]>
Commit | Line | Data |
---|---|---|
5880f2d7 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" | |
9e88b5d8 | 25 | #include "xfs_bit.h" |
5880f2d7 DW |
26 | #include "xfs_mount.h" |
27 | #include "xfs_trans.h" | |
28 | #include "xfs_trans_priv.h" | |
29 | #include "xfs_buf_item.h" | |
30 | #include "xfs_rmap_item.h" | |
31 | #include "xfs_log.h" | |
32 | ||
33 | ||
34 | kmem_zone_t *xfs_rui_zone; | |
35 | kmem_zone_t *xfs_rud_zone; | |
36 | ||
37 | static inline struct xfs_rui_log_item *RUI_ITEM(struct xfs_log_item *lip) | |
38 | { | |
39 | return container_of(lip, struct xfs_rui_log_item, rui_item); | |
40 | } | |
41 | ||
42 | void | |
43 | xfs_rui_item_free( | |
44 | struct xfs_rui_log_item *ruip) | |
45 | { | |
46 | if (ruip->rui_format.rui_nextents > XFS_RUI_MAX_FAST_EXTENTS) | |
47 | kmem_free(ruip); | |
48 | else | |
49 | kmem_zone_free(xfs_rui_zone, ruip); | |
50 | } | |
51 | ||
52 | /* | |
53 | * This returns the number of iovecs needed to log the given rui item. | |
54 | * We only need 1 iovec for an rui item. It just logs the rui_log_format | |
55 | * structure. | |
56 | */ | |
57 | static inline int | |
58 | xfs_rui_item_sizeof( | |
59 | struct xfs_rui_log_item *ruip) | |
60 | { | |
61 | return sizeof(struct xfs_rui_log_format) + | |
62 | (ruip->rui_format.rui_nextents - 1) * | |
63 | sizeof(struct xfs_map_extent); | |
64 | } | |
65 | ||
66 | STATIC void | |
67 | xfs_rui_item_size( | |
68 | struct xfs_log_item *lip, | |
69 | int *nvecs, | |
70 | int *nbytes) | |
71 | { | |
72 | *nvecs += 1; | |
73 | *nbytes += xfs_rui_item_sizeof(RUI_ITEM(lip)); | |
74 | } | |
75 | ||
76 | /* | |
77 | * This is called to fill in the vector of log iovecs for the | |
78 | * given rui log item. We use only 1 iovec, and we point that | |
79 | * at the rui_log_format structure embedded in the rui item. | |
80 | * It is at this point that we assert that all of the extent | |
81 | * slots in the rui item have been filled. | |
82 | */ | |
83 | STATIC void | |
84 | xfs_rui_item_format( | |
85 | struct xfs_log_item *lip, | |
86 | struct xfs_log_vec *lv) | |
87 | { | |
88 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); | |
89 | struct xfs_log_iovec *vecp = NULL; | |
90 | ||
91 | ASSERT(atomic_read(&ruip->rui_next_extent) == | |
92 | ruip->rui_format.rui_nextents); | |
93 | ||
94 | ruip->rui_format.rui_type = XFS_LI_RUI; | |
95 | ruip->rui_format.rui_size = 1; | |
96 | ||
97 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUI_FORMAT, &ruip->rui_format, | |
98 | xfs_rui_item_sizeof(ruip)); | |
99 | } | |
100 | ||
101 | /* | |
102 | * Pinning has no meaning for an rui item, so just return. | |
103 | */ | |
104 | STATIC void | |
105 | xfs_rui_item_pin( | |
106 | struct xfs_log_item *lip) | |
107 | { | |
108 | } | |
109 | ||
110 | /* | |
111 | * The unpin operation is the last place an RUI is manipulated in the log. It is | |
112 | * either inserted in the AIL or aborted in the event of a log I/O error. In | |
113 | * either case, the RUI transaction has been successfully committed to make it | |
114 | * this far. Therefore, we expect whoever committed the RUI to either construct | |
115 | * and commit the RUD or drop the RUD's reference in the event of error. Simply | |
116 | * drop the log's RUI reference now that the log is done with it. | |
117 | */ | |
118 | STATIC void | |
119 | xfs_rui_item_unpin( | |
120 | struct xfs_log_item *lip, | |
121 | int remove) | |
122 | { | |
123 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); | |
124 | ||
125 | xfs_rui_release(ruip); | |
126 | } | |
127 | ||
128 | /* | |
129 | * RUI items have no locking or pushing. However, since RUIs are pulled from | |
130 | * the AIL when their corresponding RUDs are committed to disk, their situation | |
131 | * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller | |
132 | * will eventually flush the log. This should help in getting the RUI out of | |
133 | * the AIL. | |
134 | */ | |
135 | STATIC uint | |
136 | xfs_rui_item_push( | |
137 | struct xfs_log_item *lip, | |
138 | struct list_head *buffer_list) | |
139 | { | |
140 | return XFS_ITEM_PINNED; | |
141 | } | |
142 | ||
143 | /* | |
144 | * The RUI has been either committed or aborted if the transaction has been | |
145 | * cancelled. If the transaction was cancelled, an RUD isn't going to be | |
146 | * constructed and thus we free the RUI here directly. | |
147 | */ | |
148 | STATIC void | |
149 | xfs_rui_item_unlock( | |
150 | struct xfs_log_item *lip) | |
151 | { | |
152 | if (lip->li_flags & XFS_LI_ABORTED) | |
153 | xfs_rui_item_free(RUI_ITEM(lip)); | |
154 | } | |
155 | ||
156 | /* | |
157 | * The RUI is logged only once and cannot be moved in the log, so simply return | |
158 | * the lsn at which it's been logged. | |
159 | */ | |
160 | STATIC xfs_lsn_t | |
161 | xfs_rui_item_committed( | |
162 | struct xfs_log_item *lip, | |
163 | xfs_lsn_t lsn) | |
164 | { | |
165 | return lsn; | |
166 | } | |
167 | ||
168 | /* | |
169 | * The RUI dependency tracking op doesn't do squat. It can't because | |
170 | * it doesn't know where the free extent is coming from. The dependency | |
171 | * tracking has to be handled by the "enclosing" metadata object. For | |
172 | * example, for inodes, the inode is locked throughout the extent freeing | |
173 | * so the dependency should be recorded there. | |
174 | */ | |
175 | STATIC void | |
176 | xfs_rui_item_committing( | |
177 | struct xfs_log_item *lip, | |
178 | xfs_lsn_t lsn) | |
179 | { | |
180 | } | |
181 | ||
182 | /* | |
183 | * This is the ops vector shared by all rui log items. | |
184 | */ | |
185 | static const struct xfs_item_ops xfs_rui_item_ops = { | |
186 | .iop_size = xfs_rui_item_size, | |
187 | .iop_format = xfs_rui_item_format, | |
188 | .iop_pin = xfs_rui_item_pin, | |
189 | .iop_unpin = xfs_rui_item_unpin, | |
190 | .iop_unlock = xfs_rui_item_unlock, | |
191 | .iop_committed = xfs_rui_item_committed, | |
192 | .iop_push = xfs_rui_item_push, | |
193 | .iop_committing = xfs_rui_item_committing, | |
194 | }; | |
195 | ||
196 | /* | |
197 | * Allocate and initialize an rui item with the given number of extents. | |
198 | */ | |
199 | struct xfs_rui_log_item * | |
200 | xfs_rui_init( | |
201 | struct xfs_mount *mp, | |
202 | uint nextents) | |
203 | ||
204 | { | |
205 | struct xfs_rui_log_item *ruip; | |
206 | uint size; | |
207 | ||
208 | ASSERT(nextents > 0); | |
209 | if (nextents > XFS_RUI_MAX_FAST_EXTENTS) { | |
210 | size = (uint)(sizeof(struct xfs_rui_log_item) + | |
211 | ((nextents - 1) * sizeof(struct xfs_map_extent))); | |
212 | ruip = kmem_zalloc(size, KM_SLEEP); | |
213 | } else { | |
214 | ruip = kmem_zone_zalloc(xfs_rui_zone, KM_SLEEP); | |
215 | } | |
216 | ||
217 | xfs_log_item_init(mp, &ruip->rui_item, XFS_LI_RUI, &xfs_rui_item_ops); | |
218 | ruip->rui_format.rui_nextents = nextents; | |
219 | ruip->rui_format.rui_id = (uintptr_t)(void *)ruip; | |
220 | atomic_set(&ruip->rui_next_extent, 0); | |
221 | atomic_set(&ruip->rui_refcount, 2); | |
222 | ||
223 | return ruip; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Copy an RUI format buffer from the given buf, and into the destination | |
228 | * RUI format structure. The RUI/RUD items were designed not to need any | |
229 | * special alignment handling. | |
230 | */ | |
231 | int | |
232 | xfs_rui_copy_format( | |
233 | struct xfs_log_iovec *buf, | |
234 | struct xfs_rui_log_format *dst_rui_fmt) | |
235 | { | |
236 | struct xfs_rui_log_format *src_rui_fmt; | |
237 | uint len; | |
238 | ||
239 | src_rui_fmt = buf->i_addr; | |
240 | len = sizeof(struct xfs_rui_log_format) + | |
241 | (src_rui_fmt->rui_nextents - 1) * | |
242 | sizeof(struct xfs_map_extent); | |
243 | ||
244 | if (buf->i_len != len) | |
245 | return -EFSCORRUPTED; | |
246 | ||
247 | memcpy((char *)dst_rui_fmt, (char *)src_rui_fmt, len); | |
248 | return 0; | |
249 | } | |
250 | ||
251 | /* | |
252 | * Freeing the RUI requires that we remove it from the AIL if it has already | |
253 | * been placed there. However, the RUI may not yet have been placed in the AIL | |
254 | * when called by xfs_rui_release() from RUD processing due to the ordering of | |
255 | * committed vs unpin operations in bulk insert operations. Hence the reference | |
256 | * count to ensure only the last caller frees the RUI. | |
257 | */ | |
258 | void | |
259 | xfs_rui_release( | |
260 | struct xfs_rui_log_item *ruip) | |
261 | { | |
262 | if (atomic_dec_and_test(&ruip->rui_refcount)) { | |
263 | xfs_trans_ail_remove(&ruip->rui_item, SHUTDOWN_LOG_IO_ERROR); | |
264 | xfs_rui_item_free(ruip); | |
265 | } | |
266 | } | |
267 | ||
268 | static inline struct xfs_rud_log_item *RUD_ITEM(struct xfs_log_item *lip) | |
269 | { | |
270 | return container_of(lip, struct xfs_rud_log_item, rud_item); | |
271 | } | |
272 | ||
273 | STATIC void | |
274 | xfs_rud_item_free(struct xfs_rud_log_item *rudp) | |
275 | { | |
276 | if (rudp->rud_format.rud_nextents > XFS_RUD_MAX_FAST_EXTENTS) | |
277 | kmem_free(rudp); | |
278 | else | |
279 | kmem_zone_free(xfs_rud_zone, rudp); | |
280 | } | |
281 | ||
282 | /* | |
283 | * This returns the number of iovecs needed to log the given rud item. | |
284 | * We only need 1 iovec for an rud item. It just logs the rud_log_format | |
285 | * structure. | |
286 | */ | |
287 | static inline int | |
288 | xfs_rud_item_sizeof( | |
289 | struct xfs_rud_log_item *rudp) | |
290 | { | |
291 | return sizeof(struct xfs_rud_log_format) + | |
292 | (rudp->rud_format.rud_nextents - 1) * | |
293 | sizeof(struct xfs_map_extent); | |
294 | } | |
295 | ||
296 | STATIC void | |
297 | xfs_rud_item_size( | |
298 | struct xfs_log_item *lip, | |
299 | int *nvecs, | |
300 | int *nbytes) | |
301 | { | |
302 | *nvecs += 1; | |
303 | *nbytes += xfs_rud_item_sizeof(RUD_ITEM(lip)); | |
304 | } | |
305 | ||
306 | /* | |
307 | * This is called to fill in the vector of log iovecs for the | |
308 | * given rud log item. We use only 1 iovec, and we point that | |
309 | * at the rud_log_format structure embedded in the rud item. | |
310 | * It is at this point that we assert that all of the extent | |
311 | * slots in the rud item have been filled. | |
312 | */ | |
313 | STATIC void | |
314 | xfs_rud_item_format( | |
315 | struct xfs_log_item *lip, | |
316 | struct xfs_log_vec *lv) | |
317 | { | |
318 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
319 | struct xfs_log_iovec *vecp = NULL; | |
320 | ||
321 | ASSERT(rudp->rud_next_extent == rudp->rud_format.rud_nextents); | |
322 | ||
323 | rudp->rud_format.rud_type = XFS_LI_RUD; | |
324 | rudp->rud_format.rud_size = 1; | |
325 | ||
326 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUD_FORMAT, &rudp->rud_format, | |
327 | xfs_rud_item_sizeof(rudp)); | |
328 | } | |
329 | ||
330 | /* | |
331 | * Pinning has no meaning for an rud item, so just return. | |
332 | */ | |
333 | STATIC void | |
334 | xfs_rud_item_pin( | |
335 | struct xfs_log_item *lip) | |
336 | { | |
337 | } | |
338 | ||
339 | /* | |
340 | * Since pinning has no meaning for an rud item, unpinning does | |
341 | * not either. | |
342 | */ | |
343 | STATIC void | |
344 | xfs_rud_item_unpin( | |
345 | struct xfs_log_item *lip, | |
346 | int remove) | |
347 | { | |
348 | } | |
349 | ||
350 | /* | |
351 | * There isn't much you can do to push on an rud item. It is simply stuck | |
352 | * waiting for the log to be flushed to disk. | |
353 | */ | |
354 | STATIC uint | |
355 | xfs_rud_item_push( | |
356 | struct xfs_log_item *lip, | |
357 | struct list_head *buffer_list) | |
358 | { | |
359 | return XFS_ITEM_PINNED; | |
360 | } | |
361 | ||
362 | /* | |
363 | * The RUD is either committed or aborted if the transaction is cancelled. If | |
364 | * the transaction is cancelled, drop our reference to the RUI and free the | |
365 | * RUD. | |
366 | */ | |
367 | STATIC void | |
368 | xfs_rud_item_unlock( | |
369 | struct xfs_log_item *lip) | |
370 | { | |
371 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
372 | ||
373 | if (lip->li_flags & XFS_LI_ABORTED) { | |
374 | xfs_rui_release(rudp->rud_ruip); | |
375 | xfs_rud_item_free(rudp); | |
376 | } | |
377 | } | |
378 | ||
379 | /* | |
380 | * When the rud item is committed to disk, all we need to do is delete our | |
381 | * reference to our partner rui item and then free ourselves. Since we're | |
382 | * freeing ourselves we must return -1 to keep the transaction code from | |
383 | * further referencing this item. | |
384 | */ | |
385 | STATIC xfs_lsn_t | |
386 | xfs_rud_item_committed( | |
387 | struct xfs_log_item *lip, | |
388 | xfs_lsn_t lsn) | |
389 | { | |
390 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
391 | ||
392 | /* | |
393 | * Drop the RUI reference regardless of whether the RUD has been | |
394 | * aborted. Once the RUD transaction is constructed, it is the sole | |
395 | * responsibility of the RUD to release the RUI (even if the RUI is | |
396 | * aborted due to log I/O error). | |
397 | */ | |
398 | xfs_rui_release(rudp->rud_ruip); | |
399 | xfs_rud_item_free(rudp); | |
400 | ||
401 | return (xfs_lsn_t)-1; | |
402 | } | |
403 | ||
404 | /* | |
405 | * The RUD dependency tracking op doesn't do squat. It can't because | |
406 | * it doesn't know where the free extent is coming from. The dependency | |
407 | * tracking has to be handled by the "enclosing" metadata object. For | |
408 | * example, for inodes, the inode is locked throughout the extent freeing | |
409 | * so the dependency should be recorded there. | |
410 | */ | |
411 | STATIC void | |
412 | xfs_rud_item_committing( | |
413 | struct xfs_log_item *lip, | |
414 | xfs_lsn_t lsn) | |
415 | { | |
416 | } | |
417 | ||
418 | /* | |
419 | * This is the ops vector shared by all rud log items. | |
420 | */ | |
421 | static const struct xfs_item_ops xfs_rud_item_ops = { | |
422 | .iop_size = xfs_rud_item_size, | |
423 | .iop_format = xfs_rud_item_format, | |
424 | .iop_pin = xfs_rud_item_pin, | |
425 | .iop_unpin = xfs_rud_item_unpin, | |
426 | .iop_unlock = xfs_rud_item_unlock, | |
427 | .iop_committed = xfs_rud_item_committed, | |
428 | .iop_push = xfs_rud_item_push, | |
429 | .iop_committing = xfs_rud_item_committing, | |
430 | }; | |
431 | ||
432 | /* | |
433 | * Allocate and initialize an rud item with the given number of extents. | |
434 | */ | |
435 | struct xfs_rud_log_item * | |
436 | xfs_rud_init( | |
437 | struct xfs_mount *mp, | |
438 | struct xfs_rui_log_item *ruip, | |
439 | uint nextents) | |
440 | ||
441 | { | |
442 | struct xfs_rud_log_item *rudp; | |
443 | uint size; | |
444 | ||
445 | ASSERT(nextents > 0); | |
446 | if (nextents > XFS_RUD_MAX_FAST_EXTENTS) { | |
447 | size = (uint)(sizeof(struct xfs_rud_log_item) + | |
448 | ((nextents - 1) * sizeof(struct xfs_map_extent))); | |
449 | rudp = kmem_zalloc(size, KM_SLEEP); | |
450 | } else { | |
451 | rudp = kmem_zone_zalloc(xfs_rud_zone, KM_SLEEP); | |
452 | } | |
453 | ||
454 | xfs_log_item_init(mp, &rudp->rud_item, XFS_LI_RUD, &xfs_rud_item_ops); | |
455 | rudp->rud_ruip = ruip; | |
456 | rudp->rud_format.rud_nextents = nextents; | |
457 | rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id; | |
458 | ||
459 | return rudp; | |
460 | } | |
9e88b5d8 DW |
461 | |
462 | /* | |
463 | * Process an rmap update intent item that was recovered from the log. | |
464 | * We need to update the rmapbt. | |
465 | */ | |
466 | int | |
467 | xfs_rui_recover( | |
468 | struct xfs_mount *mp, | |
469 | struct xfs_rui_log_item *ruip) | |
470 | { | |
471 | int i; | |
472 | int error = 0; | |
473 | struct xfs_map_extent *rmap; | |
474 | xfs_fsblock_t startblock_fsb; | |
475 | bool op_ok; | |
476 | ||
477 | ASSERT(!test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags)); | |
478 | ||
479 | /* | |
480 | * First check the validity of the extents described by the | |
481 | * RUI. If any are bad, then assume that all are bad and | |
482 | * just toss the RUI. | |
483 | */ | |
484 | for (i = 0; i < ruip->rui_format.rui_nextents; i++) { | |
485 | rmap = &(ruip->rui_format.rui_extents[i]); | |
486 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
487 | XFS_FSB_TO_DADDR(mp, rmap->me_startblock)); | |
488 | switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) { | |
489 | case XFS_RMAP_EXTENT_MAP: | |
490 | case XFS_RMAP_EXTENT_UNMAP: | |
491 | case XFS_RMAP_EXTENT_CONVERT: | |
492 | case XFS_RMAP_EXTENT_ALLOC: | |
493 | case XFS_RMAP_EXTENT_FREE: | |
494 | op_ok = true; | |
495 | break; | |
496 | default: | |
497 | op_ok = false; | |
498 | break; | |
499 | } | |
500 | if (!op_ok || (startblock_fsb == 0) || | |
501 | (rmap->me_len == 0) || | |
502 | (startblock_fsb >= mp->m_sb.sb_dblocks) || | |
503 | (rmap->me_len >= mp->m_sb.sb_agblocks) || | |
504 | (rmap->me_flags & ~XFS_RMAP_EXTENT_FLAGS)) { | |
505 | /* | |
506 | * This will pull the RUI from the AIL and | |
507 | * free the memory associated with it. | |
508 | */ | |
509 | set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags); | |
510 | xfs_rui_release(ruip); | |
511 | return -EIO; | |
512 | } | |
513 | } | |
514 | ||
515 | /* XXX: do nothing for now */ | |
516 | set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags); | |
517 | xfs_rui_release(ruip); | |
518 | return error; | |
519 | } |