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"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_log_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_alloc.h"
34 * Perform initial CIL structure initialisation. If the CIL is not
35 * enabled in this filesystem, ensure the log->l_cilp is null so
36 * we can check this conditional to determine if we are doing delayed
44 struct xfs_cil_ctx
*ctx
;
47 if (!(log
->l_mp
->m_flags
& XFS_MOUNT_DELAYLOG
))
50 cil
= kmem_zalloc(sizeof(*cil
), KM_SLEEP
|KM_MAYFAIL
);
54 ctx
= kmem_zalloc(sizeof(*ctx
), KM_SLEEP
|KM_MAYFAIL
);
60 INIT_LIST_HEAD(&cil
->xc_cil
);
61 INIT_LIST_HEAD(&cil
->xc_committing
);
62 spin_lock_init(&cil
->xc_cil_lock
);
63 init_rwsem(&cil
->xc_ctx_lock
);
64 sv_init(&cil
->xc_commit_wait
, SV_DEFAULT
, "cilwait");
66 INIT_LIST_HEAD(&ctx
->committing
);
67 INIT_LIST_HEAD(&ctx
->busy_extents
);
84 if (log
->l_cilp
->xc_ctx
) {
85 if (log
->l_cilp
->xc_ctx
->ticket
)
86 xfs_log_ticket_put(log
->l_cilp
->xc_ctx
->ticket
);
87 kmem_free(log
->l_cilp
->xc_ctx
);
90 ASSERT(list_empty(&log
->l_cilp
->xc_cil
));
91 kmem_free(log
->l_cilp
);
95 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
96 * recover, so we don't allow failure here. Also, we allocate in a context that
97 * we don't want to be issuing transactions from, so we need to tell the
98 * allocation code this as well.
100 * We don't reserve any space for the ticket - we are going to steal whatever
101 * space we require from transactions as they commit. To ensure we reserve all
102 * the space required, we need to set the current reservation of the ticket to
103 * zero so that we know to steal the initial transaction overhead from the
104 * first transaction commit.
106 static struct xlog_ticket
*
107 xlog_cil_ticket_alloc(
110 struct xlog_ticket
*tic
;
112 tic
= xlog_ticket_alloc(log
, 0, 1, XFS_TRANSACTION
, 0,
114 tic
->t_trans_type
= XFS_TRANS_CHECKPOINT
;
117 * set the current reservation to zero so we know to steal the basic
118 * transaction overhead reservation from the first transaction commit.
125 * After the first stage of log recovery is done, we know where the head and
126 * tail of the log are. We need this log initialisation done before we can
127 * initialise the first CIL checkpoint context.
129 * Here we allocate a log ticket to track space usage during a CIL push. This
130 * ticket is passed to xlog_write() directly so that we don't slowly leak log
131 * space by failing to account for space used by log headers and additional
132 * region headers for split regions.
135 xlog_cil_init_post_recovery(
141 log
->l_cilp
->xc_ctx
->ticket
= xlog_cil_ticket_alloc(log
);
142 log
->l_cilp
->xc_ctx
->sequence
= 1;
143 log
->l_cilp
->xc_ctx
->commit_lsn
= xlog_assign_lsn(log
->l_curr_cycle
,
148 * Insert the log item into the CIL and calculate the difference in space
149 * consumed by the item. Add the space to the checkpoint ticket and calculate
150 * if the change requires additional log metadata. If it does, take that space
151 * as well. Remove the amount of space we addded to the checkpoint ticket from
152 * the current transaction ticket so that the accounting works out correctly.
154 * If this is the first time the item is being placed into the CIL in this
155 * context, pin it so it can't be written to disk until the CIL is flushed to
156 * the iclog and the iclog written to disk.
161 struct xlog_ticket
*ticket
,
162 struct xfs_log_item
*item
,
163 struct xfs_log_vec
*lv
)
165 struct xfs_cil
*cil
= log
->l_cilp
;
166 struct xfs_log_vec
*old
= lv
->lv_item
->li_lv
;
167 struct xfs_cil_ctx
*ctx
= cil
->xc_ctx
;
173 /* existing lv on log item, space used is a delta */
174 ASSERT(!list_empty(&item
->li_cil
));
175 ASSERT(old
->lv_buf
&& old
->lv_buf_len
&& old
->lv_niovecs
);
177 len
= lv
->lv_buf_len
- old
->lv_buf_len
;
178 diff_iovecs
= lv
->lv_niovecs
- old
->lv_niovecs
;
179 kmem_free(old
->lv_buf
);
182 /* new lv, must pin the log item */
183 ASSERT(!lv
->lv_item
->li_lv
);
184 ASSERT(list_empty(&item
->li_cil
));
186 len
= lv
->lv_buf_len
;
187 diff_iovecs
= lv
->lv_niovecs
;
188 IOP_PIN(lv
->lv_item
);
191 len
+= diff_iovecs
* sizeof(xlog_op_header_t
);
193 /* attach new log vector to log item */
194 lv
->lv_item
->li_lv
= lv
;
196 spin_lock(&cil
->xc_cil_lock
);
197 list_move_tail(&item
->li_cil
, &cil
->xc_cil
);
198 ctx
->nvecs
+= diff_iovecs
;
201 * If this is the first time the item is being committed to the CIL,
202 * store the sequence number on the log item so we can tell
203 * in future commits whether this is the first checkpoint the item is
204 * being committed into.
207 item
->li_seq
= ctx
->sequence
;
210 * Now transfer enough transaction reservation to the context ticket
211 * for the checkpoint. The context ticket is special - the unit
212 * reservation has to grow as well as the current reservation as we
213 * steal from tickets so we can correctly determine the space used
214 * during the transaction commit.
216 if (ctx
->ticket
->t_curr_res
== 0) {
217 /* first commit in checkpoint, steal the header reservation */
218 ASSERT(ticket
->t_curr_res
>= ctx
->ticket
->t_unit_res
+ len
);
219 ctx
->ticket
->t_curr_res
= ctx
->ticket
->t_unit_res
;
220 ticket
->t_curr_res
-= ctx
->ticket
->t_unit_res
;
223 /* do we need space for more log record headers? */
224 iclog_space
= log
->l_iclog_size
- log
->l_iclog_hsize
;
225 if (len
> 0 && (ctx
->space_used
/ iclog_space
!=
226 (ctx
->space_used
+ len
) / iclog_space
)) {
229 hdrs
= (len
+ iclog_space
- 1) / iclog_space
;
230 /* need to take into account split region headers, too */
231 hdrs
*= log
->l_iclog_hsize
+ sizeof(struct xlog_op_header
);
232 ctx
->ticket
->t_unit_res
+= hdrs
;
233 ctx
->ticket
->t_curr_res
+= hdrs
;
234 ticket
->t_curr_res
-= hdrs
;
235 ASSERT(ticket
->t_curr_res
>= len
);
237 ticket
->t_curr_res
-= len
;
238 ctx
->space_used
+= len
;
240 spin_unlock(&cil
->xc_cil_lock
);
244 * Format log item into a flat buffers
246 * For delayed logging, we need to hold a formatted buffer containing all the
247 * changes on the log item. This enables us to relog the item in memory and
248 * write it out asynchronously without needing to relock the object that was
249 * modified at the time it gets written into the iclog.
251 * This function builds a vector for the changes in each log item in the
252 * transaction. It then works out the length of the buffer needed for each log
253 * item, allocates them and formats the vector for the item into the buffer.
254 * The buffer is then attached to the log item are then inserted into the
255 * Committed Item List for tracking until the next checkpoint is written out.
257 * We don't set up region headers during this process; we simply copy the
258 * regions into the flat buffer. We can do this because we still have to do a
259 * formatting step to write the regions into the iclog buffer. Writing the
260 * ophdrs during the iclog write means that we can support splitting large
261 * regions across iclog boundares without needing a change in the format of the
262 * item/region encapsulation.
264 * Hence what we need to do now is change the rewrite the vector array to point
265 * to the copied region inside the buffer we just allocated. This allows us to
266 * format the regions into the iclog as though they are being formatted
267 * directly out of the objects themselves.
270 xlog_cil_format_items(
272 struct xfs_log_vec
*log_vector
,
273 struct xlog_ticket
*ticket
,
274 xfs_lsn_t
*start_lsn
)
276 struct xfs_log_vec
*lv
;
279 *start_lsn
= log
->l_cilp
->xc_ctx
->sequence
;
282 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
287 /* build the vector array and calculate it's length */
288 IOP_FORMAT(lv
->lv_item
, lv
->lv_iovecp
);
289 for (index
= 0; index
< lv
->lv_niovecs
; index
++)
290 len
+= lv
->lv_iovecp
[index
].i_len
;
292 lv
->lv_buf_len
= len
;
293 lv
->lv_buf
= kmem_zalloc(lv
->lv_buf_len
, KM_SLEEP
|KM_NOFS
);
296 for (index
= 0; index
< lv
->lv_niovecs
; index
++) {
297 struct xfs_log_iovec
*vec
= &lv
->lv_iovecp
[index
];
299 memcpy(ptr
, vec
->i_addr
, vec
->i_len
);
303 ASSERT(ptr
== lv
->lv_buf
+ lv
->lv_buf_len
);
305 xlog_cil_insert(log
, ticket
, lv
->lv_item
, lv
);
310 xlog_cil_free_logvec(
311 struct xfs_log_vec
*log_vector
)
313 struct xfs_log_vec
*lv
;
315 for (lv
= log_vector
; lv
; ) {
316 struct xfs_log_vec
*next
= lv
->lv_next
;
317 kmem_free(lv
->lv_buf
);
324 * Commit a transaction with the given vector to the Committed Item List.
326 * To do this, we need to format the item, pin it in memory if required and
327 * account for the space used by the transaction. Once we have done that we
328 * need to release the unused reservation for the transaction, attach the
329 * transaction to the checkpoint context so we carry the busy extents through
330 * to checkpoint completion, and then unlock all the items in the transaction.
332 * For more specific information about the order of operations in
333 * xfs_log_commit_cil() please refer to the comments in
334 * xfs_trans_commit_iclog().
336 * Called with the context lock already held in read mode to lock out
337 * background commit, returns without it held once background commits are
342 struct xfs_mount
*mp
,
343 struct xfs_trans
*tp
,
344 struct xfs_log_vec
*log_vector
,
345 xfs_lsn_t
*commit_lsn
,
348 struct log
*log
= mp
->m_log
;
352 if (flags
& XFS_TRANS_RELEASE_LOG_RES
)
353 log_flags
= XFS_LOG_REL_PERM_RESERV
;
355 if (XLOG_FORCED_SHUTDOWN(log
)) {
356 xlog_cil_free_logvec(log_vector
);
357 return XFS_ERROR(EIO
);
360 /* lock out background commit */
361 down_read(&log
->l_cilp
->xc_ctx_lock
);
362 xlog_cil_format_items(log
, log_vector
, tp
->t_ticket
, commit_lsn
);
364 /* check we didn't blow the reservation */
365 if (tp
->t_ticket
->t_curr_res
< 0)
366 xlog_print_tic_res(log
->l_mp
, tp
->t_ticket
);
368 /* attach the transaction to the CIL if it has any busy extents */
369 if (!list_empty(&tp
->t_busy
)) {
370 spin_lock(&log
->l_cilp
->xc_cil_lock
);
371 list_splice_init(&tp
->t_busy
,
372 &log
->l_cilp
->xc_ctx
->busy_extents
);
373 spin_unlock(&log
->l_cilp
->xc_cil_lock
);
376 tp
->t_commit_lsn
= *commit_lsn
;
377 xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
378 xfs_trans_unreserve_and_mod_sb(tp
);
380 /* check for background commit before unlock */
381 if (log
->l_cilp
->xc_ctx
->space_used
> XLOG_CIL_SPACE_LIMIT(log
))
383 up_read(&log
->l_cilp
->xc_ctx_lock
);
386 * We need to push CIL every so often so we don't cache more than we
387 * can fit in the log. The limit really is that a checkpoint can't be
388 * more than half the log (the current checkpoint is not allowed to
389 * overwrite the previous checkpoint), but commit latency and memory
390 * usage limit this to a smaller size in most cases.
393 xlog_cil_push(log
, 0);
398 * Mark all items committed and clear busy extents. We free the log vector
399 * chains in a separate pass so that we unpin the log items as quickly as
407 struct xfs_cil_ctx
*ctx
= args
;
408 struct xfs_log_vec
*lv
;
409 int abortflag
= abort
? XFS_LI_ABORTED
: 0;
410 struct xfs_busy_extent
*busyp
, *n
;
412 /* unpin all the log items */
413 for (lv
= ctx
->lv_chain
; lv
; lv
= lv
->lv_next
) {
414 xfs_trans_item_committed(lv
->lv_item
, ctx
->start_lsn
,
418 list_for_each_entry_safe(busyp
, n
, &ctx
->busy_extents
, list
)
419 xfs_alloc_busy_clear(ctx
->cil
->xc_log
->l_mp
, busyp
);
421 spin_lock(&ctx
->cil
->xc_cil_lock
);
422 list_del(&ctx
->committing
);
423 spin_unlock(&ctx
->cil
->xc_cil_lock
);
425 xlog_cil_free_logvec(ctx
->lv_chain
);
430 * Push the Committed Item List to the log. If the push_now flag is not set,
431 * then it is a background flush and so we can chose to ignore it.
438 struct xfs_cil
*cil
= log
->l_cilp
;
439 struct xfs_log_vec
*lv
;
440 struct xfs_cil_ctx
*ctx
;
441 struct xfs_cil_ctx
*new_ctx
;
442 struct xlog_in_core
*commit_iclog
;
443 struct xlog_ticket
*tic
;
448 struct xfs_trans_header thdr
;
449 struct xfs_log_iovec lhdr
;
450 struct xfs_log_vec lvhdr
= { NULL
};
451 xfs_lsn_t commit_lsn
;
456 new_ctx
= kmem_zalloc(sizeof(*new_ctx
), KM_SLEEP
|KM_NOFS
);
457 new_ctx
->ticket
= xlog_cil_ticket_alloc(log
);
459 /* lock out transaction commit, but don't block on background push */
460 if (!down_write_trylock(&cil
->xc_ctx_lock
)) {
462 goto out_free_ticket
;
463 down_write(&cil
->xc_ctx_lock
);
467 /* check if we've anything to push */
468 if (list_empty(&cil
->xc_cil
))
471 /* check for spurious background flush */
472 if (!push_now
&& cil
->xc_ctx
->space_used
< XLOG_CIL_SPACE_LIMIT(log
))
476 * pull all the log vectors off the items in the CIL, and
477 * remove the items from the CIL. We don't need the CIL lock
478 * here because it's only needed on the transaction commit
479 * side which is currently locked out by the flush lock.
485 while (!list_empty(&cil
->xc_cil
)) {
486 struct xfs_log_item
*item
;
489 item
= list_first_entry(&cil
->xc_cil
,
490 struct xfs_log_item
, li_cil
);
491 list_del_init(&item
->li_cil
);
493 ctx
->lv_chain
= item
->li_lv
;
495 lv
->lv_next
= item
->li_lv
;
500 num_iovecs
+= lv
->lv_niovecs
;
501 for (i
= 0; i
< lv
->lv_niovecs
; i
++)
502 len
+= lv
->lv_iovecp
[i
].i_len
;
506 * initialise the new context and attach it to the CIL. Then attach
507 * the current context to the CIL committing lsit so it can be found
508 * during log forces to extract the commit lsn of the sequence that
509 * needs to be forced.
511 INIT_LIST_HEAD(&new_ctx
->committing
);
512 INIT_LIST_HEAD(&new_ctx
->busy_extents
);
513 new_ctx
->sequence
= ctx
->sequence
+ 1;
515 cil
->xc_ctx
= new_ctx
;
518 * The switch is now done, so we can drop the context lock and move out
519 * of a shared context. We can't just go straight to the commit record,
520 * though - we need to synchronise with previous and future commits so
521 * that the commit records are correctly ordered in the log to ensure
522 * that we process items during log IO completion in the correct order.
524 * For example, if we get an EFI in one checkpoint and the EFD in the
525 * next (e.g. due to log forces), we do not want the checkpoint with
526 * the EFD to be committed before the checkpoint with the EFI. Hence
527 * we must strictly order the commit records of the checkpoints so
528 * that: a) the checkpoint callbacks are attached to the iclogs in the
529 * correct order; and b) the checkpoints are replayed in correct order
532 * Hence we need to add this context to the committing context list so
533 * that higher sequences will wait for us to write out a commit record
536 spin_lock(&cil
->xc_cil_lock
);
537 list_add(&ctx
->committing
, &cil
->xc_committing
);
538 spin_unlock(&cil
->xc_cil_lock
);
539 up_write(&cil
->xc_ctx_lock
);
542 * Build a checkpoint transaction header and write it to the log to
543 * begin the transaction. We need to account for the space used by the
544 * transaction header here as it is not accounted for in xlog_write().
546 * The LSN we need to pass to the log items on transaction commit is
547 * the LSN reported by the first log vector write. If we use the commit
548 * record lsn then we can move the tail beyond the grant write head.
551 thdr
.th_magic
= XFS_TRANS_HEADER_MAGIC
;
552 thdr
.th_type
= XFS_TRANS_CHECKPOINT
;
553 thdr
.th_tid
= tic
->t_tid
;
554 thdr
.th_num_items
= num_iovecs
;
556 lhdr
.i_len
= sizeof(xfs_trans_header_t
);
557 lhdr
.i_type
= XLOG_REG_TYPE_TRANSHDR
;
558 tic
->t_curr_res
-= lhdr
.i_len
+ sizeof(xlog_op_header_t
);
560 lvhdr
.lv_niovecs
= 1;
561 lvhdr
.lv_iovecp
= &lhdr
;
562 lvhdr
.lv_next
= ctx
->lv_chain
;
564 error
= xlog_write(log
, &lvhdr
, tic
, &ctx
->start_lsn
, NULL
, 0);
569 * now that we've written the checkpoint into the log, strictly
570 * order the commit records so replay will get them in the right order.
573 spin_lock(&cil
->xc_cil_lock
);
574 list_for_each_entry(new_ctx
, &cil
->xc_committing
, committing
) {
576 * Higher sequences will wait for this one so skip them.
577 * Don't wait for own own sequence, either.
579 if (new_ctx
->sequence
>= ctx
->sequence
)
581 if (!new_ctx
->commit_lsn
) {
583 * It is still being pushed! Wait for the push to
584 * complete, then start again from the beginning.
586 sv_wait(&cil
->xc_commit_wait
, 0, &cil
->xc_cil_lock
, 0);
590 spin_unlock(&cil
->xc_cil_lock
);
592 commit_lsn
= xfs_log_done(log
->l_mp
, tic
, &commit_iclog
, 0);
593 if (error
|| commit_lsn
== -1)
596 /* attach all the transactions w/ busy extents to iclog */
597 ctx
->log_cb
.cb_func
= xlog_cil_committed
;
598 ctx
->log_cb
.cb_arg
= ctx
;
599 error
= xfs_log_notify(log
->l_mp
, commit_iclog
, &ctx
->log_cb
);
604 * now the checkpoint commit is complete and we've attached the
605 * callbacks to the iclog we can assign the commit LSN to the context
606 * and wake up anyone who is waiting for the commit to complete.
608 spin_lock(&cil
->xc_cil_lock
);
609 ctx
->commit_lsn
= commit_lsn
;
610 sv_broadcast(&cil
->xc_commit_wait
);
611 spin_unlock(&cil
->xc_cil_lock
);
613 /* release the hounds! */
614 return xfs_log_release_iclog(log
->l_mp
, commit_iclog
);
617 up_write(&cil
->xc_ctx_lock
);
619 xfs_log_ticket_put(new_ctx
->ticket
);
624 xlog_cil_committed(ctx
, XFS_LI_ABORTED
);
625 return XFS_ERROR(EIO
);
629 * Conditionally push the CIL based on the sequence passed in.
631 * We only need to push if we haven't already pushed the sequence
632 * number given. Hence the only time we will trigger a push here is
633 * if the push sequence is the same as the current context.
635 * We return the current commit lsn to allow the callers to determine if a
636 * iclog flush is necessary following this call.
638 * XXX: Initially, just push the CIL unconditionally and return whatever
639 * commit lsn is there. It'll be empty, so this is broken for now.
646 struct xfs_cil
*cil
= log
->l_cilp
;
647 struct xfs_cil_ctx
*ctx
;
648 xfs_lsn_t commit_lsn
= NULLCOMMITLSN
;
651 down_write(&cil
->xc_ctx_lock
);
652 ASSERT(push_seq
<= cil
->xc_ctx
->sequence
);
654 /* check to see if we need to force out the current context */
655 if (push_seq
== cil
->xc_ctx
->sequence
) {
656 up_write(&cil
->xc_ctx_lock
);
657 xlog_cil_push(log
, 1);
662 * See if we can find a previous sequence still committing.
663 * We can drop the flush lock as soon as we have the cil lock
664 * because we are now only comparing contexts protected by
667 * We need to wait for all previous sequence commits to complete
668 * before allowing the force of push_seq to go ahead. Hence block
669 * on commits for those as well.
671 spin_lock(&cil
->xc_cil_lock
);
672 up_write(&cil
->xc_ctx_lock
);
673 list_for_each_entry(ctx
, &cil
->xc_committing
, committing
) {
674 if (ctx
->sequence
> push_seq
)
676 if (!ctx
->commit_lsn
) {
678 * It is still being pushed! Wait for the push to
679 * complete, then start again from the beginning.
681 sv_wait(&cil
->xc_commit_wait
, 0, &cil
->xc_cil_lock
, 0);
684 if (ctx
->sequence
!= push_seq
)
687 commit_lsn
= ctx
->commit_lsn
;
689 spin_unlock(&cil
->xc_cil_lock
);
694 * Check if the current log item was first committed in this sequence.
695 * We can't rely on just the log item being in the CIL, we have to check
696 * the recorded commit sequence number.
698 * Note: for this to be used in a non-racy manner, it has to be called with
699 * CIL flushing locked out. As a result, it should only be used during the
700 * transaction commit process when deciding what to format into the item.
703 xfs_log_item_in_current_chkpt(
704 struct xfs_log_item
*lip
)
706 struct xfs_cil_ctx
*ctx
;
708 if (!(lip
->li_mountp
->m_flags
& XFS_MOUNT_DELAYLOG
))
710 if (list_empty(&lip
->li_cil
))
713 ctx
= lip
->li_mountp
->m_log
->l_cilp
->xc_ctx
;
716 * li_seq is written on the first commit of a log item to record the
717 * first checkpoint it is written to. Hence if it is different to the
718 * current sequence, we're in a new checkpoint.
720 if (XFS_LSN_CMP(lip
->li_seq
, ctx
->sequence
) != 0)