committing list.
1102 if (sequence == cil->xc_current_sequence &&
1103 !list_empty(&cil->xc_cil)) {
1104 spin_unlock(&cil->xc_push_lock);
1108 spin_unlock(&cil->xc_push_lock);
1112 * We detected a shutdown in progress. We need to trigger the log force
1113 * to pass through it's iclog state machine error handling, even though
1114 * we are already in a shutdown state. Hence we can't return
1115 * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
1116 * LSN is already stable), so we return a zero LSN instead.
1119 spin_unlock(&cil->xc_push_lock);
1124 * Check if the current log item was first committed in this sequence.
1125 * We can't rely on just the log item being in the CIL, we have to check
1126 * the recorded commit sequence number.
1128 * Note: for this to be used in a non-racy manner, it has to be called with
1129 * CIL flushing locked out. As a result, it should only be used during the
1130 * transaction commit process when deciding what to format into the item.
1133 xfs_log_item_in_current_chkpt(
1134 struct xfs_log_item *lip)
1136 struct xfs_cil_ctx *ctx;
1138 if (list_empty(&lip->li_cil))
1141 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
1144 * li_seq is written on the first commit of a log item to record the
1145 * first checkpoint it is written to. Hence if it is different to the
1146 * current sequence, we're in a new checkpoint.
1148 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
1154 * Perform initial CIL structure initialisation.
1160 struct xfs_cil *cil;
1161 struct xfs_cil_ctx *ctx;
1163 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
1167 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
1173 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
1174 INIT_LIST_HEAD(&cil->xc_cil);
1175 INIT_LIST_HEAD(&cil->xc_committing);
1176 spin_lock_init(&cil->xc_cil_lock);
1177 spin_lock_init(&cil->xc_push_lock);
1178 init_rwsem(&cil->xc_ctx_lock);
1179 init_waitqueue_head(&cil->xc_commit_wait);
1181 INIT_LIST_HEAD(&ctx->committing);
1182 INIT_LIST_HEAD(&ctx->busy_extents);
1186 cil->xc_current_sequence = ctx->sequence;
1197 if (log->l_cilp->xc_ctx) {
1198 if (log->l_cilp->xc_ctx->ticket)
1199 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
1200 kmem_free(log->l_cilp->xc_ctx);
1203 ASSERT(list_empty(&log->l_cilp->xc_cil));
1204 kmem_free(log->l_cilp);