Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
5467b34b | 8 | #include "xfs_shared.h" |
4fb6e8ad | 9 | #include "xfs_format.h" |
239880ef DC |
10 | #include "xfs_log_format.h" |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_mount.h" |
239880ef | 14 | #include "xfs_trans.h" |
3536b61e | 15 | #include "xfs_trans_priv.h" |
a844f451 | 16 | #include "xfs_buf_item.h" |
aac855ab DC |
17 | #include "xfs_inode.h" |
18 | #include "xfs_inode_item.h" | |
6f5de180 DC |
19 | #include "xfs_quota.h" |
20 | #include "xfs_dquot_item.h" | |
21 | #include "xfs_dquot.h" | |
0b1b213f | 22 | #include "xfs_trace.h" |
239880ef | 23 | #include "xfs_log.h" |
d86142dd | 24 | #include "xfs_log_priv.h" |
1da177e4 LT |
25 | |
26 | ||
182696fb | 27 | struct kmem_cache *xfs_buf_item_cache; |
1da177e4 | 28 | |
7bfa31d8 CH |
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 | ||
8a6453a8 DW |
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 | ||
166d1368 DC |
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 | ||
c81ea11e DC |
59 | static inline bool |
60 | xfs_buf_item_straddle( | |
61 | struct xfs_buf *bp, | |
62 | uint offset, | |
929f8b0d DC |
63 | int first_bit, |
64 | int nbits) | |
c81ea11e | 65 | { |
929f8b0d DC |
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; | |
c81ea11e DC |
75 | } |
76 | ||
1da177e4 | 77 | /* |
19f4e7cc DC |
78 | * Return the number of log iovecs and space needed to log the given buf log |
79 | * item segment. | |
1da177e4 | 80 | * |
19f4e7cc DC |
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. | |
1da177e4 | 84 | */ |
166d1368 | 85 | STATIC void |
372cc85e | 86 | xfs_buf_item_size_segment( |
70a20655 CM |
87 | struct xfs_buf_log_item *bip, |
88 | struct xfs_buf_log_format *blfp, | |
c81ea11e | 89 | uint offset, |
70a20655 CM |
90 | int *nvecs, |
91 | int *nbytes) | |
1da177e4 | 92 | { |
70a20655 | 93 | struct xfs_buf *bp = bip->bli_buf; |
929f8b0d DC |
94 | int first_bit; |
95 | int nbits; | |
70a20655 CM |
96 | int next_bit; |
97 | int last_bit; | |
1da177e4 | 98 | |
929f8b0d DC |
99 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
100 | if (first_bit == -1) | |
166d1368 | 101 | return; |
372cc85e | 102 | |
929f8b0d DC |
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); | |
1da177e4 | 131 | |
929f8b0d DC |
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; | |
1da177e4 LT |
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 | */ | |
372cc85e DC |
146 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
147 | last_bit + 1); | |
1da177e4 LT |
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) { | |
372cc85e | 154 | break; |
c81ea11e | 155 | } else if (next_bit != last_bit + 1 || |
929f8b0d | 156 | xfs_buf_item_straddle(bp, offset, first_bit, nbits)) { |
1da177e4 | 157 | last_bit = next_bit; |
929f8b0d | 158 | first_bit = next_bit; |
166d1368 | 159 | (*nvecs)++; |
929f8b0d | 160 | nbits = 1; |
1da177e4 LT |
161 | } else { |
162 | last_bit++; | |
929f8b0d | 163 | nbits++; |
1da177e4 | 164 | } |
166d1368 | 165 | *nbytes += XFS_BLF_CHUNK; |
1da177e4 | 166 | } |
1da177e4 LT |
167 | } |
168 | ||
169 | /* | |
19f4e7cc DC |
170 | * Return the number of log iovecs and space needed to log the given buf log |
171 | * item. | |
372cc85e | 172 | * |
b63da6c8 | 173 | * Discontiguous buffers need a format structure per region that is being |
372cc85e DC |
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 | |
19f4e7cc DC |
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. | |
1da177e4 | 186 | */ |
166d1368 | 187 | STATIC void |
372cc85e | 188 | xfs_buf_item_size( |
166d1368 DC |
189 | struct xfs_log_item *lip, |
190 | int *nvecs, | |
191 | int *nbytes) | |
1da177e4 | 192 | { |
7bfa31d8 | 193 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
c81ea11e | 194 | struct xfs_buf *bp = bip->bli_buf; |
372cc85e | 195 | int i; |
accc661b | 196 | int bytes; |
c81ea11e | 197 | uint offset = 0; |
372cc85e DC |
198 | |
199 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
200 | if (bip->bli_flags & XFS_BLI_STALE) { | |
201 | /* | |
19f4e7cc DC |
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. | |
372cc85e DC |
205 | */ |
206 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 207 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
166d1368 DC |
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; | |
372cc85e DC |
213 | } |
214 | ||
215 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
216 | ||
5f6bed76 DC |
217 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
218 | /* | |
19f4e7cc DC |
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. | |
5f6bed76 DC |
222 | */ |
223 | trace_xfs_buf_item_size_ordered(bip); | |
166d1368 DC |
224 | *nvecs = XFS_LOG_VEC_ORDERED; |
225 | return; | |
5f6bed76 DC |
226 | } |
227 | ||
372cc85e | 228 | /* |
accc661b | 229 | * The vector count is based on the number of buffer vectors we have |
372cc85e DC |
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 | */ | |
accc661b | 237 | bytes = 0; |
372cc85e | 238 | for (i = 0; i < bip->bli_format_count; i++) { |
c81ea11e | 239 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], offset, |
accc661b | 240 | nvecs, &bytes); |
c81ea11e | 241 | offset += BBTOB(bp->b_maps[i].bm_len); |
372cc85e | 242 | } |
accc661b DC |
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); | |
372cc85e | 250 | trace_xfs_buf_item_size(bip); |
372cc85e DC |
251 | } |
252 | ||
1234351c | 253 | static inline void |
7aeb7222 | 254 | xfs_buf_item_copy_iovec( |
bde7cff6 | 255 | struct xfs_log_vec *lv, |
1234351c | 256 | struct xfs_log_iovec **vecp, |
7aeb7222 CH |
257 | struct xfs_buf *bp, |
258 | uint offset, | |
259 | int first_bit, | |
260 | uint nbits) | |
261 | { | |
262 | offset += first_bit * XFS_BLF_CHUNK; | |
bde7cff6 | 263 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
1234351c CH |
264 | xfs_buf_offset(bp, offset), |
265 | nbits * XFS_BLF_CHUNK); | |
7aeb7222 CH |
266 | } |
267 | ||
1234351c | 268 | static void |
372cc85e DC |
269 | xfs_buf_item_format_segment( |
270 | struct xfs_buf_log_item *bip, | |
bde7cff6 | 271 | struct xfs_log_vec *lv, |
1234351c | 272 | struct xfs_log_iovec **vecp, |
372cc85e DC |
273 | uint offset, |
274 | struct xfs_buf_log_format *blfp) | |
275 | { | |
70a20655 CM |
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; | |
1da177e4 | 282 | |
372cc85e | 283 | /* copy the flags across from the base format item */ |
b9438173 | 284 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
285 | |
286 | /* | |
77c1a08f DC |
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. | |
1da177e4 | 290 | */ |
166d1368 | 291 | base_size = xfs_buf_log_format_size(blfp); |
820a554f | 292 | |
820a554f MT |
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 | */ | |
bde7cff6 | 299 | return; |
820a554f MT |
300 | } |
301 | ||
bde7cff6 CH |
302 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
303 | blfp->blf_size = 1; | |
1da177e4 LT |
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 | */ | |
0b1b213f | 311 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 312 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
bde7cff6 | 313 | return; |
1da177e4 LT |
314 | } |
315 | ||
5f6bed76 | 316 | |
1da177e4 LT |
317 | /* |
318 | * Fill in an iovec for each set of contiguous chunks. | |
319 | */ | |
929f8b0d DC |
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); | |
1da177e4 LT |
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 | */ | |
372cc85e DC |
361 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
362 | (uint)last_bit + 1); | |
1da177e4 | 363 | /* |
7aeb7222 CH |
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. | |
1da177e4 LT |
369 | */ |
370 | if (next_bit == -1) { | |
bde7cff6 | 371 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
7aeb7222 | 372 | first_bit, nbits); |
bde7cff6 | 373 | blfp->blf_size++; |
1da177e4 | 374 | break; |
7aeb7222 | 375 | } else if (next_bit != last_bit + 1 || |
929f8b0d | 376 | xfs_buf_item_straddle(bp, offset, first_bit, nbits)) { |
bde7cff6 | 377 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
1234351c | 378 | first_bit, nbits); |
bde7cff6 | 379 | blfp->blf_size++; |
1da177e4 LT |
380 | first_bit = next_bit; |
381 | last_bit = next_bit; | |
382 | nbits = 1; | |
383 | } else { | |
384 | last_bit++; | |
385 | nbits++; | |
386 | } | |
387 | } | |
372cc85e DC |
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, | |
bde7cff6 | 399 | struct xfs_log_vec *lv) |
372cc85e DC |
400 | { |
401 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
402 | struct xfs_buf *bp = bip->bli_buf; | |
bde7cff6 | 403 | struct xfs_log_iovec *vecp = NULL; |
372cc85e DC |
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)); | |
0d612fb5 DC |
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)); | |
e9385cc6 BF |
413 | ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) || |
414 | (bip->bli_flags & XFS_BLI_STALE)); | |
0d612fb5 | 415 | |
372cc85e DC |
416 | |
417 | /* | |
418 | * If it is an inode buffer, transfer the in-memory state to the | |
ddf6ad01 DC |
419 | * format flags and clear the in-memory state. |
420 | * | |
421 | * For buffer based inode allocation, we do not transfer | |
372cc85e DC |
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. | |
ddf6ad01 DC |
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. | |
372cc85e DC |
430 | */ |
431 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
d86142dd | 432 | if (xfs_has_v3inodes(lip->li_log->l_mp) || |
ddf6ad01 | 433 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
372cc85e | 434 | xfs_log_item_in_current_chkpt(lip))) |
b9438173 | 435 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
436 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
437 | } | |
438 | ||
439 | for (i = 0; i < bip->bli_format_count; i++) { | |
bde7cff6 | 440 | xfs_buf_item_format_segment(bip, lv, &vecp, offset, |
1234351c | 441 | &bip->bli_formats[i]); |
a3916e52 | 442 | offset += BBTOB(bp->b_maps[i].bm_len); |
372cc85e | 443 | } |
1da177e4 LT |
444 | |
445 | /* | |
446 | * Check to make sure everything is consistent. | |
447 | */ | |
0b1b213f | 448 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
449 | } |
450 | ||
451 | /* | |
64fc35de | 452 | * This is called to pin the buffer associated with the buf log item in memory |
4d16e924 | 453 | * so it cannot be written out. |
64fc35de DC |
454 | * |
455 | * We also always take a reference to the buffer log item here so that the bli | |
456 | * is held while the item is pinned in memory. This means that we can | |
457 | * unconditionally drop the reference count a transaction holds when the | |
458 | * transaction is completed. | |
1da177e4 | 459 | */ |
ba0f32d4 | 460 | STATIC void |
1da177e4 | 461 | xfs_buf_item_pin( |
7bfa31d8 | 462 | struct xfs_log_item *lip) |
1da177e4 | 463 | { |
7bfa31d8 | 464 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
1da177e4 | 465 | |
1da177e4 LT |
466 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
467 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
5f6bed76 | 468 | (bip->bli_flags & XFS_BLI_ORDERED) || |
1da177e4 | 469 | (bip->bli_flags & XFS_BLI_STALE)); |
7bfa31d8 | 470 | |
0b1b213f | 471 | trace_xfs_buf_item_pin(bip); |
4d16e924 CH |
472 | |
473 | atomic_inc(&bip->bli_refcount); | |
474 | atomic_inc(&bip->bli_buf->b_pin_count); | |
1da177e4 LT |
475 | } |
476 | ||
1da177e4 | 477 | /* |
84d8949e BF |
478 | * This is called to unpin the buffer associated with the buf log item which |
479 | * was previously pinned with a call to xfs_buf_item_pin(). | |
1da177e4 | 480 | */ |
ba0f32d4 | 481 | STATIC void |
1da177e4 | 482 | xfs_buf_item_unpin( |
7bfa31d8 | 483 | struct xfs_log_item *lip, |
9412e318 | 484 | int remove) |
1da177e4 | 485 | { |
7bfa31d8 | 486 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
e8222613 | 487 | struct xfs_buf *bp = bip->bli_buf; |
70a20655 CM |
488 | int stale = bip->bli_flags & XFS_BLI_STALE; |
489 | int freed; | |
1da177e4 | 490 | |
fb1755a6 | 491 | ASSERT(bp->b_log_item == bip); |
1da177e4 | 492 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
9412e318 | 493 | |
0b1b213f | 494 | trace_xfs_buf_item_unpin(bip); |
1da177e4 | 495 | |
84d8949e BF |
496 | /* |
497 | * Drop the bli ref associated with the pin and grab the hold required | |
498 | * for the I/O simulation failure in the abort case. We have to do this | |
499 | * before the pin count drops because the AIL doesn't acquire a bli | |
500 | * reference. Therefore if the refcount drops to zero, the bli could | |
501 | * still be AIL resident and the buffer submitted for I/O (and freed on | |
502 | * completion) at any point before we return. This can be removed once | |
503 | * the AIL properly holds a reference on the bli. | |
504 | */ | |
1da177e4 | 505 | freed = atomic_dec_and_test(&bip->bli_refcount); |
84d8949e BF |
506 | if (freed && !stale && remove) |
507 | xfs_buf_hold(bp); | |
4d16e924 CH |
508 | if (atomic_dec_and_test(&bp->b_pin_count)) |
509 | wake_up_all(&bp->b_waiters); | |
7bfa31d8 | 510 | |
84d8949e BF |
511 | /* nothing to do but drop the pin count if the bli is active */ |
512 | if (!freed) | |
513 | return; | |
514 | ||
515 | if (stale) { | |
1da177e4 | 516 | ASSERT(bip->bli_flags & XFS_BLI_STALE); |
0c842ad4 | 517 | ASSERT(xfs_buf_islocked(bp)); |
5cfd28b6 | 518 | ASSERT(bp->b_flags & XBF_STALE); |
b9438173 | 519 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
e53d3aa0 BF |
520 | ASSERT(list_empty(&lip->li_trans)); |
521 | ASSERT(!bp->b_transp); | |
9412e318 | 522 | |
0b1b213f CH |
523 | trace_xfs_buf_item_unpin_stale(bip); |
524 | ||
1da177e4 | 525 | /* |
849274c1 BF |
526 | * If we get called here because of an IO error, we may or may |
527 | * not have the item on the AIL. xfs_trans_ail_delete() will | |
528 | * take care of that situation. xfs_trans_ail_delete() drops | |
529 | * the AIL lock. | |
1da177e4 LT |
530 | */ |
531 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
fec671cd | 532 | xfs_buf_item_done(bp); |
664ffb8a | 533 | xfs_buf_inode_iodone(bp); |
48d55e2a | 534 | ASSERT(list_empty(&bp->b_li_list)); |
1da177e4 | 535 | } else { |
849274c1 | 536 | xfs_trans_ail_delete(lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 537 | xfs_buf_item_relse(bp); |
fb1755a6 | 538 | ASSERT(bp->b_log_item == NULL); |
1da177e4 LT |
539 | } |
540 | xfs_buf_relse(bp); | |
84d8949e | 541 | } else if (remove) { |
137fff09 | 542 | /* |
54b3b1f6 | 543 | * The buffer must be locked and held by the caller to simulate |
84d8949e BF |
544 | * an async I/O failure. We acquired the hold for this case |
545 | * before the buffer was unpinned. | |
137fff09 | 546 | */ |
960c60af | 547 | xfs_buf_lock(bp); |
137fff09 | 548 | bp->b_flags |= XBF_ASYNC; |
54b3b1f6 | 549 | xfs_buf_ioend_fail(bp); |
1da177e4 LT |
550 | } |
551 | } | |
552 | ||
ba0f32d4 | 553 | STATIC uint |
43ff2122 CH |
554 | xfs_buf_item_push( |
555 | struct xfs_log_item *lip, | |
556 | struct list_head *buffer_list) | |
1da177e4 | 557 | { |
7bfa31d8 CH |
558 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
559 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 560 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 561 | |
811e64c7 | 562 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 563 | return XFS_ITEM_PINNED; |
5337fe9b BF |
564 | if (!xfs_buf_trylock(bp)) { |
565 | /* | |
566 | * If we have just raced with a buffer being pinned and it has | |
567 | * been marked stale, we could end up stalling until someone else | |
568 | * issues a log force to unpin the stale buffer. Check for the | |
569 | * race condition here so xfsaild recognizes the buffer is pinned | |
570 | * and queues a log force to move it along. | |
571 | */ | |
572 | if (xfs_buf_ispinned(bp)) | |
573 | return XFS_ITEM_PINNED; | |
1da177e4 | 574 | return XFS_ITEM_LOCKED; |
5337fe9b | 575 | } |
1da177e4 | 576 | |
1da177e4 | 577 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
578 | |
579 | trace_xfs_buf_item_push(bip); | |
580 | ||
ac8809f9 | 581 | /* has a previous flush failed due to IO errors? */ |
f9bccfcc BF |
582 | if (bp->b_flags & XBF_WRITE_FAIL) { |
583 | xfs_buf_alert_ratelimited(bp, "XFS: Failing async write", | |
584 | "Failing async write on buffer block 0x%llx. Retrying async write.", | |
9343ee76 | 585 | (long long)xfs_buf_daddr(bp)); |
ac8809f9 DC |
586 | } |
587 | ||
43ff2122 CH |
588 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
589 | rval = XFS_ITEM_FLUSHING; | |
590 | xfs_buf_unlock(bp); | |
591 | return rval; | |
1da177e4 LT |
592 | } |
593 | ||
95808459 BF |
594 | /* |
595 | * Drop the buffer log item refcount and take appropriate action. This helper | |
596 | * determines whether the bli must be freed or not, since a decrement to zero | |
597 | * does not necessarily mean the bli is unused. | |
598 | * | |
599 | * Return true if the bli is freed, false otherwise. | |
600 | */ | |
601 | bool | |
602 | xfs_buf_item_put( | |
603 | struct xfs_buf_log_item *bip) | |
604 | { | |
605 | struct xfs_log_item *lip = &bip->bli_item; | |
606 | bool aborted; | |
607 | bool dirty; | |
608 | ||
609 | /* drop the bli ref and return if it wasn't the last one */ | |
610 | if (!atomic_dec_and_test(&bip->bli_refcount)) | |
611 | return false; | |
612 | ||
613 | /* | |
614 | * We dropped the last ref and must free the item if clean or aborted. | |
615 | * If the bli is dirty and non-aborted, the buffer was clean in the | |
616 | * transaction but still awaiting writeback from previous changes. In | |
617 | * that case, the bli is freed on buffer writeback completion. | |
618 | */ | |
619 | aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) || | |
d86142dd | 620 | xlog_is_shutdown(lip->li_log); |
95808459 BF |
621 | dirty = bip->bli_flags & XFS_BLI_DIRTY; |
622 | if (dirty && !aborted) | |
623 | return false; | |
624 | ||
625 | /* | |
626 | * The bli is aborted or clean. An aborted item may be in the AIL | |
627 | * regardless of dirty state. For example, consider an aborted | |
628 | * transaction that invalidated a dirty bli and cleared the dirty | |
629 | * state. | |
630 | */ | |
631 | if (aborted) | |
2b3cf093 | 632 | xfs_trans_ail_delete(lip, 0); |
95808459 BF |
633 | xfs_buf_item_relse(bip->bli_buf); |
634 | return true; | |
635 | } | |
636 | ||
1da177e4 | 637 | /* |
64fc35de DC |
638 | * Release the buffer associated with the buf log item. If there is no dirty |
639 | * logged data associated with the buffer recorded in the buf log item, then | |
640 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 641 | * |
64fc35de DC |
642 | * This call ignores the recursion count. It is only called when the buffer |
643 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 644 | * |
64fc35de DC |
645 | * We unconditionally drop the transaction's reference to the log item. If the |
646 | * item was logged, then another reference was taken when it was pinned, so we | |
647 | * can safely drop the transaction reference now. This also allows us to avoid | |
648 | * potential races with the unpin code freeing the bli by not referencing the | |
649 | * bli after we've dropped the reference count. | |
650 | * | |
651 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
652 | * if necessary but do not unlock the buffer. This is for support of | |
653 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
654 | * free the item. | |
1da177e4 | 655 | */ |
ba0f32d4 | 656 | STATIC void |
ddf92053 | 657 | xfs_buf_item_release( |
7bfa31d8 | 658 | struct xfs_log_item *lip) |
1da177e4 | 659 | { |
7bfa31d8 CH |
660 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
661 | struct xfs_buf *bp = bip->bli_buf; | |
95808459 | 662 | bool released; |
d9183105 | 663 | bool hold = bip->bli_flags & XFS_BLI_HOLD; |
d9183105 | 664 | bool stale = bip->bli_flags & XFS_BLI_STALE; |
7bf7a193 | 665 | #if defined(DEBUG) || defined(XFS_WARN) |
d9183105 | 666 | bool ordered = bip->bli_flags & XFS_BLI_ORDERED; |
95808459 | 667 | bool dirty = bip->bli_flags & XFS_BLI_DIRTY; |
4d09807f BF |
668 | bool aborted = test_bit(XFS_LI_ABORTED, |
669 | &lip->li_flags); | |
7bf7a193 | 670 | #endif |
1da177e4 | 671 | |
ddf92053 | 672 | trace_xfs_buf_item_release(bip); |
1da177e4 LT |
673 | |
674 | /* | |
6453c65d BF |
675 | * The bli dirty state should match whether the blf has logged segments |
676 | * except for ordered buffers, where only the bli should be dirty. | |
1da177e4 | 677 | */ |
6453c65d BF |
678 | ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) || |
679 | (ordered && dirty && !xfs_buf_item_dirty_format(bip))); | |
d9183105 BF |
680 | ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
681 | ||
46f9d2eb | 682 | /* |
d9183105 BF |
683 | * Clear the buffer's association with this transaction and |
684 | * per-transaction state from the bli, which has been copied above. | |
685 | */ | |
686 | bp->b_transp = NULL; | |
687 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); | |
688 | ||
689 | /* | |
95808459 BF |
690 | * Unref the item and unlock the buffer unless held or stale. Stale |
691 | * buffers remain locked until final unpin unless the bli is freed by | |
692 | * the unref call. The latter implies shutdown because buffer | |
693 | * invalidation dirties the bli and transaction. | |
46f9d2eb | 694 | */ |
95808459 BF |
695 | released = xfs_buf_item_put(bip); |
696 | if (hold || (stale && !released)) | |
d9183105 | 697 | return; |
4d09807f | 698 | ASSERT(!stale || aborted); |
95808459 | 699 | xfs_buf_relse(bp); |
1da177e4 LT |
700 | } |
701 | ||
ddf92053 CH |
702 | STATIC void |
703 | xfs_buf_item_committing( | |
704 | struct xfs_log_item *lip, | |
5f9b4b0d | 705 | xfs_csn_t seq) |
ddf92053 CH |
706 | { |
707 | return xfs_buf_item_release(lip); | |
708 | } | |
709 | ||
1da177e4 LT |
710 | /* |
711 | * This is called to find out where the oldest active copy of the | |
712 | * buf log item in the on disk log resides now that the last log | |
713 | * write of it completed at the given lsn. | |
714 | * We always re-log all the dirty data in a buffer, so usually the | |
715 | * latest copy in the on disk log is the only one that matters. For | |
716 | * those cases we simply return the given lsn. | |
717 | * | |
718 | * The one exception to this is for buffers full of newly allocated | |
719 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
720 | * flag set, indicating that only the di_next_unlinked fields from the | |
721 | * inodes in the buffers will be replayed during recovery. If the | |
722 | * original newly allocated inode images have not yet been flushed | |
723 | * when the buffer is so relogged, then we need to make sure that we | |
724 | * keep the old images in the 'active' portion of the log. We do this | |
725 | * by returning the original lsn of that transaction here rather than | |
726 | * the current one. | |
727 | */ | |
ba0f32d4 | 728 | STATIC xfs_lsn_t |
1da177e4 | 729 | xfs_buf_item_committed( |
7bfa31d8 | 730 | struct xfs_log_item *lip, |
1da177e4 LT |
731 | xfs_lsn_t lsn) |
732 | { | |
7bfa31d8 CH |
733 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
734 | ||
0b1b213f CH |
735 | trace_xfs_buf_item_committed(bip); |
736 | ||
7bfa31d8 CH |
737 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
738 | return lip->li_lsn; | |
739 | return lsn; | |
1da177e4 LT |
740 | } |
741 | ||
272e42b2 | 742 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
743 | .iop_size = xfs_buf_item_size, |
744 | .iop_format = xfs_buf_item_format, | |
745 | .iop_pin = xfs_buf_item_pin, | |
746 | .iop_unpin = xfs_buf_item_unpin, | |
ddf92053 CH |
747 | .iop_release = xfs_buf_item_release, |
748 | .iop_committing = xfs_buf_item_committing, | |
7bfa31d8 CH |
749 | .iop_committed = xfs_buf_item_committed, |
750 | .iop_push = xfs_buf_item_push, | |
1da177e4 LT |
751 | }; |
752 | ||
c64dd49b | 753 | STATIC void |
372cc85e DC |
754 | xfs_buf_item_get_format( |
755 | struct xfs_buf_log_item *bip, | |
756 | int count) | |
757 | { | |
758 | ASSERT(bip->bli_formats == NULL); | |
759 | bip->bli_format_count = count; | |
760 | ||
761 | if (count == 1) { | |
b9438173 | 762 | bip->bli_formats = &bip->__bli_format; |
c64dd49b | 763 | return; |
372cc85e DC |
764 | } |
765 | ||
766 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
707e0dda | 767 | 0); |
372cc85e DC |
768 | } |
769 | ||
770 | STATIC void | |
771 | xfs_buf_item_free_format( | |
772 | struct xfs_buf_log_item *bip) | |
773 | { | |
b9438173 | 774 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
775 | kmem_free(bip->bli_formats); |
776 | bip->bli_formats = NULL; | |
777 | } | |
778 | } | |
1da177e4 LT |
779 | |
780 | /* | |
781 | * Allocate a new buf log item to go with the given buffer. | |
fb1755a6 CM |
782 | * Set the buffer's b_log_item field to point to the new |
783 | * buf log item. | |
1da177e4 | 784 | */ |
f79af0b9 | 785 | int |
1da177e4 | 786 | xfs_buf_item_init( |
f79af0b9 DC |
787 | struct xfs_buf *bp, |
788 | struct xfs_mount *mp) | |
1da177e4 | 789 | { |
fb1755a6 | 790 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1da177e4 LT |
791 | int chunks; |
792 | int map_size; | |
372cc85e | 793 | int i; |
1da177e4 LT |
794 | |
795 | /* | |
796 | * Check to see if there is already a buf log item for | |
fb1755a6 | 797 | * this buffer. If we do already have one, there is |
1da177e4 LT |
798 | * nothing to do here so return. |
799 | */ | |
dbd329f1 | 800 | ASSERT(bp->b_mount == mp); |
1a2ebf83 | 801 | if (bip) { |
fb1755a6 | 802 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
1a2ebf83 DC |
803 | ASSERT(!bp->b_transp); |
804 | ASSERT(bip->bli_buf == bp); | |
f79af0b9 | 805 | return 0; |
fb1755a6 | 806 | } |
1da177e4 | 807 | |
182696fb | 808 | bip = kmem_cache_zalloc(xfs_buf_item_cache, GFP_KERNEL | __GFP_NOFAIL); |
43f5efc5 | 809 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 810 | bip->bli_buf = bp; |
372cc85e DC |
811 | |
812 | /* | |
813 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
814 | * can be divided into. Make sure not to truncate any pieces. | |
815 | * map_size is the size of the bitmap needed to describe the | |
816 | * chunks of the buffer. | |
817 | * | |
818 | * Discontiguous buffer support follows the layout of the underlying | |
819 | * buffer. This makes the implementation as simple as possible. | |
820 | */ | |
c64dd49b | 821 | xfs_buf_item_get_format(bip, bp->b_map_count); |
372cc85e DC |
822 | |
823 | for (i = 0; i < bip->bli_format_count; i++) { | |
824 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
825 | XFS_BLF_CHUNK); | |
826 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
827 | ||
c3d5f0c2 | 828 | if (map_size > XFS_BLF_DATAMAP_SIZE) { |
182696fb | 829 | kmem_cache_free(xfs_buf_item_cache, bip); |
c3d5f0c2 DW |
830 | xfs_err(mp, |
831 | "buffer item dirty bitmap (%u uints) too small to reflect %u bytes!", | |
832 | map_size, | |
833 | BBTOB(bp->b_maps[i].bm_len)); | |
834 | return -EFSCORRUPTED; | |
835 | } | |
836 | ||
372cc85e DC |
837 | bip->bli_formats[i].blf_type = XFS_LI_BUF; |
838 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
839 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
840 | bip->bli_formats[i].blf_map_size = map_size; | |
841 | } | |
1da177e4 | 842 | |
fb1755a6 | 843 | bp->b_log_item = bip; |
f79af0b9 DC |
844 | xfs_buf_hold(bp); |
845 | return 0; | |
1da177e4 LT |
846 | } |
847 | ||
848 | ||
849 | /* | |
850 | * Mark bytes first through last inclusive as dirty in the buf | |
851 | * item's bitmap. | |
852 | */ | |
632b89e8 | 853 | static void |
372cc85e | 854 | xfs_buf_item_log_segment( |
1da177e4 | 855 | uint first, |
372cc85e DC |
856 | uint last, |
857 | uint *map) | |
1da177e4 LT |
858 | { |
859 | uint first_bit; | |
860 | uint last_bit; | |
861 | uint bits_to_set; | |
862 | uint bits_set; | |
863 | uint word_num; | |
864 | uint *wordp; | |
865 | uint bit; | |
866 | uint end_bit; | |
867 | uint mask; | |
868 | ||
c3d5f0c2 DW |
869 | ASSERT(first < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD); |
870 | ASSERT(last < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD); | |
871 | ||
1da177e4 LT |
872 | /* |
873 | * Convert byte offsets to bit numbers. | |
874 | */ | |
c1155410 DC |
875 | first_bit = first >> XFS_BLF_SHIFT; |
876 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
877 | |
878 | /* | |
879 | * Calculate the total number of bits to be set. | |
880 | */ | |
881 | bits_to_set = last_bit - first_bit + 1; | |
882 | ||
883 | /* | |
884 | * Get a pointer to the first word in the bitmap | |
885 | * to set a bit in. | |
886 | */ | |
887 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 888 | wordp = &map[word_num]; |
1da177e4 LT |
889 | |
890 | /* | |
891 | * Calculate the starting bit in the first word. | |
892 | */ | |
893 | bit = first_bit & (uint)(NBWORD - 1); | |
894 | ||
895 | /* | |
896 | * First set any bits in the first word of our range. | |
897 | * If it starts at bit 0 of the word, it will be | |
898 | * set below rather than here. That is what the variable | |
899 | * bit tells us. The variable bits_set tracks the number | |
900 | * of bits that have been set so far. End_bit is the number | |
901 | * of the last bit to be set in this word plus one. | |
902 | */ | |
903 | if (bit) { | |
9bb54cb5 | 904 | end_bit = min(bit + bits_to_set, (uint)NBWORD); |
79c350e4 | 905 | mask = ((1U << (end_bit - bit)) - 1) << bit; |
1da177e4 LT |
906 | *wordp |= mask; |
907 | wordp++; | |
908 | bits_set = end_bit - bit; | |
909 | } else { | |
910 | bits_set = 0; | |
911 | } | |
912 | ||
913 | /* | |
914 | * Now set bits a whole word at a time that are between | |
915 | * first_bit and last_bit. | |
916 | */ | |
917 | while ((bits_to_set - bits_set) >= NBWORD) { | |
12025460 | 918 | *wordp = 0xffffffff; |
1da177e4 LT |
919 | bits_set += NBWORD; |
920 | wordp++; | |
921 | } | |
922 | ||
923 | /* | |
924 | * Finally, set any bits left to be set in one last partial word. | |
925 | */ | |
926 | end_bit = bits_to_set - bits_set; | |
927 | if (end_bit) { | |
79c350e4 | 928 | mask = (1U << end_bit) - 1; |
1da177e4 LT |
929 | *wordp |= mask; |
930 | } | |
1da177e4 LT |
931 | } |
932 | ||
372cc85e DC |
933 | /* |
934 | * Mark bytes first through last inclusive as dirty in the buf | |
935 | * item's bitmap. | |
936 | */ | |
937 | void | |
938 | xfs_buf_item_log( | |
70a20655 | 939 | struct xfs_buf_log_item *bip, |
372cc85e DC |
940 | uint first, |
941 | uint last) | |
942 | { | |
943 | int i; | |
944 | uint start; | |
945 | uint end; | |
946 | struct xfs_buf *bp = bip->bli_buf; | |
947 | ||
372cc85e DC |
948 | /* |
949 | * walk each buffer segment and mark them dirty appropriately. | |
950 | */ | |
951 | start = 0; | |
952 | for (i = 0; i < bip->bli_format_count; i++) { | |
953 | if (start > last) | |
954 | break; | |
a3916e52 BF |
955 | end = start + BBTOB(bp->b_maps[i].bm_len) - 1; |
956 | ||
957 | /* skip to the map that includes the first byte to log */ | |
372cc85e DC |
958 | if (first > end) { |
959 | start += BBTOB(bp->b_maps[i].bm_len); | |
960 | continue; | |
961 | } | |
a3916e52 BF |
962 | |
963 | /* | |
964 | * Trim the range to this segment and mark it in the bitmap. | |
965 | * Note that we must convert buffer offsets to segment relative | |
966 | * offsets (e.g., the first byte of each segment is byte 0 of | |
967 | * that segment). | |
968 | */ | |
372cc85e DC |
969 | if (first < start) |
970 | first = start; | |
971 | if (end > last) | |
972 | end = last; | |
a3916e52 | 973 | xfs_buf_item_log_segment(first - start, end - start, |
372cc85e DC |
974 | &bip->bli_formats[i].blf_data_map[0]); |
975 | ||
a3916e52 | 976 | start += BBTOB(bp->b_maps[i].bm_len); |
372cc85e DC |
977 | } |
978 | } | |
979 | ||
1da177e4 | 980 | |
6453c65d BF |
981 | /* |
982 | * Return true if the buffer has any ranges logged/dirtied by a transaction, | |
983 | * false otherwise. | |
984 | */ | |
985 | bool | |
986 | xfs_buf_item_dirty_format( | |
987 | struct xfs_buf_log_item *bip) | |
988 | { | |
989 | int i; | |
990 | ||
991 | for (i = 0; i < bip->bli_format_count; i++) { | |
992 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
993 | bip->bli_formats[i].blf_map_size)) | |
994 | return true; | |
995 | } | |
996 | ||
997 | return false; | |
998 | } | |
999 | ||
e1f5dbd7 LM |
1000 | STATIC void |
1001 | xfs_buf_item_free( | |
70a20655 | 1002 | struct xfs_buf_log_item *bip) |
e1f5dbd7 | 1003 | { |
372cc85e | 1004 | xfs_buf_item_free_format(bip); |
b1c5ebb2 | 1005 | kmem_free(bip->bli_item.li_lv_shadow); |
182696fb | 1006 | kmem_cache_free(xfs_buf_item_cache, bip); |
e1f5dbd7 LM |
1007 | } |
1008 | ||
1da177e4 | 1009 | /* |
b01d1461 | 1010 | * xfs_buf_item_relse() is called when the buf log item is no longer needed. |
1da177e4 LT |
1011 | */ |
1012 | void | |
1013 | xfs_buf_item_relse( | |
e8222613 | 1014 | struct xfs_buf *bp) |
1da177e4 | 1015 | { |
fb1755a6 | 1016 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1da177e4 | 1017 | |
0b1b213f | 1018 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
826f7e34 | 1019 | ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags)); |
0b1b213f | 1020 | |
575689fc GX |
1021 | if (atomic_read(&bip->bli_refcount)) |
1022 | return; | |
fb1755a6 | 1023 | bp->b_log_item = NULL; |
e1f5dbd7 LM |
1024 | xfs_buf_rele(bp); |
1025 | xfs_buf_item_free(bip); | |
1da177e4 LT |
1026 | } |
1027 | ||
664ffb8a | 1028 | void |
fec671cd | 1029 | xfs_buf_item_done( |
aac855ab DC |
1030 | struct xfs_buf *bp) |
1031 | { | |
fec671cd DC |
1032 | /* |
1033 | * If we are forcibly shutting down, this may well be off the AIL | |
1034 | * already. That's because we simulate the log-committed callbacks to | |
1035 | * unpin these buffers. Or we may never have put this item on AIL | |
1036 | * because of the transaction was aborted forcibly. | |
1037 | * xfs_trans_ail_delete() takes care of these. | |
1038 | * | |
1039 | * Either way, AIL is useless if we're forcing a shutdown. | |
22c10589 CH |
1040 | * |
1041 | * Note that log recovery writes might have buffer items that are not on | |
1042 | * the AIL even when the file system is not shut down. | |
fec671cd | 1043 | */ |
b840e2ad | 1044 | xfs_trans_ail_delete(&bp->b_log_item->bli_item, |
22c10589 | 1045 | (bp->b_flags & _XBF_LOGRECOVERY) ? 0 : |
b840e2ad CH |
1046 | SHUTDOWN_CORRUPT_INCORE); |
1047 | xfs_buf_item_relse(bp); | |
f593bf14 | 1048 | } |