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#include "xfs_ag.h"
24#include "xfs_btree.h"
25#include "xfs_trace.h"
26
27struct kmem_cache *xfs_cui_cache;
28struct kmem_cache *xfs_cud_cache;
29
30static const struct xfs_item_ops xfs_cui_item_ops;
31
32static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip)
33{
34 return container_of(lip, struct xfs_cui_log_item, cui_item);
35}
36
37STATIC void
38xfs_cui_item_free(
39 struct xfs_cui_log_item *cuip)
40{
41 kvfree(cuip->cui_item.li_lv_shadow);
42 if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS)
43 kfree(cuip);
44 else
45 kmem_cache_free(xfs_cui_cache, cuip);
46}
47
48/*
49 * Freeing the CUI requires that we remove it from the AIL if it has already
50 * been placed there. However, the CUI may not yet have been placed in the AIL
51 * when called by xfs_cui_release() from CUD processing due to the ordering of
52 * committed vs unpin operations in bulk insert operations. Hence the reference
53 * count to ensure only the last caller frees the CUI.
54 */
55STATIC void
56xfs_cui_release(
57 struct xfs_cui_log_item *cuip)
58{
59 ASSERT(atomic_read(&cuip->cui_refcount) > 0);
60 if (!atomic_dec_and_test(&cuip->cui_refcount))
61 return;
62
63 xfs_trans_ail_delete(&cuip->cui_item, 0);
64 xfs_cui_item_free(cuip);
65}
66
67
68STATIC void
69xfs_cui_item_size(
70 struct xfs_log_item *lip,
71 int *nvecs,
72 int *nbytes)
73{
74 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
75
76 *nvecs += 1;
77 *nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents);
78}
79
80/*
81 * This is called to fill in the vector of log iovecs for the
82 * given cui log item. We use only 1 iovec, and we point that
83 * at the cui_log_format structure embedded in the cui item.
84 * It is at this point that we assert that all of the extent
85 * slots in the cui item have been filled.
86 */
87STATIC void
88xfs_cui_item_format(
89 struct xfs_log_item *lip,
90 struct xfs_log_vec *lv)
91{
92 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
93 struct xfs_log_iovec *vecp = NULL;
94
95 ASSERT(atomic_read(&cuip->cui_next_extent) ==
96 cuip->cui_format.cui_nextents);
97
98 cuip->cui_format.cui_type = XFS_LI_CUI;
99 cuip->cui_format.cui_size = 1;
100
101 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format,
102 xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents));
103}
104
105/*
106 * The unpin operation is the last place an CUI is manipulated in the log. It is
107 * either inserted in the AIL or aborted in the event of a log I/O error. In
108 * either case, the CUI transaction has been successfully committed to make it
109 * this far. Therefore, we expect whoever committed the CUI to either construct
110 * and commit the CUD or drop the CUD's reference in the event of error. Simply
111 * drop the log's CUI reference now that the log is done with it.
112 */
113STATIC void
114xfs_cui_item_unpin(
115 struct xfs_log_item *lip,
116 int remove)
117{
118 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
119
120 xfs_cui_release(cuip);
121}
122
123/*
124 * The CUI has been either committed or aborted if the transaction has been
125 * cancelled. If the transaction was cancelled, an CUD isn't going to be
126 * constructed and thus we free the CUI here directly.
127 */
128STATIC void
129xfs_cui_item_release(
130 struct xfs_log_item *lip)
131{
132 xfs_cui_release(CUI_ITEM(lip));
133}
134
135/*
136 * Allocate and initialize an cui item with the given number of extents.
137 */
138STATIC struct xfs_cui_log_item *
139xfs_cui_init(
140 struct xfs_mount *mp,
141 uint nextents)
142
143{
144 struct xfs_cui_log_item *cuip;
145
146 ASSERT(nextents > 0);
147 if (nextents > XFS_CUI_MAX_FAST_EXTENTS)
148 cuip = kzalloc(xfs_cui_log_item_sizeof(nextents),
149 GFP_KERNEL | __GFP_NOFAIL);
150 else
151 cuip = kmem_cache_zalloc(xfs_cui_cache,
152 GFP_KERNEL | __GFP_NOFAIL);
153
154 xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops);
155 cuip->cui_format.cui_nextents = nextents;
156 cuip->cui_format.cui_id = (uintptr_t)(void *)cuip;
157 atomic_set(&cuip->cui_next_extent, 0);
158 atomic_set(&cuip->cui_refcount, 2);
159
160 return cuip;
161}
162
163static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip)
164{
165 return container_of(lip, struct xfs_cud_log_item, cud_item);
166}
167
168STATIC void
169xfs_cud_item_size(
170 struct xfs_log_item *lip,
171 int *nvecs,
172 int *nbytes)
173{
174 *nvecs += 1;
175 *nbytes += sizeof(struct xfs_cud_log_format);
176}
177
178/*
179 * This is called to fill in the vector of log iovecs for the
180 * given cud log item. We use only 1 iovec, and we point that
181 * at the cud_log_format structure embedded in the cud item.
182 * It is at this point that we assert that all of the extent
183 * slots in the cud item have been filled.
184 */
185STATIC void
186xfs_cud_item_format(
187 struct xfs_log_item *lip,
188 struct xfs_log_vec *lv)
189{
190 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
191 struct xfs_log_iovec *vecp = NULL;
192
193 cudp->cud_format.cud_type = XFS_LI_CUD;
194 cudp->cud_format.cud_size = 1;
195
196 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format,
197 sizeof(struct xfs_cud_log_format));
198}
199
200/*
201 * The CUD is either committed or aborted if the transaction is cancelled. If
202 * the transaction is cancelled, drop our reference to the CUI and free the
203 * CUD.
204 */
205STATIC void
206xfs_cud_item_release(
207 struct xfs_log_item *lip)
208{
209 struct xfs_cud_log_item *cudp = CUD_ITEM(lip);
210
211 xfs_cui_release(cudp->cud_cuip);
212 kvfree(cudp->cud_item.li_lv_shadow);
213 kmem_cache_free(xfs_cud_cache, cudp);
214}
215
216static struct xfs_log_item *
217xfs_cud_item_intent(
218 struct xfs_log_item *lip)
219{
220 return &CUD_ITEM(lip)->cud_cuip->cui_item;
221}
222
223static const struct xfs_item_ops xfs_cud_item_ops = {
224 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED |
225 XFS_ITEM_INTENT_DONE,
226 .iop_size = xfs_cud_item_size,
227 .iop_format = xfs_cud_item_format,
228 .iop_release = xfs_cud_item_release,
229 .iop_intent = xfs_cud_item_intent,
230};
231
232static inline struct xfs_refcount_intent *ci_entry(const struct list_head *e)
233{
234 return list_entry(e, struct xfs_refcount_intent, ri_list);
235}
236
237/* Sort refcount intents by AG. */
238static int
239xfs_refcount_update_diff_items(
240 void *priv,
241 const struct list_head *a,
242 const struct list_head *b)
243{
244 struct xfs_refcount_intent *ra = ci_entry(a);
245 struct xfs_refcount_intent *rb = ci_entry(b);
246
247 return ra->ri_group->xg_gno - rb->ri_group->xg_gno;
248}
249
250/* Log refcount updates in the intent item. */
251STATIC void
252xfs_refcount_update_log_item(
253 struct xfs_trans *tp,
254 struct xfs_cui_log_item *cuip,
255 struct xfs_refcount_intent *ri)
256{
257 uint next_extent;
258 struct xfs_phys_extent *pmap;
259
260 /*
261 * atomic_inc_return gives us the value after the increment;
262 * we want to use it as an array index so we need to subtract 1 from
263 * it.
264 */
265 next_extent = atomic_inc_return(&cuip->cui_next_extent) - 1;
266 ASSERT(next_extent < cuip->cui_format.cui_nextents);
267 pmap = &cuip->cui_format.cui_extents[next_extent];
268 pmap->pe_startblock = ri->ri_startblock;
269 pmap->pe_len = ri->ri_blockcount;
270
271 pmap->pe_flags = 0;
272 switch (ri->ri_type) {
273 case XFS_REFCOUNT_INCREASE:
274 case XFS_REFCOUNT_DECREASE:
275 case XFS_REFCOUNT_ALLOC_COW:
276 case XFS_REFCOUNT_FREE_COW:
277 pmap->pe_flags |= ri->ri_type;
278 break;
279 default:
280 ASSERT(0);
281 }
282}
283
284static struct xfs_log_item *
285xfs_refcount_update_create_intent(
286 struct xfs_trans *tp,
287 struct list_head *items,
288 unsigned int count,
289 bool sort)
290{
291 struct xfs_mount *mp = tp->t_mountp;
292 struct xfs_cui_log_item *cuip = xfs_cui_init(mp, count);
293 struct xfs_refcount_intent *ri;
294
295 ASSERT(count > 0);
296
297 if (sort)
298 list_sort(mp, items, xfs_refcount_update_diff_items);
299 list_for_each_entry(ri, items, ri_list)
300 xfs_refcount_update_log_item(tp, cuip, ri);
301 return &cuip->cui_item;
302}
303
304/* Get an CUD so we can process all the deferred refcount updates. */
305static struct xfs_log_item *
306xfs_refcount_update_create_done(
307 struct xfs_trans *tp,
308 struct xfs_log_item *intent,
309 unsigned int count)
310{
311 struct xfs_cui_log_item *cuip = CUI_ITEM(intent);
312 struct xfs_cud_log_item *cudp;
313
314 cudp = kmem_cache_zalloc(xfs_cud_cache, GFP_KERNEL | __GFP_NOFAIL);
315 xfs_log_item_init(tp->t_mountp, &cudp->cud_item, XFS_LI_CUD,
316 &xfs_cud_item_ops);
317 cudp->cud_cuip = cuip;
318 cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id;
319
320 return &cudp->cud_item;
321}
322
323/* Add this deferred CUI to the transaction. */
324void
325xfs_refcount_defer_add(
326 struct xfs_trans *tp,
327 struct xfs_refcount_intent *ri)
328{
329 struct xfs_mount *mp = tp->t_mountp;
330
331 trace_xfs_refcount_defer(mp, ri);
332
333 ri->ri_group = xfs_group_intent_get(mp, ri->ri_startblock, XG_TYPE_AG);
334 xfs_defer_add(tp, &ri->ri_list, &xfs_refcount_update_defer_type);
335}
336
337/* Cancel a deferred refcount update. */
338STATIC void
339xfs_refcount_update_cancel_item(
340 struct list_head *item)
341{
342 struct xfs_refcount_intent *ri = ci_entry(item);
343
344 xfs_group_intent_put(ri->ri_group);
345 kmem_cache_free(xfs_refcount_intent_cache, ri);
346}
347
348/* Process a deferred refcount update. */
349STATIC int
350xfs_refcount_update_finish_item(
351 struct xfs_trans *tp,
352 struct xfs_log_item *done,
353 struct list_head *item,
354 struct xfs_btree_cur **state)
355{
356 struct xfs_refcount_intent *ri = ci_entry(item);
357 int error;
358
359 /* Did we run out of reservation? Requeue what we didn't finish. */
360 error = xfs_refcount_finish_one(tp, ri, state);
361 if (!error && ri->ri_blockcount > 0) {
362 ASSERT(ri->ri_type == XFS_REFCOUNT_INCREASE ||
363 ri->ri_type == XFS_REFCOUNT_DECREASE);
364 return -EAGAIN;
365 }
366
367 xfs_refcount_update_cancel_item(item);
368 return error;
369}
370
371/* Clean up after calling xfs_refcount_finish_one. */
372STATIC void
373xfs_refcount_finish_one_cleanup(
374 struct xfs_trans *tp,
375 struct xfs_btree_cur *rcur,
376 int error)
377{
378 struct xfs_buf *agbp;
379
380 if (rcur == NULL)
381 return;
382 agbp = rcur->bc_ag.agbp;
383 xfs_btree_del_cursor(rcur, error);
384 if (error)
385 xfs_trans_brelse(tp, agbp);
386}
387
388/* Abort all pending CUIs. */
389STATIC void
390xfs_refcount_update_abort_intent(
391 struct xfs_log_item *intent)
392{
393 xfs_cui_release(CUI_ITEM(intent));
394}
395
396/* Is this recovered CUI ok? */
397static inline bool
398xfs_cui_validate_phys(
399 struct xfs_mount *mp,
400 struct xfs_phys_extent *pmap)
401{
402 if (!xfs_has_reflink(mp))
403 return false;
404
405 if (pmap->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)
406 return false;
407
408 switch (pmap->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) {
409 case XFS_REFCOUNT_INCREASE:
410 case XFS_REFCOUNT_DECREASE:
411 case XFS_REFCOUNT_ALLOC_COW:
412 case XFS_REFCOUNT_FREE_COW:
413 break;
414 default:
415 return false;
416 }
417
418 return xfs_verify_fsbext(mp, pmap->pe_startblock, pmap->pe_len);
419}
420
421static inline void
422xfs_cui_recover_work(
423 struct xfs_mount *mp,
424 struct xfs_defer_pending *dfp,
425 struct xfs_phys_extent *pmap)
426{
427 struct xfs_refcount_intent *ri;
428
429 ri = kmem_cache_alloc(xfs_refcount_intent_cache,
430 GFP_KERNEL | __GFP_NOFAIL);
431 ri->ri_type = pmap->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK;
432 ri->ri_startblock = pmap->pe_startblock;
433 ri->ri_blockcount = pmap->pe_len;
434 ri->ri_group = xfs_group_intent_get(mp, pmap->pe_startblock,
435 XG_TYPE_AG);
436
437 xfs_defer_add_item(dfp, &ri->ri_list);
438}
439
440/*
441 * Process a refcount update intent item that was recovered from the log.
442 * We need to update the refcountbt.
443 */
444STATIC int
445xfs_refcount_recover_work(
446 struct xfs_defer_pending *dfp,
447 struct list_head *capture_list)
448{
449 struct xfs_trans_res resv;
450 struct xfs_log_item *lip = dfp->dfp_intent;
451 struct xfs_cui_log_item *cuip = CUI_ITEM(lip);
452 struct xfs_trans *tp;
453 struct xfs_mount *mp = lip->li_log->l_mp;
454 int i;
455 int error = 0;
456
457 /*
458 * First check the validity of the extents described by the
459 * CUI. If any are bad, then assume that all are bad and
460 * just toss the CUI.
461 */
462 for (i = 0; i < cuip->cui_format.cui_nextents; i++) {
463 if (!xfs_cui_validate_phys(mp,
464 &cuip->cui_format.cui_extents[i])) {
465 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
466 &cuip->cui_format,
467 sizeof(cuip->cui_format));
468 return -EFSCORRUPTED;
469 }
470
471 xfs_cui_recover_work(mp, dfp, &cuip->cui_format.cui_extents[i]);
472 }
473
474 /*
475 * Under normal operation, refcount updates are deferred, so we
476 * wouldn't be adding them directly to a transaction. All
477 * refcount updates manage reservation usage internally and
478 * dynamically by deferring work that won't fit in the
479 * transaction. Normally, any work that needs to be deferred
480 * gets attached to the same defer_ops that scheduled the
481 * refcount update. However, we're in log recovery here, so we
482 * use the passed in defer_ops and to finish up any work that
483 * doesn't fit. We need to reserve enough blocks to handle a
484 * full btree split on either end of the refcount range.
485 */
486 resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
487 error = xfs_trans_alloc(mp, &resv, mp->m_refc_maxlevels * 2, 0,
488 XFS_TRANS_RESERVE, &tp);
489 if (error)
490 return error;
491
492 error = xlog_recover_finish_intent(tp, dfp);
493 if (error == -EFSCORRUPTED)
494 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
495 &cuip->cui_format,
496 sizeof(cuip->cui_format));
497 if (error)
498 goto abort_error;
499
500 return xfs_defer_ops_capture_and_commit(tp, capture_list);
501
502abort_error:
503 xfs_trans_cancel(tp);
504 return error;
505}
506
507/* Relog an intent item to push the log tail forward. */
508static struct xfs_log_item *
509xfs_refcount_relog_intent(
510 struct xfs_trans *tp,
511 struct xfs_log_item *intent,
512 struct xfs_log_item *done_item)
513{
514 struct xfs_cui_log_item *cuip;
515 struct xfs_phys_extent *pmap;
516 unsigned int count;
517
518 count = CUI_ITEM(intent)->cui_format.cui_nextents;
519 pmap = CUI_ITEM(intent)->cui_format.cui_extents;
520
521 cuip = xfs_cui_init(tp->t_mountp, count);
522 memcpy(cuip->cui_format.cui_extents, pmap, count * sizeof(*pmap));
523 atomic_set(&cuip->cui_next_extent, count);
524
525 return &cuip->cui_item;
526}
527
528const struct xfs_defer_op_type xfs_refcount_update_defer_type = {
529 .name = "refcount",
530 .max_items = XFS_CUI_MAX_FAST_EXTENTS,
531 .create_intent = xfs_refcount_update_create_intent,
532 .abort_intent = xfs_refcount_update_abort_intent,
533 .create_done = xfs_refcount_update_create_done,
534 .finish_item = xfs_refcount_update_finish_item,
535 .finish_cleanup = xfs_refcount_finish_one_cleanup,
536 .cancel_item = xfs_refcount_update_cancel_item,
537 .recover_work = xfs_refcount_recover_work,
538 .relog_intent = xfs_refcount_relog_intent,
539};
540
541STATIC bool
542xfs_cui_item_match(
543 struct xfs_log_item *lip,
544 uint64_t intent_id)
545{
546 return CUI_ITEM(lip)->cui_format.cui_id == intent_id;
547}
548
549static const struct xfs_item_ops xfs_cui_item_ops = {
550 .flags = XFS_ITEM_INTENT,
551 .iop_size = xfs_cui_item_size,
552 .iop_format = xfs_cui_item_format,
553 .iop_unpin = xfs_cui_item_unpin,
554 .iop_release = xfs_cui_item_release,
555 .iop_match = xfs_cui_item_match,
556};
557
558static inline void
559xfs_cui_copy_format(
560 struct xfs_cui_log_format *dst,
561 const struct xfs_cui_log_format *src)
562{
563 unsigned int i;
564
565 memcpy(dst, src, offsetof(struct xfs_cui_log_format, cui_extents));
566
567 for (i = 0; i < src->cui_nextents; i++)
568 memcpy(&dst->cui_extents[i], &src->cui_extents[i],
569 sizeof(struct xfs_phys_extent));
570}
571
572/*
573 * This routine is called to create an in-core extent refcount update
574 * item from the cui format structure which was logged on disk.
575 * It allocates an in-core cui, copies the extents from the format
576 * structure into it, and adds the cui to the AIL with the given
577 * LSN.
578 */
579STATIC int
580xlog_recover_cui_commit_pass2(
581 struct xlog *log,
582 struct list_head *buffer_list,
583 struct xlog_recover_item *item,
584 xfs_lsn_t lsn)
585{
586 struct xfs_mount *mp = log->l_mp;
587 struct xfs_cui_log_item *cuip;
588 struct xfs_cui_log_format *cui_formatp;
589 size_t len;
590
591 cui_formatp = item->ri_buf[0].i_addr;
592
593 if (item->ri_buf[0].i_len < xfs_cui_log_format_sizeof(0)) {
594 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
595 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
596 return -EFSCORRUPTED;
597 }
598
599 len = xfs_cui_log_format_sizeof(cui_formatp->cui_nextents);
600 if (item->ri_buf[0].i_len != len) {
601 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
602 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
603 return -EFSCORRUPTED;
604 }
605
606 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
607 xfs_cui_copy_format(&cuip->cui_format, cui_formatp);
608 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
609
610 xlog_recover_intent_item(log, &cuip->cui_item, lsn,
611 &xfs_refcount_update_defer_type);
612 return 0;
613}
614
615const struct xlog_recover_item_ops xlog_cui_item_ops = {
616 .item_type = XFS_LI_CUI,
617 .commit_pass2 = xlog_recover_cui_commit_pass2,
618};
619
620/*
621 * This routine is called when an CUD format structure is found in a committed
622 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
623 * was still in the log. To do this it searches the AIL for the CUI with an id
624 * equal to that in the CUD format structure. If we find it we drop the CUD
625 * reference, which removes the CUI from the AIL and frees it.
626 */
627STATIC int
628xlog_recover_cud_commit_pass2(
629 struct xlog *log,
630 struct list_head *buffer_list,
631 struct xlog_recover_item *item,
632 xfs_lsn_t lsn)
633{
634 struct xfs_cud_log_format *cud_formatp;
635
636 cud_formatp = item->ri_buf[0].i_addr;
637 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format)) {
638 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
639 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
640 return -EFSCORRUPTED;
641 }
642
643 xlog_recover_release_intent(log, XFS_LI_CUI, cud_formatp->cud_cui_id);
644 return 0;
645}
646
647const struct xlog_recover_item_ops xlog_cud_item_ops = {
648 .item_type = XFS_LI_CUD,
649 .commit_pass2 = xlog_recover_cud_commit_pass2,
650};