]>
Commit | Line | Data |
---|---|---|
baf4bcac DW |
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" | |
f997ee21 | 25 | #include "xfs_bit.h" |
baf4bcac | 26 | #include "xfs_mount.h" |
f997ee21 | 27 | #include "xfs_defer.h" |
baf4bcac DW |
28 | #include "xfs_trans.h" |
29 | #include "xfs_trans_priv.h" | |
30 | #include "xfs_buf_item.h" | |
31 | #include "xfs_refcount_item.h" | |
32 | #include "xfs_log.h" | |
f997ee21 | 33 | #include "xfs_refcount.h" |
baf4bcac DW |
34 | |
35 | ||
36 | kmem_zone_t *xfs_cui_zone; | |
37 | kmem_zone_t *xfs_cud_zone; | |
38 | ||
39 | static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip) | |
40 | { | |
41 | return container_of(lip, struct xfs_cui_log_item, cui_item); | |
42 | } | |
43 | ||
44 | void | |
45 | xfs_cui_item_free( | |
46 | struct xfs_cui_log_item *cuip) | |
47 | { | |
48 | if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS) | |
49 | kmem_free(cuip); | |
50 | else | |
51 | kmem_zone_free(xfs_cui_zone, cuip); | |
52 | } | |
53 | ||
54 | STATIC void | |
55 | xfs_cui_item_size( | |
56 | struct xfs_log_item *lip, | |
57 | int *nvecs, | |
58 | int *nbytes) | |
59 | { | |
60 | struct xfs_cui_log_item *cuip = CUI_ITEM(lip); | |
61 | ||
62 | *nvecs += 1; | |
63 | *nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents); | |
64 | } | |
65 | ||
66 | /* | |
67 | * This is called to fill in the vector of log iovecs for the | |
68 | * given cui log item. We use only 1 iovec, and we point that | |
69 | * at the cui_log_format structure embedded in the cui item. | |
70 | * It is at this point that we assert that all of the extent | |
71 | * slots in the cui item have been filled. | |
72 | */ | |
73 | STATIC void | |
74 | xfs_cui_item_format( | |
75 | struct xfs_log_item *lip, | |
76 | struct xfs_log_vec *lv) | |
77 | { | |
78 | struct xfs_cui_log_item *cuip = CUI_ITEM(lip); | |
79 | struct xfs_log_iovec *vecp = NULL; | |
80 | ||
81 | ASSERT(atomic_read(&cuip->cui_next_extent) == | |
82 | cuip->cui_format.cui_nextents); | |
83 | ||
84 | cuip->cui_format.cui_type = XFS_LI_CUI; | |
85 | cuip->cui_format.cui_size = 1; | |
86 | ||
87 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format, | |
88 | xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents)); | |
89 | } | |
90 | ||
91 | /* | |
92 | * Pinning has no meaning for an cui item, so just return. | |
93 | */ | |
94 | STATIC void | |
95 | xfs_cui_item_pin( | |
96 | struct xfs_log_item *lip) | |
97 | { | |
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 | */ | |
108 | STATIC void | |
109 | xfs_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 | * CUI items have no locking or pushing. However, since CUIs are pulled from | |
120 | * the AIL when their corresponding CUDs are committed to disk, their situation | |
121 | * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller | |
122 | * will eventually flush the log. This should help in getting the CUI out of | |
123 | * the AIL. | |
124 | */ | |
125 | STATIC uint | |
126 | xfs_cui_item_push( | |
127 | struct xfs_log_item *lip, | |
128 | struct list_head *buffer_list) | |
129 | { | |
130 | return XFS_ITEM_PINNED; | |
131 | } | |
132 | ||
133 | /* | |
134 | * The CUI has been either committed or aborted if the transaction has been | |
135 | * cancelled. If the transaction was cancelled, an CUD isn't going to be | |
136 | * constructed and thus we free the CUI here directly. | |
137 | */ | |
138 | STATIC void | |
139 | xfs_cui_item_unlock( | |
140 | struct xfs_log_item *lip) | |
141 | { | |
142 | if (lip->li_flags & XFS_LI_ABORTED) | |
143 | xfs_cui_item_free(CUI_ITEM(lip)); | |
144 | } | |
145 | ||
146 | /* | |
147 | * The CUI is logged only once and cannot be moved in the log, so simply return | |
148 | * the lsn at which it's been logged. | |
149 | */ | |
150 | STATIC xfs_lsn_t | |
151 | xfs_cui_item_committed( | |
152 | struct xfs_log_item *lip, | |
153 | xfs_lsn_t lsn) | |
154 | { | |
155 | return lsn; | |
156 | } | |
157 | ||
158 | /* | |
159 | * The CUI dependency tracking op doesn't do squat. It can't because | |
160 | * it doesn't know where the free extent is coming from. The dependency | |
161 | * tracking has to be handled by the "enclosing" metadata object. For | |
162 | * example, for inodes, the inode is locked throughout the extent freeing | |
163 | * so the dependency should be recorded there. | |
164 | */ | |
165 | STATIC void | |
166 | xfs_cui_item_committing( | |
167 | struct xfs_log_item *lip, | |
168 | xfs_lsn_t lsn) | |
169 | { | |
170 | } | |
171 | ||
172 | /* | |
173 | * This is the ops vector shared by all cui log items. | |
174 | */ | |
175 | static const struct xfs_item_ops xfs_cui_item_ops = { | |
176 | .iop_size = xfs_cui_item_size, | |
177 | .iop_format = xfs_cui_item_format, | |
178 | .iop_pin = xfs_cui_item_pin, | |
179 | .iop_unpin = xfs_cui_item_unpin, | |
180 | .iop_unlock = xfs_cui_item_unlock, | |
181 | .iop_committed = xfs_cui_item_committed, | |
182 | .iop_push = xfs_cui_item_push, | |
183 | .iop_committing = xfs_cui_item_committing, | |
184 | }; | |
185 | ||
186 | /* | |
187 | * Allocate and initialize an cui item with the given number of extents. | |
188 | */ | |
189 | struct xfs_cui_log_item * | |
190 | xfs_cui_init( | |
191 | struct xfs_mount *mp, | |
192 | uint nextents) | |
193 | ||
194 | { | |
195 | struct xfs_cui_log_item *cuip; | |
196 | ||
197 | ASSERT(nextents > 0); | |
198 | if (nextents > XFS_CUI_MAX_FAST_EXTENTS) | |
199 | cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents), | |
200 | KM_SLEEP); | |
201 | else | |
202 | cuip = kmem_zone_zalloc(xfs_cui_zone, KM_SLEEP); | |
203 | ||
204 | xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops); | |
205 | cuip->cui_format.cui_nextents = nextents; | |
206 | cuip->cui_format.cui_id = (uintptr_t)(void *)cuip; | |
207 | atomic_set(&cuip->cui_next_extent, 0); | |
208 | atomic_set(&cuip->cui_refcount, 2); | |
209 | ||
210 | return cuip; | |
211 | } | |
212 | ||
213 | /* | |
214 | * Freeing the CUI requires that we remove it from the AIL if it has already | |
215 | * been placed there. However, the CUI may not yet have been placed in the AIL | |
216 | * when called by xfs_cui_release() from CUD processing due to the ordering of | |
217 | * committed vs unpin operations in bulk insert operations. Hence the reference | |
218 | * count to ensure only the last caller frees the CUI. | |
219 | */ | |
220 | void | |
221 | xfs_cui_release( | |
222 | struct xfs_cui_log_item *cuip) | |
223 | { | |
c4cf1acd | 224 | ASSERT(atomic_read(&cuip->cui_refcount) > 0); |
baf4bcac DW |
225 | if (atomic_dec_and_test(&cuip->cui_refcount)) { |
226 | xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR); | |
227 | xfs_cui_item_free(cuip); | |
228 | } | |
229 | } | |
230 | ||
231 | static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip) | |
232 | { | |
233 | return container_of(lip, struct xfs_cud_log_item, cud_item); | |
234 | } | |
235 | ||
236 | STATIC void | |
237 | xfs_cud_item_size( | |
238 | struct xfs_log_item *lip, | |
239 | int *nvecs, | |
240 | int *nbytes) | |
241 | { | |
242 | *nvecs += 1; | |
243 | *nbytes += sizeof(struct xfs_cud_log_format); | |
244 | } | |
245 | ||
246 | /* | |
247 | * This is called to fill in the vector of log iovecs for the | |
248 | * given cud log item. We use only 1 iovec, and we point that | |
249 | * at the cud_log_format structure embedded in the cud item. | |
250 | * It is at this point that we assert that all of the extent | |
251 | * slots in the cud item have been filled. | |
252 | */ | |
253 | STATIC void | |
254 | xfs_cud_item_format( | |
255 | struct xfs_log_item *lip, | |
256 | struct xfs_log_vec *lv) | |
257 | { | |
258 | struct xfs_cud_log_item *cudp = CUD_ITEM(lip); | |
259 | struct xfs_log_iovec *vecp = NULL; | |
260 | ||
261 | cudp->cud_format.cud_type = XFS_LI_CUD; | |
262 | cudp->cud_format.cud_size = 1; | |
263 | ||
264 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format, | |
265 | sizeof(struct xfs_cud_log_format)); | |
266 | } | |
267 | ||
268 | /* | |
269 | * Pinning has no meaning for an cud item, so just return. | |
270 | */ | |
271 | STATIC void | |
272 | xfs_cud_item_pin( | |
273 | struct xfs_log_item *lip) | |
274 | { | |
275 | } | |
276 | ||
277 | /* | |
278 | * Since pinning has no meaning for an cud item, unpinning does | |
279 | * not either. | |
280 | */ | |
281 | STATIC void | |
282 | xfs_cud_item_unpin( | |
283 | struct xfs_log_item *lip, | |
284 | int remove) | |
285 | { | |
286 | } | |
287 | ||
288 | /* | |
289 | * There isn't much you can do to push on an cud item. It is simply stuck | |
290 | * waiting for the log to be flushed to disk. | |
291 | */ | |
292 | STATIC uint | |
293 | xfs_cud_item_push( | |
294 | struct xfs_log_item *lip, | |
295 | struct list_head *buffer_list) | |
296 | { | |
297 | return XFS_ITEM_PINNED; | |
298 | } | |
299 | ||
300 | /* | |
301 | * The CUD is either committed or aborted if the transaction is cancelled. If | |
302 | * the transaction is cancelled, drop our reference to the CUI and free the | |
303 | * CUD. | |
304 | */ | |
305 | STATIC void | |
306 | xfs_cud_item_unlock( | |
307 | struct xfs_log_item *lip) | |
308 | { | |
309 | struct xfs_cud_log_item *cudp = CUD_ITEM(lip); | |
310 | ||
311 | if (lip->li_flags & XFS_LI_ABORTED) { | |
312 | xfs_cui_release(cudp->cud_cuip); | |
313 | kmem_zone_free(xfs_cud_zone, cudp); | |
314 | } | |
315 | } | |
316 | ||
317 | /* | |
318 | * When the cud item is committed to disk, all we need to do is delete our | |
319 | * reference to our partner cui item and then free ourselves. Since we're | |
320 | * freeing ourselves we must return -1 to keep the transaction code from | |
321 | * further referencing this item. | |
322 | */ | |
323 | STATIC xfs_lsn_t | |
324 | xfs_cud_item_committed( | |
325 | struct xfs_log_item *lip, | |
326 | xfs_lsn_t lsn) | |
327 | { | |
328 | struct xfs_cud_log_item *cudp = CUD_ITEM(lip); | |
329 | ||
330 | /* | |
331 | * Drop the CUI reference regardless of whether the CUD has been | |
332 | * aborted. Once the CUD transaction is constructed, it is the sole | |
333 | * responsibility of the CUD to release the CUI (even if the CUI is | |
334 | * aborted due to log I/O error). | |
335 | */ | |
336 | xfs_cui_release(cudp->cud_cuip); | |
337 | kmem_zone_free(xfs_cud_zone, cudp); | |
338 | ||
339 | return (xfs_lsn_t)-1; | |
340 | } | |
341 | ||
342 | /* | |
343 | * The CUD dependency tracking op doesn't do squat. It can't because | |
344 | * it doesn't know where the free extent is coming from. The dependency | |
345 | * tracking has to be handled by the "enclosing" metadata object. For | |
346 | * example, for inodes, the inode is locked throughout the extent freeing | |
347 | * so the dependency should be recorded there. | |
348 | */ | |
349 | STATIC void | |
350 | xfs_cud_item_committing( | |
351 | struct xfs_log_item *lip, | |
352 | xfs_lsn_t lsn) | |
353 | { | |
354 | } | |
355 | ||
356 | /* | |
357 | * This is the ops vector shared by all cud log items. | |
358 | */ | |
359 | static const struct xfs_item_ops xfs_cud_item_ops = { | |
360 | .iop_size = xfs_cud_item_size, | |
361 | .iop_format = xfs_cud_item_format, | |
362 | .iop_pin = xfs_cud_item_pin, | |
363 | .iop_unpin = xfs_cud_item_unpin, | |
364 | .iop_unlock = xfs_cud_item_unlock, | |
365 | .iop_committed = xfs_cud_item_committed, | |
366 | .iop_push = xfs_cud_item_push, | |
367 | .iop_committing = xfs_cud_item_committing, | |
368 | }; | |
369 | ||
370 | /* | |
371 | * Allocate and initialize an cud item with the given number of extents. | |
372 | */ | |
373 | struct xfs_cud_log_item * | |
374 | xfs_cud_init( | |
375 | struct xfs_mount *mp, | |
376 | struct xfs_cui_log_item *cuip) | |
377 | ||
378 | { | |
379 | struct xfs_cud_log_item *cudp; | |
380 | ||
381 | cudp = kmem_zone_zalloc(xfs_cud_zone, KM_SLEEP); | |
382 | xfs_log_item_init(mp, &cudp->cud_item, XFS_LI_CUD, &xfs_cud_item_ops); | |
383 | cudp->cud_cuip = cuip; | |
384 | cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id; | |
385 | ||
386 | return cudp; | |
387 | } | |
f997ee21 DW |
388 | |
389 | /* | |
390 | * Process a refcount update intent item that was recovered from the log. | |
391 | * We need to update the refcountbt. | |
392 | */ | |
393 | int | |
394 | xfs_cui_recover( | |
395 | struct xfs_mount *mp, | |
396 | struct xfs_cui_log_item *cuip) | |
397 | { | |
398 | int i; | |
399 | int error = 0; | |
33ba6129 | 400 | unsigned int refc_type; |
f997ee21 DW |
401 | struct xfs_phys_extent *refc; |
402 | xfs_fsblock_t startblock_fsb; | |
403 | bool op_ok; | |
33ba6129 DW |
404 | struct xfs_cud_log_item *cudp; |
405 | struct xfs_trans *tp; | |
406 | struct xfs_btree_cur *rcur = NULL; | |
407 | enum xfs_refcount_intent_type type; | |
408 | xfs_fsblock_t firstfsb; | |
409 | xfs_fsblock_t new_fsb; | |
410 | xfs_extlen_t new_len; | |
411 | struct xfs_bmbt_irec irec; | |
412 | struct xfs_defer_ops dfops; | |
413 | bool requeue_only = false; | |
f997ee21 DW |
414 | |
415 | ASSERT(!test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags)); | |
416 | ||
417 | /* | |
418 | * First check the validity of the extents described by the | |
419 | * CUI. If any are bad, then assume that all are bad and | |
420 | * just toss the CUI. | |
421 | */ | |
422 | for (i = 0; i < cuip->cui_format.cui_nextents; i++) { | |
423 | refc = &cuip->cui_format.cui_extents[i]; | |
424 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
425 | XFS_FSB_TO_DADDR(mp, refc->pe_startblock)); | |
426 | switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) { | |
427 | case XFS_REFCOUNT_INCREASE: | |
428 | case XFS_REFCOUNT_DECREASE: | |
429 | case XFS_REFCOUNT_ALLOC_COW: | |
430 | case XFS_REFCOUNT_FREE_COW: | |
431 | op_ok = true; | |
432 | break; | |
433 | default: | |
434 | op_ok = false; | |
435 | break; | |
436 | } | |
437 | if (!op_ok || startblock_fsb == 0 || | |
438 | refc->pe_len == 0 || | |
439 | startblock_fsb >= mp->m_sb.sb_dblocks || | |
440 | refc->pe_len >= mp->m_sb.sb_agblocks || | |
441 | (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)) { | |
442 | /* | |
443 | * This will pull the CUI from the AIL and | |
444 | * free the memory associated with it. | |
445 | */ | |
446 | set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags); | |
447 | xfs_cui_release(cuip); | |
448 | return -EIO; | |
449 | } | |
450 | } | |
451 | ||
33ba6129 DW |
452 | /* |
453 | * Under normal operation, refcount updates are deferred, so we | |
454 | * wouldn't be adding them directly to a transaction. All | |
455 | * refcount updates manage reservation usage internally and | |
456 | * dynamically by deferring work that won't fit in the | |
457 | * transaction. Normally, any work that needs to be deferred | |
458 | * gets attached to the same defer_ops that scheduled the | |
459 | * refcount update. However, we're in log recovery here, so we | |
460 | * we create our own defer_ops and use that to finish up any | |
461 | * work that doesn't fit. | |
462 | */ | |
463 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); | |
464 | if (error) | |
465 | return error; | |
466 | cudp = xfs_trans_get_cud(tp, cuip); | |
467 | ||
468 | xfs_defer_init(&dfops, &firstfsb); | |
469 | for (i = 0; i < cuip->cui_format.cui_nextents; i++) { | |
470 | refc = &cuip->cui_format.cui_extents[i]; | |
471 | refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK; | |
472 | switch (refc_type) { | |
473 | case XFS_REFCOUNT_INCREASE: | |
474 | case XFS_REFCOUNT_DECREASE: | |
475 | case XFS_REFCOUNT_ALLOC_COW: | |
476 | case XFS_REFCOUNT_FREE_COW: | |
477 | type = refc_type; | |
478 | break; | |
479 | default: | |
480 | error = -EFSCORRUPTED; | |
481 | goto abort_error; | |
482 | } | |
483 | if (requeue_only) { | |
484 | new_fsb = refc->pe_startblock; | |
485 | new_len = refc->pe_len; | |
486 | } else | |
487 | error = xfs_trans_log_finish_refcount_update(tp, cudp, | |
488 | &dfops, type, refc->pe_startblock, refc->pe_len, | |
489 | &new_fsb, &new_len, &rcur); | |
490 | if (error) | |
491 | goto abort_error; | |
492 | ||
493 | /* Requeue what we didn't finish. */ | |
494 | if (new_len > 0) { | |
495 | irec.br_startblock = new_fsb; | |
496 | irec.br_blockcount = new_len; | |
497 | switch (type) { | |
498 | case XFS_REFCOUNT_INCREASE: | |
499 | error = xfs_refcount_increase_extent( | |
500 | tp->t_mountp, &dfops, &irec); | |
501 | break; | |
502 | case XFS_REFCOUNT_DECREASE: | |
503 | error = xfs_refcount_decrease_extent( | |
504 | tp->t_mountp, &dfops, &irec); | |
505 | break; | |
174edb0e DW |
506 | case XFS_REFCOUNT_ALLOC_COW: |
507 | error = xfs_refcount_alloc_cow_extent( | |
508 | tp->t_mountp, &dfops, | |
509 | irec.br_startblock, | |
510 | irec.br_blockcount); | |
511 | break; | |
512 | case XFS_REFCOUNT_FREE_COW: | |
513 | error = xfs_refcount_free_cow_extent( | |
514 | tp->t_mountp, &dfops, | |
515 | irec.br_startblock, | |
516 | irec.br_blockcount); | |
517 | break; | |
33ba6129 DW |
518 | default: |
519 | ASSERT(0); | |
520 | } | |
521 | if (error) | |
522 | goto abort_error; | |
523 | requeue_only = true; | |
524 | } | |
525 | } | |
526 | ||
527 | xfs_refcount_finish_one_cleanup(tp, rcur, error); | |
528 | error = xfs_defer_finish(&tp, &dfops, NULL); | |
529 | if (error) | |
7a21272b | 530 | goto abort_defer; |
f997ee21 | 531 | set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags); |
33ba6129 DW |
532 | error = xfs_trans_commit(tp); |
533 | return error; | |
534 | ||
535 | abort_error: | |
536 | xfs_refcount_finish_one_cleanup(tp, rcur, error); | |
7a21272b | 537 | abort_defer: |
33ba6129 DW |
538 | xfs_defer_cancel(&dfops); |
539 | xfs_trans_cancel(tp); | |
f997ee21 DW |
540 | return error; |
541 | } |