Loading...
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_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
25#include "xfs_bit.h"
26#include "xfs_shared.h"
27#include "xfs_mount.h"
28#include "xfs_defer.h"
29#include "xfs_trans.h"
30#include "xfs_trans_priv.h"
31#include "xfs_buf_item.h"
32#include "xfs_refcount_item.h"
33#include "xfs_log.h"
34#include "xfs_refcount.h"
35
36
37kmem_zone_t *xfs_cui_zone;
38kmem_zone_t *xfs_cud_zone;
39
40static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip)
41{
42 return container_of(lip, struct xfs_cui_log_item, cui_item);
43}
44
45void
46xfs_cui_item_free(
47 struct xfs_cui_log_item *cuip)
48{
49 if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS)
50 kmem_free(cuip);
51 else
52 kmem_zone_free(xfs_cui_zone, cuip);
53}
54
55/*
56 * Freeing the CUI requires that we remove it from the AIL if it has already
57 * been placed there. However, the CUI may not yet have been placed in the AIL
58 * when called by xfs_cui_release() from CUD processing due to the ordering of
59 * committed vs unpin operations in bulk insert operations. Hence the reference
60 * count to ensure only the last caller frees the CUI.
61 */
62void
63xfs_cui_release(
64 struct xfs_cui_log_item *cuip)
65{
66 ASSERT(atomic_read(&cuip->cui_refcount) > 0);
67 if (atomic_dec_and_test(&cuip->cui_refcount)) {
68 xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR);
69 xfs_cui_item_free(cuip);
70 }
71}
72
73
74STATIC void
75xfs_cui_item_size(
76 struct xfs_log_item *lip,
77 int *nvecs,
78 int *nbytes)
79{
80 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
81
82 *nvecs += 1;
83 *nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents);
84}
85
86/*
87 * This is called to fill in the vector of log iovecs for the
88 * given cui log item. We use only 1 iovec, and we point that
89 * at the cui_log_format structure embedded in the cui item.
90 * It is at this point that we assert that all of the extent
91 * slots in the cui item have been filled.
92 */
93STATIC void
94xfs_cui_item_format(
95 struct xfs_log_item *lip,
96 struct xfs_log_vec *lv)
97{
98 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
99 struct xfs_log_iovec *vecp = NULL;
100
101 ASSERT(atomic_read(&cuip->cui_next_extent) ==
102 cuip->cui_format.cui_nextents);
103
104 cuip->cui_format.cui_type = XFS_LI_CUI;
105 cuip->cui_format.cui_size = 1;
106
107 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format,
108 xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents));
109}
110
111/*
112 * Pinning has no meaning for an cui item, so just return.
113 */
114STATIC void
115xfs_cui_item_pin(
116 struct xfs_log_item *lip)
117{
118}
119
120/*
121 * The unpin operation is the last place an CUI is manipulated in the log. It is
122 * either inserted in the AIL or aborted in the event of a log I/O error. In
123 * either case, the CUI transaction has been successfully committed to make it
124 * this far. Therefore, we expect whoever committed the CUI to either construct
125 * and commit the CUD or drop the CUD's reference in the event of error. Simply
126 * drop the log's CUI reference now that the log is done with it.
127 */
128STATIC void
129xfs_cui_item_unpin(
130 struct xfs_log_item *lip,
131 int remove)
132{
133 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
134
135 xfs_cui_release(cuip);
136}
137
138/*
139 * CUI items have no locking or pushing. However, since CUIs are pulled from
140 * the AIL when their corresponding CUDs are committed to disk, their situation
141 * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller
142 * will eventually flush the log. This should help in getting the CUI out of
143 * the AIL.
144 */
145STATIC uint
146xfs_cui_item_push(
147 struct xfs_log_item *lip,
148 struct list_head *buffer_list)
149{
150 return XFS_ITEM_PINNED;
151}
152
153/*
154 * The CUI has been either committed or aborted if the transaction has been
155 * cancelled. If the transaction was cancelled, an CUD isn't going to be
156 * constructed and thus we free the CUI here directly.
157 */
158STATIC void
159xfs_cui_item_unlock(
160 struct xfs_log_item *lip)
161{
162 if (lip->li_flags & XFS_LI_ABORTED)
163 xfs_cui_release(CUI_ITEM(lip));
164}
165
166/*
167 * The CUI is logged only once and cannot be moved in the log, so simply return
168 * the lsn at which it's been logged.
169 */
170STATIC xfs_lsn_t
171xfs_cui_item_committed(
172 struct xfs_log_item *lip,
173 xfs_lsn_t lsn)
174{
175 return lsn;
176}
177
178/*
179 * The CUI dependency tracking op doesn't do squat. It can't because
180 * it doesn't know where the free extent is coming from. The dependency
181 * tracking has to be handled by the "enclosing" metadata object. For
182 * example, for inodes, the inode is locked throughout the extent freeing
183 * so the dependency should be recorded there.
184 */
185STATIC void
186xfs_cui_item_committing(
187 struct xfs_log_item *lip,
188 xfs_lsn_t lsn)
189{
190}
191
192/*
193 * This is the ops vector shared by all cui log items.
194 */
195static const struct xfs_item_ops xfs_cui_item_ops = {
196 .iop_size = xfs_cui_item_size,
197 .iop_format = xfs_cui_item_format,
198 .iop_pin = xfs_cui_item_pin,
199 .iop_unpin = xfs_cui_item_unpin,
200 .iop_unlock = xfs_cui_item_unlock,
201 .iop_committed = xfs_cui_item_committed,
202 .iop_push = xfs_cui_item_push,
203 .iop_committing = xfs_cui_item_committing,
204};
205
206/*
207 * Allocate and initialize an cui item with the given number of extents.
208 */
209struct xfs_cui_log_item *
210xfs_cui_init(
211 struct xfs_mount *mp,
212 uint nextents)
213
214{
215 struct xfs_cui_log_item *cuip;
216
217 ASSERT(nextents > 0);
218 if (nextents > XFS_CUI_MAX_FAST_EXTENTS)
219 cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents),
220 KM_SLEEP);
221 else
222 cuip = kmem_zone_zalloc(xfs_cui_zone, KM_SLEEP);
223
224 xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops);
225 cuip->cui_format.cui_nextents = nextents;
226 cuip->cui_format.cui_id = (uintptr_t)(void *)cuip;
227 atomic_set(&cuip->cui_next_extent, 0);
228 atomic_set(&cuip->cui_refcount, 2);
229
230 return cuip;
231}
232
233static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip)
234{
235 return container_of(lip, struct xfs_cud_log_item, cud_item);
236}
237
238STATIC void
239xfs_cud_item_size(
240 struct xfs_log_item *lip,
241 int *nvecs,
242 int *nbytes)
243{
244 *nvecs += 1;
245 *nbytes += sizeof(struct xfs_cud_log_format);
246}
247
248/*
249 * This is called to fill in the vector of log iovecs for the
250 * given cud log item. We use only 1 iovec, and we point that
251 * at the cud_log_format structure embedded in the cud item.
252 * It is at this point that we assert that all of the extent
253 * slots in the cud item have been filled.
254 */
255STATIC void
256xfs_cud_item_format(
257 struct xfs_log_item *lip,
258 struct xfs_log_vec *lv)
259{
260 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
261 struct xfs_log_iovec *vecp = NULL;
262
263 cudp->cud_format.cud_type = XFS_LI_CUD;
264 cudp->cud_format.cud_size = 1;
265
266 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format,
267 sizeof(struct xfs_cud_log_format));
268}
269
270/*
271 * Pinning has no meaning for an cud item, so just return.
272 */
273STATIC void
274xfs_cud_item_pin(
275 struct xfs_log_item *lip)
276{
277}
278
279/*
280 * Since pinning has no meaning for an cud item, unpinning does
281 * not either.
282 */
283STATIC void
284xfs_cud_item_unpin(
285 struct xfs_log_item *lip,
286 int remove)
287{
288}
289
290/*
291 * There isn't much you can do to push on an cud item. It is simply stuck
292 * waiting for the log to be flushed to disk.
293 */
294STATIC uint
295xfs_cud_item_push(
296 struct xfs_log_item *lip,
297 struct list_head *buffer_list)
298{
299 return XFS_ITEM_PINNED;
300}
301
302/*
303 * The CUD is either committed or aborted if the transaction is cancelled. If
304 * the transaction is cancelled, drop our reference to the CUI and free the
305 * CUD.
306 */
307STATIC void
308xfs_cud_item_unlock(
309 struct xfs_log_item *lip)
310{
311 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
312
313 if (lip->li_flags & XFS_LI_ABORTED) {
314 xfs_cui_release(cudp->cud_cuip);
315 kmem_zone_free(xfs_cud_zone, cudp);
316 }
317}
318
319/*
320 * When the cud item is committed to disk, all we need to do is delete our
321 * reference to our partner cui item and then free ourselves. Since we're
322 * freeing ourselves we must return -1 to keep the transaction code from
323 * further referencing this item.
324 */
325STATIC xfs_lsn_t
326xfs_cud_item_committed(
327 struct xfs_log_item *lip,
328 xfs_lsn_t lsn)
329{
330 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
331
332 /*
333 * Drop the CUI reference regardless of whether the CUD has been
334 * aborted. Once the CUD transaction is constructed, it is the sole
335 * responsibility of the CUD to release the CUI (even if the CUI is
336 * aborted due to log I/O error).
337 */
338 xfs_cui_release(cudp->cud_cuip);
339 kmem_zone_free(xfs_cud_zone, cudp);
340
341 return (xfs_lsn_t)-1;
342}
343
344/*
345 * The CUD dependency tracking op doesn't do squat. It can't because
346 * it doesn't know where the free extent is coming from. The dependency
347 * tracking has to be handled by the "enclosing" metadata object. For
348 * example, for inodes, the inode is locked throughout the extent freeing
349 * so the dependency should be recorded there.
350 */
351STATIC void
352xfs_cud_item_committing(
353 struct xfs_log_item *lip,
354 xfs_lsn_t lsn)
355{
356}
357
358/*
359 * This is the ops vector shared by all cud log items.
360 */
361static const struct xfs_item_ops xfs_cud_item_ops = {
362 .iop_size = xfs_cud_item_size,
363 .iop_format = xfs_cud_item_format,
364 .iop_pin = xfs_cud_item_pin,
365 .iop_unpin = xfs_cud_item_unpin,
366 .iop_unlock = xfs_cud_item_unlock,
367 .iop_committed = xfs_cud_item_committed,
368 .iop_push = xfs_cud_item_push,
369 .iop_committing = xfs_cud_item_committing,
370};
371
372/*
373 * Allocate and initialize an cud item with the given number of extents.
374 */
375struct xfs_cud_log_item *
376xfs_cud_init(
377 struct xfs_mount *mp,
378 struct xfs_cui_log_item *cuip)
379
380{
381 struct xfs_cud_log_item *cudp;
382
383 cudp = kmem_zone_zalloc(xfs_cud_zone, KM_SLEEP);
384 xfs_log_item_init(mp, &cudp->cud_item, XFS_LI_CUD, &xfs_cud_item_ops);
385 cudp->cud_cuip = cuip;
386 cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id;
387
388 return cudp;
389}
390
391/*
392 * Process a refcount update intent item that was recovered from the log.
393 * We need to update the refcountbt.
394 */
395int
396xfs_cui_recover(
397 struct xfs_mount *mp,
398 struct xfs_cui_log_item *cuip,
399 struct xfs_defer_ops *dfops)
400{
401 int i;
402 int error = 0;
403 unsigned int refc_type;
404 struct xfs_phys_extent *refc;
405 xfs_fsblock_t startblock_fsb;
406 bool op_ok;
407 struct xfs_cud_log_item *cudp;
408 struct xfs_trans *tp;
409 struct xfs_btree_cur *rcur = NULL;
410 enum xfs_refcount_intent_type type;
411 xfs_fsblock_t new_fsb;
412 xfs_extlen_t new_len;
413 struct xfs_bmbt_irec irec;
414 bool requeue_only = false;
415
416 ASSERT(!test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags));
417
418 /*
419 * First check the validity of the extents described by the
420 * CUI. If any are bad, then assume that all are bad and
421 * just toss the CUI.
422 */
423 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
424 refc = &cuip->cui_format.cui_extents[i];
425 startblock_fsb = XFS_BB_TO_FSB(mp,
426 XFS_FSB_TO_DADDR(mp, refc->pe_startblock));
427 switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) {
428 case XFS_REFCOUNT_INCREASE:
429 case XFS_REFCOUNT_DECREASE:
430 case XFS_REFCOUNT_ALLOC_COW:
431 case XFS_REFCOUNT_FREE_COW:
432 op_ok = true;
433 break;
434 default:
435 op_ok = false;
436 break;
437 }
438 if (!op_ok || startblock_fsb == 0 ||
439 refc->pe_len == 0 ||
440 startblock_fsb >= mp->m_sb.sb_dblocks ||
441 refc->pe_len >= mp->m_sb.sb_agblocks ||
442 (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)) {
443 /*
444 * This will pull the CUI from the AIL and
445 * free the memory associated with it.
446 */
447 set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
448 xfs_cui_release(cuip);
449 return -EIO;
450 }
451 }
452
453 /*
454 * Under normal operation, refcount updates are deferred, so we
455 * wouldn't be adding them directly to a transaction. All
456 * refcount updates manage reservation usage internally and
457 * dynamically by deferring work that won't fit in the
458 * transaction. Normally, any work that needs to be deferred
459 * gets attached to the same defer_ops that scheduled the
460 * refcount update. However, we're in log recovery here, so we
461 * we use the passed in defer_ops and to finish up any work that
462 * doesn't fit. We need to reserve enough blocks to handle a
463 * full btree split on either end of the refcount range.
464 */
465 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
466 mp->m_refc_maxlevels * 2, 0, XFS_TRANS_RESERVE, &tp);
467 if (error)
468 return error;
469 cudp = xfs_trans_get_cud(tp, cuip);
470
471 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
472 refc = &cuip->cui_format.cui_extents[i];
473 refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK;
474 switch (refc_type) {
475 case XFS_REFCOUNT_INCREASE:
476 case XFS_REFCOUNT_DECREASE:
477 case XFS_REFCOUNT_ALLOC_COW:
478 case XFS_REFCOUNT_FREE_COW:
479 type = refc_type;
480 break;
481 default:
482 error = -EFSCORRUPTED;
483 goto abort_error;
484 }
485 if (requeue_only) {
486 new_fsb = refc->pe_startblock;
487 new_len = refc->pe_len;
488 } else
489 error = xfs_trans_log_finish_refcount_update(tp, cudp,
490 dfops, type, refc->pe_startblock, refc->pe_len,
491 &new_fsb, &new_len, &rcur);
492 if (error)
493 goto abort_error;
494
495 /* Requeue what we didn't finish. */
496 if (new_len > 0) {
497 irec.br_startblock = new_fsb;
498 irec.br_blockcount = new_len;
499 switch (type) {
500 case XFS_REFCOUNT_INCREASE:
501 error = xfs_refcount_increase_extent(
502 tp->t_mountp, dfops, &irec);
503 break;
504 case XFS_REFCOUNT_DECREASE:
505 error = xfs_refcount_decrease_extent(
506 tp->t_mountp, dfops, &irec);
507 break;
508 case XFS_REFCOUNT_ALLOC_COW:
509 error = xfs_refcount_alloc_cow_extent(
510 tp->t_mountp, dfops,
511 irec.br_startblock,
512 irec.br_blockcount);
513 break;
514 case XFS_REFCOUNT_FREE_COW:
515 error = xfs_refcount_free_cow_extent(
516 tp->t_mountp, dfops,
517 irec.br_startblock,
518 irec.br_blockcount);
519 break;
520 default:
521 ASSERT(0);
522 }
523 if (error)
524 goto abort_error;
525 requeue_only = true;
526 }
527 }
528
529 xfs_refcount_finish_one_cleanup(tp, rcur, error);
530 set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags);
531 error = xfs_trans_commit(tp);
532 return error;
533
534abort_error:
535 xfs_refcount_finish_one_cleanup(tp, rcur, error);
536 xfs_trans_cancel(tp);
537 return error;
538}
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_format.h"
9#include "xfs_log_format.h"
10#include "xfs_trans_resv.h"
11#include "xfs_bit.h"
12#include "xfs_shared.h"
13#include "xfs_mount.h"
14#include "xfs_defer.h"
15#include "xfs_trans.h"
16#include "xfs_trans_priv.h"
17#include "xfs_refcount_item.h"
18#include "xfs_log.h"
19#include "xfs_refcount.h"
20#include "xfs_error.h"
21#include "xfs_log_priv.h"
22#include "xfs_log_recover.h"
23
24kmem_zone_t *xfs_cui_zone;
25kmem_zone_t *xfs_cud_zone;
26
27static const struct xfs_item_ops xfs_cui_item_ops;
28
29static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip)
30{
31 return container_of(lip, struct xfs_cui_log_item, cui_item);
32}
33
34STATIC void
35xfs_cui_item_free(
36 struct xfs_cui_log_item *cuip)
37{
38 if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS)
39 kmem_free(cuip);
40 else
41 kmem_cache_free(xfs_cui_zone, cuip);
42}
43
44/*
45 * Freeing the CUI requires that we remove it from the AIL if it has already
46 * been placed there. However, the CUI may not yet have been placed in the AIL
47 * when called by xfs_cui_release() from CUD processing due to the ordering of
48 * committed vs unpin operations in bulk insert operations. Hence the reference
49 * count to ensure only the last caller frees the CUI.
50 */
51STATIC void
52xfs_cui_release(
53 struct xfs_cui_log_item *cuip)
54{
55 ASSERT(atomic_read(&cuip->cui_refcount) > 0);
56 if (atomic_dec_and_test(&cuip->cui_refcount)) {
57 xfs_trans_ail_delete(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR);
58 xfs_cui_item_free(cuip);
59 }
60}
61
62
63STATIC void
64xfs_cui_item_size(
65 struct xfs_log_item *lip,
66 int *nvecs,
67 int *nbytes)
68{
69 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
70
71 *nvecs += 1;
72 *nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents);
73}
74
75/*
76 * This is called to fill in the vector of log iovecs for the
77 * given cui log item. We use only 1 iovec, and we point that
78 * at the cui_log_format structure embedded in the cui item.
79 * It is at this point that we assert that all of the extent
80 * slots in the cui item have been filled.
81 */
82STATIC void
83xfs_cui_item_format(
84 struct xfs_log_item *lip,
85 struct xfs_log_vec *lv)
86{
87 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
88 struct xfs_log_iovec *vecp = NULL;
89
90 ASSERT(atomic_read(&cuip->cui_next_extent) ==
91 cuip->cui_format.cui_nextents);
92
93 cuip->cui_format.cui_type = XFS_LI_CUI;
94 cuip->cui_format.cui_size = 1;
95
96 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format,
97 xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents));
98}
99
100/*
101 * The unpin operation is the last place an CUI is manipulated in the log. It is
102 * either inserted in the AIL or aborted in the event of a log I/O error. In
103 * either case, the CUI transaction has been successfully committed to make it
104 * this far. Therefore, we expect whoever committed the CUI to either construct
105 * and commit the CUD or drop the CUD's reference in the event of error. Simply
106 * drop the log's CUI reference now that the log is done with it.
107 */
108STATIC void
109xfs_cui_item_unpin(
110 struct xfs_log_item *lip,
111 int remove)
112{
113 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
114
115 xfs_cui_release(cuip);
116}
117
118/*
119 * The CUI has been either committed or aborted if the transaction has been
120 * cancelled. If the transaction was cancelled, an CUD isn't going to be
121 * constructed and thus we free the CUI here directly.
122 */
123STATIC void
124xfs_cui_item_release(
125 struct xfs_log_item *lip)
126{
127 xfs_cui_release(CUI_ITEM(lip));
128}
129
130/*
131 * Allocate and initialize an cui item with the given number of extents.
132 */
133STATIC struct xfs_cui_log_item *
134xfs_cui_init(
135 struct xfs_mount *mp,
136 uint nextents)
137
138{
139 struct xfs_cui_log_item *cuip;
140
141 ASSERT(nextents > 0);
142 if (nextents > XFS_CUI_MAX_FAST_EXTENTS)
143 cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents),
144 0);
145 else
146 cuip = kmem_cache_zalloc(xfs_cui_zone,
147 GFP_KERNEL | __GFP_NOFAIL);
148
149 xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops);
150 cuip->cui_format.cui_nextents = nextents;
151 cuip->cui_format.cui_id = (uintptr_t)(void *)cuip;
152 atomic_set(&cuip->cui_next_extent, 0);
153 atomic_set(&cuip->cui_refcount, 2);
154
155 return cuip;
156}
157
158static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip)
159{
160 return container_of(lip, struct xfs_cud_log_item, cud_item);
161}
162
163STATIC void
164xfs_cud_item_size(
165 struct xfs_log_item *lip,
166 int *nvecs,
167 int *nbytes)
168{
169 *nvecs += 1;
170 *nbytes += sizeof(struct xfs_cud_log_format);
171}
172
173/*
174 * This is called to fill in the vector of log iovecs for the
175 * given cud log item. We use only 1 iovec, and we point that
176 * at the cud_log_format structure embedded in the cud item.
177 * It is at this point that we assert that all of the extent
178 * slots in the cud item have been filled.
179 */
180STATIC void
181xfs_cud_item_format(
182 struct xfs_log_item *lip,
183 struct xfs_log_vec *lv)
184{
185 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
186 struct xfs_log_iovec *vecp = NULL;
187
188 cudp->cud_format.cud_type = XFS_LI_CUD;
189 cudp->cud_format.cud_size = 1;
190
191 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format,
192 sizeof(struct xfs_cud_log_format));
193}
194
195/*
196 * The CUD is either committed or aborted if the transaction is cancelled. If
197 * the transaction is cancelled, drop our reference to the CUI and free the
198 * CUD.
199 */
200STATIC void
201xfs_cud_item_release(
202 struct xfs_log_item *lip)
203{
204 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
205
206 xfs_cui_release(cudp->cud_cuip);
207 kmem_cache_free(xfs_cud_zone, cudp);
208}
209
210static const struct xfs_item_ops xfs_cud_item_ops = {
211 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED,
212 .iop_size = xfs_cud_item_size,
213 .iop_format = xfs_cud_item_format,
214 .iop_release = xfs_cud_item_release,
215};
216
217static struct xfs_cud_log_item *
218xfs_trans_get_cud(
219 struct xfs_trans *tp,
220 struct xfs_cui_log_item *cuip)
221{
222 struct xfs_cud_log_item *cudp;
223
224 cudp = kmem_cache_zalloc(xfs_cud_zone, GFP_KERNEL | __GFP_NOFAIL);
225 xfs_log_item_init(tp->t_mountp, &cudp->cud_item, XFS_LI_CUD,
226 &xfs_cud_item_ops);
227 cudp->cud_cuip = cuip;
228 cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id;
229
230 xfs_trans_add_item(tp, &cudp->cud_item);
231 return cudp;
232}
233
234/*
235 * Finish an refcount update and log it to the CUD. Note that the
236 * transaction is marked dirty regardless of whether the refcount
237 * update succeeds or fails to support the CUI/CUD lifecycle rules.
238 */
239static int
240xfs_trans_log_finish_refcount_update(
241 struct xfs_trans *tp,
242 struct xfs_cud_log_item *cudp,
243 enum xfs_refcount_intent_type type,
244 xfs_fsblock_t startblock,
245 xfs_extlen_t blockcount,
246 xfs_fsblock_t *new_fsb,
247 xfs_extlen_t *new_len,
248 struct xfs_btree_cur **pcur)
249{
250 int error;
251
252 error = xfs_refcount_finish_one(tp, type, startblock,
253 blockcount, new_fsb, new_len, pcur);
254
255 /*
256 * Mark the transaction dirty, even on error. This ensures the
257 * transaction is aborted, which:
258 *
259 * 1.) releases the CUI and frees the CUD
260 * 2.) shuts down the filesystem
261 */
262 tp->t_flags |= XFS_TRANS_DIRTY;
263 set_bit(XFS_LI_DIRTY, &cudp->cud_item.li_flags);
264
265 return error;
266}
267
268/* Sort refcount intents by AG. */
269static int
270xfs_refcount_update_diff_items(
271 void *priv,
272 const struct list_head *a,
273 const struct list_head *b)
274{
275 struct xfs_mount *mp = priv;
276 struct xfs_refcount_intent *ra;
277 struct xfs_refcount_intent *rb;
278
279 ra = container_of(a, struct xfs_refcount_intent, ri_list);
280 rb = container_of(b, struct xfs_refcount_intent, ri_list);
281 return XFS_FSB_TO_AGNO(mp, ra->ri_startblock) -
282 XFS_FSB_TO_AGNO(mp, rb->ri_startblock);
283}
284
285/* Set the phys extent flags for this reverse mapping. */
286static void
287xfs_trans_set_refcount_flags(
288 struct xfs_phys_extent *refc,
289 enum xfs_refcount_intent_type type)
290{
291 refc->pe_flags = 0;
292 switch (type) {
293 case XFS_REFCOUNT_INCREASE:
294 case XFS_REFCOUNT_DECREASE:
295 case XFS_REFCOUNT_ALLOC_COW:
296 case XFS_REFCOUNT_FREE_COW:
297 refc->pe_flags |= type;
298 break;
299 default:
300 ASSERT(0);
301 }
302}
303
304/* Log refcount updates in the intent item. */
305STATIC void
306xfs_refcount_update_log_item(
307 struct xfs_trans *tp,
308 struct xfs_cui_log_item *cuip,
309 struct xfs_refcount_intent *refc)
310{
311 uint next_extent;
312 struct xfs_phys_extent *ext;
313
314 tp->t_flags |= XFS_TRANS_DIRTY;
315 set_bit(XFS_LI_DIRTY, &cuip->cui_item.li_flags);
316
317 /*
318 * atomic_inc_return gives us the value after the increment;
319 * we want to use it as an array index so we need to subtract 1 from
320 * it.
321 */
322 next_extent = atomic_inc_return(&cuip->cui_next_extent) - 1;
323 ASSERT(next_extent < cuip->cui_format.cui_nextents);
324 ext = &cuip->cui_format.cui_extents[next_extent];
325 ext->pe_startblock = refc->ri_startblock;
326 ext->pe_len = refc->ri_blockcount;
327 xfs_trans_set_refcount_flags(ext, refc->ri_type);
328}
329
330static struct xfs_log_item *
331xfs_refcount_update_create_intent(
332 struct xfs_trans *tp,
333 struct list_head *items,
334 unsigned int count,
335 bool sort)
336{
337 struct xfs_mount *mp = tp->t_mountp;
338 struct xfs_cui_log_item *cuip = xfs_cui_init(mp, count);
339 struct xfs_refcount_intent *refc;
340
341 ASSERT(count > 0);
342
343 xfs_trans_add_item(tp, &cuip->cui_item);
344 if (sort)
345 list_sort(mp, items, xfs_refcount_update_diff_items);
346 list_for_each_entry(refc, items, ri_list)
347 xfs_refcount_update_log_item(tp, cuip, refc);
348 return &cuip->cui_item;
349}
350
351/* Get an CUD so we can process all the deferred refcount updates. */
352static struct xfs_log_item *
353xfs_refcount_update_create_done(
354 struct xfs_trans *tp,
355 struct xfs_log_item *intent,
356 unsigned int count)
357{
358 return &xfs_trans_get_cud(tp, CUI_ITEM(intent))->cud_item;
359}
360
361/* Process a deferred refcount update. */
362STATIC int
363xfs_refcount_update_finish_item(
364 struct xfs_trans *tp,
365 struct xfs_log_item *done,
366 struct list_head *item,
367 struct xfs_btree_cur **state)
368{
369 struct xfs_refcount_intent *refc;
370 xfs_fsblock_t new_fsb;
371 xfs_extlen_t new_aglen;
372 int error;
373
374 refc = container_of(item, struct xfs_refcount_intent, ri_list);
375 error = xfs_trans_log_finish_refcount_update(tp, CUD_ITEM(done),
376 refc->ri_type, refc->ri_startblock, refc->ri_blockcount,
377 &new_fsb, &new_aglen, state);
378
379 /* Did we run out of reservation? Requeue what we didn't finish. */
380 if (!error && new_aglen > 0) {
381 ASSERT(refc->ri_type == XFS_REFCOUNT_INCREASE ||
382 refc->ri_type == XFS_REFCOUNT_DECREASE);
383 refc->ri_startblock = new_fsb;
384 refc->ri_blockcount = new_aglen;
385 return -EAGAIN;
386 }
387 kmem_free(refc);
388 return error;
389}
390
391/* Abort all pending CUIs. */
392STATIC void
393xfs_refcount_update_abort_intent(
394 struct xfs_log_item *intent)
395{
396 xfs_cui_release(CUI_ITEM(intent));
397}
398
399/* Cancel a deferred refcount update. */
400STATIC void
401xfs_refcount_update_cancel_item(
402 struct list_head *item)
403{
404 struct xfs_refcount_intent *refc;
405
406 refc = container_of(item, struct xfs_refcount_intent, ri_list);
407 kmem_free(refc);
408}
409
410const struct xfs_defer_op_type xfs_refcount_update_defer_type = {
411 .max_items = XFS_CUI_MAX_FAST_EXTENTS,
412 .create_intent = xfs_refcount_update_create_intent,
413 .abort_intent = xfs_refcount_update_abort_intent,
414 .create_done = xfs_refcount_update_create_done,
415 .finish_item = xfs_refcount_update_finish_item,
416 .finish_cleanup = xfs_refcount_finish_one_cleanup,
417 .cancel_item = xfs_refcount_update_cancel_item,
418};
419
420/* Is this recovered CUI ok? */
421static inline bool
422xfs_cui_validate_phys(
423 struct xfs_mount *mp,
424 struct xfs_phys_extent *refc)
425{
426 if (!xfs_sb_version_hasreflink(&mp->m_sb))
427 return false;
428
429 if (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)
430 return false;
431
432 switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) {
433 case XFS_REFCOUNT_INCREASE:
434 case XFS_REFCOUNT_DECREASE:
435 case XFS_REFCOUNT_ALLOC_COW:
436 case XFS_REFCOUNT_FREE_COW:
437 break;
438 default:
439 return false;
440 }
441
442 return xfs_verify_fsbext(mp, refc->pe_startblock, refc->pe_len);
443}
444
445/*
446 * Process a refcount update intent item that was recovered from the log.
447 * We need to update the refcountbt.
448 */
449STATIC int
450xfs_cui_item_recover(
451 struct xfs_log_item *lip,
452 struct list_head *capture_list)
453{
454 struct xfs_bmbt_irec irec;
455 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
456 struct xfs_phys_extent *refc;
457 struct xfs_cud_log_item *cudp;
458 struct xfs_trans *tp;
459 struct xfs_btree_cur *rcur = NULL;
460 struct xfs_mount *mp = lip->li_mountp;
461 xfs_fsblock_t new_fsb;
462 xfs_extlen_t new_len;
463 unsigned int refc_type;
464 bool requeue_only = false;
465 enum xfs_refcount_intent_type type;
466 int i;
467 int error = 0;
468
469 /*
470 * First check the validity of the extents described by the
471 * CUI. If any are bad, then assume that all are bad and
472 * just toss the CUI.
473 */
474 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
475 if (!xfs_cui_validate_phys(mp,
476 &cuip->cui_format.cui_extents[i])) {
477 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
478 &cuip->cui_format,
479 sizeof(cuip->cui_format));
480 return -EFSCORRUPTED;
481 }
482 }
483
484 /*
485 * Under normal operation, refcount updates are deferred, so we
486 * wouldn't be adding them directly to a transaction. All
487 * refcount updates manage reservation usage internally and
488 * dynamically by deferring work that won't fit in the
489 * transaction. Normally, any work that needs to be deferred
490 * gets attached to the same defer_ops that scheduled the
491 * refcount update. However, we're in log recovery here, so we
492 * use the passed in defer_ops and to finish up any work that
493 * doesn't fit. We need to reserve enough blocks to handle a
494 * full btree split on either end of the refcount range.
495 */
496 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
497 mp->m_refc_maxlevels * 2, 0, XFS_TRANS_RESERVE, &tp);
498 if (error)
499 return error;
500
501 cudp = xfs_trans_get_cud(tp, cuip);
502
503 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
504 refc = &cuip->cui_format.cui_extents[i];
505 refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK;
506 switch (refc_type) {
507 case XFS_REFCOUNT_INCREASE:
508 case XFS_REFCOUNT_DECREASE:
509 case XFS_REFCOUNT_ALLOC_COW:
510 case XFS_REFCOUNT_FREE_COW:
511 type = refc_type;
512 break;
513 default:
514 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
515 error = -EFSCORRUPTED;
516 goto abort_error;
517 }
518 if (requeue_only) {
519 new_fsb = refc->pe_startblock;
520 new_len = refc->pe_len;
521 } else
522 error = xfs_trans_log_finish_refcount_update(tp, cudp,
523 type, refc->pe_startblock, refc->pe_len,
524 &new_fsb, &new_len, &rcur);
525 if (error)
526 goto abort_error;
527
528 /* Requeue what we didn't finish. */
529 if (new_len > 0) {
530 irec.br_startblock = new_fsb;
531 irec.br_blockcount = new_len;
532 switch (type) {
533 case XFS_REFCOUNT_INCREASE:
534 xfs_refcount_increase_extent(tp, &irec);
535 break;
536 case XFS_REFCOUNT_DECREASE:
537 xfs_refcount_decrease_extent(tp, &irec);
538 break;
539 case XFS_REFCOUNT_ALLOC_COW:
540 xfs_refcount_alloc_cow_extent(tp,
541 irec.br_startblock,
542 irec.br_blockcount);
543 break;
544 case XFS_REFCOUNT_FREE_COW:
545 xfs_refcount_free_cow_extent(tp,
546 irec.br_startblock,
547 irec.br_blockcount);
548 break;
549 default:
550 ASSERT(0);
551 }
552 requeue_only = true;
553 }
554 }
555
556 xfs_refcount_finish_one_cleanup(tp, rcur, error);
557 return xfs_defer_ops_capture_and_commit(tp, NULL, capture_list);
558
559abort_error:
560 xfs_refcount_finish_one_cleanup(tp, rcur, error);
561 xfs_trans_cancel(tp);
562 return error;
563}
564
565STATIC bool
566xfs_cui_item_match(
567 struct xfs_log_item *lip,
568 uint64_t intent_id)
569{
570 return CUI_ITEM(lip)->cui_format.cui_id == intent_id;
571}
572
573/* Relog an intent item to push the log tail forward. */
574static struct xfs_log_item *
575xfs_cui_item_relog(
576 struct xfs_log_item *intent,
577 struct xfs_trans *tp)
578{
579 struct xfs_cud_log_item *cudp;
580 struct xfs_cui_log_item *cuip;
581 struct xfs_phys_extent *extp;
582 unsigned int count;
583
584 count = CUI_ITEM(intent)->cui_format.cui_nextents;
585 extp = CUI_ITEM(intent)->cui_format.cui_extents;
586
587 tp->t_flags |= XFS_TRANS_DIRTY;
588 cudp = xfs_trans_get_cud(tp, CUI_ITEM(intent));
589 set_bit(XFS_LI_DIRTY, &cudp->cud_item.li_flags);
590
591 cuip = xfs_cui_init(tp->t_mountp, count);
592 memcpy(cuip->cui_format.cui_extents, extp, count * sizeof(*extp));
593 atomic_set(&cuip->cui_next_extent, count);
594 xfs_trans_add_item(tp, &cuip->cui_item);
595 set_bit(XFS_LI_DIRTY, &cuip->cui_item.li_flags);
596 return &cuip->cui_item;
597}
598
599static const struct xfs_item_ops xfs_cui_item_ops = {
600 .iop_size = xfs_cui_item_size,
601 .iop_format = xfs_cui_item_format,
602 .iop_unpin = xfs_cui_item_unpin,
603 .iop_release = xfs_cui_item_release,
604 .iop_recover = xfs_cui_item_recover,
605 .iop_match = xfs_cui_item_match,
606 .iop_relog = xfs_cui_item_relog,
607};
608
609/*
610 * Copy an CUI format buffer from the given buf, and into the destination
611 * CUI format structure. The CUI/CUD items were designed not to need any
612 * special alignment handling.
613 */
614static int
615xfs_cui_copy_format(
616 struct xfs_log_iovec *buf,
617 struct xfs_cui_log_format *dst_cui_fmt)
618{
619 struct xfs_cui_log_format *src_cui_fmt;
620 uint len;
621
622 src_cui_fmt = buf->i_addr;
623 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
624
625 if (buf->i_len == len) {
626 memcpy(dst_cui_fmt, src_cui_fmt, len);
627 return 0;
628 }
629 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
630 return -EFSCORRUPTED;
631}
632
633/*
634 * This routine is called to create an in-core extent refcount update
635 * item from the cui format structure which was logged on disk.
636 * It allocates an in-core cui, copies the extents from the format
637 * structure into it, and adds the cui to the AIL with the given
638 * LSN.
639 */
640STATIC int
641xlog_recover_cui_commit_pass2(
642 struct xlog *log,
643 struct list_head *buffer_list,
644 struct xlog_recover_item *item,
645 xfs_lsn_t lsn)
646{
647 int error;
648 struct xfs_mount *mp = log->l_mp;
649 struct xfs_cui_log_item *cuip;
650 struct xfs_cui_log_format *cui_formatp;
651
652 cui_formatp = item->ri_buf[0].i_addr;
653
654 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
655 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
656 if (error) {
657 xfs_cui_item_free(cuip);
658 return error;
659 }
660 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
661 /*
662 * Insert the intent into the AIL directly and drop one reference so
663 * that finishing or canceling the work will drop the other.
664 */
665 xfs_trans_ail_insert(log->l_ailp, &cuip->cui_item, lsn);
666 xfs_cui_release(cuip);
667 return 0;
668}
669
670const struct xlog_recover_item_ops xlog_cui_item_ops = {
671 .item_type = XFS_LI_CUI,
672 .commit_pass2 = xlog_recover_cui_commit_pass2,
673};
674
675/*
676 * This routine is called when an CUD format structure is found in a committed
677 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
678 * was still in the log. To do this it searches the AIL for the CUI with an id
679 * equal to that in the CUD format structure. If we find it we drop the CUD
680 * reference, which removes the CUI from the AIL and frees it.
681 */
682STATIC int
683xlog_recover_cud_commit_pass2(
684 struct xlog *log,
685 struct list_head *buffer_list,
686 struct xlog_recover_item *item,
687 xfs_lsn_t lsn)
688{
689 struct xfs_cud_log_format *cud_formatp;
690
691 cud_formatp = item->ri_buf[0].i_addr;
692 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format)) {
693 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
694 return -EFSCORRUPTED;
695 }
696
697 xlog_recover_release_intent(log, XFS_LI_CUI, cud_formatp->cud_cui_id);
698 return 0;
699}
700
701const struct xlog_recover_item_ops xlog_cud_item_ops = {
702 .item_type = XFS_LI_CUD,
703 .commit_pass2 = xlog_recover_cud_commit_pass2,
704};