1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. |
5 | */ |
6 | #include "xfs.h" |
7 | #include "xfs_fs.h" |
8 | #include "xfs_shared.h" |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" |
11 | #include "xfs_trans_resv.h" |
12 | #include "xfs_bit.h" |
13 | #include "xfs_mount.h" |
14 | #include "xfs_trans.h" |
15 | #include "xfs_trans_priv.h" |
16 | #include "xfs_buf_item.h" |
17 | #include "xfs_inode.h" |
18 | #include "xfs_inode_item.h" |
19 | #include "xfs_quota.h" |
20 | #include "xfs_dquot_item.h" |
21 | #include "xfs_dquot.h" |
22 | #include "xfs_trace.h" |
23 | #include "xfs_log.h" |
24 | #include "xfs_log_priv.h" |
25 | |
26 | |
27 | struct kmem_cache *xfs_buf_item_cache; |
28 | |
29 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
30 | { |
31 | return container_of(lip, struct xfs_buf_log_item, bli_item); |
32 | } |
33 | |
34 | /* Is this log iovec plausibly large enough to contain the buffer log format? */ |
35 | bool |
36 | xfs_buf_log_check_iovec( |
37 | struct xfs_log_iovec *iovec) |
38 | { |
39 | struct xfs_buf_log_format *blfp = iovec->i_addr; |
40 | char *bmp_end; |
41 | char *item_end; |
42 | |
43 | if (offsetof(struct xfs_buf_log_format, blf_data_map) > iovec->i_len) |
44 | return false; |
45 | |
46 | item_end = (char *)iovec->i_addr + iovec->i_len; |
47 | bmp_end = (char *)&blfp->blf_data_map[blfp->blf_map_size]; |
48 | return bmp_end <= item_end; |
49 | } |
50 | |
51 | static inline int |
52 | xfs_buf_log_format_size( |
53 | struct xfs_buf_log_format *blfp) |
54 | { |
55 | return offsetof(struct xfs_buf_log_format, blf_data_map) + |
56 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); |
57 | } |
58 | |
59 | static inline bool |
60 | xfs_buf_item_straddle( |
61 | struct xfs_buf *bp, |
62 | uint offset, |
63 | int first_bit, |
64 | int nbits) |
65 | { |
66 | void *first, *last; |
67 | |
68 | first = xfs_buf_offset(bp, offset + (first_bit << XFS_BLF_SHIFT)); |
69 | last = xfs_buf_offset(bp, |
70 | offset + ((first_bit + nbits) << XFS_BLF_SHIFT)); |
71 | |
72 | if (last - first != nbits * XFS_BLF_CHUNK) |
73 | return true; |
74 | return false; |
75 | } |
76 | |
77 | /* |
78 | * Return the number of log iovecs and space needed to log the given buf log |
79 | * item segment. |
80 | * |
81 | * It calculates this as 1 iovec for the buf log format structure and 1 for each |
82 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged |
83 | * in a single iovec. |
84 | */ |
85 | STATIC void |
86 | xfs_buf_item_size_segment( |
87 | struct xfs_buf_log_item *bip, |
88 | struct xfs_buf_log_format *blfp, |
89 | uint offset, |
90 | int *nvecs, |
91 | int *nbytes) |
92 | { |
93 | struct xfs_buf *bp = bip->bli_buf; |
94 | int first_bit; |
95 | int nbits; |
96 | int next_bit; |
97 | int last_bit; |
98 | |
99 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
100 | if (first_bit == -1) |
101 | return; |
102 | |
103 | (*nvecs)++; |
104 | *nbytes += xfs_buf_log_format_size(blfp); |
105 | |
106 | do { |
107 | nbits = xfs_contig_bits(blfp->blf_data_map, |
108 | blfp->blf_map_size, first_bit); |
109 | ASSERT(nbits > 0); |
110 | |
111 | /* |
112 | * Straddling a page is rare because we don't log contiguous |
113 | * chunks of unmapped buffers anywhere. |
114 | */ |
115 | if (nbits > 1 && |
116 | xfs_buf_item_straddle(bp, offset, first_bit, nbits)) |
117 | goto slow_scan; |
118 | |
119 | (*nvecs)++; |
120 | *nbytes += nbits * XFS_BLF_CHUNK; |
121 | |
122 | /* |
123 | * This takes the bit number to start looking from and |
124 | * returns the next set bit from there. It returns -1 |
125 | * if there are no more bits set or the start bit is |
126 | * beyond the end of the bitmap. |
127 | */ |
128 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
129 | (uint)first_bit + nbits + 1); |
130 | } while (first_bit != -1); |
131 | |
132 | return; |
133 | |
134 | slow_scan: |
135 | /* Count the first bit we jumped out of the above loop from */ |
136 | (*nvecs)++; |
137 | *nbytes += XFS_BLF_CHUNK; |
138 | last_bit = first_bit; |
139 | while (last_bit != -1) { |
140 | /* |
141 | * This takes the bit number to start looking from and |
142 | * returns the next set bit from there. It returns -1 |
143 | * if there are no more bits set or the start bit is |
144 | * beyond the end of the bitmap. |
145 | */ |
146 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
147 | last_bit + 1); |
148 | /* |
149 | * If we run out of bits, leave the loop, |
150 | * else if we find a new set of bits bump the number of vecs, |
151 | * else keep scanning the current set of bits. |
152 | */ |
153 | if (next_bit == -1) { |
154 | break; |
155 | } else if (next_bit != last_bit + 1 || |
156 | xfs_buf_item_straddle(bp, offset, first_bit, nbits)) { |
157 | last_bit = next_bit; |
158 | first_bit = next_bit; |
159 | (*nvecs)++; |
160 | nbits = 1; |
161 | } else { |
162 | last_bit++; |
163 | nbits++; |
164 | } |
165 | *nbytes += XFS_BLF_CHUNK; |
166 | } |
167 | } |
168 | |
169 | /* |
170 | * Return the number of log iovecs and space needed to log the given buf log |
171 | * item. |
172 | * |
173 | * Discontiguous buffers need a format structure per region that is being |
174 | * logged. This makes the changes in the buffer appear to log recovery as though |
175 | * they came from separate buffers, just like would occur if multiple buffers |
176 | * were used instead of a single discontiguous buffer. This enables |
177 | * discontiguous buffers to be in-memory constructs, completely transparent to |
178 | * what ends up on disk. |
179 | * |
180 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log |
181 | * format structures. If the item has previously been logged and has dirty |
182 | * regions, we do not relog them in stale buffers. This has the effect of |
183 | * reducing the size of the relogged item by the amount of dirty data tracked |
184 | * by the log item. This can result in the committing transaction reducing the |
185 | * amount of space being consumed by the CIL. |
186 | */ |
187 | STATIC void |
188 | xfs_buf_item_size( |
189 | struct xfs_log_item *lip, |
190 | int *nvecs, |
191 | int *nbytes) |
192 | { |
193 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
194 | struct xfs_buf *bp = bip->bli_buf; |
195 | int i; |
196 | int bytes; |
197 | uint offset = 0; |
198 | |
199 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
200 | if (bip->bli_flags & XFS_BLI_STALE) { |
201 | /* |
202 | * The buffer is stale, so all we need to log is the buf log |
203 | * format structure with the cancel flag in it as we are never |
204 | * going to replay the changes tracked in the log item. |
205 | */ |
206 | trace_xfs_buf_item_size_stale(bip); |
207 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
208 | *nvecs += bip->bli_format_count; |
209 | for (i = 0; i < bip->bli_format_count; i++) { |
210 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); |
211 | } |
212 | return; |
213 | } |
214 | |
215 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); |
216 | |
217 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
218 | /* |
219 | * The buffer has been logged just to order it. It is not being |
220 | * included in the transaction commit, so no vectors are used at |
221 | * all. |
222 | */ |
223 | trace_xfs_buf_item_size_ordered(bip); |
224 | *nvecs = XFS_LOG_VEC_ORDERED; |
225 | return; |
226 | } |
227 | |
228 | /* |
229 | * The vector count is based on the number of buffer vectors we have |
230 | * dirty bits in. This will only be greater than one when we have a |
231 | * compound buffer with more than one segment dirty. Hence for compound |
232 | * buffers we need to track which segment the dirty bits correspond to, |
233 | * and when we move from one segment to the next increment the vector |
234 | * count for the extra buf log format structure that will need to be |
235 | * written. |
236 | */ |
237 | bytes = 0; |
238 | for (i = 0; i < bip->bli_format_count; i++) { |
239 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], offset, |
240 | nvecs, &bytes); |
241 | offset += BBTOB(bp->b_maps[i].bm_len); |
242 | } |
243 | |
244 | /* |
245 | * Round up the buffer size required to minimise the number of memory |
246 | * allocations that need to be done as this item grows when relogged by |
247 | * repeated modifications. |
248 | */ |
249 | *nbytes = round_up(bytes, 512); |
250 | trace_xfs_buf_item_size(bip); |
251 | } |
252 | |
253 | static inline void |
254 | xfs_buf_item_copy_iovec( |
255 | struct xfs_log_vec *lv, |
256 | struct xfs_log_iovec **vecp, |
257 | struct xfs_buf *bp, |
258 | uint offset, |
259 | int first_bit, |
260 | uint nbits) |
261 | { |
262 | offset += first_bit * XFS_BLF_CHUNK; |
263 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
264 | xfs_buf_offset(bp, offset), |
265 | nbits * XFS_BLF_CHUNK); |
266 | } |
267 | |
268 | static void |
269 | xfs_buf_item_format_segment( |
270 | struct xfs_buf_log_item *bip, |
271 | struct xfs_log_vec *lv, |
272 | struct xfs_log_iovec **vecp, |
273 | uint offset, |
274 | struct xfs_buf_log_format *blfp) |
275 | { |
276 | struct xfs_buf *bp = bip->bli_buf; |
277 | uint base_size; |
278 | int first_bit; |
279 | int last_bit; |
280 | int next_bit; |
281 | uint nbits; |
282 | |
283 | /* copy the flags across from the base format item */ |
284 | blfp->blf_flags = bip->__bli_format.blf_flags; |
285 | |
286 | /* |
287 | * Base size is the actual size of the ondisk structure - it reflects |
288 | * the actual size of the dirty bitmap rather than the size of the in |
289 | * memory structure. |
290 | */ |
291 | base_size = xfs_buf_log_format_size(blfp); |
292 | |
293 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
294 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { |
295 | /* |
296 | * If the map is not be dirty in the transaction, mark |
297 | * the size as zero and do not advance the vector pointer. |
298 | */ |
299 | return; |
300 | } |
301 | |
302 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
303 | blfp->blf_size = 1; |
304 | |
305 | if (bip->bli_flags & XFS_BLI_STALE) { |
306 | /* |
307 | * The buffer is stale, so all we need to log |
308 | * is the buf log format structure with the |
309 | * cancel flag in it. |
310 | */ |
311 | trace_xfs_buf_item_format_stale(bip); |
312 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
313 | return; |
314 | } |
315 | |
316 | |
317 | /* |
318 | * Fill in an iovec for each set of contiguous chunks. |
319 | */ |
320 | do { |
321 | ASSERT(first_bit >= 0); |
322 | nbits = xfs_contig_bits(blfp->blf_data_map, |
323 | blfp->blf_map_size, first_bit); |
324 | ASSERT(nbits > 0); |
325 | |
326 | /* |
327 | * Straddling a page is rare because we don't log contiguous |
328 | * chunks of unmapped buffers anywhere. |
329 | */ |
330 | if (nbits > 1 && |
331 | xfs_buf_item_straddle(bp, offset, first_bit, nbits)) |
332 | goto slow_scan; |
333 | |
334 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
335 | first_bit, nbits); |
336 | blfp->blf_size++; |
337 | |
338 | /* |
339 | * This takes the bit number to start looking from and |
340 | * returns the next set bit from there. It returns -1 |
341 | * if there are no more bits set or the start bit is |
342 | * beyond the end of the bitmap. |
343 | */ |
344 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
345 | (uint)first_bit + nbits + 1); |
346 | } while (first_bit != -1); |
347 | |
348 | return; |
349 | |
350 | slow_scan: |
351 | ASSERT(bp->b_addr == NULL); |
352 | last_bit = first_bit; |
353 | nbits = 1; |
354 | for (;;) { |
355 | /* |
356 | * This takes the bit number to start looking from and |
357 | * returns the next set bit from there. It returns -1 |
358 | * if there are no more bits set or the start bit is |
359 | * beyond the end of the bitmap. |
360 | */ |
361 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
362 | (uint)last_bit + 1); |
363 | /* |
364 | * If we run out of bits fill in the last iovec and get out of |
365 | * the loop. Else if we start a new set of bits then fill in |
366 | * the iovec for the series we were looking at and start |
367 | * counting the bits in the new one. Else we're still in the |
368 | * same set of bits so just keep counting and scanning. |
369 | */ |
370 | if (next_bit == -1) { |
371 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
372 | first_bit, nbits); |
373 | blfp->blf_size++; |
374 | break; |
375 | } else if (next_bit != last_bit + 1 || |
376 | xfs_buf_item_straddle(bp, offset, first_bit, nbits)) { |
377 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
378 | first_bit, nbits); |
379 | blfp->blf_size++; |
380 | first_bit = next_bit; |
381 | last_bit = next_bit; |
382 | nbits = 1; |
383 | } else { |
384 | last_bit++; |
385 | nbits++; |
386 | } |
387 | } |
388 | } |
389 | |
390 | /* |
391 | * This is called to fill in the vector of log iovecs for the |
392 | * given log buf item. It fills the first entry with a buf log |
393 | * format structure, and the rest point to contiguous chunks |
394 | * within the buffer. |
395 | */ |
396 | STATIC void |
397 | xfs_buf_item_format( |
398 | struct xfs_log_item *lip, |
399 | struct xfs_log_vec *lv) |
400 | { |
401 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
402 | struct xfs_buf *bp = bip->bli_buf; |
403 | struct xfs_log_iovec *vecp = NULL; |
404 | uint offset = 0; |
405 | int i; |
406 | |
407 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
408 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
409 | (bip->bli_flags & XFS_BLI_STALE)); |
410 | ASSERT((bip->bli_flags & XFS_BLI_STALE) || |
411 | (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF |
412 | && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF)); |
413 | ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) || |
414 | (bip->bli_flags & XFS_BLI_STALE)); |
415 | |
416 | |
417 | /* |
418 | * If it is an inode buffer, transfer the in-memory state to the |
419 | * format flags and clear the in-memory state. |
420 | * |
421 | * For buffer based inode allocation, we do not transfer |
422 | * this state if the inode buffer allocation has not yet been committed |
423 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent |
424 | * correct replay of the inode allocation. |
425 | * |
426 | * For icreate item based inode allocation, the buffers aren't written |
427 | * to the journal during allocation, and hence we should always tag the |
428 | * buffer as an inode buffer so that the correct unlinked list replay |
429 | * occurs during recovery. |
430 | */ |
431 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { |
432 | if (xfs_has_v3inodes(lip->li_log->l_mp) || |
433 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
434 | xfs_log_item_in_current_chkpt(lip))) |
435 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
436 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
437 | } |
438 | |
439 | for (i = 0; i < bip->bli_format_count; i++) { |
440 | xfs_buf_item_format_segment(bip, lv, &vecp, offset, |
441 | &bip->bli_formats[i]); |
442 | offset += BBTOB(bp->b_maps[i].bm_len); |
443 | } |
444 | |
445 | /* |
446 | * Check to make sure everything is consistent. |
447 | */ |
448 | trace_xfs_buf_item_format(bip); |
449 | } |
450 | |
451 | /* |
452 | * This is called to pin the buffer associated with the buf log item in memory |
453 | * so it cannot be written out. |
454 | * |
455 | * We take a reference to the buffer log item here so that the BLI life cycle |
456 | * extends at least until the buffer is unpinned via xfs_buf_item_unpin() and |
457 | * inserted into the AIL. |
458 | * |
459 | * We also need to take a reference to the buffer itself as the BLI unpin |
460 | * processing requires accessing the buffer after the BLI has dropped the final |
461 | * BLI reference. See xfs_buf_item_unpin() for an explanation. |
462 | * If unpins race to drop the final BLI reference and only the |
463 | * BLI owns a reference to the buffer, then the loser of the race can have the |
464 | * buffer fgreed from under it (e.g. on shutdown). Taking a buffer reference per |
465 | * pin count ensures the life cycle of the buffer extends for as |
466 | * long as we hold the buffer pin reference in xfs_buf_item_unpin(). |
467 | */ |
468 | STATIC void |
469 | xfs_buf_item_pin( |
470 | struct xfs_log_item *lip) |
471 | { |
472 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
473 | |
474 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
475 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
476 | (bip->bli_flags & XFS_BLI_ORDERED) || |
477 | (bip->bli_flags & XFS_BLI_STALE)); |
478 | |
479 | trace_xfs_buf_item_pin(bip); |
480 | |
481 | xfs_buf_hold(bp: bip->bli_buf); |
482 | atomic_inc(v: &bip->bli_refcount); |
483 | atomic_inc(v: &bip->bli_buf->b_pin_count); |
484 | } |
485 | |
486 | /* |
487 | * This is called to unpin the buffer associated with the buf log item which was |
488 | * previously pinned with a call to xfs_buf_item_pin(). We enter this function |
489 | * with a buffer pin count, a buffer reference and a BLI reference. |
490 | * |
491 | * We must drop the BLI reference before we unpin the buffer because the AIL |
492 | * doesn't acquire a BLI reference whenever it accesses it. Therefore if the |
493 | * refcount drops to zero, the bli could still be AIL resident and the buffer |
494 | * submitted for I/O at any point before we return. This can result in IO |
495 | * completion freeing the buffer while we are still trying to access it here. |
496 | * This race condition can also occur in shutdown situations where we abort and |
497 | * unpin buffers from contexts other that journal IO completion. |
498 | * |
499 | * Hence we have to hold a buffer reference per pin count to ensure that the |
500 | * buffer cannot be freed until we have finished processing the unpin operation. |
501 | * The reference is taken in xfs_buf_item_pin(), and we must hold it until we |
502 | * are done processing the buffer state. In the case of an abort (remove = |
503 | * true) then we re-use the current pin reference as the IO reference we hand |
504 | * off to IO failure handling. |
505 | */ |
506 | STATIC void |
507 | xfs_buf_item_unpin( |
508 | struct xfs_log_item *lip, |
509 | int remove) |
510 | { |
511 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
512 | struct xfs_buf *bp = bip->bli_buf; |
513 | int stale = bip->bli_flags & XFS_BLI_STALE; |
514 | int freed; |
515 | |
516 | ASSERT(bp->b_log_item == bip); |
517 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
518 | |
519 | trace_xfs_buf_item_unpin(bip); |
520 | |
521 | freed = atomic_dec_and_test(v: &bip->bli_refcount); |
522 | if (atomic_dec_and_test(v: &bp->b_pin_count)) |
523 | wake_up_all(&bp->b_waiters); |
524 | |
525 | /* |
526 | * Nothing to do but drop the buffer pin reference if the BLI is |
527 | * still active. |
528 | */ |
529 | if (!freed) { |
530 | xfs_buf_rele(bp); |
531 | return; |
532 | } |
533 | |
534 | if (stale) { |
535 | ASSERT(bip->bli_flags & XFS_BLI_STALE); |
536 | ASSERT(xfs_buf_islocked(bp)); |
537 | ASSERT(bp->b_flags & XBF_STALE); |
538 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
539 | ASSERT(list_empty(&lip->li_trans)); |
540 | ASSERT(!bp->b_transp); |
541 | |
542 | trace_xfs_buf_item_unpin_stale(bip); |
543 | |
544 | /* |
545 | * The buffer has been locked and referenced since it was marked |
546 | * stale so we own both lock and reference exclusively here. We |
547 | * do not need the pin reference any more, so drop it now so |
548 | * that we only have one reference to drop once item completion |
549 | * processing is complete. |
550 | */ |
551 | xfs_buf_rele(bp); |
552 | |
553 | /* |
554 | * If we get called here because of an IO error, we may or may |
555 | * not have the item on the AIL. xfs_trans_ail_delete() will |
556 | * take care of that situation. xfs_trans_ail_delete() drops |
557 | * the AIL lock. |
558 | */ |
559 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { |
560 | xfs_buf_item_done(bp); |
561 | xfs_buf_inode_iodone(bp); |
562 | ASSERT(list_empty(&bp->b_li_list)); |
563 | } else { |
564 | xfs_trans_ail_delete(lip, SHUTDOWN_LOG_IO_ERROR); |
565 | xfs_buf_item_relse(bp); |
566 | ASSERT(bp->b_log_item == NULL); |
567 | } |
568 | xfs_buf_relse(bp); |
569 | return; |
570 | } |
571 | |
572 | if (remove) { |
573 | /* |
574 | * We need to simulate an async IO failures here to ensure that |
575 | * the correct error completion is run on this buffer. This |
576 | * requires a reference to the buffer and for the buffer to be |
577 | * locked. We can safely pass ownership of the pin reference to |
578 | * the IO to ensure that nothing can free the buffer while we |
579 | * wait for the lock and then run the IO failure completion. |
580 | */ |
581 | xfs_buf_lock(bp); |
582 | bp->b_flags |= XBF_ASYNC; |
583 | xfs_buf_ioend_fail(bp); |
584 | return; |
585 | } |
586 | |
587 | /* |
588 | * BLI has no more active references - it will be moved to the AIL to |
589 | * manage the remaining BLI/buffer life cycle. There is nothing left for |
590 | * us to do here so drop the pin reference to the buffer. |
591 | */ |
592 | xfs_buf_rele(bp); |
593 | } |
594 | |
595 | STATIC uint |
596 | xfs_buf_item_push( |
597 | struct xfs_log_item *lip, |
598 | struct list_head *buffer_list) |
599 | { |
600 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
601 | struct xfs_buf *bp = bip->bli_buf; |
602 | uint rval = XFS_ITEM_SUCCESS; |
603 | |
604 | if (xfs_buf_ispinned(bp)) |
605 | return XFS_ITEM_PINNED; |
606 | if (!xfs_buf_trylock(bp)) { |
607 | /* |
608 | * If we have just raced with a buffer being pinned and it has |
609 | * been marked stale, we could end up stalling until someone else |
610 | * issues a log force to unpin the stale buffer. Check for the |
611 | * race condition here so xfsaild recognizes the buffer is pinned |
612 | * and queues a log force to move it along. |
613 | */ |
614 | if (xfs_buf_ispinned(bp)) |
615 | return XFS_ITEM_PINNED; |
616 | return XFS_ITEM_LOCKED; |
617 | } |
618 | |
619 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
620 | |
621 | trace_xfs_buf_item_push(bip); |
622 | |
623 | /* has a previous flush failed due to IO errors? */ |
624 | if (bp->b_flags & XBF_WRITE_FAIL) { |
625 | xfs_buf_alert_ratelimited(bp, rlmsg: "XFS: Failing async write" , |
626 | fmt: "Failing async write on buffer block 0x%llx. Retrying async write." , |
627 | (long long)xfs_buf_daddr(bp)); |
628 | } |
629 | |
630 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
631 | rval = XFS_ITEM_FLUSHING; |
632 | xfs_buf_unlock(bp); |
633 | return rval; |
634 | } |
635 | |
636 | /* |
637 | * Drop the buffer log item refcount and take appropriate action. This helper |
638 | * determines whether the bli must be freed or not, since a decrement to zero |
639 | * does not necessarily mean the bli is unused. |
640 | * |
641 | * Return true if the bli is freed, false otherwise. |
642 | */ |
643 | bool |
644 | xfs_buf_item_put( |
645 | struct xfs_buf_log_item *bip) |
646 | { |
647 | struct xfs_log_item *lip = &bip->bli_item; |
648 | bool aborted; |
649 | bool dirty; |
650 | |
651 | /* drop the bli ref and return if it wasn't the last one */ |
652 | if (!atomic_dec_and_test(v: &bip->bli_refcount)) |
653 | return false; |
654 | |
655 | /* |
656 | * We dropped the last ref and must free the item if clean or aborted. |
657 | * If the bli is dirty and non-aborted, the buffer was clean in the |
658 | * transaction but still awaiting writeback from previous changes. In |
659 | * that case, the bli is freed on buffer writeback completion. |
660 | */ |
661 | aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) || |
662 | xlog_is_shutdown(log: lip->li_log); |
663 | dirty = bip->bli_flags & XFS_BLI_DIRTY; |
664 | if (dirty && !aborted) |
665 | return false; |
666 | |
667 | /* |
668 | * The bli is aborted or clean. An aborted item may be in the AIL |
669 | * regardless of dirty state. For example, consider an aborted |
670 | * transaction that invalidated a dirty bli and cleared the dirty |
671 | * state. |
672 | */ |
673 | if (aborted) |
674 | xfs_trans_ail_delete(lip, shutdown_type: 0); |
675 | xfs_buf_item_relse(bip->bli_buf); |
676 | return true; |
677 | } |
678 | |
679 | /* |
680 | * Release the buffer associated with the buf log item. If there is no dirty |
681 | * logged data associated with the buffer recorded in the buf log item, then |
682 | * free the buf log item and remove the reference to it in the buffer. |
683 | * |
684 | * This call ignores the recursion count. It is only called when the buffer |
685 | * should REALLY be unlocked, regardless of the recursion count. |
686 | * |
687 | * We unconditionally drop the transaction's reference to the log item. If the |
688 | * item was logged, then another reference was taken when it was pinned, so we |
689 | * can safely drop the transaction reference now. This also allows us to avoid |
690 | * potential races with the unpin code freeing the bli by not referencing the |
691 | * bli after we've dropped the reference count. |
692 | * |
693 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item |
694 | * if necessary but do not unlock the buffer. This is for support of |
695 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't |
696 | * free the item. |
697 | */ |
698 | STATIC void |
699 | xfs_buf_item_release( |
700 | struct xfs_log_item *lip) |
701 | { |
702 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
703 | struct xfs_buf *bp = bip->bli_buf; |
704 | bool released; |
705 | bool hold = bip->bli_flags & XFS_BLI_HOLD; |
706 | bool stale = bip->bli_flags & XFS_BLI_STALE; |
707 | #if defined(DEBUG) || defined(XFS_WARN) |
708 | bool ordered = bip->bli_flags & XFS_BLI_ORDERED; |
709 | bool dirty = bip->bli_flags & XFS_BLI_DIRTY; |
710 | bool aborted = test_bit(XFS_LI_ABORTED, |
711 | &lip->li_flags); |
712 | #endif |
713 | |
714 | trace_xfs_buf_item_release(bip); |
715 | |
716 | /* |
717 | * The bli dirty state should match whether the blf has logged segments |
718 | * except for ordered buffers, where only the bli should be dirty. |
719 | */ |
720 | ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) || |
721 | (ordered && dirty && !xfs_buf_item_dirty_format(bip))); |
722 | ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
723 | |
724 | /* |
725 | * Clear the buffer's association with this transaction and |
726 | * per-transaction state from the bli, which has been copied above. |
727 | */ |
728 | bp->b_transp = NULL; |
729 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); |
730 | |
731 | /* |
732 | * Unref the item and unlock the buffer unless held or stale. Stale |
733 | * buffers remain locked until final unpin unless the bli is freed by |
734 | * the unref call. The latter implies shutdown because buffer |
735 | * invalidation dirties the bli and transaction. |
736 | */ |
737 | released = xfs_buf_item_put(bip); |
738 | if (hold || (stale && !released)) |
739 | return; |
740 | ASSERT(!stale || aborted); |
741 | xfs_buf_relse(bp); |
742 | } |
743 | |
744 | STATIC void |
745 | xfs_buf_item_committing( |
746 | struct xfs_log_item *lip, |
747 | xfs_csn_t seq) |
748 | { |
749 | return xfs_buf_item_release(lip); |
750 | } |
751 | |
752 | /* |
753 | * This is called to find out where the oldest active copy of the |
754 | * buf log item in the on disk log resides now that the last log |
755 | * write of it completed at the given lsn. |
756 | * We always re-log all the dirty data in a buffer, so usually the |
757 | * latest copy in the on disk log is the only one that matters. For |
758 | * those cases we simply return the given lsn. |
759 | * |
760 | * The one exception to this is for buffers full of newly allocated |
761 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF |
762 | * flag set, indicating that only the di_next_unlinked fields from the |
763 | * inodes in the buffers will be replayed during recovery. If the |
764 | * original newly allocated inode images have not yet been flushed |
765 | * when the buffer is so relogged, then we need to make sure that we |
766 | * keep the old images in the 'active' portion of the log. We do this |
767 | * by returning the original lsn of that transaction here rather than |
768 | * the current one. |
769 | */ |
770 | STATIC xfs_lsn_t |
771 | xfs_buf_item_committed( |
772 | struct xfs_log_item *lip, |
773 | xfs_lsn_t lsn) |
774 | { |
775 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
776 | |
777 | trace_xfs_buf_item_committed(bip); |
778 | |
779 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
780 | return lip->li_lsn; |
781 | return lsn; |
782 | } |
783 | |
784 | static const struct xfs_item_ops xfs_buf_item_ops = { |
785 | .iop_size = xfs_buf_item_size, |
786 | .iop_format = xfs_buf_item_format, |
787 | .iop_pin = xfs_buf_item_pin, |
788 | .iop_unpin = xfs_buf_item_unpin, |
789 | .iop_release = xfs_buf_item_release, |
790 | .iop_committing = xfs_buf_item_committing, |
791 | .iop_committed = xfs_buf_item_committed, |
792 | .iop_push = xfs_buf_item_push, |
793 | }; |
794 | |
795 | STATIC void |
796 | xfs_buf_item_get_format( |
797 | struct xfs_buf_log_item *bip, |
798 | int count) |
799 | { |
800 | ASSERT(bip->bli_formats == NULL); |
801 | bip->bli_format_count = count; |
802 | |
803 | if (count == 1) { |
804 | bip->bli_formats = &bip->__bli_format; |
805 | return; |
806 | } |
807 | |
808 | bip->bli_formats = kzalloc(count * sizeof(struct xfs_buf_log_format), |
809 | GFP_KERNEL | __GFP_NOFAIL); |
810 | } |
811 | |
812 | STATIC void |
813 | xfs_buf_item_free_format( |
814 | struct xfs_buf_log_item *bip) |
815 | { |
816 | if (bip->bli_formats != &bip->__bli_format) { |
817 | kfree(objp: bip->bli_formats); |
818 | bip->bli_formats = NULL; |
819 | } |
820 | } |
821 | |
822 | /* |
823 | * Allocate a new buf log item to go with the given buffer. |
824 | * Set the buffer's b_log_item field to point to the new |
825 | * buf log item. |
826 | */ |
827 | int |
828 | xfs_buf_item_init( |
829 | struct xfs_buf *bp, |
830 | struct xfs_mount *mp) |
831 | { |
832 | struct xfs_buf_log_item *bip = bp->b_log_item; |
833 | int chunks; |
834 | int map_size; |
835 | int i; |
836 | |
837 | /* |
838 | * Check to see if there is already a buf log item for |
839 | * this buffer. If we do already have one, there is |
840 | * nothing to do here so return. |
841 | */ |
842 | ASSERT(bp->b_mount == mp); |
843 | if (bip) { |
844 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
845 | ASSERT(!bp->b_transp); |
846 | ASSERT(bip->bli_buf == bp); |
847 | return 0; |
848 | } |
849 | |
850 | bip = kmem_cache_zalloc(k: xfs_buf_item_cache, GFP_KERNEL | __GFP_NOFAIL); |
851 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
852 | bip->bli_buf = bp; |
853 | |
854 | /* |
855 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer |
856 | * can be divided into. Make sure not to truncate any pieces. |
857 | * map_size is the size of the bitmap needed to describe the |
858 | * chunks of the buffer. |
859 | * |
860 | * Discontiguous buffer support follows the layout of the underlying |
861 | * buffer. This makes the implementation as simple as possible. |
862 | */ |
863 | xfs_buf_item_get_format(bip, count: bp->b_map_count); |
864 | |
865 | for (i = 0; i < bip->bli_format_count; i++) { |
866 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), |
867 | XFS_BLF_CHUNK); |
868 | map_size = DIV_ROUND_UP(chunks, NBWORD); |
869 | |
870 | if (map_size > XFS_BLF_DATAMAP_SIZE) { |
871 | kmem_cache_free(s: xfs_buf_item_cache, objp: bip); |
872 | xfs_err(mp, |
873 | "buffer item dirty bitmap (%u uints) too small to reflect %u bytes!" , |
874 | map_size, |
875 | BBTOB(bp->b_maps[i].bm_len)); |
876 | return -EFSCORRUPTED; |
877 | } |
878 | |
879 | bip->bli_formats[i].blf_type = XFS_LI_BUF; |
880 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; |
881 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; |
882 | bip->bli_formats[i].blf_map_size = map_size; |
883 | } |
884 | |
885 | bp->b_log_item = bip; |
886 | xfs_buf_hold(bp); |
887 | return 0; |
888 | } |
889 | |
890 | |
891 | /* |
892 | * Mark bytes first through last inclusive as dirty in the buf |
893 | * item's bitmap. |
894 | */ |
895 | static void |
896 | xfs_buf_item_log_segment( |
897 | uint first, |
898 | uint last, |
899 | uint *map) |
900 | { |
901 | uint first_bit; |
902 | uint last_bit; |
903 | uint bits_to_set; |
904 | uint bits_set; |
905 | uint word_num; |
906 | uint *wordp; |
907 | uint bit; |
908 | uint end_bit; |
909 | uint mask; |
910 | |
911 | ASSERT(first < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD); |
912 | ASSERT(last < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD); |
913 | |
914 | /* |
915 | * Convert byte offsets to bit numbers. |
916 | */ |
917 | first_bit = first >> XFS_BLF_SHIFT; |
918 | last_bit = last >> XFS_BLF_SHIFT; |
919 | |
920 | /* |
921 | * Calculate the total number of bits to be set. |
922 | */ |
923 | bits_to_set = last_bit - first_bit + 1; |
924 | |
925 | /* |
926 | * Get a pointer to the first word in the bitmap |
927 | * to set a bit in. |
928 | */ |
929 | word_num = first_bit >> BIT_TO_WORD_SHIFT; |
930 | wordp = &map[word_num]; |
931 | |
932 | /* |
933 | * Calculate the starting bit in the first word. |
934 | */ |
935 | bit = first_bit & (uint)(NBWORD - 1); |
936 | |
937 | /* |
938 | * First set any bits in the first word of our range. |
939 | * If it starts at bit 0 of the word, it will be |
940 | * set below rather than here. That is what the variable |
941 | * bit tells us. The variable bits_set tracks the number |
942 | * of bits that have been set so far. End_bit is the number |
943 | * of the last bit to be set in this word plus one. |
944 | */ |
945 | if (bit) { |
946 | end_bit = min(bit + bits_to_set, (uint)NBWORD); |
947 | mask = ((1U << (end_bit - bit)) - 1) << bit; |
948 | *wordp |= mask; |
949 | wordp++; |
950 | bits_set = end_bit - bit; |
951 | } else { |
952 | bits_set = 0; |
953 | } |
954 | |
955 | /* |
956 | * Now set bits a whole word at a time that are between |
957 | * first_bit and last_bit. |
958 | */ |
959 | while ((bits_to_set - bits_set) >= NBWORD) { |
960 | *wordp = 0xffffffff; |
961 | bits_set += NBWORD; |
962 | wordp++; |
963 | } |
964 | |
965 | /* |
966 | * Finally, set any bits left to be set in one last partial word. |
967 | */ |
968 | end_bit = bits_to_set - bits_set; |
969 | if (end_bit) { |
970 | mask = (1U << end_bit) - 1; |
971 | *wordp |= mask; |
972 | } |
973 | } |
974 | |
975 | /* |
976 | * Mark bytes first through last inclusive as dirty in the buf |
977 | * item's bitmap. |
978 | */ |
979 | void |
980 | xfs_buf_item_log( |
981 | struct xfs_buf_log_item *bip, |
982 | uint first, |
983 | uint last) |
984 | { |
985 | int i; |
986 | uint start; |
987 | uint end; |
988 | struct xfs_buf *bp = bip->bli_buf; |
989 | |
990 | /* |
991 | * walk each buffer segment and mark them dirty appropriately. |
992 | */ |
993 | start = 0; |
994 | for (i = 0; i < bip->bli_format_count; i++) { |
995 | if (start > last) |
996 | break; |
997 | end = start + BBTOB(bp->b_maps[i].bm_len) - 1; |
998 | |
999 | /* skip to the map that includes the first byte to log */ |
1000 | if (first > end) { |
1001 | start += BBTOB(bp->b_maps[i].bm_len); |
1002 | continue; |
1003 | } |
1004 | |
1005 | /* |
1006 | * Trim the range to this segment and mark it in the bitmap. |
1007 | * Note that we must convert buffer offsets to segment relative |
1008 | * offsets (e.g., the first byte of each segment is byte 0 of |
1009 | * that segment). |
1010 | */ |
1011 | if (first < start) |
1012 | first = start; |
1013 | if (end > last) |
1014 | end = last; |
1015 | xfs_buf_item_log_segment(first - start, end - start, |
1016 | &bip->bli_formats[i].blf_data_map[0]); |
1017 | |
1018 | start += BBTOB(bp->b_maps[i].bm_len); |
1019 | } |
1020 | } |
1021 | |
1022 | |
1023 | /* |
1024 | * Return true if the buffer has any ranges logged/dirtied by a transaction, |
1025 | * false otherwise. |
1026 | */ |
1027 | bool |
1028 | xfs_buf_item_dirty_format( |
1029 | struct xfs_buf_log_item *bip) |
1030 | { |
1031 | int i; |
1032 | |
1033 | for (i = 0; i < bip->bli_format_count; i++) { |
1034 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, |
1035 | bip->bli_formats[i].blf_map_size)) |
1036 | return true; |
1037 | } |
1038 | |
1039 | return false; |
1040 | } |
1041 | |
1042 | STATIC void |
1043 | xfs_buf_item_free( |
1044 | struct xfs_buf_log_item *bip) |
1045 | { |
1046 | xfs_buf_item_free_format(bip); |
1047 | kvfree(addr: bip->bli_item.li_lv_shadow); |
1048 | kmem_cache_free(s: xfs_buf_item_cache, objp: bip); |
1049 | } |
1050 | |
1051 | /* |
1052 | * xfs_buf_item_relse() is called when the buf log item is no longer needed. |
1053 | */ |
1054 | void |
1055 | xfs_buf_item_relse( |
1056 | struct xfs_buf *bp) |
1057 | { |
1058 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1059 | |
1060 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
1061 | ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags)); |
1062 | |
1063 | if (atomic_read(v: &bip->bli_refcount)) |
1064 | return; |
1065 | bp->b_log_item = NULL; |
1066 | xfs_buf_rele(bp); |
1067 | xfs_buf_item_free(bip); |
1068 | } |
1069 | |
1070 | void |
1071 | xfs_buf_item_done( |
1072 | struct xfs_buf *bp) |
1073 | { |
1074 | /* |
1075 | * If we are forcibly shutting down, this may well be off the AIL |
1076 | * already. That's because we simulate the log-committed callbacks to |
1077 | * unpin these buffers. Or we may never have put this item on AIL |
1078 | * because of the transaction was aborted forcibly. |
1079 | * xfs_trans_ail_delete() takes care of these. |
1080 | * |
1081 | * Either way, AIL is useless if we're forcing a shutdown. |
1082 | * |
1083 | * Note that log recovery writes might have buffer items that are not on |
1084 | * the AIL even when the file system is not shut down. |
1085 | */ |
1086 | xfs_trans_ail_delete(lip: &bp->b_log_item->bli_item, |
1087 | shutdown_type: (bp->b_flags & _XBF_LOGRECOVERY) ? 0 : |
1088 | SHUTDOWN_CORRUPT_INCORE); |
1089 | xfs_buf_item_relse(bp); |
1090 | } |
1091 | |