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.
85 * The key here is that we must log an intent item for all pending
86 * work items every time we roll the transaction, and that we must log
87 * a done item as soon as the work is completed. With this mechanism
88 * we can perform complex remapping operations, chaining intent items
91 * This is an example of remapping the extent (E, E+B) into file X at
92 * offset A and dealing with the extent (C, C+B) already being mapped
94 * +-------------------------------------------------+
95 * | Unmap file X startblock C offset A length B | t0
96 * | Intent to reduce refcount for extent (C, B) |
97 * | Intent to remove rmap (X, C, A, B) |
98 * | Intent to free extent (D, 1) (bmbt block) |
99 * | Intent to map (X, A, B) at startblock E |
100 * +-------------------------------------------------+
101 * | Map file X startblock E offset A length B | t1
102 * | Done mapping (X, E, A, B) |
103 * | Intent to increase refcount for extent (E, B) |
104 * | Intent to add rmap (X, E, A, B) |
105 * +-------------------------------------------------+
106 * | Reduce refcount for extent (C, B) | t2
107 * | Done reducing refcount for extent (C, B) |
108 * | Increase refcount for extent (E, B) |
109 * | Done increasing refcount for extent (E, B) |
110 * | Intent to free extent (C, B) |
111 * | Intent to free extent (F, 1) (refcountbt block) |
112 * | Intent to remove rmap (F, 1, REFC) |
113 * +-------------------------------------------------+
114 * | Remove rmap (X, C, A, B) | t3
115 * | Done removing rmap (X, C, A, B) |
116 * | Add rmap (X, E, A, B) |
117 * | Done adding rmap (X, E, A, B) |
118 * | Remove rmap (F, 1, REFC) |
119 * | Done removing rmap (F, 1, REFC) |
120 * +-------------------------------------------------+
121 * | Free extent (C, B) | t4
122 * | Done freeing extent (C, B) |
123 * | Free extent (D, 1) |
124 * | Done freeing extent (D, 1) |
125 * | Free extent (F, 1) |
126 * | Done freeing extent (F, 1) |
127 * +-------------------------------------------------+
129 * If we should crash before t2 commits, log recovery replays
130 * the following intent items:
132 * - Intent to reduce refcount for extent (C, B)
133 * - Intent to remove rmap (X, C, A, B)
134 * - Intent to free extent (D, 1) (bmbt block)
135 * - Intent to increase refcount for extent (E, B)
136 * - Intent to add rmap (X, E, A, B)
138 * In the process of recovering, it should also generate and take care
139 * of these intent items:
141 * - Intent to free extent (C, B)
142 * - Intent to free extent (F, 1) (refcountbt block)
143 * - Intent to remove rmap (F, 1, REFC)
146 static const struct xfs_defer_op_type *defer_op_types[XFS_DEFER_OPS_TYPE_MAX];
149 * For each pending item in the intake list, log its intent item and the
150 * associated extents, then add the entire intake list to the end of
154 xfs_defer_intake_work(
155 struct xfs_trans *tp,
156 struct xfs_defer_ops *dop)
158 struct list_head *li;
159 struct xfs_defer_pending *dfp;
161 list_for_each_entry(dfp, &dop->dop_intake, dfp_list) {
162 trace_xfs_defer_intake_work(tp->t_mountp, dfp);
163 dfp->dfp_intent = dfp->dfp_type->create_intent(tp,
165 list_sort(tp->t_mountp, &dfp->dfp_work,
166 dfp->dfp_type->diff_items);
167 list_for_each(li, &dfp->dfp_work)
168 dfp->dfp_type->log_item(tp, dfp->dfp_intent, li);
171 list_splice_tail_init(&dop->dop_intake, &dop->dop_pending);
174 /* Abort all the intents that were committed. */
176 xfs_defer_trans_abort(
177 struct xfs_trans *tp,
178 struct xfs_defer_ops *dop,
181 struct xfs_defer_pending *dfp;
183 trace_xfs_defer_trans_abort(tp->t_mountp, dop);
185 * If the transaction was committed, drop the intent reference
186 * since we're bailing out of here. The other reference is
187 * dropped when the intent hits the AIL. If the transaction
188 * was not committed, the intent is freed by the intent item
189 * unlock handler on abort.
191 if (!dop->dop_committed)
194 /* Abort intent items. */
195 list_for_each_entry(dfp, &dop->dop_pending, dfp_list) {
196 trace_xfs_defer_pending_abort(tp->t_mountp, dfp);
198 dfp->dfp_type->abort_intent(dfp->dfp_intent);
202 xfs_force_shutdown(tp->t_mountp, (error == -EFSCORRUPTED) ?
203 SHUTDOWN_CORRUPT_INCORE : SHUTDOWN_META_IO_ERROR);
206 /* Roll a transaction so we can do some deferred op processing. */
208 xfs_defer_trans_roll(
209 struct xfs_trans **tp,
210 struct xfs_defer_ops *dop,
211 struct xfs_inode *ip)
216 /* Log all the joined inodes except the one we passed in. */
217 for (i = 0; i < XFS_DEFER_OPS_NR_INODES && dop->dop_inodes[i]; i++) {
218 if (dop->dop_inodes[i] == ip)
220 xfs_trans_log_inode(*tp, dop->dop_inodes[i], XFS_ILOG_CORE);
223 trace_xfs_defer_trans_roll((*tp)->t_mountp, dop);
225 /* Roll the transaction. */
226 error = xfs_trans_roll(tp, ip);
228 trace_xfs_defer_trans_roll_error((*tp)->t_mountp, dop, error);
229 xfs_defer_trans_abort(*tp, dop, error);
232 dop->dop_committed = true;
234 /* Rejoin the joined inodes except the one we passed in. */
235 for (i = 0; i < XFS_DEFER_OPS_NR_INODES && dop->dop_inodes[i]; i++) {
236 if (dop->dop_inodes[i] == ip)
238 xfs_trans_ijoin(*tp, dop->dop_inodes[i], 0);
244 /* Do we have any work items to finish? */
246 xfs_defer_has_unfinished_work(
247 struct xfs_defer_ops *dop)
249 return !list_empty(&dop->dop_pending) || !list_empty(&dop->dop_intake);
253 * Add this inode to the deferred op. Each joined inode is relogged
254 * each time we roll the transaction, in addition to any inode passed
255 * to xfs_defer_finish().
259 struct xfs_defer_ops *dop,
260 struct xfs_inode *ip)
264 for (i = 0; i < XFS_DEFER_OPS_NR_INODES; i++) {
265 if (dop->dop_inodes[i] == ip)
267 else if (dop->dop_inodes[i] == NULL) {
268 dop->dop_inodes[i] = ip;
273 return -EFSCORRUPTED;
277 * Finish all the pending work. This involves logging intent items for
278 * any work items that wandered in since the last transaction roll (if
279 * one has even happened), rolling the transaction, and finishing the
280 * work items in the first item on the logged-and-pending list.
282 * If an inode is provided, relog it to the new transaction.
286 struct xfs_trans **tp,
287 struct xfs_defer_ops *dop,
288 struct xfs_inode *ip)
290 struct xfs_defer_pending *dfp;
291 struct list_head *li;
295 void (*cleanup_fn)(struct xfs_trans *, void *, int);
297 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
299 trace_xfs_defer_finish((*tp)->t_mountp, dop);
301 /* Until we run out of pending work to finish... */
302 while (xfs_defer_has_unfinished_work(dop)) {
303 /* Log intents for work items sitting in the intake. */
304 xfs_defer_intake_work(*tp, dop);
306 /* Roll the transaction. */
307 error = xfs_defer_trans_roll(tp, dop, ip);
311 /* Log an intent-done item for the first pending item. */
312 dfp = list_first_entry(&dop->dop_pending,
313 struct xfs_defer_pending, dfp_list);
314 trace_xfs_defer_pending_finish((*tp)->t_mountp, dfp);
315 dfp->dfp_done = dfp->dfp_type->create_done(*tp, dfp->dfp_intent,
317 cleanup_fn = dfp->dfp_type->finish_cleanup;
319 /* Finish the work items. */
321 list_for_each_safe(li, n, &dfp->dfp_work) {
324 error = dfp->dfp_type->finish_item(*tp, dop, li,
325 dfp->dfp_done, &state);
328 * Clean up after ourselves and jump out.
329 * xfs_defer_cancel will take care of freeing
330 * all these lists and stuff.
333 cleanup_fn(*tp, state, error);
334 xfs_defer_trans_abort(*tp, dop, error);
338 /* Done with the dfp, free it. */
339 list_del(&dfp->dfp_list);
343 cleanup_fn(*tp, state, error);
348 trace_xfs_defer_finish_error((*tp)->t_mountp, dop, error);
350 trace_xfs_defer_finish_done((*tp)->t_mountp, dop);
355 * Free up any items left in the list.
359 struct xfs_defer_ops *dop)
361 struct xfs_defer_pending *dfp;
362 struct xfs_defer_pending *pli;
363 struct list_head *pwi;
366 trace_xfs_defer_cancel(NULL, dop);
369 * Free the pending items. Caller should already have arranged
370 * for the intent items to be released.
372 list_for_each_entry_safe(dfp, pli, &dop->dop_intake, dfp_list) {
373 trace_xfs_defer_intake_cancel(NULL, dfp);
374 list_del(&dfp->dfp_list);
375 list_for_each_safe(pwi, n, &dfp->dfp_work) {
378 dfp->dfp_type->cancel_item(pwi);
380 ASSERT(dfp->dfp_count == 0);
383 list_for_each_entry_safe(dfp, pli, &dop->dop_pending, dfp_list) {
384 trace_xfs_defer_pending_cancel(NULL, dfp);
385 list_del(&dfp->dfp_list);
386 list_for_each_safe(pwi, n, &dfp->dfp_work) {
389 dfp->dfp_type->cancel_item(pwi);
391 ASSERT(dfp->dfp_count == 0);
396 /* Add an item for later deferred processing. */
399 struct xfs_defer_ops *dop,
400 enum xfs_defer_ops_type type,
401 struct list_head *li)
403 struct xfs_defer_pending *dfp = NULL;
406 * Add the item to a pending item at the end of the intake list.
407 * If the last pending item has the same type, reuse it. Else,
408 * create a new pending item at the end of the intake list.
410 if (!list_empty(&dop->dop_intake)) {
411 dfp = list_last_entry(&dop->dop_intake,
412 struct xfs_defer_pending, dfp_list);
413 if (dfp->dfp_type->type != type ||
414 (dfp->dfp_type->max_items &&
415 dfp->dfp_count >= dfp->dfp_type->max_items))
419 dfp = kmem_alloc(sizeof(struct xfs_defer_pending),
421 dfp->dfp_type = defer_op_types[type];
422 dfp->dfp_intent = NULL;
423 dfp->dfp_done = NULL;
425 INIT_LIST_HEAD(&dfp->dfp_work);
426 list_add_tail(&dfp->dfp_list, &dop->dop_intake);
429 list_add_tail(li, &dfp->dfp_work);
433 /* Initialize a deferred operation list. */
435 xfs_defer_init_op_type(
436 const struct xfs_defer_op_type *type)
438 defer_op_types[type->type] = type;
441 /* Initialize a deferred operation. */
444 struct xfs_defer_ops *dop,
447 dop->dop_committed = false;
448 dop->dop_low = false;
449 memset(&dop->dop_inodes, 0, sizeof(dop->dop_inodes));
451 INIT_LIST_HEAD(&dop->dop_intake);
452 INIT_LIST_HEAD(&dop->dop_pending);
453 trace_xfs_defer_init(NULL, dop);