xref: /linux/fs/xfs/libxfs/xfs_defer.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1 /*
2  * Copyright (C) 2016 Oracle.  All Rights Reserved.
3  *
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  *
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.
10  *
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.
15  *
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.
19  */
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_bit.h"
27 #include "xfs_sb.h"
28 #include "xfs_mount.h"
29 #include "xfs_defer.h"
30 #include "xfs_trans.h"
31 #include "xfs_trace.h"
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[XFS_DEFER_OPS_TYPE_MAX];
187 
188 /*
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
191  * the pending list.
192  */
193 STATIC void
194 xfs_defer_intake_work(
195 	struct xfs_trans		*tp,
196 	struct xfs_defer_ops		*dop)
197 {
198 	struct list_head		*li;
199 	struct xfs_defer_pending	*dfp;
200 
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,
204 				dfp->dfp_count);
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);
209 	}
210 
211 	list_splice_tail_init(&dop->dop_intake, &dop->dop_pending);
212 }
213 
214 /* Abort all the intents that were committed. */
215 STATIC void
216 xfs_defer_trans_abort(
217 	struct xfs_trans		*tp,
218 	struct xfs_defer_ops		*dop,
219 	int				error)
220 {
221 	struct xfs_defer_pending	*dfp;
222 
223 	trace_xfs_defer_trans_abort(tp->t_mountp, dop);
224 	/*
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.
230 	 */
231 	if (!dop->dop_committed)
232 		return;
233 
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);
237 		if (!dfp->dfp_done)
238 			dfp->dfp_type->abort_intent(dfp->dfp_intent);
239 	}
240 
241 	/* Shut down FS. */
242 	xfs_force_shutdown(tp->t_mountp, (error == -EFSCORRUPTED) ?
243 			SHUTDOWN_CORRUPT_INCORE : SHUTDOWN_META_IO_ERROR);
244 }
245 
246 /* Roll a transaction so we can do some deferred op processing. */
247 STATIC int
248 xfs_defer_trans_roll(
249 	struct xfs_trans		**tp,
250 	struct xfs_defer_ops		*dop,
251 	struct xfs_inode		*ip)
252 {
253 	int				i;
254 	int				error;
255 
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)
259 			continue;
260 		xfs_trans_log_inode(*tp, dop->dop_inodes[i], XFS_ILOG_CORE);
261 	}
262 
263 	trace_xfs_defer_trans_roll((*tp)->t_mountp, dop);
264 
265 	/* Roll the transaction. */
266 	error = xfs_trans_roll(tp, ip);
267 	if (error) {
268 		trace_xfs_defer_trans_roll_error((*tp)->t_mountp, dop, error);
269 		xfs_defer_trans_abort(*tp, dop, error);
270 		return error;
271 	}
272 	dop->dop_committed = true;
273 
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)
277 			continue;
278 		xfs_trans_ijoin(*tp, dop->dop_inodes[i], 0);
279 	}
280 
281 	return error;
282 }
283 
284 /* Do we have any work items to finish? */
285 bool
286 xfs_defer_has_unfinished_work(
287 	struct xfs_defer_ops		*dop)
288 {
289 	return !list_empty(&dop->dop_pending) || !list_empty(&dop->dop_intake);
290 }
291 
292 /*
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().
296  */
297 int
298 xfs_defer_join(
299 	struct xfs_defer_ops		*dop,
300 	struct xfs_inode		*ip)
301 {
302 	int				i;
303 
304 	for (i = 0; i < XFS_DEFER_OPS_NR_INODES; i++) {
305 		if (dop->dop_inodes[i] == ip)
306 			return 0;
307 		else if (dop->dop_inodes[i] == NULL) {
308 			dop->dop_inodes[i] = ip;
309 			return 0;
310 		}
311 	}
312 
313 	return -EFSCORRUPTED;
314 }
315 
316 /*
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.
321  *
322  * If an inode is provided, relog it to the new transaction.
323  */
324 int
325 xfs_defer_finish(
326 	struct xfs_trans		**tp,
327 	struct xfs_defer_ops		*dop,
328 	struct xfs_inode		*ip)
329 {
330 	struct xfs_defer_pending	*dfp;
331 	struct list_head		*li;
332 	struct list_head		*n;
333 	void				*state;
334 	int				error = 0;
335 	void				(*cleanup_fn)(struct xfs_trans *, void *, int);
336 
337 	ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
338 
339 	trace_xfs_defer_finish((*tp)->t_mountp, dop);
340 
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);
345 
346 		/* Roll the transaction. */
347 		error = xfs_defer_trans_roll(tp, dop, ip);
348 		if (error)
349 			goto out;
350 
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,
356 				dfp->dfp_count);
357 		cleanup_fn = dfp->dfp_type->finish_cleanup;
358 
359 		/* Finish the work items. */
360 		state = NULL;
361 		list_for_each_safe(li, n, &dfp->dfp_work) {
362 			list_del(li);
363 			dfp->dfp_count--;
364 			error = dfp->dfp_type->finish_item(*tp, dop, li,
365 					dfp->dfp_done, &state);
366 			if (error == -EAGAIN) {
367 				/*
368 				 * Caller wants a fresh transaction;
369 				 * put the work item back on the list
370 				 * and jump out.
371 				 */
372 				list_add(li, &dfp->dfp_work);
373 				dfp->dfp_count++;
374 				break;
375 			} else if (error) {
376 				/*
377 				 * Clean up after ourselves and jump out.
378 				 * xfs_defer_cancel will take care of freeing
379 				 * all these lists and stuff.
380 				 */
381 				if (cleanup_fn)
382 					cleanup_fn(*tp, state, error);
383 				xfs_defer_trans_abort(*tp, dop, error);
384 				goto out;
385 			}
386 		}
387 		if (error == -EAGAIN) {
388 			/*
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.
394 			 */
395 			dfp->dfp_intent = dfp->dfp_type->create_intent(*tp,
396 					dfp->dfp_count);
397 			dfp->dfp_done = NULL;
398 			list_for_each(li, &dfp->dfp_work)
399 				dfp->dfp_type->log_item(*tp, dfp->dfp_intent,
400 						li);
401 		} else {
402 			/* Done with the dfp, free it. */
403 			list_del(&dfp->dfp_list);
404 			kmem_free(dfp);
405 		}
406 
407 		if (cleanup_fn)
408 			cleanup_fn(*tp, state, error);
409 	}
410 
411 out:
412 	if (error)
413 		trace_xfs_defer_finish_error((*tp)->t_mountp, dop, error);
414 	else
415 		trace_xfs_defer_finish_done((*tp)->t_mountp, dop);
416 	return error;
417 }
418 
419 /*
420  * Free up any items left in the list.
421  */
422 void
423 xfs_defer_cancel(
424 	struct xfs_defer_ops		*dop)
425 {
426 	struct xfs_defer_pending	*dfp;
427 	struct xfs_defer_pending	*pli;
428 	struct list_head		*pwi;
429 	struct list_head		*n;
430 
431 	trace_xfs_defer_cancel(NULL, dop);
432 
433 	/*
434 	 * Free the pending items.  Caller should already have arranged
435 	 * for the intent items to be released.
436 	 */
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) {
441 			list_del(pwi);
442 			dfp->dfp_count--;
443 			dfp->dfp_type->cancel_item(pwi);
444 		}
445 		ASSERT(dfp->dfp_count == 0);
446 		kmem_free(dfp);
447 	}
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) {
452 			list_del(pwi);
453 			dfp->dfp_count--;
454 			dfp->dfp_type->cancel_item(pwi);
455 		}
456 		ASSERT(dfp->dfp_count == 0);
457 		kmem_free(dfp);
458 	}
459 }
460 
461 /* Add an item for later deferred processing. */
462 void
463 xfs_defer_add(
464 	struct xfs_defer_ops		*dop,
465 	enum xfs_defer_ops_type		type,
466 	struct list_head		*li)
467 {
468 	struct xfs_defer_pending	*dfp = NULL;
469 
470 	/*
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.
474 	 */
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))
481 			dfp = NULL;
482 	}
483 	if (!dfp) {
484 		dfp = kmem_alloc(sizeof(struct xfs_defer_pending),
485 				KM_SLEEP | KM_NOFS);
486 		dfp->dfp_type = defer_op_types[type];
487 		dfp->dfp_intent = NULL;
488 		dfp->dfp_done = NULL;
489 		dfp->dfp_count = 0;
490 		INIT_LIST_HEAD(&dfp->dfp_work);
491 		list_add_tail(&dfp->dfp_list, &dop->dop_intake);
492 	}
493 
494 	list_add_tail(li, &dfp->dfp_work);
495 	dfp->dfp_count++;
496 }
497 
498 /* Initialize a deferred operation list. */
499 void
500 xfs_defer_init_op_type(
501 	const struct xfs_defer_op_type	*type)
502 {
503 	defer_op_types[type->type] = type;
504 }
505 
506 /* Initialize a deferred operation. */
507 void
508 xfs_defer_init(
509 	struct xfs_defer_ops		*dop,
510 	xfs_fsblock_t			*fbp)
511 {
512 	dop->dop_committed = false;
513 	dop->dop_low = false;
514 	memset(&dop->dop_inodes, 0, sizeof(dop->dop_inodes));
515 	*fbp = NULLFSBLOCK;
516 	INIT_LIST_HEAD(&dop->dop_intake);
517 	INIT_LIST_HEAD(&dop->dop_pending);
518 	trace_xfs_defer_init(NULL, dop);
519 }
520