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" |
b31c2bdc | 26 | #include "xfs_shared.h" |
baf4bcac | 27 | #include "xfs_mount.h" |
f997ee21 | 28 | #include "xfs_defer.h" |
baf4bcac DW |
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" | |
f997ee21 | 34 | #include "xfs_refcount.h" |
baf4bcac DW |
35 | |
36 | ||
37 | kmem_zone_t *xfs_cui_zone; | |
38 | kmem_zone_t *xfs_cud_zone; | |
39 | ||
40 | static 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 | ||
45 | void | |
46 | xfs_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 | ||
0612d116 DC |
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 | */ | |
62 | void | |
63 | xfs_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 | ||
baf4bcac DW |
74 | STATIC void |
75 | xfs_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 | */ | |
93 | STATIC void | |
94 | xfs_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 | */ | |
114 | STATIC void | |
115 | xfs_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 | */ | |
128 | STATIC void | |
129 | xfs_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 | */ | |
145 | STATIC uint | |
146 | xfs_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 | */ | |
158 | STATIC void | |
159 | xfs_cui_item_unlock( | |
160 | struct xfs_log_item *lip) | |
161 | { | |
162 | if (lip->li_flags & XFS_LI_ABORTED) | |
0612d116 | 163 | xfs_cui_release(CUI_ITEM(lip)); |
baf4bcac DW |
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 | */ | |
170 | STATIC xfs_lsn_t | |
171 | xfs_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 | */ | |
185 | STATIC void | |
186 | xfs_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 | */ | |
195 | static 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 | */ | |
209 | struct xfs_cui_log_item * | |
210 | xfs_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 | ||
baf4bcac DW |
233 | static 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 | ||
238 | STATIC void | |
239 | xfs_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 | */ | |
255 | STATIC void | |
256 | xfs_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 | */ | |
273 | STATIC void | |
274 | xfs_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 | */ | |
283 | STATIC void | |
284 | xfs_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 | */ | |
294 | STATIC uint | |
295 | xfs_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 | */ | |
307 | STATIC void | |
308 | xfs_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 | */ | |
325 | STATIC xfs_lsn_t | |
326 | xfs_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 | */ | |
351 | STATIC void | |
352 | xfs_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 | */ | |
361 | static 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 | */ | |
375 | struct xfs_cud_log_item * | |
376 | xfs_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 | } | |
f997ee21 DW |
390 | |
391 | /* | |
392 | * Process a refcount update intent item that was recovered from the log. | |
393 | * We need to update the refcountbt. | |
394 | */ | |
395 | int | |
396 | xfs_cui_recover( | |
397 | struct xfs_mount *mp, | |
50995582 DW |
398 | struct xfs_cui_log_item *cuip, |
399 | struct xfs_defer_ops *dfops) | |
f997ee21 DW |
400 | { |
401 | int i; | |
402 | int error = 0; | |
33ba6129 | 403 | unsigned int refc_type; |
f997ee21 DW |
404 | struct xfs_phys_extent *refc; |
405 | xfs_fsblock_t startblock_fsb; | |
406 | bool op_ok; | |
33ba6129 DW |
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; | |
33ba6129 DW |
411 | xfs_fsblock_t new_fsb; |
412 | xfs_extlen_t new_len; | |
413 | struct xfs_bmbt_irec irec; | |
33ba6129 | 414 | bool requeue_only = false; |
f997ee21 DW |
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 | ||
33ba6129 DW |
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 | |
b31c2bdc DW |
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. | |
33ba6129 | 464 | */ |
b31c2bdc DW |
465 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, |
466 | mp->m_refc_maxlevels * 2, 0, XFS_TRANS_RESERVE, &tp); | |
33ba6129 DW |
467 | if (error) |
468 | return error; | |
469 | cudp = xfs_trans_get_cud(tp, cuip); | |
470 | ||
33ba6129 DW |
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, | |
50995582 | 490 | dfops, type, refc->pe_startblock, refc->pe_len, |
33ba6129 DW |
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( | |
50995582 | 502 | tp->t_mountp, dfops, &irec); |
33ba6129 DW |
503 | break; |
504 | case XFS_REFCOUNT_DECREASE: | |
505 | error = xfs_refcount_decrease_extent( | |
50995582 | 506 | tp->t_mountp, dfops, &irec); |
33ba6129 | 507 | break; |
174edb0e DW |
508 | case XFS_REFCOUNT_ALLOC_COW: |
509 | error = xfs_refcount_alloc_cow_extent( | |
50995582 | 510 | tp->t_mountp, dfops, |
174edb0e DW |
511 | irec.br_startblock, |
512 | irec.br_blockcount); | |
513 | break; | |
514 | case XFS_REFCOUNT_FREE_COW: | |
515 | error = xfs_refcount_free_cow_extent( | |
50995582 | 516 | tp->t_mountp, dfops, |
174edb0e DW |
517 | irec.br_startblock, |
518 | irec.br_blockcount); | |
519 | break; | |
33ba6129 DW |
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); | |
f997ee21 | 530 | set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags); |
33ba6129 DW |
531 | error = xfs_trans_commit(tp); |
532 | return error; | |
533 | ||
534 | abort_error: | |
535 | xfs_refcount_finish_one_cleanup(tp, rcur, error); | |
33ba6129 | 536 | xfs_trans_cancel(tp); |
f997ee21 DW |
537 | return error; |
538 | } |