xref: /linux/fs/xfs/libxfs/xfs_defer.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
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  *
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
89  * as needed.
90  *
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
93  * there:
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  * +-------------------------------------------------+
128  *
129  * If we should crash before t2 commits, log recovery replays
130  * the following intent items:
131  *
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)
137  *
138  * In the process of recovering, it should also generate and take care
139  * of these intent items:
140  *
141  * - Intent to free extent (C, B)
142  * - Intent to free extent (F, 1) (refcountbt block)
143  * - Intent to remove rmap (F, 1, REFC)
144  */
145 
146 static const struct xfs_defer_op_type *defer_op_types[XFS_DEFER_OPS_TYPE_MAX];
147 
148 /*
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
151  * the pending list.
152  */
153 STATIC void
154 xfs_defer_intake_work(
155 	struct xfs_trans		*tp,
156 	struct xfs_defer_ops		*dop)
157 {
158 	struct list_head		*li;
159 	struct xfs_defer_pending	*dfp;
160 
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,
164 				dfp->dfp_count);
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);
169 	}
170 
171 	list_splice_tail_init(&dop->dop_intake, &dop->dop_pending);
172 }
173 
174 /* Abort all the intents that were committed. */
175 STATIC void
176 xfs_defer_trans_abort(
177 	struct xfs_trans		*tp,
178 	struct xfs_defer_ops		*dop,
179 	int				error)
180 {
181 	struct xfs_defer_pending	*dfp;
182 
183 	trace_xfs_defer_trans_abort(tp->t_mountp, dop);
184 	/*
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.
190 	 */
191 	if (!dop->dop_committed)
192 		return;
193 
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);
197 		if (dfp->dfp_committed)
198 			dfp->dfp_type->abort_intent(dfp->dfp_intent);
199 	}
200 
201 	/* Shut down FS. */
202 	xfs_force_shutdown(tp->t_mountp, (error == -EFSCORRUPTED) ?
203 			SHUTDOWN_CORRUPT_INCORE : SHUTDOWN_META_IO_ERROR);
204 }
205 
206 /* Roll a transaction so we can do some deferred op processing. */
207 STATIC int
208 xfs_defer_trans_roll(
209 	struct xfs_trans		**tp,
210 	struct xfs_defer_ops		*dop,
211 	struct xfs_inode		*ip)
212 {
213 	int				i;
214 	int				error;
215 
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)
219 			continue;
220 		xfs_trans_log_inode(*tp, dop->dop_inodes[i], XFS_ILOG_CORE);
221 	}
222 
223 	trace_xfs_defer_trans_roll((*tp)->t_mountp, dop);
224 
225 	/* Roll the transaction. */
226 	error = xfs_trans_roll(tp, ip);
227 	if (error) {
228 		trace_xfs_defer_trans_roll_error((*tp)->t_mountp, dop, error);
229 		xfs_defer_trans_abort(*tp, dop, error);
230 		return error;
231 	}
232 	dop->dop_committed = true;
233 
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)
237 			continue;
238 		xfs_trans_ijoin(*tp, dop->dop_inodes[i], 0);
239 	}
240 
241 	return error;
242 }
243 
244 /* Do we have any work items to finish? */
245 bool
246 xfs_defer_has_unfinished_work(
247 	struct xfs_defer_ops		*dop)
248 {
249 	return !list_empty(&dop->dop_pending) || !list_empty(&dop->dop_intake);
250 }
251 
252 /*
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().
256  */
257 int
258 xfs_defer_join(
259 	struct xfs_defer_ops		*dop,
260 	struct xfs_inode		*ip)
261 {
262 	int				i;
263 
264 	for (i = 0; i < XFS_DEFER_OPS_NR_INODES; i++) {
265 		if (dop->dop_inodes[i] == ip)
266 			return 0;
267 		else if (dop->dop_inodes[i] == NULL) {
268 			dop->dop_inodes[i] = ip;
269 			return 0;
270 		}
271 	}
272 
273 	return -EFSCORRUPTED;
274 }
275 
276 /*
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.
281  *
282  * If an inode is provided, relog it to the new transaction.
283  */
284 int
285 xfs_defer_finish(
286 	struct xfs_trans		**tp,
287 	struct xfs_defer_ops		*dop,
288 	struct xfs_inode		*ip)
289 {
290 	struct xfs_defer_pending	*dfp;
291 	struct list_head		*li;
292 	struct list_head		*n;
293 	void				*done_item = NULL;
294 	void				*state;
295 	int				error = 0;
296 	void				(*cleanup_fn)(struct xfs_trans *, void *, int);
297 
298 	ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
299 
300 	trace_xfs_defer_finish((*tp)->t_mountp, dop);
301 
302 	/* Until we run out of pending work to finish... */
303 	while (xfs_defer_has_unfinished_work(dop)) {
304 		/* Log intents for work items sitting in the intake. */
305 		xfs_defer_intake_work(*tp, dop);
306 
307 		/* Roll the transaction. */
308 		error = xfs_defer_trans_roll(tp, dop, ip);
309 		if (error)
310 			goto out;
311 
312 		/* Mark all pending intents as committed. */
313 		list_for_each_entry_reverse(dfp, &dop->dop_pending, dfp_list) {
314 			if (dfp->dfp_committed)
315 				break;
316 			trace_xfs_defer_pending_commit((*tp)->t_mountp, dfp);
317 			dfp->dfp_committed = true;
318 		}
319 
320 		/* Log an intent-done item for the first pending item. */
321 		dfp = list_first_entry(&dop->dop_pending,
322 				struct xfs_defer_pending, dfp_list);
323 		trace_xfs_defer_pending_finish((*tp)->t_mountp, dfp);
324 		done_item = dfp->dfp_type->create_done(*tp, dfp->dfp_intent,
325 				dfp->dfp_count);
326 		cleanup_fn = dfp->dfp_type->finish_cleanup;
327 
328 		/* Finish the work items. */
329 		state = NULL;
330 		list_for_each_safe(li, n, &dfp->dfp_work) {
331 			list_del(li);
332 			dfp->dfp_count--;
333 			error = dfp->dfp_type->finish_item(*tp, dop, li,
334 					done_item, &state);
335 			if (error) {
336 				/*
337 				 * Clean up after ourselves and jump out.
338 				 * xfs_defer_cancel will take care of freeing
339 				 * all these lists and stuff.
340 				 */
341 				if (cleanup_fn)
342 					cleanup_fn(*tp, state, error);
343 				xfs_defer_trans_abort(*tp, dop, error);
344 				goto out;
345 			}
346 		}
347 		/* Done with the dfp, free it. */
348 		list_del(&dfp->dfp_list);
349 		kmem_free(dfp);
350 
351 		if (cleanup_fn)
352 			cleanup_fn(*tp, state, error);
353 	}
354 
355 out:
356 	if (error)
357 		trace_xfs_defer_finish_error((*tp)->t_mountp, dop, error);
358 	else
359 		trace_xfs_defer_finish_done((*tp)->t_mountp, dop);
360 	return error;
361 }
362 
363 /*
364  * Free up any items left in the list.
365  */
366 void
367 xfs_defer_cancel(
368 	struct xfs_defer_ops		*dop)
369 {
370 	struct xfs_defer_pending	*dfp;
371 	struct xfs_defer_pending	*pli;
372 	struct list_head		*pwi;
373 	struct list_head		*n;
374 
375 	trace_xfs_defer_cancel(NULL, dop);
376 
377 	/*
378 	 * Free the pending items.  Caller should already have arranged
379 	 * for the intent items to be released.
380 	 */
381 	list_for_each_entry_safe(dfp, pli, &dop->dop_intake, dfp_list) {
382 		trace_xfs_defer_intake_cancel(NULL, dfp);
383 		list_del(&dfp->dfp_list);
384 		list_for_each_safe(pwi, n, &dfp->dfp_work) {
385 			list_del(pwi);
386 			dfp->dfp_count--;
387 			dfp->dfp_type->cancel_item(pwi);
388 		}
389 		ASSERT(dfp->dfp_count == 0);
390 		kmem_free(dfp);
391 	}
392 	list_for_each_entry_safe(dfp, pli, &dop->dop_pending, dfp_list) {
393 		trace_xfs_defer_pending_cancel(NULL, dfp);
394 		list_del(&dfp->dfp_list);
395 		list_for_each_safe(pwi, n, &dfp->dfp_work) {
396 			list_del(pwi);
397 			dfp->dfp_count--;
398 			dfp->dfp_type->cancel_item(pwi);
399 		}
400 		ASSERT(dfp->dfp_count == 0);
401 		kmem_free(dfp);
402 	}
403 }
404 
405 /* Add an item for later deferred processing. */
406 void
407 xfs_defer_add(
408 	struct xfs_defer_ops		*dop,
409 	enum xfs_defer_ops_type		type,
410 	struct list_head		*li)
411 {
412 	struct xfs_defer_pending	*dfp = NULL;
413 
414 	/*
415 	 * Add the item to a pending item at the end of the intake list.
416 	 * If the last pending item has the same type, reuse it.  Else,
417 	 * create a new pending item at the end of the intake list.
418 	 */
419 	if (!list_empty(&dop->dop_intake)) {
420 		dfp = list_last_entry(&dop->dop_intake,
421 				struct xfs_defer_pending, dfp_list);
422 		if (dfp->dfp_type->type != type ||
423 		    (dfp->dfp_type->max_items &&
424 		     dfp->dfp_count >= dfp->dfp_type->max_items))
425 			dfp = NULL;
426 	}
427 	if (!dfp) {
428 		dfp = kmem_alloc(sizeof(struct xfs_defer_pending),
429 				KM_SLEEP | KM_NOFS);
430 		dfp->dfp_type = defer_op_types[type];
431 		dfp->dfp_committed = false;
432 		dfp->dfp_intent = NULL;
433 		dfp->dfp_count = 0;
434 		INIT_LIST_HEAD(&dfp->dfp_work);
435 		list_add_tail(&dfp->dfp_list, &dop->dop_intake);
436 	}
437 
438 	list_add_tail(li, &dfp->dfp_work);
439 	dfp->dfp_count++;
440 }
441 
442 /* Initialize a deferred operation list. */
443 void
444 xfs_defer_init_op_type(
445 	const struct xfs_defer_op_type	*type)
446 {
447 	defer_op_types[type->type] = type;
448 }
449 
450 /* Initialize a deferred operation. */
451 void
452 xfs_defer_init(
453 	struct xfs_defer_ops		*dop,
454 	xfs_fsblock_t			*fbp)
455 {
456 	dop->dop_committed = false;
457 	dop->dop_low = false;
458 	memset(&dop->dop_inodes, 0, sizeof(dop->dop_inodes));
459 	*fbp = NULLFSBLOCK;
460 	INIT_LIST_HEAD(&dop->dop_intake);
461 	INIT_LIST_HEAD(&dop->dop_pending);
462 	trace_xfs_defer_init(NULL, dop);
463 }
464