1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved.
4 */
5
6 #include "xfs.h"
7 #include "xfs_fs.h"
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_extent_busy.h"
14 #include "xfs_trans.h"
15 #include "xfs_trans_priv.h"
16 #include "xfs_log.h"
17 #include "xfs_log_priv.h"
18 #include "xfs_trace.h"
19
20 struct workqueue_struct *xfs_discard_wq;
21
22 /*
23 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
24 * recover, so we don't allow failure here. Also, we allocate in a context that
25 * we don't want to be issuing transactions from, so we need to tell the
26 * allocation code this as well.
27 *
28 * We don't reserve any space for the ticket - we are going to steal whatever
29 * space we require from transactions as they commit. To ensure we reserve all
30 * the space required, we need to set the current reservation of the ticket to
31 * zero so that we know to steal the initial transaction overhead from the
32 * first transaction commit.
33 */
34 static struct xlog_ticket *
35 xlog_cil_ticket_alloc(
36 struct xlog *log)
37 {
38 struct xlog_ticket *tic;
39
40 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
41 KM_NOFS);
42
43 /*
44 * set the current reservation to zero so we know to steal the basic
45 * transaction overhead reservation from the first transaction commit.
46 */
47 tic->t_curr_res = 0;
48 return tic;
49 }
50
51 /*
52 * After the first stage of log recovery is done, we know where the head and
53 * tail of the log are. We need this log initialisation done before we can
54 * initialise the first CIL checkpoint context.
55 *
56 * Here we allocate a log ticket to track space usage during a CIL push. This
57 * ticket is passed to xlog_write() directly so that we don't slowly leak log
58 * space by failing to account for space used by log headers and additional
59 * region headers for split regions.
60 */
61 void
62 xlog_cil_init_post_recovery(
63 struct xlog *log)
64 {
65 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
66 log->l_cilp->xc_ctx->sequence = 1;
67 }
68
69 static inline int
70 xlog_cil_iovec_space(
71 uint niovecs)
72 {
73 return round_up((sizeof(struct xfs_log_vec) +
74 niovecs * sizeof(struct xfs_log_iovec)),
75 sizeof(uint64_t));
76 }
77
78 /*
79 * Allocate or pin log vector buffers for CIL insertion.
80 *
81 * The CIL currently uses disposable buffers for copying a snapshot of the
82 * modified items into the log during a push. The biggest problem with this is
83 * the requirement to allocate the disposable buffer during the commit if:
84 * a) does not exist; or
85 * b) it is too small
86 *
87 * If we do this allocation within xlog_cil_insert_format_items(), it is done
88 * under the xc_ctx_lock, which means that a CIL push cannot occur during
89 * the memory allocation. This means that we have a potential deadlock situation
90 * under low memory conditions when we have lots of dirty metadata pinned in
91 * the CIL and we need a CIL commit to occur to free memory.
92 *
93 * To avoid this, we need to move the memory allocation outside the
94 * xc_ctx_lock, but because the log vector buffers are disposable, that opens
95 * up a TOCTOU race condition w.r.t. the CIL committing and removing the log
96 * vector buffers between the check and the formatting of the item into the
97 * log vector buffer within the xc_ctx_lock.
98 *
99 * Because the log vector buffer needs to be unchanged during the CIL push
100 * process, we cannot share the buffer between the transaction commit (which
101 * modifies the buffer) and the CIL push context that is writing the changes
102 * into the log. This means skipping preallocation of buffer space is
103 * unreliable, but we most definitely do not want to be allocating and freeing
104 * buffers unnecessarily during commits when overwrites can be done safely.
105 *
106 * The simplest solution to this problem is to allocate a shadow buffer when a
107 * log item is committed for the second time, and then to only use this buffer
108 * if necessary. The buffer can remain attached to the log item until such time
109 * it is needed, and this is the buffer that is reallocated to match the size of
110 * the incoming modification. Then during the formatting of the item we can swap
111 * the active buffer with the new one if we can't reuse the existing buffer. We
112 * don't free the old buffer as it may be reused on the next modification if
113 * it's size is right, otherwise we'll free and reallocate it at that point.
114 *
115 * This function builds a vector for the changes in each log item in the
116 * transaction. It then works out the length of the buffer needed for each log
117 * item, allocates them and attaches the vector to the log item in preparation
118 * for the formatting step which occurs under the xc_ctx_lock.
119 *
120 * While this means the memory footprint goes up, it avoids the repeated
121 * alloc/free pattern that repeated modifications of an item would otherwise
122 * cause, and hence minimises the CPU overhead of such behaviour.
123 */
124 static void
125 xlog_cil_alloc_shadow_bufs(
126 struct xlog *log,
127 struct xfs_trans *tp)
128 {
129 struct xfs_log_item *lip;
130
131 list_for_each_entry(lip, &tp->t_items, li_trans) {
132 struct xfs_log_vec *lv;
133 int niovecs = 0;
134 int nbytes = 0;
135 int buf_size;
136 bool ordered = false;
137
138 /* Skip items which aren't dirty in this transaction. */
139 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
140 continue;
141
142 /* get number of vecs and size of data to be stored */
143 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
144
145 /*
146 * Ordered items need to be tracked but we do not wish to write
147 * them. We need a logvec to track the object, but we do not
148 * need an iovec or buffer to be allocated for copying data.
149 */
150 if (niovecs == XFS_LOG_VEC_ORDERED) {
151 ordered = true;
152 niovecs = 0;
153 nbytes = 0;
154 }
155
156 /*
157 * We 64-bit align the length of each iovec so that the start
158 * of the next one is naturally aligned. We'll need to
159 * account for that slack space here. Then round nbytes up
160 * to 64-bit alignment so that the initial buffer alignment is
161 * easy to calculate and verify.
162 */
163 nbytes += niovecs * sizeof(uint64_t);
164 nbytes = round_up(nbytes, sizeof(uint64_t));
165
166 /*
167 * The data buffer needs to start 64-bit aligned, so round up
168 * that space to ensure we can align it appropriately and not
169 * overrun the buffer.
170 */
171 buf_size = nbytes + xlog_cil_iovec_space(niovecs);
172
173 /*
174 * if we have no shadow buffer, or it is too small, we need to
175 * reallocate it.
176 */
177 if (!lip->li_lv_shadow ||
178 buf_size > lip->li_lv_shadow->lv_size) {
179
180 /*
181 * We free and allocate here as a realloc would copy
182 * unecessary data. We don't use kmem_zalloc() for the
183 * same reason - we don't need to zero the data area in
184 * the buffer, only the log vector header and the iovec
185 * storage.
186 */
187 kmem_free(lip->li_lv_shadow);
188
189 lv = kmem_alloc_large(buf_size, KM_NOFS);
190 memset(lv, 0, xlog_cil_iovec_space(niovecs));
191
192 lv->lv_item = lip;
193 lv->lv_size = buf_size;
194 if (ordered)
195 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
196 else
197 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
198 lip->li_lv_shadow = lv;
199 } else {
200 /* same or smaller, optimise common overwrite case */
201 lv = lip->li_lv_shadow;
202 if (ordered)
203 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
204 else
205 lv->lv_buf_len = 0;
206 lv->lv_bytes = 0;
207 lv->lv_next = NULL;
208 }
209
210 /* Ensure the lv is set up according to ->iop_size */
211 lv->lv_niovecs = niovecs;
212
213 /* The allocated data region lies beyond the iovec region */
214 lv->lv_buf = (char *)lv + xlog_cil_iovec_space(niovecs);
215 }
216
217 }
218
219 /*
220 * Prepare the log item for insertion into the CIL. Calculate the difference in
221 * log space and vectors it will consume, and if it is a new item pin it as
222 * well.
223 */
224 STATIC void
225 xfs_cil_prepare_item(
226 struct xlog *log,
227 struct xfs_log_vec *lv,
228 struct xfs_log_vec *old_lv,
229 int *diff_len,
230 int *diff_iovecs)
231 {
232 /* Account for the new LV being passed in */
233 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
234 *diff_len += lv->lv_bytes;
235 *diff_iovecs += lv->lv_niovecs;
236 }
237
238 /*
239 * If there is no old LV, this is the first time we've seen the item in
240 * this CIL context and so we need to pin it. If we are replacing the
241 * old_lv, then remove the space it accounts for and make it the shadow
242 * buffer for later freeing. In both cases we are now switching to the
243 * shadow buffer, so update the the pointer to it appropriately.
244 */
245 if (!old_lv) {
246 if (lv->lv_item->li_ops->iop_pin)
247 lv->lv_item->li_ops->iop_pin(lv->lv_item);
248 lv->lv_item->li_lv_shadow = NULL;
249 } else if (old_lv != lv) {
250 ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
251
252 *diff_len -= old_lv->lv_bytes;
253 *diff_iovecs -= old_lv->lv_niovecs;
254 lv->lv_item->li_lv_shadow = old_lv;
255 }
256
257 /* attach new log vector to log item */
258 lv->lv_item->li_lv = lv;
259
260 /*
261 * If this is the first time the item is being committed to the
262 * CIL, store the sequence number on the log item so we can
263 * tell in future commits whether this is the first checkpoint
264 * the item is being committed into.
265 */
266 if (!lv->lv_item->li_seq)
267 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
268 }
269
270 /*
271 * Format log item into a flat buffers
272 *
273 * For delayed logging, we need to hold a formatted buffer containing all the
274 * changes on the log item. This enables us to relog the item in memory and
275 * write it out asynchronously without needing to relock the object that was
276 * modified at the time it gets written into the iclog.
277 *
278 * This function takes the prepared log vectors attached to each log item, and
279 * formats the changes into the log vector buffer. The buffer it uses is
280 * dependent on the current state of the vector in the CIL - the shadow lv is
281 * guaranteed to be large enough for the current modification, but we will only
282 * use that if we can't reuse the existing lv. If we can't reuse the existing
283 * lv, then simple swap it out for the shadow lv. We don't free it - that is
284 * done lazily either by th enext modification or the freeing of the log item.
285 *
286 * We don't set up region headers during this process; we simply copy the
287 * regions into the flat buffer. We can do this because we still have to do a
288 * formatting step to write the regions into the iclog buffer. Writing the
289 * ophdrs during the iclog write means that we can support splitting large
290 * regions across iclog boundares without needing a change in the format of the
291 * item/region encapsulation.
292 *
293 * Hence what we need to do now is change the rewrite the vector array to point
294 * to the copied region inside the buffer we just allocated. This allows us to
295 * format the regions into the iclog as though they are being formatted
296 * directly out of the objects themselves.
297 */
298 static void
299 xlog_cil_insert_format_items(
300 struct xlog *log,
301 struct xfs_trans *tp,
302 int *diff_len,
303 int *diff_iovecs)
304 {
305 struct xfs_log_item *lip;
306
307
308 /* Bail out if we didn't find a log item. */
309 if (list_empty(&tp->t_items)) {
310 ASSERT(0);
311 return;
312 }
313
314 list_for_each_entry(lip, &tp->t_items, li_trans) {
315 struct xfs_log_vec *lv;
316 struct xfs_log_vec *old_lv = NULL;
317 struct xfs_log_vec *shadow;
318 bool ordered = false;
319
320 /* Skip items which aren't dirty in this transaction. */
321 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
322 continue;
323
324 /*
325 * The formatting size information is already attached to
326 * the shadow lv on the log item.
327 */
328 shadow = lip->li_lv_shadow;
329 if (shadow->lv_buf_len == XFS_LOG_VEC_ORDERED)
330 ordered = true;
331
332 /* Skip items that do not have any vectors for writing */
333 if (!shadow->lv_niovecs && !ordered)
334 continue;
335
336 /* compare to existing item size */
337 old_lv = lip->li_lv;
338 if (lip->li_lv && shadow->lv_size <= lip->li_lv->lv_size) {
339 /* same or smaller, optimise common overwrite case */
340 lv = lip->li_lv;
341 lv->lv_next = NULL;
342
343 if (ordered)
344 goto insert;
345
346 /*
347 * set the item up as though it is a new insertion so
348 * that the space reservation accounting is correct.
349 */
350 *diff_iovecs -= lv->lv_niovecs;
351 *diff_len -= lv->lv_bytes;
352
353 /* Ensure the lv is set up according to ->iop_size */
354 lv->lv_niovecs = shadow->lv_niovecs;
355
356 /* reset the lv buffer information for new formatting */
357 lv->lv_buf_len = 0;
358 lv->lv_bytes = 0;
359 lv->lv_buf = (char *)lv +
360 xlog_cil_iovec_space(lv->lv_niovecs);
361 } else {
362 /* switch to shadow buffer! */
363 lv = shadow;
364 lv->lv_item = lip;
365 if (ordered) {
366 /* track as an ordered logvec */
367 ASSERT(lip->li_lv == NULL);
368 goto insert;
369 }
370 }
371
372 ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
373 lip->li_ops->iop_format(lip, lv);
374 insert:
375 xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
376 }
377 }
378
379 /*
380 * Insert the log items into the CIL and calculate the difference in space
381 * consumed by the item. Add the space to the checkpoint ticket and calculate
382 * if the change requires additional log metadata. If it does, take that space
383 * as well. Remove the amount of space we added to the checkpoint ticket from
384 * the current transaction ticket so that the accounting works out correctly.
385 */
386 static void
387 xlog_cil_insert_items(
388 struct xlog *log,
389 struct xfs_trans *tp)
390 {
391 struct xfs_cil *cil = log->l_cilp;
392 struct xfs_cil_ctx *ctx = cil->xc_ctx;
393 struct xfs_log_item *lip;
394 int len = 0;
395 int diff_iovecs = 0;
396 int iclog_space;
397 int iovhdr_res = 0, split_res = 0, ctx_res = 0;
398
399 ASSERT(tp);
400
401 /*
402 * We can do this safely because the context can't checkpoint until we
403 * are done so it doesn't matter exactly how we update the CIL.
404 */
405 xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
406
407 spin_lock(&cil->xc_cil_lock);
408
409 /* account for space used by new iovec headers */
410 iovhdr_res = diff_iovecs * sizeof(xlog_op_header_t);
411 len += iovhdr_res;
412 ctx->nvecs += diff_iovecs;
413
414 /* attach the transaction to the CIL if it has any busy extents */
415 if (!list_empty(&tp->t_busy))
416 list_splice_init(&tp->t_busy, &ctx->busy_extents);
417
418 /*
419 * Now transfer enough transaction reservation to the context ticket
420 * for the checkpoint. The context ticket is special - the unit
421 * reservation has to grow as well as the current reservation as we
422 * steal from tickets so we can correctly determine the space used
423 * during the transaction commit.
424 */
425 if (ctx->ticket->t_curr_res == 0) {
426 ctx_res = ctx->ticket->t_unit_res;
427 ctx->ticket->t_curr_res = ctx_res;
428 tp->t_ticket->t_curr_res -= ctx_res;
429 }
430
431 /* do we need space for more log record headers? */
432 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
433 if (len > 0 && (ctx->space_used / iclog_space !=
434 (ctx->space_used + len) / iclog_space)) {
435 split_res = (len + iclog_space - 1) / iclog_space;
436 /* need to take into account split region headers, too */
437 split_res *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
438 ctx->ticket->t_unit_res += split_res;
439 ctx->ticket->t_curr_res += split_res;
440 tp->t_ticket->t_curr_res -= split_res;
441 ASSERT(tp->t_ticket->t_curr_res >= len);
442 }
443 tp->t_ticket->t_curr_res -= len;
444 ctx->space_used += len;
445
446 /*
447 * If we've overrun the reservation, dump the tx details before we move
448 * the log items. Shutdown is imminent...
449 */
450 if (WARN_ON(tp->t_ticket->t_curr_res < 0)) {
451 xfs_warn(log->l_mp, "Transaction log reservation overrun:");
452 xfs_warn(log->l_mp,
453 " log items: %d bytes (iov hdrs: %d bytes)",
454 len, iovhdr_res);
455 xfs_warn(log->l_mp, " split region headers: %d bytes",
456 split_res);
457 xfs_warn(log->l_mp, " ctx ticket: %d bytes", ctx_res);
458 xlog_print_trans(tp);
459 }
460
461 /*
462 * Now (re-)position everything modified at the tail of the CIL.
463 * We do this here so we only need to take the CIL lock once during
464 * the transaction commit.
465 */
466 list_for_each_entry(lip, &tp->t_items, li_trans) {
467
468 /* Skip items which aren't dirty in this transaction. */
469 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
470 continue;
471
472 /*
473 * Only move the item if it isn't already at the tail. This is
474 * to prevent a transient list_empty() state when reinserting
475 * an item that is already the only item in the CIL.
476 */
477 if (!list_is_last(&lip->li_cil, &cil->xc_cil))
478 list_move_tail(&lip->li_cil, &cil->xc_cil);
479 }
480
481 spin_unlock(&cil->xc_cil_lock);
482
483 if (tp->t_ticket->t_curr_res < 0)
484 xfs_force_shutdown(log->l_mp, SHUTDOWN_LOG_IO_ERROR);
485 }
486
487 static void
488 xlog_cil_free_logvec(
489 struct xfs_log_vec *log_vector)
490 {
491 struct xfs_log_vec *lv;
492
493 for (lv = log_vector; lv; ) {
494 struct xfs_log_vec *next = lv->lv_next;
495 kmem_free(lv);
496 lv = next;
497 }
498 }
499
500 static void
501 xlog_discard_endio_work(
502 struct work_struct *work)
503 {
504 struct xfs_cil_ctx *ctx =
505 container_of(work, struct xfs_cil_ctx, discard_endio_work);
506 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
507
508 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
509 kmem_free(ctx);
510 }
511
512 /*
513 * Queue up the actual completion to a thread to avoid IRQ-safe locking for
514 * pagb_lock. Note that we need a unbounded workqueue, otherwise we might
515 * get the execution delayed up to 30 seconds for weird reasons.
516 */
517 static void
518 xlog_discard_endio(
519 struct bio *bio)
520 {
521 struct xfs_cil_ctx *ctx = bio->bi_private;
522
523 INIT_WORK(&ctx->discard_endio_work, xlog_discard_endio_work);
524 queue_work(xfs_discard_wq, &ctx->discard_endio_work);
525 bio_put(bio);
526 }
527
528 static void
529 xlog_discard_busy_extents(
530 struct xfs_mount *mp,
531 struct xfs_cil_ctx *ctx)
532 {
533 struct list_head *list = &ctx->busy_extents;
534 struct xfs_extent_busy *busyp;
535 struct bio *bio = NULL;
536 struct blk_plug plug;
537 int error = 0;
538
539 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
540
541 blk_start_plug(&plug);
542 list_for_each_entry(busyp, list, list) {
543 trace_xfs_discard_extent(mp, busyp->agno, busyp->bno,
544 busyp->length);
545
546 error = __blkdev_issue_discard(mp->m_ddev_targp->bt_bdev,
547 XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno),
548 XFS_FSB_TO_BB(mp, busyp->length),
549 GFP_NOFS, 0, &bio);
550 if (error && error != -EOPNOTSUPP) {
551 xfs_info(mp,
552 "discard failed for extent [0x%llx,%u], error %d",
553 (unsigned long long)busyp->bno,
554 busyp->length,
555 error);
556 break;
557 }
558 }
559
560 if (bio) {
561 bio->bi_private = ctx;
562 bio->bi_end_io = xlog_discard_endio;
563 submit_bio(bio);
564 } else {
565 xlog_discard_endio_work(&ctx->discard_endio_work);
566 }
567 blk_finish_plug(&plug);
568 }
569
570 /*
571 * Mark all items committed and clear busy extents. We free the log vector
572 * chains in a separate pass so that we unpin the log items as quickly as
573 * possible.
574 */
575 static void
576 xlog_cil_committed(
577 struct xfs_cil_ctx *ctx,
578 bool abort)
579 {
580 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
581
582 /*
583 * If the I/O failed, we're aborting the commit and already shutdown.
584 * Wake any commit waiters before aborting the log items so we don't
585 * block async log pushers on callbacks. Async log pushers explicitly do
586 * not wait on log force completion because they may be holding locks
587 * required to unpin items.
588 */
589 if (abort) {
590 spin_lock(&ctx->cil->xc_push_lock);
591 wake_up_all(&ctx->cil->xc_commit_wait);
592 spin_unlock(&ctx->cil->xc_push_lock);
593 }
594
595 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
596 ctx->start_lsn, abort);
597
598 xfs_extent_busy_sort(&ctx->busy_extents);
599 xfs_extent_busy_clear(mp, &ctx->busy_extents,
600 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
601
602 spin_lock(&ctx->cil->xc_push_lock);
603 list_del(&ctx->committing);
604 spin_unlock(&ctx->cil->xc_push_lock);
605
606 xlog_cil_free_logvec(ctx->lv_chain);
607
608 if (!list_empty(&ctx->busy_extents))
609 xlog_discard_busy_extents(mp, ctx);
610 else
611 kmem_free(ctx);
612 }
613
614 void
615 xlog_cil_process_committed(
616 struct list_head *list,
617 bool aborted)
618 {
619 struct xfs_cil_ctx *ctx;
620
621 while ((ctx = list_first_entry_or_null(list,
622 struct xfs_cil_ctx, iclog_entry))) {
623 list_del(&ctx->iclog_entry);
624 xlog_cil_committed(ctx, aborted);
625 }
626 }
627
628 /*
629 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
630 * is a background flush and so we can chose to ignore it. Otherwise, if the
631 * current sequence is the same as @push_seq we need to do a flush. If
632 * @push_seq is less than the current sequence, then it has already been
633 * flushed and we don't need to do anything - the caller will wait for it to
634 * complete if necessary.
635 *
636 * @push_seq is a value rather than a flag because that allows us to do an
637 * unlocked check of the sequence number for a match. Hence we can allows log
638 * forces to run racily and not issue pushes for the same sequence twice. If we
639 * get a race between multiple pushes for the same sequence they will block on
640 * the first one and then abort, hence avoiding needless pushes.
641 */
642 STATIC int
643 xlog_cil_push(
644 struct xlog *log)
645 {
646 struct xfs_cil *cil = log->l_cilp;
647 struct xfs_log_vec *lv;
648 struct xfs_cil_ctx *ctx;
649 struct xfs_cil_ctx *new_ctx;
650 struct xlog_in_core *commit_iclog;
651 struct xlog_ticket *tic;
652 int num_iovecs;
653 int error = 0;
654 struct xfs_trans_header thdr;
655 struct xfs_log_iovec lhdr;
656 struct xfs_log_vec lvhdr = { NULL };
657 xfs_lsn_t commit_lsn;
658 xfs_lsn_t push_seq;
659
660 if (!cil)
661 return 0;
662
663 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_NOFS);
664 new_ctx->ticket = xlog_cil_ticket_alloc(log);
665
666 down_write(&cil->xc_ctx_lock);
667 ctx = cil->xc_ctx;
668
669 spin_lock(&cil->xc_push_lock);
670 push_seq = cil->xc_push_seq;
671 ASSERT(push_seq <= ctx->sequence);
672
673 /*
674 * Check if we've anything to push. If there is nothing, then we don't
675 * move on to a new sequence number and so we have to be able to push
676 * this sequence again later.
677 */
678 if (list_empty(&cil->xc_cil)) {
679 cil->xc_push_seq = 0;
680 spin_unlock(&cil->xc_push_lock);
681 goto out_skip;
682 }
683
684
685 /* check for a previously pushed seqeunce */
686 if (push_seq < cil->xc_ctx->sequence) {
687 spin_unlock(&cil->xc_push_lock);
688 goto out_skip;
689 }
690
691 /*
692 * We are now going to push this context, so add it to the committing
693 * list before we do anything else. This ensures that anyone waiting on
694 * this push can easily detect the difference between a "push in
695 * progress" and "CIL is empty, nothing to do".
696 *
697 * IOWs, a wait loop can now check for:
698 * the current sequence not being found on the committing list;
699 * an empty CIL; and
700 * an unchanged sequence number
701 * to detect a push that had nothing to do and therefore does not need
702 * waiting on. If the CIL is not empty, we get put on the committing
703 * list before emptying the CIL and bumping the sequence number. Hence
704 * an empty CIL and an unchanged sequence number means we jumped out
705 * above after doing nothing.
706 *
707 * Hence the waiter will either find the commit sequence on the
708 * committing list or the sequence number will be unchanged and the CIL
709 * still dirty. In that latter case, the push has not yet started, and
710 * so the waiter will have to continue trying to check the CIL
711 * committing list until it is found. In extreme cases of delay, the
712 * sequence may fully commit between the attempts the wait makes to wait
713 * on the commit sequence.
714 */
715 list_add(&ctx->committing, &cil->xc_committing);
716 spin_unlock(&cil->xc_push_lock);
717
718 /*
719 * pull all the log vectors off the items in the CIL, and
720 * remove the items from the CIL. We don't need the CIL lock
721 * here because it's only needed on the transaction commit
722 * side which is currently locked out by the flush lock.
723 */
724 lv = NULL;
725 num_iovecs = 0;
726 while (!list_empty(&cil->xc_cil)) {
727 struct xfs_log_item *item;
728
729 item = list_first_entry(&cil->xc_cil,
730 struct xfs_log_item, li_cil);
731 list_del_init(&item->li_cil);
732 if (!ctx->lv_chain)
733 ctx->lv_chain = item->li_lv;
734 else
735 lv->lv_next = item->li_lv;
736 lv = item->li_lv;
737 item->li_lv = NULL;
738 num_iovecs += lv->lv_niovecs;
739 }
740
741 /*
742 * initialise the new context and attach it to the CIL. Then attach
743 * the current context to the CIL committing lsit so it can be found
744 * during log forces to extract the commit lsn of the sequence that
745 * needs to be forced.
746 */
747 INIT_LIST_HEAD(&new_ctx->committing);
748 INIT_LIST_HEAD(&new_ctx->busy_extents);
749 new_ctx->sequence = ctx->sequence + 1;
750 new_ctx->cil = cil;
751 cil->xc_ctx = new_ctx;
752
753 /*
754 * The switch is now done, so we can drop the context lock and move out
755 * of a shared context. We can't just go straight to the commit record,
756 * though - we need to synchronise with previous and future commits so
757 * that the commit records are correctly ordered in the log to ensure
758 * that we process items during log IO completion in the correct order.
759 *
760 * For example, if we get an EFI in one checkpoint and the EFD in the
761 * next (e.g. due to log forces), we do not want the checkpoint with
762 * the EFD to be committed before the checkpoint with the EFI. Hence
763 * we must strictly order the commit records of the checkpoints so
764 * that: a) the checkpoint callbacks are attached to the iclogs in the
765 * correct order; and b) the checkpoints are replayed in correct order
766 * in log recovery.
767 *
768 * Hence we need to add this context to the committing context list so
769 * that higher sequences will wait for us to write out a commit record
770 * before they do.
771 *
772 * xfs_log_force_lsn requires us to mirror the new sequence into the cil
773 * structure atomically with the addition of this sequence to the
774 * committing list. This also ensures that we can do unlocked checks
775 * against the current sequence in log forces without risking
776 * deferencing a freed context pointer.
777 */
778 spin_lock(&cil->xc_push_lock);
779 cil->xc_current_sequence = new_ctx->sequence;
780 spin_unlock(&cil->xc_push_lock);
781 up_write(&cil->xc_ctx_lock);
782
783 /*
784 * Build a checkpoint transaction header and write it to the log to
785 * begin the transaction. We need to account for the space used by the
786 * transaction header here as it is not accounted for in xlog_write().
787 *
788 * The LSN we need to pass to the log items on transaction commit is
789 * the LSN reported by the first log vector write. If we use the commit
790 * record lsn then we can move the tail beyond the grant write head.
791 */
792 tic = ctx->ticket;
793 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
794 thdr.th_type = XFS_TRANS_CHECKPOINT;
795 thdr.th_tid = tic->t_tid;
796 thdr.th_num_items = num_iovecs;
797 lhdr.i_addr = &thdr;
798 lhdr.i_len = sizeof(xfs_trans_header_t);
799 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
800 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
801
802 lvhdr.lv_niovecs = 1;
803 lvhdr.lv_iovecp = &lhdr;
804 lvhdr.lv_next = ctx->lv_chain;
805
806 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
807 if (error)
808 goto out_abort_free_ticket;
809
810 /*
811 * now that we've written the checkpoint into the log, strictly
812 * order the commit records so replay will get them in the right order.
813 */
814 restart:
815 spin_lock(&cil->xc_push_lock);
816 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
817 /*
818 * Avoid getting stuck in this loop because we were woken by the
819 * shutdown, but then went back to sleep once already in the
820 * shutdown state.
821 */
822 if (XLOG_FORCED_SHUTDOWN(log)) {
823 spin_unlock(&cil->xc_push_lock);
824 goto out_abort_free_ticket;
825 }
826
827 /*
828 * Higher sequences will wait for this one so skip them.
829 * Don't wait for our own sequence, either.
830 */
831 if (new_ctx->sequence >= ctx->sequence)
832 continue;
833 if (!new_ctx->commit_lsn) {
834 /*
835 * It is still being pushed! Wait for the push to
836 * complete, then start again from the beginning.
837 */
838 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
839 goto restart;
840 }
841 }
842 spin_unlock(&cil->xc_push_lock);
843
844 /* xfs_log_done always frees the ticket on error. */
845 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, false);
846 if (commit_lsn == -1)
847 goto out_abort;
848
849 spin_lock(&commit_iclog->ic_callback_lock);
850 if (commit_iclog->ic_state & XLOG_STATE_IOERROR) {
851 spin_unlock(&commit_iclog->ic_callback_lock);
852 goto out_abort;
853 }
854 ASSERT_ALWAYS(commit_iclog->ic_state == XLOG_STATE_ACTIVE ||
855 commit_iclog->ic_state == XLOG_STATE_WANT_SYNC);
856 list_add_tail(&ctx->iclog_entry, &commit_iclog->ic_callbacks);
857 spin_unlock(&commit_iclog->ic_callback_lock);
858
859 /*
860 * now the checkpoint commit is complete and we've attached the
861 * callbacks to the iclog we can assign the commit LSN to the context
862 * and wake up anyone who is waiting for the commit to complete.
863 */
864 spin_lock(&cil->xc_push_lock);
865 ctx->commit_lsn = commit_lsn;
866 wake_up_all(&cil->xc_commit_wait);
867 spin_unlock(&cil->xc_push_lock);
868
869 /* release the hounds! */
870 return xfs_log_release_iclog(log->l_mp, commit_iclog);
871
872 out_skip:
873 up_write(&cil->xc_ctx_lock);
874 xfs_log_ticket_put(new_ctx->ticket);
875 kmem_free(new_ctx);
876 return 0;
877
878 out_abort_free_ticket:
879 xfs_log_ticket_put(tic);
880 out_abort:
881 xlog_cil_committed(ctx, true);
882 return -EIO;
883 }
884
885 static void
886 xlog_cil_push_work(
887 struct work_struct *work)
888 {
889 struct xfs_cil *cil = container_of(work, struct xfs_cil,
890 xc_push_work);
891 xlog_cil_push(cil->xc_log);
892 }
893
894 /*
895 * We need to push CIL every so often so we don't cache more than we can fit in
896 * the log. The limit really is that a checkpoint can't be more than half the
897 * log (the current checkpoint is not allowed to overwrite the previous
898 * checkpoint), but commit latency and memory usage limit this to a smaller
899 * size.
900 */
901 static void
902 xlog_cil_push_background(
903 struct xlog *log)
904 {
905 struct xfs_cil *cil = log->l_cilp;
906
907 /*
908 * The cil won't be empty because we are called while holding the
909 * context lock so whatever we added to the CIL will still be there
910 */
911 ASSERT(!list_empty(&cil->xc_cil));
912
913 /*
914 * don't do a background push if we haven't used up all the
915 * space available yet.
916 */
917 if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
918 return;
919
920 spin_lock(&cil->xc_push_lock);
921 if (cil->xc_push_seq < cil->xc_current_sequence) {
922 cil->xc_push_seq = cil->xc_current_sequence;
923 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
924 }
925 spin_unlock(&cil->xc_push_lock);
926
927 }
928
929 /*
930 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
931 * number that is passed. When it returns, the work will be queued for
932 * @push_seq, but it won't be completed. The caller is expected to do any
933 * waiting for push_seq to complete if it is required.
934 */
935 static void
936 xlog_cil_push_now(
937 struct xlog *log,
938 xfs_lsn_t push_seq)
939 {
940 struct xfs_cil *cil = log->l_cilp;
941
942 if (!cil)
943 return;
944
945 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
946
947 /* start on any pending background push to minimise wait time on it */
948 flush_work(&cil->xc_push_work);
949
950 /*
951 * If the CIL is empty or we've already pushed the sequence then
952 * there's no work we need to do.
953 */
954 spin_lock(&cil->xc_push_lock);
955 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
956 spin_unlock(&cil->xc_push_lock);
957 return;
958 }
959
960 cil->xc_push_seq = push_seq;
961 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
962 spin_unlock(&cil->xc_push_lock);
963 }
964
965 bool
966 xlog_cil_empty(
967 struct xlog *log)
968 {
969 struct xfs_cil *cil = log->l_cilp;
970 bool empty = false;
971
972 spin_lock(&cil->xc_push_lock);
973 if (list_empty(&cil->xc_cil))
974 empty = true;
975 spin_unlock(&cil->xc_push_lock);
976 return empty;
977 }
978
979 /*
980 * Commit a transaction with the given vector to the Committed Item List.
981 *
982 * To do this, we need to format the item, pin it in memory if required and
983 * account for the space used by the transaction. Once we have done that we
984 * need to release the unused reservation for the transaction, attach the
985 * transaction to the checkpoint context so we carry the busy extents through
986 * to checkpoint completion, and then unlock all the items in the transaction.
987 *
988 * Called with the context lock already held in read mode to lock out
989 * background commit, returns without it held once background commits are
990 * allowed again.
991 */
992 void
993 xfs_log_commit_cil(
994 struct xfs_mount *mp,
995 struct xfs_trans *tp,
996 xfs_lsn_t *commit_lsn,
997 bool regrant)
998 {
999 struct xlog *log = mp->m_log;
1000 struct xfs_cil *cil = log->l_cilp;
1001 struct xfs_log_item *lip, *next;
1002 xfs_lsn_t xc_commit_lsn;
1003
1004 /*
1005 * Do all necessary memory allocation before we lock the CIL.
1006 * This ensures the allocation does not deadlock with a CIL
1007 * push in memory reclaim (e.g. from kswapd).
1008 */
1009 xlog_cil_alloc_shadow_bufs(log, tp);
1010
1011 /* lock out background commit */
1012 down_read(&cil->xc_ctx_lock);
1013
1014 xlog_cil_insert_items(log, tp);
1015
1016 xc_commit_lsn = cil->xc_ctx->sequence;
1017 if (commit_lsn)
1018 *commit_lsn = xc_commit_lsn;
1019
1020 xfs_log_done(mp, tp->t_ticket, NULL, regrant);
1021 tp->t_ticket = NULL;
1022 xfs_trans_unreserve_and_mod_sb(tp);
1023
1024 /*
1025 * Once all the items of the transaction have been copied to the CIL,
1026 * the items can be unlocked and possibly freed.
1027 *
1028 * This needs to be done before we drop the CIL context lock because we
1029 * have to update state in the log items and unlock them before they go
1030 * to disk. If we don't, then the CIL checkpoint can race with us and
1031 * we can run checkpoint completion before we've updated and unlocked
1032 * the log items. This affects (at least) processing of stale buffers,
1033 * inodes and EFIs.
1034 */
1035 trace_xfs_trans_commit_items(tp, _RET_IP_);
1036 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
1037 xfs_trans_del_item(lip);
1038 if (lip->li_ops->iop_committing)
1039 lip->li_ops->iop_committing(lip, xc_commit_lsn);
1040 }
1041 xlog_cil_push_background(log);
1042
1043 up_read(&cil->xc_ctx_lock);
1044 }
1045
1046 /*
1047 * Conditionally push the CIL based on the sequence passed in.
1048 *
1049 * We only need to push if we haven't already pushed the sequence
1050 * number given. Hence the only time we will trigger a push here is
1051 * if the push sequence is the same as the current context.
1052 *
1053 * We return the current commit lsn to allow the callers to determine if a
1054 * iclog flush is necessary following this call.
1055 */
1056 xfs_lsn_t
1057 xlog_cil_force_lsn(
1058 struct xlog *log,
1059 xfs_lsn_t sequence)
1060 {
1061 struct xfs_cil *cil = log->l_cilp;
1062 struct xfs_cil_ctx *ctx;
1063 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
1064
1065 ASSERT(sequence <= cil->xc_current_sequence);
1066
1067 /*
1068 * check to see if we need to force out the current context.
1069 * xlog_cil_push() handles racing pushes for the same sequence,
1070 * so no need to deal with it here.
1071 */
1072 restart:
1073 xlog_cil_push_now(log, sequence);
1074
1075 /*
1076 * See if we can find a previous sequence still committing.
1077 * We need to wait for all previous sequence commits to complete
1078 * before allowing the force of push_seq to go ahead. Hence block
1079 * on commits for those as well.
1080 */
1081 spin_lock(&cil->xc_push_lock);
1082 list_for_each_entry(ctx, &cil->xc_committing, committing) {
1083 /*
1084 * Avoid getting stuck in this loop because we were woken by the
1085 * shutdown, but then went back to sleep once already in the
1086 * shutdown state.
1087 */
1088 if (XLOG_FORCED_SHUTDOWN(log))
1089 goto out_shutdown;
1090 if (ctx->sequence > sequence)
1091 continue;
1092 if (!ctx->commit_lsn) {
1093 /*
1094 * It is still being pushed! Wait for the push to
1095 * complete, then start again from the beginning.
1096 */
1097 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
1098 goto restart;
1099 }
1100 if (ctx->sequence != sequence)
1101 continue;
1102 /* found it! */
1103 commit_lsn = ctx->commit_lsn;
1104 }
1105
1106 /*
1107 * The call to xlog_cil_push_now() executes the push in the background.
1108 * Hence by the time we have got here it our sequence may not have been
1109 * pushed yet. This is true if the current sequence still matches the
1110 * push sequence after the above wait loop and the CIL still contains
1111 * dirty objects. This is guaranteed by the push code first adding the
1112 * context to the committing list before emptying the CIL.
1113 *
1114 * Hence if we don't find the context in the committing list and the
1115 * current sequence number is unchanged then the CIL contents are
1116 * significant. If the CIL is empty, if means there was nothing to push
1117 * and that means there is nothing to wait for. If the CIL is not empty,
1118 * it means we haven't yet started the push, because if it had started
1119 * we would have found the context on the committing list.
1120 */
1121 if (sequence == cil->xc_current_sequence &&
1122 !list_empty(&cil->xc_cil)) {
1123 spin_unlock(&cil->xc_push_lock);
1124 goto restart;
1125 }
1126
1127 spin_unlock(&cil->xc_push_lock);
1128 return commit_lsn;
1129
1130 /*
1131 * We detected a shutdown in progress. We need to trigger the log force
1132 * to pass through it's iclog state machine error handling, even though
1133 * we are already in a shutdown state. Hence we can't return
1134 * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
1135 * LSN is already stable), so we return a zero LSN instead.
1136 */
1137 out_shutdown:
1138 spin_unlock(&cil->xc_push_lock);
1139 return 0;
1140 }
1141
1142 /*
1143 * Check if the current log item was first committed in this sequence.
1144 * We can't rely on just the log item being in the CIL, we have to check
1145 * the recorded commit sequence number.
1146 *
1147 * Note: for this to be used in a non-racy manner, it has to be called with
1148 * CIL flushing locked out. As a result, it should only be used during the
1149 * transaction commit process when deciding what to format into the item.
1150 */
1151 bool
1152 xfs_log_item_in_current_chkpt(
1153 struct xfs_log_item *lip)
1154 {
1155 struct xfs_cil_ctx *ctx;
1156
1157 if (list_empty(&lip->li_cil))
1158 return false;
1159
1160 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
1161
1162 /*
1163 * li_seq is written on the first commit of a log item to record the
1164 * first checkpoint it is written to. Hence if it is different to the
1165 * current sequence, we're in a new checkpoint.
1166 */
1167 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
1168 return false;
1169 return true;
1170 }
1171
1172 /*
1173 * Perform initial CIL structure initialisation.
1174 */
1175 int
1176 xlog_cil_init(
1177 struct xlog *log)
1178 {
1179 struct xfs_cil *cil;
1180 struct xfs_cil_ctx *ctx;
1181
1182 cil = kmem_zalloc(sizeof(*cil), KM_MAYFAIL);
1183 if (!cil)
1184 return -ENOMEM;
1185
1186 ctx = kmem_zalloc(sizeof(*ctx), KM_MAYFAIL);
1187 if (!ctx) {
1188 kmem_free(cil);
1189 return -ENOMEM;
1190 }
1191
1192 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
1193 INIT_LIST_HEAD(&cil->xc_cil);
1194 INIT_LIST_HEAD(&cil->xc_committing);
1195 spin_lock_init(&cil->xc_cil_lock);
1196 spin_lock_init(&cil->xc_push_lock);
1197 init_rwsem(&cil->xc_ctx_lock);
1198 init_waitqueue_head(&cil->xc_commit_wait);
1199
1200 INIT_LIST_HEAD(&ctx->committing);
1201 INIT_LIST_HEAD(&ctx->busy_extents);
1202 ctx->sequence = 1;
1203 ctx->cil = cil;
1204 cil->xc_ctx = ctx;
1205 cil->xc_current_sequence = ctx->sequence;
1206
1207 cil->xc_log = log;
1208 log->l_cilp = cil;
1209 return 0;
1210 }
1211
1212 void
1213 xlog_cil_destroy(
1214 struct xlog *log)
1215 {
1216 if (log->l_cilp->xc_ctx) {
1217 if (log->l_cilp->xc_ctx->ticket)
1218 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
1219 kmem_free(log->l_cilp->xc_ctx);
1220 }
1221
1222 ASSERT(list_empty(&log->l_cilp->xc_cil));
1223 kmem_free(log->l_cilp);
1224 }
1225