2 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version 2
9 * of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it would be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_defer.h"
30 #include "xfs_trans.h"
31 #include "xfs_trace.h"
34 * Deferred Operations in XFS
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
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
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.
60 * Finishing a set of deferred operations is an involved process. To
61 * start, we define "rolling a deferred-op transaction" as follows:
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
70 * > Roll the transaction.
72 * NOTE: To avoid exceeding the transaction reservation, we limit the
73 * number of items that we attach to a given xfs_defer_pending.
75 * The actual finishing process looks like this:
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
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
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
95 * Requesting a Fresh Transaction while Finishing Deferred Work
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.
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
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.
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
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 * +-------------------------------------------------+
166 * If we should crash before t2 commits, log recovery replays
167 * the following intent items:
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)
175 * In the process of recovering, it should also generate and take care
176 * of these intent items:
178 * - Intent to free extent (C, B)
179 * - Intent to free extent (F, 1) (refcountbt block)
180 * - Intent to remove rmap (F, 1, REFC)
182 * Note that the continuation requested between t2 and t3 is likely to
186 static const struct xfs_defer_op_type *defer_op_types[XFS_DEFER_OPS_TYPE_MAX];
189 * For each pending item in the intake list, log its intent item and the
190 * associated extents, then add the entire intake list to the end of
194 xfs_defer_intake_work(
195 struct xfs_trans *tp,
196 struct xfs_defer_ops *dop)
198 struct list_head *li;
199 struct xfs_defer_pending *dfp;
201 list_for_each_entry(dfp, &dop->dop_intake, dfp_list) {
202 trace_xfs_defer_intake_work(tp->t_mountp, dfp);
203 dfp->dfp_intent = dfp->dfp_type->create_intent(tp,
205 list_sort(tp->t_mountp, &dfp->dfp_work,
206 dfp->dfp_type->diff_items);
207 list_for_each(li, &dfp->dfp_work)
208 dfp->dfp_type->log_item(tp, dfp->dfp_intent, li);
211 list_splice_tail_init(&dop->dop_intake, &dop->dop_pending);
214 /* Abort all the intents that were committed. */
216 xfs_defer_trans_abort(
217 struct xfs_trans *tp,
218 struct xfs_defer_ops *dop,
221 struct xfs_defer_pending *dfp;
223 trace_xfs_defer_trans_abort(tp->t_mountp, dop);
225 * If the transaction was committed, drop the intent reference
226 * since we're bailing out of here. The other reference is
227 * dropped when the intent hits the AIL. If the transaction
228 * was not committed, the intent is freed by the intent item
229 * unlock handler on abort.
231 if (!dop->dop_committed)
234 /* Abort intent items. */
235 list_for_each_entry(dfp, &dop->dop_pending, dfp_list) {
236 trace_xfs_defer_pending_abort(tp->t_mountp, dfp);
238 dfp->dfp_type->abort_intent(dfp->dfp_intent);
242 xfs_force_shutdown(tp->t_mountp, (error == -EFSCORRUPTED) ?
243 SHUTDOWN_CORRUPT_INCORE : SHUTDOWN_META_IO_ERROR);
246 /* Roll a transaction so we can do some deferred op processing. */
248 xfs_defer_trans_roll(
249 struct xfs_trans **tp,
250 struct xfs_defer_ops *dop,
251 struct xfs_inode *ip)
256 /* Log all the joined inodes except the one we passed in. */
257 for (i = 0; i < XFS_DEFER_OPS_NR_INODES && dop->dop_inodes[i]; i++) {
258 if (dop->dop_inodes[i] == ip)
260 xfs_trans_log_inode(*tp, dop->dop_inodes[i], XFS_ILOG_CORE);
263 trace_xfs_defer_trans_roll((*tp)->t_mountp, dop);
265 /* Roll the transaction. */
266 error = xfs_trans_roll(tp, ip);
268 trace_xfs_defer_trans_roll_error((*tp)->t_mountp, dop, error);
269 xfs_defer_trans_abort(*tp, dop, error);
272 dop->dop_committed = true;
274 /* Rejoin the joined inodes except the one we passed in. */
275 for (i = 0; i < XFS_DEFER_OPS_NR_INODES && dop->dop_inodes[i]; i++) {
276 if (dop->dop_inodes[i] == ip)
278 xfs_trans_ijoin(*tp, dop->dop_inodes[i], 0);
284 /* Do we have any work items to finish? */
286 xfs_defer_has_unfinished_work(
287 struct xfs_defer_ops *dop)
289 return !list_empty(&dop->dop_pending) || !list_empty(&dop->dop_intake);
293 * Add this inode to the deferred op. Each joined inode is relogged
294 * each time we roll the transaction, in addition to any inode passed
295 * to xfs_defer_finish().
299 struct xfs_defer_ops *dop,
300 struct xfs_inode *ip)
304 for (i = 0; i < XFS_DEFER_OPS_NR_INODES; i++) {
305 if (dop->dop_inodes[i] == ip)
307 else if (dop->dop_inodes[i] == NULL) {
308 dop->dop_inodes[i] = ip;
313 return -EFSCORRUPTED;
317 * Finish all the pending work. This involves logging intent items for
318 * any work items that wandered in since the last transaction roll (if
319 * one has even happened), rolling the transaction, and finishing the
320 * work items in the first item on the logged-and-pending list.
322 * If an inode is provided, relog it to the new transaction.
326 struct xfs_trans **tp,
327 struct xfs_defer_ops *dop,
328 struct xfs_inode *ip)
330 struct xfs_defer_pending *dfp;
331 struct list_head *li;
335 void (*cleanup_fn)(struct xfs_trans *, void *, int);
337 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
339 trace_xfs_defer_finish((*tp)->t_mountp, dop);
341 /* Until we run out of pending work to finish... */
342 while (xfs_defer_has_unfinished_work(dop)) {
343 /* Log intents for work items sitting in the intake. */
344 xfs_defer_intake_work(*tp, dop);
346 /* Roll the transaction. */
347 error = xfs_defer_trans_roll(tp, dop, ip);
351 /* Log an intent-done item for the first pending item. */
352 dfp = list_first_entry(&dop->dop_pending,
353 struct xfs_defer_pending, dfp_list);
354 trace_xfs_defer_pending_finish((*tp)->t_mountp, dfp);
355 dfp->dfp_done = dfp->dfp_type->create_done(*tp, dfp->dfp_intent,
357 cleanup_fn = dfp->dfp_type->finish_cleanup;
359 /* Finish the work items. */
361 list_for_each_safe(li, n, &dfp->dfp_work) {
364 error = dfp->dfp_type->finish_item(*tp, dop, li,
365 dfp->dfp_done, &state);
366 if (error == -EAGAIN) {
368 * Caller wants a fresh transaction;
369 * put the work item back on the list
372 list_add(li, &dfp->dfp_work);
377 * Clean up after ourselves and jump out.
378 * xfs_defer_cancel will take care of freeing
379 * all these lists and stuff.
382 cleanup_fn(*tp, state, error);
383 xfs_defer_trans_abort(*tp, dop, error);
387 if (error == -EAGAIN) {
389 * Caller wants a fresh transaction, so log a
390 * new log intent item to replace the old one
391 * and roll the transaction. See "Requesting
392 * a Fresh Transaction while Finishing
393 * Deferred Work" above.
395 dfp->dfp_intent = dfp->dfp_type->create_intent(*tp,
397 dfp->dfp_done = NULL;
398 list_for_each(li, &dfp->dfp_work)
399 dfp->dfp_type->log_item(*tp, dfp->dfp_intent,
402 /* Done with the dfp, free it. */
403 list_del(&dfp->dfp_list);
408 cleanup_fn(*tp, state, error);
413 trace_xfs_defer_finish_error((*tp)->t_mountp, dop, error);
415 trace_xfs_defer_finish_done((*tp)->t_mountp, dop);
420 * Free up any items left in the list.
424 struct xfs_defer_ops *dop)
426 struct xfs_defer_pending *dfp;
427 struct xfs_defer_pending *pli;
428 struct list_head *pwi;
431 trace_xfs_defer_cancel(NULL, dop);
434 * Free the pending items. Caller should already have arranged
435 * for the intent items to be released.
437 list_for_each_entry_safe(dfp, pli, &dop->dop_intake, dfp_list) {
438 trace_xfs_defer_intake_cancel(NULL, dfp);
439 list_del(&dfp->dfp_list);
440 list_for_each_safe(pwi, n, &dfp->dfp_work) {
443 dfp->dfp_type->cancel_item(pwi);
445 ASSERT(dfp->dfp_count == 0);
448 list_for_each_entry_safe(dfp, pli, &dop->dop_pending, dfp_list) {
449 trace_xfs_defer_pending_cancel(NULL, dfp);
450 list_del(&dfp->dfp_list);
451 list_for_each_safe(pwi, n, &dfp->dfp_work) {
454 dfp->dfp_type->cancel_item(pwi);
456 ASSERT(dfp->dfp_count == 0);
461 /* Add an item for later deferred processing. */
464 struct xfs_defer_ops *dop,
465 enum xfs_defer_ops_type type,
466 struct list_head *li)
468 struct xfs_defer_pending *dfp = NULL;
471 * Add the item to a pending item at the end of the intake list.
472 * If the last pending item has the same type, reuse it. Else,
473 * create a new pending item at the end of the intake list.
475 if (!list_empty(&dop->dop_intake)) {
476 dfp = list_last_entry(&dop->dop_intake,
477 struct xfs_defer_pending, dfp_list);
478 if (dfp->dfp_type->type != type ||
479 (dfp->dfp_type->max_items &&
480 dfp->dfp_count >= dfp->dfp_type->max_items))
484 dfp = kmem_alloc(sizeof(struct xfs_defer_pending),
486 dfp->dfp_type = defer_op_types[type];
487 dfp->dfp_intent = NULL;
488 dfp->dfp_done = NULL;
490 INIT_LIST_HEAD(&dfp->dfp_work);
491 list_add_tail(&dfp->dfp_list, &dop->dop_intake);
494 list_add_tail(li, &dfp->dfp_work);
498 /* Initialize a deferred operation list. */
500 xfs_defer_init_op_type(
501 const struct xfs_defer_op_type *type)
503 defer_op_types[type->type] = type;
506 /* Initialize a deferred operation. */
509 struct xfs_defer_ops *dop,
512 dop->dop_committed = false;
513 dop->dop_low = false;
514 memset(&dop->dop_inodes, 0, sizeof(dop->dop_inodes));
516 INIT_LIST_HEAD(&dop->dop_intake);
517 INIT_LIST_HEAD(&dop->dop_pending);
518 trace_xfs_defer_init(NULL, dop);