Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0+
2 : /*
3 : * Copyright (C) 2016 Oracle. All Rights Reserved.
4 : * Author: Darrick J. Wong <darrick.wong@oracle.com>
5 : */
6 : #include "xfs.h"
7 : #include "xfs_fs.h"
8 : #include "xfs_shared.h"
9 : #include "xfs_format.h"
10 : #include "xfs_log_format.h"
11 : #include "xfs_trans_resv.h"
12 : #include "xfs_mount.h"
13 : #include "xfs_defer.h"
14 : #include "xfs_trans.h"
15 : #include "xfs_buf_item.h"
16 : #include "xfs_inode.h"
17 : #include "xfs_inode_item.h"
18 : #include "xfs_trace.h"
19 : #include "xfs_icache.h"
20 : #include "xfs_log.h"
21 : #include "xfs_rmap.h"
22 : #include "xfs_refcount.h"
23 : #include "xfs_bmap.h"
24 : #include "xfs_alloc.h"
25 : #include "xfs_buf.h"
26 : #include "xfs_da_format.h"
27 : #include "xfs_da_btree.h"
28 : #include "xfs_attr.h"
29 : #include "xfs_swapext.h"
30 :
31 : static struct kmem_cache *xfs_defer_pending_cache;
32 :
33 : /*
34 : * Deferred Operations in XFS
35 : *
36 : * Due to the way locking rules work in XFS, certain transactions (block
37 : * mapping and unmapping, typically) have permanent reservations so that
38 : * we can roll the transaction to adhere to AG locking order rules and
39 : * to unlock buffers between metadata updates. Prior to rmap/reflink,
40 : * the mapping code had a mechanism to perform these deferrals for
41 : * extents that were going to be freed; this code makes that facility
42 : * more generic.
43 : *
44 : * When adding the reverse mapping and reflink features, it became
45 : * necessary to perform complex remapping multi-transactions to comply
46 : * with AG locking order rules, and to be able to spread a single
47 : * refcount update operation (an operation on an n-block extent can
48 : * update as many as n records!) among multiple transactions. XFS can
49 : * roll a transaction to facilitate this, but using this facility
50 : * requires us to log "intent" items in case log recovery needs to
51 : * redo the operation, and to log "done" items to indicate that redo
52 : * is not necessary.
53 : *
54 : * Deferred work is tracked in xfs_defer_pending items. Each pending
55 : * item tracks one type of deferred work. Incoming work items (which
56 : * have not yet had an intent logged) are attached to a pending item
57 : * on the dop_intake list, where they wait for the caller to finish
58 : * the deferred operations.
59 : *
60 : * Finishing a set of deferred operations is an involved process. To
61 : * start, we define "rolling a deferred-op transaction" as follows:
62 : *
63 : * > For each xfs_defer_pending item on the dop_intake list,
64 : * - Sort the work items in AG order. XFS locking
65 : * order rules require us to lock buffers in AG order.
66 : * - Create a log intent item for that type.
67 : * - Attach it to the pending item.
68 : * - Move the pending item from the dop_intake list to the
69 : * dop_pending list.
70 : * > Roll the transaction.
71 : *
72 : * NOTE: To avoid exceeding the transaction reservation, we limit the
73 : * number of items that we attach to a given xfs_defer_pending.
74 : *
75 : * The actual finishing process looks like this:
76 : *
77 : * > For each xfs_defer_pending in the dop_pending list,
78 : * - Roll the deferred-op transaction as above.
79 : * - Create a log done item for that type, and attach it to the
80 : * log intent item.
81 : * - For each work item attached to the log intent item,
82 : * * Perform the described action.
83 : * * Attach the work item to the log done item.
84 : * * If the result of doing the work was -EAGAIN, ->finish work
85 : * wants a new transaction. See the "Requesting a Fresh
86 : * Transaction while Finishing Deferred Work" section below for
87 : * details.
88 : *
89 : * The key here is that we must log an intent item for all pending
90 : * work items every time we roll the transaction, and that we must log
91 : * a done item as soon as the work is completed. With this mechanism
92 : * we can perform complex remapping operations, chaining intent items
93 : * as needed.
94 : *
95 : * Requesting a Fresh Transaction while Finishing Deferred Work
96 : *
97 : * If ->finish_item decides that it needs a fresh transaction to
98 : * finish the work, it must ask its caller (xfs_defer_finish) for a
99 : * continuation. The most likely cause of this circumstance are the
100 : * refcount adjust functions deciding that they've logged enough items
101 : * to be at risk of exceeding the transaction reservation.
102 : *
103 : * To get a fresh transaction, we want to log the existing log done
104 : * item to prevent the log intent item from replaying, immediately log
105 : * a new log intent item with the unfinished work items, roll the
106 : * transaction, and re-call ->finish_item wherever it left off. The
107 : * log done item and the new log intent item must be in the same
108 : * transaction or atomicity cannot be guaranteed; defer_finish ensures
109 : * that this happens.
110 : *
111 : * This requires some coordination between ->finish_item and
112 : * defer_finish. Upon deciding to request a new transaction,
113 : * ->finish_item should update the current work item to reflect the
114 : * unfinished work. Next, it should reset the log done item's list
115 : * count to the number of items finished, and return -EAGAIN.
116 : * defer_finish sees the -EAGAIN, logs the new log intent item
117 : * with the remaining work items, and leaves the xfs_defer_pending
118 : * item at the head of the dop_work queue. Then it rolls the
119 : * transaction and picks up processing where it left off. It is
120 : * required that ->finish_item must be careful to leave enough
121 : * transaction reservation to fit the new log intent item.
122 : *
123 : * This is an example of remapping the extent (E, E+B) into file X at
124 : * offset A and dealing with the extent (C, C+B) already being mapped
125 : * there:
126 : * +-------------------------------------------------+
127 : * | Unmap file X startblock C offset A length B | t0
128 : * | Intent to reduce refcount for extent (C, B) |
129 : * | Intent to remove rmap (X, C, A, B) |
130 : * | Intent to free extent (D, 1) (bmbt block) |
131 : * | Intent to map (X, A, B) at startblock E |
132 : * +-------------------------------------------------+
133 : * | Map file X startblock E offset A length B | t1
134 : * | Done mapping (X, E, A, B) |
135 : * | Intent to increase refcount for extent (E, B) |
136 : * | Intent to add rmap (X, E, A, B) |
137 : * +-------------------------------------------------+
138 : * | Reduce refcount for extent (C, B) | t2
139 : * | Done reducing refcount for extent (C, 9) |
140 : * | Intent to reduce refcount for extent (C+9, B-9) |
141 : * | (ran out of space after 9 refcount updates) |
142 : * +-------------------------------------------------+
143 : * | Reduce refcount for extent (C+9, B+9) | t3
144 : * | Done reducing refcount for extent (C+9, B-9) |
145 : * | Increase refcount for extent (E, B) |
146 : * | Done increasing refcount for extent (E, B) |
147 : * | Intent to free extent (C, B) |
148 : * | Intent to free extent (F, 1) (refcountbt block) |
149 : * | Intent to remove rmap (F, 1, REFC) |
150 : * +-------------------------------------------------+
151 : * | Remove rmap (X, C, A, B) | t4
152 : * | Done removing rmap (X, C, A, B) |
153 : * | Add rmap (X, E, A, B) |
154 : * | Done adding rmap (X, E, A, B) |
155 : * | Remove rmap (F, 1, REFC) |
156 : * | Done removing rmap (F, 1, REFC) |
157 : * +-------------------------------------------------+
158 : * | Free extent (C, B) | t5
159 : * | Done freeing extent (C, B) |
160 : * | Free extent (D, 1) |
161 : * | Done freeing extent (D, 1) |
162 : * | Free extent (F, 1) |
163 : * | Done freeing extent (F, 1) |
164 : * +-------------------------------------------------+
165 : *
166 : * If we should crash before t2 commits, log recovery replays
167 : * the following intent items:
168 : *
169 : * - Intent to reduce refcount for extent (C, B)
170 : * - Intent to remove rmap (X, C, A, B)
171 : * - Intent to free extent (D, 1) (bmbt block)
172 : * - Intent to increase refcount for extent (E, B)
173 : * - Intent to add rmap (X, E, A, B)
174 : *
175 : * In the process of recovering, it should also generate and take care
176 : * of these intent items:
177 : *
178 : * - Intent to free extent (C, B)
179 : * - Intent to free extent (F, 1) (refcountbt block)
180 : * - Intent to remove rmap (F, 1, REFC)
181 : *
182 : * Note that the continuation requested between t2 and t3 is likely to
183 : * reoccur.
184 : */
185 :
186 : static const struct xfs_defer_op_type *defer_op_types[] = {
187 : [XFS_DEFER_OPS_TYPE_BMAP] = &xfs_bmap_update_defer_type,
188 : [XFS_DEFER_OPS_TYPE_REFCOUNT] = &xfs_refcount_update_defer_type,
189 : [XFS_DEFER_OPS_TYPE_RMAP] = &xfs_rmap_update_defer_type,
190 : [XFS_DEFER_OPS_TYPE_FREE] = &xfs_extent_free_defer_type,
191 : [XFS_DEFER_OPS_TYPE_AGFL_FREE] = &xfs_agfl_free_defer_type,
192 : [XFS_DEFER_OPS_TYPE_ATTR] = &xfs_attr_defer_type,
193 : [XFS_DEFER_OPS_TYPE_SWAPEXT] = &xfs_swapext_defer_type,
194 : };
195 :
196 : /*
197 : * Ensure there's a log intent item associated with this deferred work item if
198 : * the operation must be restarted on crash. Returns 1 if there's a log item;
199 : * 0 if there isn't; or a negative errno.
200 : */
201 : static int
202 1290335106 : xfs_defer_create_intent(
203 : struct xfs_trans *tp,
204 : struct xfs_defer_pending *dfp,
205 : bool sort)
206 : {
207 1290335106 : const struct xfs_defer_op_type *ops = defer_op_types[dfp->dfp_type];
208 1290313122 : struct xfs_log_item *lip;
209 :
210 1290313122 : if (dfp->dfp_intent)
211 : return 1;
212 :
213 1290412010 : lip = ops->create_intent(tp, &dfp->dfp_work, dfp->dfp_count, sort);
214 1290935974 : if (!lip)
215 : return 0;
216 1105756380 : if (IS_ERR(lip))
217 0 : return PTR_ERR(lip);
218 :
219 1105756380 : dfp->dfp_intent = lip;
220 1105756380 : return 1;
221 : }
222 :
223 : /*
224 : * For each pending item in the intake list, log its intent item and the
225 : * associated extents, then add the entire intake list to the end of
226 : * the pending list.
227 : *
228 : * Returns 1 if at least one log item was associated with the deferred work;
229 : * 0 if there are no log items; or a negative errno.
230 : */
231 : static int
232 1290512248 : xfs_defer_create_intents(
233 : struct xfs_trans *tp)
234 : {
235 1290512248 : struct xfs_defer_pending *dfp;
236 1290512248 : int ret = 0;
237 :
238 2478653419 : list_for_each_entry(dfp, &tp->t_dfops, dfp_list) {
239 1187802631 : int ret2;
240 :
241 1187802631 : trace_xfs_defer_create_intent(tp->t_mountp, dfp);
242 1187485182 : ret2 = xfs_defer_create_intent(tp, dfp, true);
243 1188141171 : if (ret2 < 0)
244 0 : return ret2;
245 1188141171 : ret |= ret2;
246 : }
247 : return ret;
248 : }
249 :
250 : /* Abort all the intents that were committed. */
251 : STATIC void
252 4777 : xfs_defer_trans_abort(
253 : struct xfs_trans *tp,
254 : struct list_head *dop_pending)
255 : {
256 4777 : struct xfs_defer_pending *dfp;
257 4777 : const struct xfs_defer_op_type *ops;
258 :
259 4777 : trace_xfs_defer_trans_abort(tp, _RET_IP_);
260 :
261 : /* Abort intent items that don't have a done item. */
262 12574 : list_for_each_entry(dfp, dop_pending, dfp_list) {
263 7797 : ops = defer_op_types[dfp->dfp_type];
264 7799 : trace_xfs_defer_pending_abort(tp->t_mountp, dfp);
265 7799 : if (dfp->dfp_intent && !dfp->dfp_done) {
266 3232 : ops->abort_intent(dfp->dfp_intent);
267 3232 : dfp->dfp_intent = NULL;
268 : }
269 : }
270 4777 : }
271 :
272 : /*
273 : * Capture resources that the caller said not to release ("held") when the
274 : * transaction commits. Caller is responsible for zero-initializing @dres.
275 : */
276 : static int
277 1318429961 : xfs_defer_save_resources(
278 : struct xfs_defer_resources *dres,
279 : struct xfs_trans *tp)
280 : {
281 1318429961 : struct xfs_buf_log_item *bli;
282 1318429961 : struct xfs_inode_log_item *ili;
283 1318429961 : struct xfs_log_item *lip;
284 :
285 1318429961 : BUILD_BUG_ON(NBBY * sizeof(dres->dr_ordered) < XFS_DEFER_OPS_NR_BUFS);
286 :
287 7415906468 : list_for_each_entry(lip, &tp->t_items, li_trans) {
288 6097073134 : switch (lip->li_type) {
289 2512909548 : case XFS_LI_BUF:
290 2512909548 : bli = container_of(lip, struct xfs_buf_log_item,
291 : bli_item);
292 2512909548 : if (bli->bli_flags & XFS_BLI_HOLD) {
293 4189240 : if (dres->dr_bufs >= XFS_DEFER_OPS_NR_BUFS) {
294 0 : ASSERT(0);
295 0 : return -EFSCORRUPTED;
296 : }
297 4189240 : if (bli->bli_flags & XFS_BLI_ORDERED)
298 73981 : dres->dr_ordered |=
299 73981 : (1U << dres->dr_bufs);
300 : else
301 4115259 : xfs_trans_dirty_buf(tp, bli->bli_buf);
302 4190849 : dres->dr_bp[dres->dr_bufs++] = bli->bli_buf;
303 : }
304 : break;
305 1821962817 : case XFS_LI_INODE:
306 1821962817 : ili = container_of(lip, struct xfs_inode_log_item,
307 : ili_item);
308 1821962817 : if (ili->ili_lock_flags == 0) {
309 1795426350 : if (dres->dr_inos >= XFS_DEFER_OPS_NR_INODES) {
310 0 : ASSERT(0);
311 0 : return -EFSCORRUPTED;
312 : }
313 1795426350 : xfs_trans_log_inode(tp, ili->ili_inode,
314 : XFS_ILOG_CORE);
315 1795847707 : dres->dr_ip[dres->dr_inos++] = ili->ili_inode;
316 : }
317 : break;
318 : default:
319 : break;
320 : }
321 : }
322 :
323 : return 0;
324 : }
325 :
326 : /* Attach the held resources to the transaction. */
327 : static void
328 1318847768 : xfs_defer_restore_resources(
329 : struct xfs_trans *tp,
330 : struct xfs_defer_resources *dres)
331 : {
332 1318847768 : unsigned short i;
333 :
334 : /* Rejoin the joined inodes. */
335 3114663018 : for (i = 0; i < dres->dr_inos; i++)
336 1795662235 : xfs_trans_ijoin(tp, dres->dr_ip[i], 0);
337 :
338 : /* Rejoin the buffers and dirty them so the log moves forward. */
339 1323179273 : for (i = 0; i < dres->dr_bufs; i++) {
340 4189739 : xfs_trans_bjoin(tp, dres->dr_bp[i]);
341 4191349 : if (dres->dr_ordered & (1U << i))
342 73980 : xfs_trans_ordered_buf(tp, dres->dr_bp[i]);
343 4191351 : xfs_trans_bhold(tp, dres->dr_bp[i]);
344 : }
345 1318989534 : }
346 :
347 : /* Roll a transaction so we can do some deferred op processing. */
348 : STATIC int
349 1318541862 : xfs_defer_trans_roll(
350 : struct xfs_trans **tpp)
351 : {
352 1318541862 : struct xfs_defer_resources dres = { };
353 1318541862 : int error;
354 :
355 1318541862 : error = xfs_defer_save_resources(&dres, *tpp);
356 1318827114 : if (error)
357 : return error;
358 :
359 1318822325 : trace_xfs_defer_trans_roll(*tpp, _RET_IP_);
360 :
361 : /*
362 : * Roll the transaction. Rolling always given a new transaction (even
363 : * if committing the old one fails!) to hand back to the caller, so we
364 : * join the held resources to the new transaction so that we always
365 : * return with the held resources joined to @tpp, no matter what
366 : * happened.
367 : */
368 1318446544 : error = xfs_trans_roll(tpp);
369 :
370 1318910876 : xfs_defer_restore_resources(*tpp, &dres);
371 :
372 1318956444 : if (error)
373 234 : trace_xfs_defer_trans_roll_error(*tpp, error);
374 : return error;
375 : }
376 :
377 : /*
378 : * Free up any items left in the list.
379 : */
380 : static void
381 9898 : xfs_defer_cancel_list(
382 : struct xfs_mount *mp,
383 : struct list_head *dop_list)
384 : {
385 9898 : struct xfs_defer_pending *dfp;
386 9898 : struct xfs_defer_pending *pli;
387 9898 : struct list_head *pwi;
388 9898 : struct list_head *n;
389 9898 : const struct xfs_defer_op_type *ops;
390 :
391 : /*
392 : * Free the pending items. Caller should already have arranged
393 : * for the intent items to be released.
394 : */
395 18050 : list_for_each_entry_safe(dfp, pli, dop_list, dfp_list) {
396 8153 : ops = defer_op_types[dfp->dfp_type];
397 8152 : trace_xfs_defer_cancel_list(mp, dfp);
398 8152 : list_del(&dfp->dfp_list);
399 11780 : list_for_each_safe(pwi, n, &dfp->dfp_work) {
400 3628 : list_del(pwi);
401 3628 : dfp->dfp_count--;
402 3628 : trace_xfs_defer_cancel_item(mp, dfp, pwi);
403 3628 : ops->cancel_item(pwi);
404 : }
405 8152 : ASSERT(dfp->dfp_count == 0);
406 8152 : kmem_cache_free(xfs_defer_pending_cache, dfp);
407 : }
408 9897 : }
409 :
410 : /*
411 : * Prevent a log intent item from pinning the tail of the log by logging a
412 : * done item to release the intent item; and then log a new intent item.
413 : * The caller should provide a fresh transaction and roll it after we're done.
414 : */
415 : static int
416 1147660031 : xfs_defer_relog(
417 : struct xfs_trans **tpp,
418 : struct list_head *dfops)
419 : {
420 1147660031 : struct xlog *log = (*tpp)->t_mountp->m_log;
421 1147660031 : struct xfs_defer_pending *dfp;
422 1147660031 : xfs_lsn_t threshold_lsn = NULLCOMMITLSN;
423 :
424 :
425 1147660031 : ASSERT((*tpp)->t_flags & XFS_TRANS_PERM_LOG_RES);
426 :
427 2968595544 : list_for_each_entry(dfp, dfops, dfp_list) {
428 : /*
429 : * If the log intent item for this deferred op is not a part of
430 : * the current log checkpoint, relog the intent item to keep
431 : * the log tail moving forward. We're ok with this being racy
432 : * because an incorrect decision means we'll be a little slower
433 : * at pushing the tail.
434 : */
435 3592398641 : if (dfp->dfp_intent == NULL ||
436 1770945735 : xfs_log_item_in_current_chkpt(dfp->dfp_intent))
437 1820857800 : continue;
438 :
439 : /*
440 : * Figure out where we need the tail to be in order to maintain
441 : * the minimum required free space in the log. Only sample
442 : * the log threshold once per call.
443 : */
444 595106 : if (threshold_lsn == NULLCOMMITLSN) {
445 564274 : threshold_lsn = xlog_grant_push_threshold(log, 0);
446 564275 : if (threshold_lsn == NULLCOMMITLSN)
447 : break;
448 : }
449 77714 : if (XFS_LSN_CMP(dfp->dfp_intent->li_lsn, threshold_lsn) >= 0)
450 4845 : continue;
451 :
452 72869 : trace_xfs_defer_relog_intent((*tpp)->t_mountp, dfp);
453 72869 : XFS_STATS_INC((*tpp)->t_mountp, defer_relog);
454 72869 : dfp->dfp_intent = xfs_trans_item_relog(dfp->dfp_intent, *tpp);
455 : }
456 :
457 1147646020 : if ((*tpp)->t_flags & XFS_TRANS_DIRTY)
458 81301 : return xfs_defer_trans_roll(tpp);
459 : return 0;
460 : }
461 :
462 : /*
463 : * Log an intent-done item for the first pending intent, and finish the work
464 : * items.
465 : */
466 : static int
467 1290884912 : xfs_defer_finish_one(
468 : struct xfs_trans *tp,
469 : struct xfs_defer_pending *dfp)
470 : {
471 1290884912 : const struct xfs_defer_op_type *ops = defer_op_types[dfp->dfp_type];
472 1290898970 : struct xfs_btree_cur *state = NULL;
473 1290898970 : struct list_head *li, *n;
474 1290898970 : int error;
475 :
476 1290898970 : trace_xfs_defer_pending_finish(tp->t_mountp, dfp);
477 :
478 1290826383 : dfp->dfp_done = ops->create_done(tp, dfp->dfp_intent, dfp->dfp_count);
479 2551970761 : list_for_each_safe(li, n, &dfp->dfp_work) {
480 1363676120 : list_del(li);
481 1363527701 : dfp->dfp_count--;
482 1363527701 : trace_xfs_defer_finish_item(tp->t_mountp, dfp, li);
483 1363437475 : error = ops->finish_item(tp, dfp->dfp_done, li, &state);
484 1363709832 : if (error == -EAGAIN) {
485 102731703 : int ret;
486 :
487 : /*
488 : * Caller wants a fresh transaction; put the work item
489 : * back on the list and log a new log intent item to
490 : * replace the old one. See "Requesting a Fresh
491 : * Transaction while Finishing Deferred Work" above.
492 : */
493 102731703 : list_add(li, &dfp->dfp_work);
494 102721688 : dfp->dfp_count++;
495 102721688 : dfp->dfp_done = NULL;
496 102721688 : dfp->dfp_intent = NULL;
497 102721688 : ret = xfs_defer_create_intent(tp, dfp, false);
498 102733533 : if (ret < 0)
499 0 : error = ret;
500 : }
501 :
502 1363711662 : if (error)
503 102739272 : goto out;
504 : }
505 :
506 : /* Done with the dfp, free it. */
507 1188294641 : list_del(&dfp->dfp_list);
508 1188169020 : kmem_cache_free(xfs_defer_pending_cache, dfp);
509 1188271609 : tp->t_dfops_nr--;
510 1188271609 : tp->t_dfops_finished++;
511 1291010881 : out:
512 1291010881 : if (ops->finish_cleanup)
513 597712322 : ops->finish_cleanup(tp, state, error);
514 1291063797 : return error;
515 : }
516 :
517 : /*
518 : * Finish all the pending work. This involves logging intent items for
519 : * any work items that wandered in since the last transaction roll (if
520 : * one has even happened), rolling the transaction, and finishing the
521 : * work items in the first item on the logged-and-pending list.
522 : *
523 : * If an inode is provided, relog it to the new transaction.
524 : */
525 : int
526 1192687630 : xfs_defer_finish_noroll(
527 : struct xfs_trans **tp)
528 : {
529 1192687630 : struct xfs_defer_pending *dfp = NULL;
530 1192687630 : int error = 0;
531 1192687630 : LIST_HEAD(dop_pending);
532 :
533 1192687630 : ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
534 :
535 1192687630 : trace_xfs_defer_finish(*tp, _RET_IP_);
536 :
537 : /* Until we run out of pending work to finish... */
538 2483008543 : while (!list_empty(&dop_pending) || !list_empty(&(*tp)->t_dfops)) {
539 : /*
540 : * Deferred items that are created in the process of finishing
541 : * other deferred work items should be queued at the head of
542 : * the pending list, which puts them ahead of the deferred work
543 : * that was created by the caller. This keeps the number of
544 : * pending work items to a minimum, which decreases the amount
545 : * of time that any one intent item can stick around in memory,
546 : * pinning the log tail.
547 : */
548 1290210776 : int has_intents = xfs_defer_create_intents(*tp);
549 :
550 1290705361 : list_splice_init(&(*tp)->t_dfops, &dop_pending);
551 :
552 1290705361 : (*tp)->t_dfops_nr_max = max((*tp)->t_dfops_nr,
553 : (*tp)->t_dfops_nr_max);
554 :
555 1290705361 : if (has_intents < 0) {
556 0 : error = has_intents;
557 0 : goto out_shutdown;
558 : }
559 1290705361 : if (has_intents || dfp) {
560 1147250700 : error = xfs_defer_trans_roll(tp);
561 1147745600 : if (error)
562 227 : goto out_shutdown;
563 :
564 : /* Relog intent items to keep the log moving. */
565 1147745373 : error = xfs_defer_relog(tp, &dop_pending);
566 1147532465 : if (error)
567 0 : goto out_shutdown;
568 : }
569 :
570 1290987126 : dfp = list_first_entry(&dop_pending, struct xfs_defer_pending,
571 : dfp_list);
572 1290987126 : error = xfs_defer_finish_one(*tp, dfp);
573 1290917998 : if (error && error != -EAGAIN)
574 4549 : goto out_shutdown;
575 : }
576 :
577 1192242839 : trace_xfs_defer_finish_done(*tp, _RET_IP_);
578 1192276238 : return 0;
579 :
580 4776 : out_shutdown:
581 4776 : xfs_defer_trans_abort(*tp, &dop_pending);
582 4777 : xfs_force_shutdown((*tp)->t_mountp, SHUTDOWN_CORRUPT_INCORE);
583 4776 : trace_xfs_defer_finish_error(*tp, error);
584 4777 : xfs_defer_cancel_list((*tp)->t_mountp, &dop_pending);
585 4777 : (*tp)->t_dfops_nr = 0;
586 4777 : xfs_defer_cancel(*tp);
587 4777 : return error;
588 : }
589 :
590 : int
591 185376121 : xfs_defer_finish(
592 : struct xfs_trans **tp)
593 : {
594 185376121 : int error;
595 :
596 : /*
597 : * Finish and roll the transaction once more to avoid returning to the
598 : * caller with a dirty transaction.
599 : */
600 185376121 : error = xfs_defer_finish_noroll(tp);
601 185350053 : if (error)
602 : return error;
603 185348469 : if ((*tp)->t_flags & XFS_TRANS_DIRTY) {
604 171093343 : error = xfs_defer_trans_roll(tp);
605 171115877 : if (error) {
606 7 : xfs_force_shutdown((*tp)->t_mountp,
607 : SHUTDOWN_CORRUPT_INCORE);
608 7 : return error;
609 : }
610 : }
611 :
612 : /* Reset LOWMODE now that we've finished all the dfops. */
613 185370996 : ASSERT(list_empty(&(*tp)->t_dfops));
614 185370996 : (*tp)->t_flags &= ~XFS_TRANS_LOWMODE;
615 185370996 : return 0;
616 : }
617 :
618 : void
619 5120 : xfs_defer_cancel(
620 : struct xfs_trans *tp)
621 : {
622 5120 : struct xfs_mount *mp = tp->t_mountp;
623 :
624 5120 : trace_xfs_defer_cancel(tp, _RET_IP_);
625 5120 : xfs_defer_cancel_list(mp, &tp->t_dfops);
626 5121 : tp->t_dfops_nr = 0;
627 5121 : }
628 :
629 : /* Add an item for later deferred processing. */
630 : void
631 1260797991 : xfs_defer_add(
632 : struct xfs_trans *tp,
633 : enum xfs_defer_ops_type type,
634 : struct list_head *li)
635 : {
636 1260797991 : struct xfs_defer_pending *dfp = NULL;
637 1260797991 : const struct xfs_defer_op_type *ops = defer_op_types[type];
638 :
639 1260692852 : ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
640 1260692852 : BUILD_BUG_ON(ARRAY_SIZE(defer_op_types) != XFS_DEFER_OPS_TYPE_MAX);
641 :
642 : /*
643 : * Add the item to a pending item at the end of the intake list.
644 : * If the last pending item has the same type, reuse it. Else,
645 : * create a new pending item at the end of the intake list.
646 : */
647 1260692852 : if (!list_empty(&tp->t_dfops)) {
648 392465514 : dfp = list_last_entry(&tp->t_dfops,
649 : struct xfs_defer_pending, dfp_list);
650 392465514 : if (dfp->dfp_type != type ||
651 97555365 : (ops->max_items && dfp->dfp_count >= ops->max_items))
652 : dfp = NULL;
653 : }
654 72620307 : if (!dfp) {
655 1188072545 : dfp = kmem_cache_zalloc(xfs_defer_pending_cache,
656 : GFP_NOFS | __GFP_NOFAIL);
657 1188188993 : dfp->dfp_type = type;
658 1188188993 : dfp->dfp_intent = NULL;
659 1188188993 : dfp->dfp_done = NULL;
660 1188188993 : dfp->dfp_count = 0;
661 1188188993 : INIT_LIST_HEAD(&dfp->dfp_work);
662 1188188993 : list_add_tail(&dfp->dfp_list, &tp->t_dfops);
663 1187966653 : tp->t_dfops_nr++;
664 : }
665 :
666 1260586960 : list_add_tail(li, &dfp->dfp_work);
667 1260676792 : trace_xfs_defer_add_item(tp->t_mountp, dfp, li);
668 1260669897 : dfp->dfp_count++;
669 1260669897 : }
670 :
671 : /*
672 : * Move deferred ops from one transaction to another and reset the source to
673 : * initial state. This is primarily used to carry state forward across
674 : * transaction rolls with pending dfops.
675 : */
676 : void
677 1330906080 : xfs_defer_move(
678 : struct xfs_trans *dtp,
679 : struct xfs_trans *stp)
680 : {
681 1330906080 : list_splice_init(&stp->t_dfops, &dtp->t_dfops);
682 1330906080 : dtp->t_dfops_nr += stp->t_dfops_nr;
683 1330906080 : dtp->t_dfops_nr_max = stp->t_dfops_nr_max;
684 1330906080 : dtp->t_dfops_finished = stp->t_dfops_finished;
685 1330906080 : stp->t_dfops_nr = 0;
686 1330906080 : stp->t_dfops_nr_max = 0;
687 1330906080 : stp->t_dfops_finished = 0;
688 :
689 : /*
690 : * Low free space mode was historically controlled by a dfops field.
691 : * This meant that low mode state potentially carried across multiple
692 : * transaction rolls. Transfer low mode on a dfops move to preserve
693 : * that behavior.
694 : */
695 1330906080 : dtp->t_flags |= (stp->t_flags & XFS_TRANS_LOWMODE);
696 1330906080 : stp->t_flags &= ~XFS_TRANS_LOWMODE;
697 1330906080 : }
698 :
699 : /*
700 : * Prepare a chain of fresh deferred ops work items to be completed later. Log
701 : * recovery requires the ability to put off until later the actual finishing
702 : * work so that it can process unfinished items recovered from the log in
703 : * correct order.
704 : *
705 : * Create and log intent items for all the work that we're capturing so that we
706 : * can be assured that the items will get replayed if the system goes down
707 : * before log recovery gets a chance to finish the work it put off. The entire
708 : * deferred ops state is transferred to the capture structure and the
709 : * transaction is then ready for the caller to commit it. If there are no
710 : * intent items to capture, this function returns NULL.
711 : *
712 : * If capture_ip is not NULL, the capture structure will obtain an extra
713 : * reference to the inode.
714 : */
715 : static struct xfs_defer_capture *
716 5215 : xfs_defer_ops_capture(
717 : struct xfs_trans *tp)
718 : {
719 5215 : struct xfs_defer_capture *dfc;
720 5215 : unsigned short i;
721 5215 : int error;
722 :
723 5215 : if (list_empty(&tp->t_dfops))
724 : return NULL;
725 :
726 1295 : error = xfs_defer_create_intents(tp);
727 1295 : if (error < 0)
728 0 : return ERR_PTR(error);
729 :
730 : /* Create an object to capture the defer ops. */
731 1295 : dfc = kmem_zalloc(sizeof(*dfc), KM_NOFS);
732 1295 : INIT_LIST_HEAD(&dfc->dfc_list);
733 1295 : INIT_LIST_HEAD(&dfc->dfc_dfops);
734 :
735 : /* Move the dfops chain and transaction state to the capture struct. */
736 1295 : list_splice_init(&tp->t_dfops, &dfc->dfc_dfops);
737 1295 : dfc->dfc_tpflags = tp->t_flags & XFS_TRANS_LOWMODE;
738 1295 : tp->t_flags &= ~XFS_TRANS_LOWMODE;
739 :
740 : /* Capture the remaining block reservations along with the dfops. */
741 1295 : dfc->dfc_blkres = tp->t_blk_res - tp->t_blk_res_used;
742 1295 : dfc->dfc_rtxres = tp->t_rtx_res - tp->t_rtx_res_used;
743 :
744 : /* Preserve the log reservation size. */
745 1295 : dfc->dfc_logres = tp->t_log_res;
746 :
747 1295 : error = xfs_defer_save_resources(&dfc->dfc_held, tp);
748 1295 : if (error) {
749 : /*
750 : * Resource capture should never fail, but if it does, we
751 : * still have to shut down the log and release things
752 : * properly.
753 : */
754 0 : xfs_force_shutdown(tp->t_mountp, SHUTDOWN_CORRUPT_INCORE);
755 : }
756 :
757 : /*
758 : * Grab extra references to the inodes and buffers because callers are
759 : * expected to release their held references after we commit the
760 : * transaction.
761 : */
762 2025 : for (i = 0; i < dfc->dfc_held.dr_inos; i++) {
763 730 : ASSERT(xfs_isilocked(dfc->dfc_held.dr_ip[i], XFS_ILOCK_EXCL));
764 730 : ihold(VFS_I(dfc->dfc_held.dr_ip[i]));
765 : }
766 :
767 1295 : for (i = 0; i < dfc->dfc_held.dr_bufs; i++)
768 0 : xfs_buf_hold(dfc->dfc_held.dr_bp[i]);
769 :
770 : return dfc;
771 : }
772 :
773 : /* Release all resources that we used to capture deferred ops. */
774 : void
775 0 : xfs_defer_ops_capture_free(
776 : struct xfs_mount *mp,
777 : struct xfs_defer_capture *dfc)
778 : {
779 0 : unsigned short i;
780 :
781 0 : xfs_defer_cancel_list(mp, &dfc->dfc_dfops);
782 :
783 0 : for (i = 0; i < dfc->dfc_held.dr_bufs; i++)
784 0 : xfs_buf_relse(dfc->dfc_held.dr_bp[i]);
785 :
786 0 : for (i = 0; i < dfc->dfc_held.dr_inos; i++)
787 0 : xfs_irele(dfc->dfc_held.dr_ip[i]);
788 :
789 0 : kmem_free(dfc);
790 0 : }
791 :
792 : /*
793 : * Capture any deferred ops and commit the transaction. This is the last step
794 : * needed to finish a log intent item that we recovered from the log. If any
795 : * of the deferred ops operate on an inode, the caller must pass in that inode
796 : * so that the reference can be transferred to the capture structure. The
797 : * caller must hold ILOCK_EXCL on the inode, and must unlock it before calling
798 : * xfs_defer_ops_continue.
799 : */
800 : int
801 5215 : xfs_defer_ops_capture_and_commit(
802 : struct xfs_trans *tp,
803 : struct list_head *capture_list)
804 : {
805 5215 : struct xfs_mount *mp = tp->t_mountp;
806 5215 : struct xfs_defer_capture *dfc;
807 5215 : int error;
808 :
809 : /* If we don't capture anything, commit transaction and exit. */
810 5215 : dfc = xfs_defer_ops_capture(tp);
811 5215 : if (IS_ERR(dfc)) {
812 0 : xfs_trans_cancel(tp);
813 0 : return PTR_ERR(dfc);
814 : }
815 5215 : if (!dfc)
816 3920 : return xfs_trans_commit(tp);
817 :
818 : /* Commit the transaction and add the capture structure to the list. */
819 1295 : error = xfs_trans_commit(tp);
820 1295 : if (error) {
821 0 : xfs_defer_ops_capture_free(mp, dfc);
822 0 : return error;
823 : }
824 :
825 1295 : list_add_tail(&dfc->dfc_list, capture_list);
826 1295 : return 0;
827 : }
828 :
829 : /*
830 : * Attach a chain of captured deferred ops to a new transaction and free the
831 : * capture structure. If an inode was captured, it will be passed back to the
832 : * caller with ILOCK_EXCL held and joined to the transaction with lockflags==0.
833 : * The caller now owns the inode reference.
834 : */
835 : void
836 1295 : xfs_defer_ops_continue(
837 : struct xfs_defer_capture *dfc,
838 : struct xfs_trans *tp,
839 : struct xfs_defer_resources *dres)
840 : {
841 1295 : unsigned int i;
842 :
843 1295 : ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
844 1295 : ASSERT(!(tp->t_flags & XFS_TRANS_DIRTY));
845 :
846 : /* Lock the captured resources to the new transaction. */
847 1295 : if (dfc->dfc_held.dr_inos > 2) {
848 0 : xfs_sort_inodes(dfc->dfc_held.dr_ip, dfc->dfc_held.dr_inos);
849 0 : xfs_lock_inodes(dfc->dfc_held.dr_ip, dfc->dfc_held.dr_inos,
850 : XFS_ILOCK_EXCL);
851 1295 : } else if (dfc->dfc_held.dr_inos == 2)
852 17 : xfs_lock_two_inodes(dfc->dfc_held.dr_ip[0], XFS_ILOCK_EXCL,
853 : dfc->dfc_held.dr_ip[1], XFS_ILOCK_EXCL);
854 1278 : else if (dfc->dfc_held.dr_inos == 1)
855 696 : xfs_ilock(dfc->dfc_held.dr_ip[0], XFS_ILOCK_EXCL);
856 :
857 1295 : for (i = 0; i < dfc->dfc_held.dr_bufs; i++)
858 0 : xfs_buf_lock(dfc->dfc_held.dr_bp[i]);
859 :
860 : /* Join the captured resources to the new transaction. */
861 1295 : xfs_defer_restore_resources(tp, &dfc->dfc_held);
862 2590 : memcpy(dres, &dfc->dfc_held, sizeof(struct xfs_defer_resources));
863 1295 : dres->dr_bufs = 0;
864 :
865 : /* Move captured dfops chain and state to the transaction. */
866 1295 : list_splice_init(&dfc->dfc_dfops, &tp->t_dfops);
867 1295 : tp->t_flags |= dfc->dfc_tpflags;
868 :
869 1295 : kmem_free(dfc);
870 1295 : }
871 :
872 : /* Release the resources captured and continued during recovery. */
873 : void
874 1295 : xfs_defer_resources_rele(
875 : struct xfs_defer_resources *dres)
876 : {
877 1295 : unsigned short i;
878 :
879 2025 : for (i = 0; i < dres->dr_inos; i++) {
880 730 : xfs_iunlock(dres->dr_ip[i], XFS_ILOCK_EXCL);
881 730 : xfs_irele(dres->dr_ip[i]);
882 730 : dres->dr_ip[i] = NULL;
883 : }
884 :
885 1295 : for (i = 0; i < dres->dr_bufs; i++) {
886 0 : xfs_buf_relse(dres->dr_bp[i]);
887 0 : dres->dr_bp[i] = NULL;
888 : }
889 :
890 1295 : dres->dr_inos = 0;
891 1295 : dres->dr_bufs = 0;
892 1295 : dres->dr_ordered = 0;
893 1295 : }
894 :
895 : static inline int __init
896 50 : xfs_defer_init_cache(void)
897 : {
898 50 : xfs_defer_pending_cache = kmem_cache_create("xfs_defer_pending",
899 : sizeof(struct xfs_defer_pending),
900 : 0, 0, NULL);
901 :
902 50 : return xfs_defer_pending_cache != NULL ? 0 : -ENOMEM;
903 : }
904 :
905 : static inline void
906 : xfs_defer_destroy_cache(void)
907 : {
908 49 : kmem_cache_destroy(xfs_defer_pending_cache);
909 49 : xfs_defer_pending_cache = NULL;
910 : }
911 :
912 : /* Set up caches for deferred work items. */
913 : int __init
914 50 : xfs_defer_init_item_caches(void)
915 : {
916 50 : int error;
917 :
918 50 : error = xfs_defer_init_cache();
919 50 : if (error)
920 : return error;
921 50 : error = xfs_rmap_intent_init_cache();
922 50 : if (error)
923 0 : goto err;
924 50 : error = xfs_refcount_intent_init_cache();
925 50 : if (error)
926 0 : goto err;
927 50 : error = xfs_bmap_intent_init_cache();
928 50 : if (error)
929 0 : goto err;
930 50 : error = xfs_extfree_intent_init_cache();
931 50 : if (error)
932 0 : goto err;
933 50 : error = xfs_attr_intent_init_cache();
934 50 : if (error)
935 0 : goto err;
936 50 : error = xfs_swapext_intent_init_cache();
937 50 : if (error)
938 0 : goto err;
939 :
940 : return 0;
941 0 : err:
942 0 : xfs_defer_destroy_item_caches();
943 0 : return error;
944 : }
945 :
946 : /* Destroy all the deferred work item caches, if they've been allocated. */
947 : void
948 49 : xfs_defer_destroy_item_caches(void)
949 : {
950 49 : xfs_swapext_intent_destroy_cache();
951 49 : xfs_attr_intent_destroy_cache();
952 49 : xfs_extfree_intent_destroy_cache();
953 49 : xfs_bmap_intent_destroy_cache();
954 49 : xfs_refcount_intent_destroy_cache();
955 49 : xfs_rmap_intent_destroy_cache();
956 49 : xfs_defer_destroy_cache();
957 49 : }
|
