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"
30 #include "xfs_dmapi.h"
31 #include "xfs_mount.h"
32 #include "xfs_error.h"
33 #include "xfs_alloc.h"
36 * Perform initial CIL structure initialisation. If the CIL is not
37 * enabled in this filesystem, ensure the log->l_cilp is null so
38 * we can check this conditional to determine if we are doing delayed
46 struct xfs_cil_ctx
*ctx
;
49 if (!(log
->l_mp
->m_flags
& XFS_MOUNT_DELAYLOG
))
52 cil
= kmem_zalloc(sizeof(*cil
), KM_SLEEP
|KM_MAYFAIL
);
56 ctx
= kmem_zalloc(sizeof(*ctx
), KM_SLEEP
|KM_MAYFAIL
);
62 INIT_LIST_HEAD(&cil
->xc_cil
);
63 INIT_LIST_HEAD(&cil
->xc_committing
);
64 spin_lock_init(&cil
->xc_cil_lock
);
65 init_rwsem(&cil
->xc_ctx_lock
);
66 sv_init(&cil
->xc_commit_wait
, SV_DEFAULT
, "cilwait");
68 INIT_LIST_HEAD(&ctx
->committing
);
69 INIT_LIST_HEAD(&ctx
->busy_extents
);
86 if (log
->l_cilp
->xc_ctx
) {
87 if (log
->l_cilp
->xc_ctx
->ticket
)
88 xfs_log_ticket_put(log
->l_cilp
->xc_ctx
->ticket
);
89 kmem_free(log
->l_cilp
->xc_ctx
);
92 ASSERT(list_empty(&log
->l_cilp
->xc_cil
));
93 kmem_free(log
->l_cilp
);
97 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
98 * recover, so we don't allow failure here. Also, we allocate in a context that
99 * we don't want to be issuing transactions from, so we need to tell the
100 * allocation code this as well.
102 * We don't reserve any space for the ticket - we are going to steal whatever
103 * space we require from transactions as they commit. To ensure we reserve all
104 * the space required, we need to set the current reservation of the ticket to
105 * zero so that we know to steal the initial transaction overhead from the
106 * first transaction commit.
108 static struct xlog_ticket
*
109 xlog_cil_ticket_alloc(
112 struct xlog_ticket
*tic
;
114 tic
= xlog_ticket_alloc(log
, 0, 1, XFS_TRANSACTION
, 0,
116 tic
->t_trans_type
= XFS_TRANS_CHECKPOINT
;
119 * set the current reservation to zero so we know to steal the basic
120 * transaction overhead reservation from the first transaction commit.
127 * After the first stage of log recovery is done, we know where the head and
128 * tail of the log are. We need this log initialisation done before we can
129 * initialise the first CIL checkpoint context.
131 * Here we allocate a log ticket to track space usage during a CIL push. This
132 * ticket is passed to xlog_write() directly so that we don't slowly leak log
133 * space by failing to account for space used by log headers and additional
134 * region headers for split regions.
137 xlog_cil_init_post_recovery(
143 log
->l_cilp
->xc_ctx
->ticket
= xlog_cil_ticket_alloc(log
);
144 log
->l_cilp
->xc_ctx
->sequence
= 1;
145 log
->l_cilp
->xc_ctx
->commit_lsn
= xlog_assign_lsn(log
->l_curr_cycle
,
150 * Insert the log item into the CIL and calculate the difference in space
151 * consumed by the item. Add the space to the checkpoint ticket and calculate
152 * if the change requires additional log metadata. If it does, take that space
153 * as well. Remove the amount of space we addded to the checkpoint ticket from
154 * the current transaction ticket so that the accounting works out correctly.
156 * If this is the first time the item is being placed into the CIL in this
157 * context, pin it so it can't be written to disk until the CIL is flushed to
158 * the iclog and the iclog written to disk.
163 struct xlog_ticket
*ticket
,
164 struct xfs_log_item
*item
,
165 struct xfs_log_vec
*lv
)
167 struct xfs_cil
*cil
= log
->l_cilp
;
168 struct xfs_log_vec
*old
= lv
->lv_item
->li_lv
;
169 struct xfs_cil_ctx
*ctx
= cil
->xc_ctx
;
175 /* existing lv on log item, space used is a delta */
176 ASSERT(!list_empty(&item
->li_cil
));
177 ASSERT(old
->lv_buf
&& old
->lv_buf_len
&& old
->lv_niovecs
);
179 len
= lv
->lv_buf_len
- old
->lv_buf_len
;
180 diff_iovecs
= lv
->lv_niovecs
- old
->lv_niovecs
;
181 kmem_free(old
->lv_buf
);
184 /* new lv, must pin the log item */
185 ASSERT(!lv
->lv_item
->li_lv
);
186 ASSERT(list_empty(&item
->li_cil
));
188 len
= lv
->lv_buf_len
;
189 diff_iovecs
= lv
->lv_niovecs
;
190 IOP_PIN(lv
->lv_item
);
193 len
+= diff_iovecs
* sizeof(xlog_op_header_t
);
195 /* attach new log vector to log item */
196 lv
->lv_item
->li_lv
= lv
;
198 spin_lock(&cil
->xc_cil_lock
);
199 list_move_tail(&item
->li_cil
, &cil
->xc_cil
);
200 ctx
->nvecs
+= diff_iovecs
;
203 * If this is the first time the item is being committed to the CIL,
204 * store the sequence number on the log item so we can tell
205 * in future commits whether this is the first checkpoint the item is
206 * being committed into.
209 item
->li_seq
= ctx
->sequence
;
212 * Now transfer enough transaction reservation to the context ticket
213 * for the checkpoint. The context ticket is special - the unit
214 * reservation has to grow as well as the current reservation as we
215 * steal from tickets so we can correctly determine the space used
216 * during the transaction commit.
218 if (ctx
->ticket
->t_curr_res
== 0) {
219 /* first commit in checkpoint, steal the header reservation */
220 ASSERT(ticket
->t_curr_res
>= ctx
->ticket
->t_unit_res
+ len
);
221 ctx
->ticket
->t_curr_res
= ctx
->ticket
->t_unit_res
;
222 ticket
->t_curr_res
-= ctx
->ticket
->t_unit_res
;
225 /* do we need space for more log record headers? */
226 iclog_space
= log
->l_iclog_size
- log
->l_iclog_hsize
;
227 if (len
> 0 && (ctx
->space_used
/ iclog_space
!=
228 (ctx
->space_used
+ len
) / iclog_space
)) {
231 hdrs
= (len
+ iclog_space
- 1) / iclog_space
;
232 /* need to take into account split region headers, too */
233 hdrs
*= log
->l_iclog_hsize
+ sizeof(struct xlog_op_header
);
234 ctx
->ticket
->t_unit_res
+= hdrs
;
235 ctx
->ticket
->t_curr_res
+= hdrs
;
236 ticket
->t_curr_res
-= hdrs
;
237 ASSERT(ticket
->t_curr_res
>= len
);
239 ticket
->t_curr_res
-= len
;
240 ctx
->space_used
+= len
;
242 spin_unlock(&cil
->xc_cil_lock
);
246 * Format log item into a flat buffers
248 * For delayed logging, we need to hold a formatted buffer containing all the
249 * changes on the log item. This enables us to relog the item in memory and
250 * write it out asynchronously without needing to relock the object that was
251 * modified at the time it gets written into the iclog.
253 * This function builds a vector for the changes in each log item in the
254 * transaction. It then works out the length of the buffer needed for each log
255 * item, allocates them and formats the vector for the item into the buffer.
256 * The buffer is then attached to the log item are then inserted into the
257 * Committed Item List for tracking until the next checkpoint is written out.
259 * We don't set up region headers during this process; we simply copy the
260 * regions into the flat buffer. We can do this because we still have to do a
261 * formatting step to write the regions into the iclog buffer. Writing the
262 * ophdrs during the iclog write means that we can support splitting large
263 * regions across iclog boundares without needing a change in the format of the
264 * item/region encapsulation.
266 * Hence what we need to do now is change the rewrite the vector array to point
267 * to the copied region inside the buffer we just allocated. This allows us to
268 * format the regions into the iclog as though they are being formatted
269 * directly out of the objects themselves.
272 xlog_cil_format_items(
274 struct xfs_log_vec
*log_vector
,
275 struct xlog_ticket
*ticket
,
276 xfs_lsn_t
*start_lsn
)
278 struct xfs_log_vec
*lv
;
281 *start_lsn
= log
->l_cilp
->xc_ctx
->sequence
;
284 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
289 /* build the vector array and calculate it's length */
290 IOP_FORMAT(lv
->lv_item
, lv
->lv_iovecp
);
291 for (index
= 0; index
< lv
->lv_niovecs
; index
++)
292 len
+= lv
->lv_iovecp
[index
].i_len
;
294 lv
->lv_buf_len
= len
;
295 lv
->lv_buf
= kmem_zalloc(lv
->lv_buf_len
, KM_SLEEP
|KM_NOFS
);
298 for (index
= 0; index
< lv
->lv_niovecs
; index
++) {
299 struct xfs_log_iovec
*vec
= &lv
->lv_iovecp
[index
];
301 memcpy(ptr
, vec
->i_addr
, vec
->i_len
);
305 ASSERT(ptr
== lv
->lv_buf
+ lv
->lv_buf_len
);
307 xlog_cil_insert(log
, ticket
, lv
->lv_item
, lv
);
312 xlog_cil_free_logvec(
313 struct xfs_log_vec
*log_vector
)
315 struct xfs_log_vec
*lv
;
317 for (lv
= log_vector
; lv
; ) {
318 struct xfs_log_vec
*next
= lv
->lv_next
;
319 kmem_free(lv
->lv_buf
);
326 * Commit a transaction with the given vector to the Committed Item List.
328 * To do this, we need to format the item, pin it in memory if required and
329 * account for the space used by the transaction. Once we have done that we
330 * need to release the unused reservation for the transaction, attach the
331 * transaction to the checkpoint context so we carry the busy extents through
332 * to checkpoint completion, and then unlock all the items in the transaction.
334 * For more specific information about the order of operations in
335 * xfs_log_commit_cil() please refer to the comments in
336 * xfs_trans_commit_iclog().
338 * Called with the context lock already held in read mode to lock out
339 * background commit, returns without it held once background commits are
344 struct xfs_mount
*mp
,
345 struct xfs_trans
*tp
,
346 struct xfs_log_vec
*log_vector
,
347 xfs_lsn_t
*commit_lsn
,
350 struct log
*log
= mp
->m_log
;
354 if (flags
& XFS_TRANS_RELEASE_LOG_RES
)
355 log_flags
= XFS_LOG_REL_PERM_RESERV
;
357 if (XLOG_FORCED_SHUTDOWN(log
)) {
358 xlog_cil_free_logvec(log_vector
);
359 return XFS_ERROR(EIO
);
362 /* lock out background commit */
363 down_read(&log
->l_cilp
->xc_ctx_lock
);
364 xlog_cil_format_items(log
, log_vector
, tp
->t_ticket
, commit_lsn
);
366 /* check we didn't blow the reservation */
367 if (tp
->t_ticket
->t_curr_res
< 0)
368 xlog_print_tic_res(log
->l_mp
, tp
->t_ticket
);
370 /* attach the transaction to the CIL if it has any busy extents */
371 if (!list_empty(&tp
->t_busy
)) {
372 spin_lock(&log
->l_cilp
->xc_cil_lock
);
373 list_splice_init(&tp
->t_busy
,
374 &log
->l_cilp
->xc_ctx
->busy_extents
);
375 spin_unlock(&log
->l_cilp
->xc_cil_lock
);
378 tp
->t_commit_lsn
= *commit_lsn
;
379 xfs_log_done(mp
, tp
->t_ticket
, NULL
, log_flags
);
380 xfs_trans_unreserve_and_mod_sb(tp
);
382 /* check for background commit before unlock */
383 if (log
->l_cilp
->xc_ctx
->space_used
> XLOG_CIL_SPACE_LIMIT(log
))
385 up_read(&log
->l_cilp
->xc_ctx_lock
);
388 * We need to push CIL every so often so we don't cache more than we
389 * can fit in the log. The limit really is that a checkpoint can't be
390 * more than half the log (the current checkpoint is not allowed to
391 * overwrite the previous checkpoint), but commit latency and memory
392 * usage limit this to a smaller size in most cases.
395 xlog_cil_push(log
, 0);
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_log_vec
*lv
;
411 int abortflag
= abort
? XFS_LI_ABORTED
: 0;
412 struct xfs_busy_extent
*busyp
, *n
;
414 /* unpin all the log items */
415 for (lv
= ctx
->lv_chain
; lv
; lv
= lv
->lv_next
) {
416 xfs_trans_item_committed(lv
->lv_item
, ctx
->start_lsn
,
420 list_for_each_entry_safe(busyp
, n
, &ctx
->busy_extents
, list
)
421 xfs_alloc_busy_clear(ctx
->cil
->xc_log
->l_mp
, busyp
);
423 spin_lock(&ctx
->cil
->xc_cil_lock
);
424 list_del(&ctx
->committing
);
425 spin_unlock(&ctx
->cil
->xc_cil_lock
);
427 xlog_cil_free_logvec(ctx
->lv_chain
);
432 * Push the Committed Item List to the log. If the push_now flag is not set,
433 * then it is a background flush and so we can chose to ignore it.
440 struct xfs_cil
*cil
= log
->l_cilp
;
441 struct xfs_log_vec
*lv
;
442 struct xfs_cil_ctx
*ctx
;
443 struct xfs_cil_ctx
*new_ctx
;
444 struct xlog_in_core
*commit_iclog
;
445 struct xlog_ticket
*tic
;
450 struct xfs_trans_header thdr
;
451 struct xfs_log_iovec lhdr
;
452 struct xfs_log_vec lvhdr
= { NULL
};
453 xfs_lsn_t commit_lsn
;
458 new_ctx
= kmem_zalloc(sizeof(*new_ctx
), KM_SLEEP
|KM_NOFS
);
459 new_ctx
->ticket
= xlog_cil_ticket_alloc(log
);
461 /* lock out transaction commit, but don't block on background push */
462 if (!down_write_trylock(&cil
->xc_ctx_lock
)) {
464 goto out_free_ticket
;
465 down_write(&cil
->xc_ctx_lock
);
469 /* check if we've anything to push */
470 if (list_empty(&cil
->xc_cil
))
473 /* check for spurious background flush */
474 if (!push_now
&& cil
->xc_ctx
->space_used
< XLOG_CIL_SPACE_LIMIT(log
))
478 * pull all the log vectors off the items in the CIL, and
479 * remove the items from the CIL. We don't need the CIL lock
480 * here because it's only needed on the transaction commit
481 * side which is currently locked out by the flush lock.
487 while (!list_empty(&cil
->xc_cil
)) {
488 struct xfs_log_item
*item
;
491 item
= list_first_entry(&cil
->xc_cil
,
492 struct xfs_log_item
, li_cil
);
493 list_del_init(&item
->li_cil
);
495 ctx
->lv_chain
= item
->li_lv
;
497 lv
->lv_next
= item
->li_lv
;
502 num_iovecs
+= lv
->lv_niovecs
;
503 for (i
= 0; i
< lv
->lv_niovecs
; i
++)
504 len
+= lv
->lv_iovecp
[i
].i_len
;
508 * initialise the new context and attach it to the CIL. Then attach
509 * the current context to the CIL committing lsit so it can be found
510 * during log forces to extract the commit lsn of the sequence that
511 * needs to be forced.
513 INIT_LIST_HEAD(&new_ctx
->committing
);
514 INIT_LIST_HEAD(&new_ctx
->busy_extents
);
515 new_ctx
->sequence
= ctx
->sequence
+ 1;
517 cil
->xc_ctx
= new_ctx
;
520 * The switch is now done, so we can drop the context lock and move out
521 * of a shared context. We can't just go straight to the commit record,
522 * though - we need to synchronise with previous and future commits so
523 * that the commit records are correctly ordered in the log to ensure
524 * that we process items during log IO completion in the correct order.
526 * For example, if we get an EFI in one checkpoint and the EFD in the
527 * next (e.g. due to log forces), we do not want the checkpoint with
528 * the EFD to be committed before the checkpoint with the EFI. Hence
529 * we must strictly order the commit records of the checkpoints so
530 * that: a) the checkpoint callbacks are attached to the iclogs in the
531 * correct order; and b) the checkpoints are replayed in correct order
534 * Hence we need to add this context to the committing context list so
535 * that higher sequences will wait for us to write out a commit record
538 spin_lock(&cil
->xc_cil_lock
);
539 list_add(&ctx
->committing
, &cil
->xc_committing
);
540 spin_unlock(&cil
->xc_cil_lock
);
541 up_write(&cil
->xc_ctx_lock
);
544 * Build a checkpoint transaction header and write it to the log to
545 * begin the transaction. We need to account for the space used by the
546 * transaction header here as it is not accounted for in xlog_write().
548 * The LSN we need to pass to the log items on transaction commit is
549 * the LSN reported by the first log vector write. If we use the commit
550 * record lsn then we can move the tail beyond the grant write head.
553 thdr
.th_magic
= XFS_TRANS_HEADER_MAGIC
;
554 thdr
.th_type
= XFS_TRANS_CHECKPOINT
;
555 thdr
.th_tid
= tic
->t_tid
;
556 thdr
.th_num_items
= num_iovecs
;
557 lhdr
.i_addr
= (xfs_caddr_t
)&thdr
;
558 lhdr
.i_len
= sizeof(xfs_trans_header_t
);
559 lhdr
.i_type
= XLOG_REG_TYPE_TRANSHDR
;
560 tic
->t_curr_res
-= lhdr
.i_len
+ sizeof(xlog_op_header_t
);
562 lvhdr
.lv_niovecs
= 1;
563 lvhdr
.lv_iovecp
= &lhdr
;
564 lvhdr
.lv_next
= ctx
->lv_chain
;
566 error
= xlog_write(log
, &lvhdr
, tic
, &ctx
->start_lsn
, NULL
, 0);
571 * now that we've written the checkpoint into the log, strictly
572 * order the commit records so replay will get them in the right order.
575 spin_lock(&cil
->xc_cil_lock
);
576 list_for_each_entry(new_ctx
, &cil
->xc_committing
, committing
) {
578 * Higher sequences will wait for this one so skip them.
579 * Don't wait for own own sequence, either.
581 if (new_ctx
->sequence
>= ctx
->sequence
)
583 if (!new_ctx
->commit_lsn
) {
585 * It is still being pushed! Wait for the push to
586 * complete, then start again from the beginning.
588 sv_wait(&cil
->xc_commit_wait
, 0, &cil
->xc_cil_lock
, 0);
592 spin_unlock(&cil
->xc_cil_lock
);
594 commit_lsn
= xfs_log_done(log
->l_mp
, tic
, &commit_iclog
, 0);
595 if (error
|| commit_lsn
== -1)
598 /* attach all the transactions w/ busy extents to iclog */
599 ctx
->log_cb
.cb_func
= xlog_cil_committed
;
600 ctx
->log_cb
.cb_arg
= ctx
;
601 error
= xfs_log_notify(log
->l_mp
, commit_iclog
, &ctx
->log_cb
);
606 * now the checkpoint commit is complete and we've attached the
607 * callbacks to the iclog we can assign the commit LSN to the context
608 * and wake up anyone who is waiting for the commit to complete.
610 spin_lock(&cil
->xc_cil_lock
);
611 ctx
->commit_lsn
= commit_lsn
;
612 sv_broadcast(&cil
->xc_commit_wait
);
613 spin_unlock(&cil
->xc_cil_lock
);
615 /* release the hounds! */
616 return xfs_log_release_iclog(log
->l_mp
, commit_iclog
);
619 up_write(&cil
->xc_ctx_lock
);
621 xfs_log_ticket_put(new_ctx
->ticket
);
626 xlog_cil_committed(ctx
, XFS_LI_ABORTED
);
627 return XFS_ERROR(EIO
);
631 * Conditionally push the CIL based on the sequence passed in.
633 * We only need to push if we haven't already pushed the sequence
634 * number given. Hence the only time we will trigger a push here is
635 * if the push sequence is the same as the current context.
637 * We return the current commit lsn to allow the callers to determine if a
638 * iclog flush is necessary following this call.
640 * XXX: Initially, just push the CIL unconditionally and return whatever
641 * commit lsn is there. It'll be empty, so this is broken for now.
648 struct xfs_cil
*cil
= log
->l_cilp
;
649 struct xfs_cil_ctx
*ctx
;
650 xfs_lsn_t commit_lsn
= NULLCOMMITLSN
;
653 down_write(&cil
->xc_ctx_lock
);
654 ASSERT(push_seq
<= cil
->xc_ctx
->sequence
);
656 /* check to see if we need to force out the current context */
657 if (push_seq
== cil
->xc_ctx
->sequence
) {
658 up_write(&cil
->xc_ctx_lock
);
659 xlog_cil_push(log
, 1);
664 * See if we can find a previous sequence still committing.
665 * We can drop the flush lock as soon as we have the cil lock
666 * because we are now only comparing contexts protected by
669 * We need to wait for all previous sequence commits to complete
670 * before allowing the force of push_seq to go ahead. Hence block
671 * on commits for those as well.
673 spin_lock(&cil
->xc_cil_lock
);
674 up_write(&cil
->xc_ctx_lock
);
675 list_for_each_entry(ctx
, &cil
->xc_committing
, committing
) {
676 if (ctx
->sequence
> push_seq
)
678 if (!ctx
->commit_lsn
) {
680 * It is still being pushed! Wait for the push to
681 * complete, then start again from the beginning.
683 sv_wait(&cil
->xc_commit_wait
, 0, &cil
->xc_cil_lock
, 0);
686 if (ctx
->sequence
!= push_seq
)
689 commit_lsn
= ctx
->commit_lsn
;
691 spin_unlock(&cil
->xc_cil_lock
);
696 * Check if the current log item was first committed in this sequence.
697 * We can't rely on just the log item being in the CIL, we have to check
698 * the recorded commit sequence number.
700 * Note: for this to be used in a non-racy manner, it has to be called with
701 * CIL flushing locked out. As a result, it should only be used during the
702 * transaction commit process when deciding what to format into the item.
705 xfs_log_item_in_current_chkpt(
706 struct xfs_log_item
*lip
)
708 struct xfs_cil_ctx
*ctx
;
710 if (!(lip
->li_mountp
->m_flags
& XFS_MOUNT_DELAYLOG
))
712 if (list_empty(&lip
->li_cil
))
715 ctx
= lip
->li_mountp
->m_log
->l_cilp
->xc_ctx
;
718 * li_seq is written on the first commit of a log item to record the
719 * first checkpoint it is written to. Hence if it is different to the
720 * current sequence, we're in a new checkpoint.
722 if (XFS_LSN_CMP(lip
->li_seq
, ctx
->sequence
) != 0)