xfs: drop dmapi hooks
[deliverable/linux.git] / fs / xfs / xfs_log_cil.c
CommitLineData
71e330b5
DC
1/*
2 * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
3 *
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.
7 *
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.
12 *
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
16 */
17
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_types.h"
21#include "xfs_bit.h"
22#include "xfs_log.h"
23#include "xfs_inum.h"
24#include "xfs_trans.h"
25#include "xfs_trans_priv.h"
26#include "xfs_log_priv.h"
27#include "xfs_sb.h"
28#include "xfs_ag.h"
29#include "xfs_dir2.h"
71e330b5
DC
30#include "xfs_mount.h"
31#include "xfs_error.h"
32#include "xfs_alloc.h"
33
34/*
35 * Perform initial CIL structure initialisation. If the CIL is not
36 * enabled in this filesystem, ensure the log->l_cilp is null so
37 * we can check this conditional to determine if we are doing delayed
38 * logging or not.
39 */
40int
41xlog_cil_init(
42 struct log *log)
43{
44 struct xfs_cil *cil;
45 struct xfs_cil_ctx *ctx;
46
47 log->l_cilp = NULL;
48 if (!(log->l_mp->m_flags & XFS_MOUNT_DELAYLOG))
49 return 0;
50
51 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
52 if (!cil)
53 return ENOMEM;
54
55 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
56 if (!ctx) {
57 kmem_free(cil);
58 return ENOMEM;
59 }
60
61 INIT_LIST_HEAD(&cil->xc_cil);
62 INIT_LIST_HEAD(&cil->xc_committing);
63 spin_lock_init(&cil->xc_cil_lock);
64 init_rwsem(&cil->xc_ctx_lock);
65 sv_init(&cil->xc_commit_wait, SV_DEFAULT, "cilwait");
66
67 INIT_LIST_HEAD(&ctx->committing);
68 INIT_LIST_HEAD(&ctx->busy_extents);
69 ctx->sequence = 1;
70 ctx->cil = cil;
71 cil->xc_ctx = ctx;
72
73 cil->xc_log = log;
74 log->l_cilp = cil;
75 return 0;
76}
77
78void
79xlog_cil_destroy(
80 struct log *log)
81{
82 if (!log->l_cilp)
83 return;
84
85 if (log->l_cilp->xc_ctx) {
86 if (log->l_cilp->xc_ctx->ticket)
87 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
88 kmem_free(log->l_cilp->xc_ctx);
89 }
90
91 ASSERT(list_empty(&log->l_cilp->xc_cil));
92 kmem_free(log->l_cilp);
93}
94
95/*
96 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
97 * recover, so we don't allow failure here. Also, we allocate in a context that
98 * we don't want to be issuing transactions from, so we need to tell the
99 * allocation code this as well.
100 *
101 * We don't reserve any space for the ticket - we are going to steal whatever
102 * space we require from transactions as they commit. To ensure we reserve all
103 * the space required, we need to set the current reservation of the ticket to
104 * zero so that we know to steal the initial transaction overhead from the
105 * first transaction commit.
106 */
107static struct xlog_ticket *
108xlog_cil_ticket_alloc(
109 struct log *log)
110{
111 struct xlog_ticket *tic;
112
113 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
114 KM_SLEEP|KM_NOFS);
115 tic->t_trans_type = XFS_TRANS_CHECKPOINT;
116
117 /*
118 * set the current reservation to zero so we know to steal the basic
119 * transaction overhead reservation from the first transaction commit.
120 */
121 tic->t_curr_res = 0;
122 return tic;
123}
124
125/*
126 * After the first stage of log recovery is done, we know where the head and
127 * tail of the log are. We need this log initialisation done before we can
128 * initialise the first CIL checkpoint context.
129 *
130 * Here we allocate a log ticket to track space usage during a CIL push. This
131 * ticket is passed to xlog_write() directly so that we don't slowly leak log
132 * space by failing to account for space used by log headers and additional
133 * region headers for split regions.
134 */
135void
136xlog_cil_init_post_recovery(
137 struct log *log)
138{
139 if (!log->l_cilp)
140 return;
141
142 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
143 log->l_cilp->xc_ctx->sequence = 1;
144 log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle,
145 log->l_curr_block);
146}
147
148/*
149 * Insert the log item into the CIL and calculate the difference in space
150 * consumed by the item. Add the space to the checkpoint ticket and calculate
151 * if the change requires additional log metadata. If it does, take that space
152 * as well. Remove the amount of space we addded to the checkpoint ticket from
153 * the current transaction ticket so that the accounting works out correctly.
154 *
155 * If this is the first time the item is being placed into the CIL in this
156 * context, pin it so it can't be written to disk until the CIL is flushed to
157 * the iclog and the iclog written to disk.
158 */
159static void
160xlog_cil_insert(
161 struct log *log,
162 struct xlog_ticket *ticket,
163 struct xfs_log_item *item,
164 struct xfs_log_vec *lv)
165{
166 struct xfs_cil *cil = log->l_cilp;
167 struct xfs_log_vec *old = lv->lv_item->li_lv;
168 struct xfs_cil_ctx *ctx = cil->xc_ctx;
169 int len;
170 int diff_iovecs;
171 int iclog_space;
172
173 if (old) {
174 /* existing lv on log item, space used is a delta */
175 ASSERT(!list_empty(&item->li_cil));
176 ASSERT(old->lv_buf && old->lv_buf_len && old->lv_niovecs);
177
178 len = lv->lv_buf_len - old->lv_buf_len;
179 diff_iovecs = lv->lv_niovecs - old->lv_niovecs;
180 kmem_free(old->lv_buf);
181 kmem_free(old);
182 } else {
183 /* new lv, must pin the log item */
184 ASSERT(!lv->lv_item->li_lv);
185 ASSERT(list_empty(&item->li_cil));
186
187 len = lv->lv_buf_len;
188 diff_iovecs = lv->lv_niovecs;
189 IOP_PIN(lv->lv_item);
190
191 }
192 len += diff_iovecs * sizeof(xlog_op_header_t);
193
194 /* attach new log vector to log item */
195 lv->lv_item->li_lv = lv;
196
197 spin_lock(&cil->xc_cil_lock);
198 list_move_tail(&item->li_cil, &cil->xc_cil);
199 ctx->nvecs += diff_iovecs;
200
ccf7c23f
DC
201 /*
202 * If this is the first time the item is being committed to the CIL,
203 * store the sequence number on the log item so we can tell
204 * in future commits whether this is the first checkpoint the item is
205 * being committed into.
206 */
207 if (!item->li_seq)
208 item->li_seq = ctx->sequence;
209
71e330b5
DC
210 /*
211 * Now transfer enough transaction reservation to the context ticket
212 * for the checkpoint. The context ticket is special - the unit
213 * reservation has to grow as well as the current reservation as we
214 * steal from tickets so we can correctly determine the space used
215 * during the transaction commit.
216 */
217 if (ctx->ticket->t_curr_res == 0) {
218 /* first commit in checkpoint, steal the header reservation */
219 ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len);
220 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
221 ticket->t_curr_res -= ctx->ticket->t_unit_res;
222 }
223
224 /* do we need space for more log record headers? */
225 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
226 if (len > 0 && (ctx->space_used / iclog_space !=
227 (ctx->space_used + len) / iclog_space)) {
228 int hdrs;
229
230 hdrs = (len + iclog_space - 1) / iclog_space;
231 /* need to take into account split region headers, too */
232 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
233 ctx->ticket->t_unit_res += hdrs;
234 ctx->ticket->t_curr_res += hdrs;
235 ticket->t_curr_res -= hdrs;
236 ASSERT(ticket->t_curr_res >= len);
237 }
238 ticket->t_curr_res -= len;
239 ctx->space_used += len;
240
241 spin_unlock(&cil->xc_cil_lock);
242}
243
244/*
245 * Format log item into a flat buffers
246 *
247 * For delayed logging, we need to hold a formatted buffer containing all the
248 * changes on the log item. This enables us to relog the item in memory and
249 * write it out asynchronously without needing to relock the object that was
250 * modified at the time it gets written into the iclog.
251 *
252 * This function builds a vector for the changes in each log item in the
253 * transaction. It then works out the length of the buffer needed for each log
254 * item, allocates them and formats the vector for the item into the buffer.
255 * The buffer is then attached to the log item are then inserted into the
256 * Committed Item List for tracking until the next checkpoint is written out.
257 *
258 * We don't set up region headers during this process; we simply copy the
259 * regions into the flat buffer. We can do this because we still have to do a
260 * formatting step to write the regions into the iclog buffer. Writing the
261 * ophdrs during the iclog write means that we can support splitting large
262 * regions across iclog boundares without needing a change in the format of the
263 * item/region encapsulation.
264 *
265 * Hence what we need to do now is change the rewrite the vector array to point
266 * to the copied region inside the buffer we just allocated. This allows us to
267 * format the regions into the iclog as though they are being formatted
268 * directly out of the objects themselves.
269 */
270static void
271xlog_cil_format_items(
272 struct log *log,
273 struct xfs_log_vec *log_vector,
274 struct xlog_ticket *ticket,
275 xfs_lsn_t *start_lsn)
276{
277 struct xfs_log_vec *lv;
278
279 if (start_lsn)
280 *start_lsn = log->l_cilp->xc_ctx->sequence;
281
282 ASSERT(log_vector);
283 for (lv = log_vector; lv; lv = lv->lv_next) {
284 void *ptr;
285 int index;
286 int len = 0;
287
288 /* build the vector array and calculate it's length */
289 IOP_FORMAT(lv->lv_item, lv->lv_iovecp);
290 for (index = 0; index < lv->lv_niovecs; index++)
291 len += lv->lv_iovecp[index].i_len;
292
293 lv->lv_buf_len = len;
294 lv->lv_buf = kmem_zalloc(lv->lv_buf_len, KM_SLEEP|KM_NOFS);
295 ptr = lv->lv_buf;
296
297 for (index = 0; index < lv->lv_niovecs; index++) {
298 struct xfs_log_iovec *vec = &lv->lv_iovecp[index];
299
300 memcpy(ptr, vec->i_addr, vec->i_len);
301 vec->i_addr = ptr;
302 ptr += vec->i_len;
303 }
304 ASSERT(ptr == lv->lv_buf + lv->lv_buf_len);
305
306 xlog_cil_insert(log, ticket, lv->lv_item, lv);
307 }
308}
309
310static void
311xlog_cil_free_logvec(
312 struct xfs_log_vec *log_vector)
313{
314 struct xfs_log_vec *lv;
315
316 for (lv = log_vector; lv; ) {
317 struct xfs_log_vec *next = lv->lv_next;
318 kmem_free(lv->lv_buf);
319 kmem_free(lv);
320 lv = next;
321 }
322}
323
324/*
325 * Commit a transaction with the given vector to the Committed Item List.
326 *
327 * To do this, we need to format the item, pin it in memory if required and
328 * account for the space used by the transaction. Once we have done that we
329 * need to release the unused reservation for the transaction, attach the
330 * transaction to the checkpoint context so we carry the busy extents through
331 * to checkpoint completion, and then unlock all the items in the transaction.
332 *
333 * For more specific information about the order of operations in
334 * xfs_log_commit_cil() please refer to the comments in
335 * xfs_trans_commit_iclog().
ccf7c23f
DC
336 *
337 * Called with the context lock already held in read mode to lock out
338 * background commit, returns without it held once background commits are
339 * allowed again.
71e330b5
DC
340 */
341int
342xfs_log_commit_cil(
343 struct xfs_mount *mp,
344 struct xfs_trans *tp,
345 struct xfs_log_vec *log_vector,
346 xfs_lsn_t *commit_lsn,
347 int flags)
348{
349 struct log *log = mp->m_log;
350 int log_flags = 0;
df806158 351 int push = 0;
71e330b5
DC
352
353 if (flags & XFS_TRANS_RELEASE_LOG_RES)
354 log_flags = XFS_LOG_REL_PERM_RESERV;
355
356 if (XLOG_FORCED_SHUTDOWN(log)) {
357 xlog_cil_free_logvec(log_vector);
358 return XFS_ERROR(EIO);
359 }
360
361 /* lock out background commit */
362 down_read(&log->l_cilp->xc_ctx_lock);
363 xlog_cil_format_items(log, log_vector, tp->t_ticket, commit_lsn);
364
365 /* check we didn't blow the reservation */
366 if (tp->t_ticket->t_curr_res < 0)
367 xlog_print_tic_res(log->l_mp, tp->t_ticket);
368
369 /* attach the transaction to the CIL if it has any busy extents */
370 if (!list_empty(&tp->t_busy)) {
371 spin_lock(&log->l_cilp->xc_cil_lock);
372 list_splice_init(&tp->t_busy,
373 &log->l_cilp->xc_ctx->busy_extents);
374 spin_unlock(&log->l_cilp->xc_cil_lock);
375 }
376
377 tp->t_commit_lsn = *commit_lsn;
378 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
379 xfs_trans_unreserve_and_mod_sb(tp);
380
df806158
DC
381 /* check for background commit before unlock */
382 if (log->l_cilp->xc_ctx->space_used > XLOG_CIL_SPACE_LIMIT(log))
383 push = 1;
71e330b5 384 up_read(&log->l_cilp->xc_ctx_lock);
df806158
DC
385
386 /*
387 * We need to push CIL every so often so we don't cache more than we
388 * can fit in the log. The limit really is that a checkpoint can't be
389 * more than half the log (the current checkpoint is not allowed to
390 * overwrite the previous checkpoint), but commit latency and memory
391 * usage limit this to a smaller size in most cases.
392 */
393 if (push)
394 xlog_cil_push(log, 0);
71e330b5
DC
395 return 0;
396}
397
398/*
399 * Mark all items committed and clear busy extents. We free the log vector
400 * chains in a separate pass so that we unpin the log items as quickly as
401 * possible.
402 */
403static void
404xlog_cil_committed(
405 void *args,
406 int abort)
407{
408 struct xfs_cil_ctx *ctx = args;
409 struct xfs_log_vec *lv;
410 int abortflag = abort ? XFS_LI_ABORTED : 0;
411 struct xfs_busy_extent *busyp, *n;
412
413 /* unpin all the log items */
414 for (lv = ctx->lv_chain; lv; lv = lv->lv_next ) {
415 xfs_trans_item_committed(lv->lv_item, ctx->start_lsn,
416 abortflag);
417 }
418
419 list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list)
420 xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, busyp);
421
422 spin_lock(&ctx->cil->xc_cil_lock);
423 list_del(&ctx->committing);
424 spin_unlock(&ctx->cil->xc_cil_lock);
425
426 xlog_cil_free_logvec(ctx->lv_chain);
427 kmem_free(ctx);
428}
429
430/*
431 * Push the Committed Item List to the log. If the push_now flag is not set,
432 * then it is a background flush and so we can chose to ignore it.
433 */
434int
435xlog_cil_push(
436 struct log *log,
437 int push_now)
438{
439 struct xfs_cil *cil = log->l_cilp;
440 struct xfs_log_vec *lv;
441 struct xfs_cil_ctx *ctx;
442 struct xfs_cil_ctx *new_ctx;
443 struct xlog_in_core *commit_iclog;
444 struct xlog_ticket *tic;
445 int num_lv;
446 int num_iovecs;
447 int len;
448 int error = 0;
449 struct xfs_trans_header thdr;
450 struct xfs_log_iovec lhdr;
451 struct xfs_log_vec lvhdr = { NULL };
452 xfs_lsn_t commit_lsn;
453
454 if (!cil)
455 return 0;
456
71e330b5
DC
457 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
458 new_ctx->ticket = xlog_cil_ticket_alloc(log);
459
df806158
DC
460 /* lock out transaction commit, but don't block on background push */
461 if (!down_write_trylock(&cil->xc_ctx_lock)) {
462 if (!push_now)
463 goto out_free_ticket;
464 down_write(&cil->xc_ctx_lock);
465 }
71e330b5
DC
466 ctx = cil->xc_ctx;
467
468 /* check if we've anything to push */
469 if (list_empty(&cil->xc_cil))
470 goto out_skip;
471
df806158
DC
472 /* check for spurious background flush */
473 if (!push_now && cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
474 goto out_skip;
475
71e330b5
DC
476 /*
477 * pull all the log vectors off the items in the CIL, and
478 * remove the items from the CIL. We don't need the CIL lock
479 * here because it's only needed on the transaction commit
480 * side which is currently locked out by the flush lock.
481 */
482 lv = NULL;
483 num_lv = 0;
484 num_iovecs = 0;
485 len = 0;
486 while (!list_empty(&cil->xc_cil)) {
487 struct xfs_log_item *item;
488 int i;
489
490 item = list_first_entry(&cil->xc_cil,
491 struct xfs_log_item, li_cil);
492 list_del_init(&item->li_cil);
493 if (!ctx->lv_chain)
494 ctx->lv_chain = item->li_lv;
495 else
496 lv->lv_next = item->li_lv;
497 lv = item->li_lv;
498 item->li_lv = NULL;
499
500 num_lv++;
501 num_iovecs += lv->lv_niovecs;
502 for (i = 0; i < lv->lv_niovecs; i++)
503 len += lv->lv_iovecp[i].i_len;
504 }
505
506 /*
507 * initialise the new context and attach it to the CIL. Then attach
508 * the current context to the CIL committing lsit so it can be found
509 * during log forces to extract the commit lsn of the sequence that
510 * needs to be forced.
511 */
512 INIT_LIST_HEAD(&new_ctx->committing);
513 INIT_LIST_HEAD(&new_ctx->busy_extents);
514 new_ctx->sequence = ctx->sequence + 1;
515 new_ctx->cil = cil;
516 cil->xc_ctx = new_ctx;
517
518 /*
519 * The switch is now done, so we can drop the context lock and move out
520 * of a shared context. We can't just go straight to the commit record,
521 * though - we need to synchronise with previous and future commits so
522 * that the commit records are correctly ordered in the log to ensure
523 * that we process items during log IO completion in the correct order.
524 *
525 * For example, if we get an EFI in one checkpoint and the EFD in the
526 * next (e.g. due to log forces), we do not want the checkpoint with
527 * the EFD to be committed before the checkpoint with the EFI. Hence
528 * we must strictly order the commit records of the checkpoints so
529 * that: a) the checkpoint callbacks are attached to the iclogs in the
530 * correct order; and b) the checkpoints are replayed in correct order
531 * in log recovery.
532 *
533 * Hence we need to add this context to the committing context list so
534 * that higher sequences will wait for us to write out a commit record
535 * before they do.
536 */
537 spin_lock(&cil->xc_cil_lock);
538 list_add(&ctx->committing, &cil->xc_committing);
539 spin_unlock(&cil->xc_cil_lock);
540 up_write(&cil->xc_ctx_lock);
541
542 /*
543 * Build a checkpoint transaction header and write it to the log to
544 * begin the transaction. We need to account for the space used by the
545 * transaction header here as it is not accounted for in xlog_write().
546 *
547 * The LSN we need to pass to the log items on transaction commit is
548 * the LSN reported by the first log vector write. If we use the commit
549 * record lsn then we can move the tail beyond the grant write head.
550 */
551 tic = ctx->ticket;
552 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
553 thdr.th_type = XFS_TRANS_CHECKPOINT;
554 thdr.th_tid = tic->t_tid;
555 thdr.th_num_items = num_iovecs;
556 lhdr.i_addr = (xfs_caddr_t)&thdr;
557 lhdr.i_len = sizeof(xfs_trans_header_t);
558 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
559 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
560
561 lvhdr.lv_niovecs = 1;
562 lvhdr.lv_iovecp = &lhdr;
563 lvhdr.lv_next = ctx->lv_chain;
564
565 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
566 if (error)
567 goto out_abort;
568
569 /*
570 * now that we've written the checkpoint into the log, strictly
571 * order the commit records so replay will get them in the right order.
572 */
573restart:
574 spin_lock(&cil->xc_cil_lock);
575 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
576 /*
577 * Higher sequences will wait for this one so skip them.
578 * Don't wait for own own sequence, either.
579 */
580 if (new_ctx->sequence >= ctx->sequence)
581 continue;
582 if (!new_ctx->commit_lsn) {
583 /*
584 * It is still being pushed! Wait for the push to
585 * complete, then start again from the beginning.
586 */
587 sv_wait(&cil->xc_commit_wait, 0, &cil->xc_cil_lock, 0);
588 goto restart;
589 }
590 }
591 spin_unlock(&cil->xc_cil_lock);
592
593 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
594 if (error || commit_lsn == -1)
595 goto out_abort;
596
597 /* attach all the transactions w/ busy extents to iclog */
598 ctx->log_cb.cb_func = xlog_cil_committed;
599 ctx->log_cb.cb_arg = ctx;
600 error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
601 if (error)
602 goto out_abort;
603
604 /*
605 * now the checkpoint commit is complete and we've attached the
606 * callbacks to the iclog we can assign the commit LSN to the context
607 * and wake up anyone who is waiting for the commit to complete.
608 */
609 spin_lock(&cil->xc_cil_lock);
610 ctx->commit_lsn = commit_lsn;
611 sv_broadcast(&cil->xc_commit_wait);
612 spin_unlock(&cil->xc_cil_lock);
613
614 /* release the hounds! */
615 return xfs_log_release_iclog(log->l_mp, commit_iclog);
616
617out_skip:
618 up_write(&cil->xc_ctx_lock);
df806158 619out_free_ticket:
71e330b5
DC
620 xfs_log_ticket_put(new_ctx->ticket);
621 kmem_free(new_ctx);
622 return 0;
623
624out_abort:
625 xlog_cil_committed(ctx, XFS_LI_ABORTED);
626 return XFS_ERROR(EIO);
627}
628
629/*
630 * Conditionally push the CIL based on the sequence passed in.
631 *
632 * We only need to push if we haven't already pushed the sequence
633 * number given. Hence the only time we will trigger a push here is
634 * if the push sequence is the same as the current context.
635 *
636 * We return the current commit lsn to allow the callers to determine if a
637 * iclog flush is necessary following this call.
638 *
639 * XXX: Initially, just push the CIL unconditionally and return whatever
640 * commit lsn is there. It'll be empty, so this is broken for now.
641 */
642xfs_lsn_t
643xlog_cil_push_lsn(
644 struct log *log,
645 xfs_lsn_t push_seq)
646{
647 struct xfs_cil *cil = log->l_cilp;
648 struct xfs_cil_ctx *ctx;
649 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
650
651restart:
652 down_write(&cil->xc_ctx_lock);
653 ASSERT(push_seq <= cil->xc_ctx->sequence);
654
655 /* check to see if we need to force out the current context */
656 if (push_seq == cil->xc_ctx->sequence) {
657 up_write(&cil->xc_ctx_lock);
658 xlog_cil_push(log, 1);
659 goto restart;
660 }
661
662 /*
663 * See if we can find a previous sequence still committing.
664 * We can drop the flush lock as soon as we have the cil lock
665 * because we are now only comparing contexts protected by
666 * the cil lock.
667 *
668 * We need to wait for all previous sequence commits to complete
669 * before allowing the force of push_seq to go ahead. Hence block
670 * on commits for those as well.
671 */
672 spin_lock(&cil->xc_cil_lock);
673 up_write(&cil->xc_ctx_lock);
674 list_for_each_entry(ctx, &cil->xc_committing, committing) {
675 if (ctx->sequence > push_seq)
676 continue;
677 if (!ctx->commit_lsn) {
678 /*
679 * It is still being pushed! Wait for the push to
680 * complete, then start again from the beginning.
681 */
682 sv_wait(&cil->xc_commit_wait, 0, &cil->xc_cil_lock, 0);
683 goto restart;
684 }
685 if (ctx->sequence != push_seq)
686 continue;
687 /* found it! */
688 commit_lsn = ctx->commit_lsn;
689 }
690 spin_unlock(&cil->xc_cil_lock);
691 return commit_lsn;
692}
ccf7c23f
DC
693
694/*
695 * Check if the current log item was first committed in this sequence.
696 * We can't rely on just the log item being in the CIL, we have to check
697 * the recorded commit sequence number.
698 *
699 * Note: for this to be used in a non-racy manner, it has to be called with
700 * CIL flushing locked out. As a result, it should only be used during the
701 * transaction commit process when deciding what to format into the item.
702 */
703bool
704xfs_log_item_in_current_chkpt(
705 struct xfs_log_item *lip)
706{
707 struct xfs_cil_ctx *ctx;
708
709 if (!(lip->li_mountp->m_flags & XFS_MOUNT_DELAYLOG))
710 return false;
711 if (list_empty(&lip->li_cil))
712 return false;
713
714 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
715
716 /*
717 * li_seq is written on the first commit of a log item to record the
718 * first checkpoint it is written to. Hence if it is different to the
719 * current sequence, we're in a new checkpoint.
720 */
721 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
722 return false;
723 return true;
724}
This page took 0.063868 seconds and 5 git commands to generate.