Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
da353b0d | 26 | #include "xfs_ag.h" |
1da177e4 LT |
27 | #include "xfs_dmapi.h" |
28 | #include "xfs_mount.h" | |
a844f451 | 29 | #include "xfs_buf_item.h" |
1da177e4 | 30 | #include "xfs_trans_priv.h" |
1da177e4 LT |
31 | #include "xfs_error.h" |
32 | ||
33 | ||
34 | kmem_zone_t *xfs_buf_item_zone; | |
35 | ||
36 | #ifdef XFS_TRANS_DEBUG | |
37 | /* | |
38 | * This function uses an alternate strategy for tracking the bytes | |
39 | * that the user requests to be logged. This can then be used | |
40 | * in conjunction with the bli_orig array in the buf log item to | |
41 | * catch bugs in our callers' code. | |
42 | * | |
43 | * We also double check the bits set in xfs_buf_item_log using a | |
44 | * simple algorithm to check that every byte is accounted for. | |
45 | */ | |
46 | STATIC void | |
47 | xfs_buf_item_log_debug( | |
48 | xfs_buf_log_item_t *bip, | |
49 | uint first, | |
50 | uint last) | |
51 | { | |
52 | uint x; | |
53 | uint byte; | |
54 | uint nbytes; | |
55 | uint chunk_num; | |
56 | uint word_num; | |
57 | uint bit_num; | |
58 | uint bit_set; | |
59 | uint *wordp; | |
60 | ||
61 | ASSERT(bip->bli_logged != NULL); | |
62 | byte = first; | |
63 | nbytes = last - first + 1; | |
64 | bfset(bip->bli_logged, first, nbytes); | |
65 | for (x = 0; x < nbytes; x++) { | |
66 | chunk_num = byte >> XFS_BLI_SHIFT; | |
67 | word_num = chunk_num >> BIT_TO_WORD_SHIFT; | |
68 | bit_num = chunk_num & (NBWORD - 1); | |
69 | wordp = &(bip->bli_format.blf_data_map[word_num]); | |
70 | bit_set = *wordp & (1 << bit_num); | |
71 | ASSERT(bit_set); | |
72 | byte++; | |
73 | } | |
74 | } | |
75 | ||
76 | /* | |
77 | * This function is called when we flush something into a buffer without | |
78 | * logging it. This happens for things like inodes which are logged | |
79 | * separately from the buffer. | |
80 | */ | |
81 | void | |
82 | xfs_buf_item_flush_log_debug( | |
83 | xfs_buf_t *bp, | |
84 | uint first, | |
85 | uint last) | |
86 | { | |
87 | xfs_buf_log_item_t *bip; | |
88 | uint nbytes; | |
89 | ||
90 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); | |
91 | if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) { | |
92 | return; | |
93 | } | |
94 | ||
95 | ASSERT(bip->bli_logged != NULL); | |
96 | nbytes = last - first + 1; | |
97 | bfset(bip->bli_logged, first, nbytes); | |
98 | } | |
99 | ||
100 | /* | |
c41564b5 | 101 | * This function is called to verify that our callers have logged |
1da177e4 LT |
102 | * all the bytes that they changed. |
103 | * | |
104 | * It does this by comparing the original copy of the buffer stored in | |
105 | * the buf log item's bli_orig array to the current copy of the buffer | |
c41564b5 | 106 | * and ensuring that all bytes which mismatch are set in the bli_logged |
1da177e4 LT |
107 | * array of the buf log item. |
108 | */ | |
109 | STATIC void | |
110 | xfs_buf_item_log_check( | |
111 | xfs_buf_log_item_t *bip) | |
112 | { | |
113 | char *orig; | |
114 | char *buffer; | |
115 | int x; | |
116 | xfs_buf_t *bp; | |
117 | ||
118 | ASSERT(bip->bli_orig != NULL); | |
119 | ASSERT(bip->bli_logged != NULL); | |
120 | ||
121 | bp = bip->bli_buf; | |
122 | ASSERT(XFS_BUF_COUNT(bp) > 0); | |
123 | ASSERT(XFS_BUF_PTR(bp) != NULL); | |
124 | orig = bip->bli_orig; | |
125 | buffer = XFS_BUF_PTR(bp); | |
126 | for (x = 0; x < XFS_BUF_COUNT(bp); x++) { | |
127 | if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) | |
128 | cmn_err(CE_PANIC, | |
129 | "xfs_buf_item_log_check bip %x buffer %x orig %x index %d", | |
130 | bip, bp, orig, x); | |
131 | } | |
132 | } | |
133 | #else | |
134 | #define xfs_buf_item_log_debug(x,y,z) | |
135 | #define xfs_buf_item_log_check(x) | |
136 | #endif | |
137 | ||
138 | STATIC void xfs_buf_error_relse(xfs_buf_t *bp); | |
139 | STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip); | |
140 | ||
141 | /* | |
142 | * This returns the number of log iovecs needed to log the | |
143 | * given buf log item. | |
144 | * | |
145 | * It calculates this as 1 iovec for the buf log format structure | |
146 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
147 | * Contiguous chunks are logged in a single iovec. | |
148 | * | |
149 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
150 | */ | |
ba0f32d4 | 151 | STATIC uint |
1da177e4 LT |
152 | xfs_buf_item_size( |
153 | xfs_buf_log_item_t *bip) | |
154 | { | |
155 | uint nvecs; | |
156 | int next_bit; | |
157 | int last_bit; | |
158 | xfs_buf_t *bp; | |
159 | ||
160 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
161 | if (bip->bli_flags & XFS_BLI_STALE) { | |
162 | /* | |
163 | * The buffer is stale, so all we need to log | |
164 | * is the buf log format structure with the | |
165 | * cancel flag in it. | |
166 | */ | |
167 | xfs_buf_item_trace("SIZE STALE", bip); | |
168 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); | |
169 | return 1; | |
170 | } | |
171 | ||
172 | bp = bip->bli_buf; | |
173 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
174 | nvecs = 1; | |
175 | last_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
176 | bip->bli_format.blf_map_size, 0); | |
177 | ASSERT(last_bit != -1); | |
178 | nvecs++; | |
179 | while (last_bit != -1) { | |
180 | /* | |
181 | * This takes the bit number to start looking from and | |
182 | * returns the next set bit from there. It returns -1 | |
183 | * if there are no more bits set or the start bit is | |
184 | * beyond the end of the bitmap. | |
185 | */ | |
186 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
187 | bip->bli_format.blf_map_size, | |
188 | last_bit + 1); | |
189 | /* | |
190 | * If we run out of bits, leave the loop, | |
191 | * else if we find a new set of bits bump the number of vecs, | |
192 | * else keep scanning the current set of bits. | |
193 | */ | |
194 | if (next_bit == -1) { | |
195 | last_bit = -1; | |
196 | } else if (next_bit != last_bit + 1) { | |
197 | last_bit = next_bit; | |
198 | nvecs++; | |
199 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLI_CHUNK) != | |
200 | (xfs_buf_offset(bp, last_bit * XFS_BLI_CHUNK) + | |
201 | XFS_BLI_CHUNK)) { | |
202 | last_bit = next_bit; | |
203 | nvecs++; | |
204 | } else { | |
205 | last_bit++; | |
206 | } | |
207 | } | |
208 | ||
209 | xfs_buf_item_trace("SIZE NORM", bip); | |
210 | return nvecs; | |
211 | } | |
212 | ||
213 | /* | |
214 | * This is called to fill in the vector of log iovecs for the | |
215 | * given log buf item. It fills the first entry with a buf log | |
216 | * format structure, and the rest point to contiguous chunks | |
217 | * within the buffer. | |
218 | */ | |
ba0f32d4 | 219 | STATIC void |
1da177e4 LT |
220 | xfs_buf_item_format( |
221 | xfs_buf_log_item_t *bip, | |
222 | xfs_log_iovec_t *log_vector) | |
223 | { | |
224 | uint base_size; | |
225 | uint nvecs; | |
226 | xfs_log_iovec_t *vecp; | |
227 | xfs_buf_t *bp; | |
228 | int first_bit; | |
229 | int last_bit; | |
230 | int next_bit; | |
231 | uint nbits; | |
232 | uint buffer_offset; | |
233 | ||
234 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
235 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
236 | (bip->bli_flags & XFS_BLI_STALE)); | |
237 | bp = bip->bli_buf; | |
1da177e4 LT |
238 | vecp = log_vector; |
239 | ||
240 | /* | |
241 | * The size of the base structure is the size of the | |
242 | * declared structure plus the space for the extra words | |
243 | * of the bitmap. We subtract one from the map size, because | |
244 | * the first element of the bitmap is accounted for in the | |
245 | * size of the base structure. | |
246 | */ | |
247 | base_size = | |
248 | (uint)(sizeof(xfs_buf_log_format_t) + | |
249 | ((bip->bli_format.blf_map_size - 1) * sizeof(uint))); | |
250 | vecp->i_addr = (xfs_caddr_t)&bip->bli_format; | |
251 | vecp->i_len = base_size; | |
7e9c6396 | 252 | XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BFORMAT); |
1da177e4 LT |
253 | vecp++; |
254 | nvecs = 1; | |
255 | ||
256 | if (bip->bli_flags & XFS_BLI_STALE) { | |
257 | /* | |
258 | * The buffer is stale, so all we need to log | |
259 | * is the buf log format structure with the | |
260 | * cancel flag in it. | |
261 | */ | |
262 | xfs_buf_item_trace("FORMAT STALE", bip); | |
263 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); | |
264 | bip->bli_format.blf_size = nvecs; | |
265 | return; | |
266 | } | |
267 | ||
268 | /* | |
269 | * Fill in an iovec for each set of contiguous chunks. | |
270 | */ | |
271 | first_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
272 | bip->bli_format.blf_map_size, 0); | |
273 | ASSERT(first_bit != -1); | |
274 | last_bit = first_bit; | |
275 | nbits = 1; | |
276 | for (;;) { | |
277 | /* | |
278 | * This takes the bit number to start looking from and | |
279 | * returns the next set bit from there. It returns -1 | |
280 | * if there are no more bits set or the start bit is | |
281 | * beyond the end of the bitmap. | |
282 | */ | |
283 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
284 | bip->bli_format.blf_map_size, | |
285 | (uint)last_bit + 1); | |
286 | /* | |
287 | * If we run out of bits fill in the last iovec and get | |
288 | * out of the loop. | |
289 | * Else if we start a new set of bits then fill in the | |
290 | * iovec for the series we were looking at and start | |
291 | * counting the bits in the new one. | |
292 | * Else we're still in the same set of bits so just | |
293 | * keep counting and scanning. | |
294 | */ | |
295 | if (next_bit == -1) { | |
296 | buffer_offset = first_bit * XFS_BLI_CHUNK; | |
297 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); | |
298 | vecp->i_len = nbits * XFS_BLI_CHUNK; | |
7e9c6396 | 299 | XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK); |
1da177e4 LT |
300 | nvecs++; |
301 | break; | |
302 | } else if (next_bit != last_bit + 1) { | |
303 | buffer_offset = first_bit * XFS_BLI_CHUNK; | |
304 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); | |
305 | vecp->i_len = nbits * XFS_BLI_CHUNK; | |
7e9c6396 | 306 | XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK); |
1da177e4 LT |
307 | nvecs++; |
308 | vecp++; | |
309 | first_bit = next_bit; | |
310 | last_bit = next_bit; | |
311 | nbits = 1; | |
312 | } else if (xfs_buf_offset(bp, next_bit << XFS_BLI_SHIFT) != | |
313 | (xfs_buf_offset(bp, last_bit << XFS_BLI_SHIFT) + | |
314 | XFS_BLI_CHUNK)) { | |
315 | buffer_offset = first_bit * XFS_BLI_CHUNK; | |
316 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); | |
317 | vecp->i_len = nbits * XFS_BLI_CHUNK; | |
7e9c6396 | 318 | XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK); |
1da177e4 LT |
319 | /* You would think we need to bump the nvecs here too, but we do not |
320 | * this number is used by recovery, and it gets confused by the boundary | |
321 | * split here | |
322 | * nvecs++; | |
323 | */ | |
324 | vecp++; | |
325 | first_bit = next_bit; | |
326 | last_bit = next_bit; | |
327 | nbits = 1; | |
328 | } else { | |
329 | last_bit++; | |
330 | nbits++; | |
331 | } | |
332 | } | |
333 | bip->bli_format.blf_size = nvecs; | |
334 | ||
335 | /* | |
336 | * Check to make sure everything is consistent. | |
337 | */ | |
338 | xfs_buf_item_trace("FORMAT NORM", bip); | |
339 | xfs_buf_item_log_check(bip); | |
340 | } | |
341 | ||
342 | /* | |
343 | * This is called to pin the buffer associated with the buf log | |
344 | * item in memory so it cannot be written out. Simply call bpin() | |
345 | * on the buffer to do this. | |
346 | */ | |
ba0f32d4 | 347 | STATIC void |
1da177e4 LT |
348 | xfs_buf_item_pin( |
349 | xfs_buf_log_item_t *bip) | |
350 | { | |
351 | xfs_buf_t *bp; | |
352 | ||
353 | bp = bip->bli_buf; | |
354 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
355 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
356 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
357 | (bip->bli_flags & XFS_BLI_STALE)); | |
358 | xfs_buf_item_trace("PIN", bip); | |
359 | xfs_buftrace("XFS_PIN", bp); | |
360 | xfs_bpin(bp); | |
361 | } | |
362 | ||
363 | ||
364 | /* | |
365 | * This is called to unpin the buffer associated with the buf log | |
366 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
367 | * Just call bunpin() on the buffer to do this. | |
368 | * | |
369 | * Also drop the reference to the buf item for the current transaction. | |
370 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
371 | * then free up the buf log item and unlock the buffer. | |
372 | */ | |
ba0f32d4 | 373 | STATIC void |
1da177e4 LT |
374 | xfs_buf_item_unpin( |
375 | xfs_buf_log_item_t *bip, | |
376 | int stale) | |
377 | { | |
378 | xfs_mount_t *mp; | |
379 | xfs_buf_t *bp; | |
380 | int freed; | |
1da177e4 LT |
381 | |
382 | bp = bip->bli_buf; | |
383 | ASSERT(bp != NULL); | |
384 | ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip); | |
385 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
386 | xfs_buf_item_trace("UNPIN", bip); | |
387 | xfs_buftrace("XFS_UNPIN", bp); | |
388 | ||
389 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
390 | mp = bip->bli_item.li_mountp; | |
391 | xfs_bunpin(bp); | |
392 | if (freed && stale) { | |
393 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
394 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
395 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); | |
396 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
397 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); | |
398 | xfs_buf_item_trace("UNPIN STALE", bip); | |
399 | xfs_buftrace("XFS_UNPIN STALE", bp); | |
400 | /* | |
401 | * If we get called here because of an IO error, we may | |
402 | * or may not have the item on the AIL. xfs_trans_delete_ail() | |
403 | * will take care of that situation. | |
404 | * xfs_trans_delete_ail() drops the AIL lock. | |
405 | */ | |
406 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
407 | xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip); | |
408 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
409 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
410 | } else { | |
287f3dad DD |
411 | spin_lock(&mp->m_ail_lock); |
412 | xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip); | |
1da177e4 LT |
413 | xfs_buf_item_relse(bp); |
414 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL); | |
415 | } | |
416 | xfs_buf_relse(bp); | |
417 | } | |
418 | } | |
419 | ||
420 | /* | |
421 | * this is called from uncommit in the forced-shutdown path. | |
422 | * we need to check to see if the reference count on the log item | |
423 | * is going to drop to zero. If so, unpin will free the log item | |
424 | * so we need to free the item's descriptor (that points to the item) | |
425 | * in the transaction. | |
426 | */ | |
ba0f32d4 | 427 | STATIC void |
1da177e4 LT |
428 | xfs_buf_item_unpin_remove( |
429 | xfs_buf_log_item_t *bip, | |
430 | xfs_trans_t *tp) | |
431 | { | |
432 | xfs_buf_t *bp; | |
433 | xfs_log_item_desc_t *lidp; | |
434 | int stale = 0; | |
435 | ||
436 | bp = bip->bli_buf; | |
437 | /* | |
438 | * will xfs_buf_item_unpin() call xfs_buf_item_relse()? | |
439 | */ | |
440 | if ((atomic_read(&bip->bli_refcount) == 1) && | |
441 | (bip->bli_flags & XFS_BLI_STALE)) { | |
442 | ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0); | |
443 | xfs_buf_item_trace("UNPIN REMOVE", bip); | |
444 | xfs_buftrace("XFS_UNPIN_REMOVE", bp); | |
445 | /* | |
446 | * yes -- clear the xaction descriptor in-use flag | |
447 | * and free the chunk if required. We can safely | |
448 | * do some work here and then call buf_item_unpin | |
449 | * to do the rest because if the if is true, then | |
450 | * we are holding the buffer locked so no one else | |
451 | * will be able to bump up the refcount. | |
452 | */ | |
453 | lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip); | |
454 | stale = lidp->lid_flags & XFS_LID_BUF_STALE; | |
455 | xfs_trans_free_item(tp, lidp); | |
456 | /* | |
457 | * Since the transaction no longer refers to the buffer, | |
458 | * the buffer should no longer refer to the transaction. | |
459 | */ | |
460 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); | |
461 | } | |
462 | ||
463 | xfs_buf_item_unpin(bip, stale); | |
464 | ||
465 | return; | |
466 | } | |
467 | ||
468 | /* | |
469 | * This is called to attempt to lock the buffer associated with this | |
470 | * buf log item. Don't sleep on the buffer lock. If we can't get | |
471 | * the lock right away, return 0. If we can get the lock, pull the | |
472 | * buffer from the free list, mark it busy, and return 1. | |
473 | */ | |
ba0f32d4 | 474 | STATIC uint |
1da177e4 LT |
475 | xfs_buf_item_trylock( |
476 | xfs_buf_log_item_t *bip) | |
477 | { | |
478 | xfs_buf_t *bp; | |
479 | ||
480 | bp = bip->bli_buf; | |
481 | ||
482 | if (XFS_BUF_ISPINNED(bp)) { | |
483 | return XFS_ITEM_PINNED; | |
484 | } | |
485 | ||
486 | if (!XFS_BUF_CPSEMA(bp)) { | |
487 | return XFS_ITEM_LOCKED; | |
488 | } | |
489 | ||
490 | /* | |
491 | * Remove the buffer from the free list. Only do this | |
492 | * if it's on the free list. Private buffers like the | |
493 | * superblock buffer are not. | |
494 | */ | |
495 | XFS_BUF_HOLD(bp); | |
496 | ||
497 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
498 | xfs_buf_item_trace("TRYLOCK SUCCESS", bip); | |
499 | return XFS_ITEM_SUCCESS; | |
500 | } | |
501 | ||
502 | /* | |
503 | * Release the buffer associated with the buf log item. | |
504 | * If there is no dirty logged data associated with the | |
505 | * buffer recorded in the buf log item, then free the | |
506 | * buf log item and remove the reference to it in the | |
507 | * buffer. | |
508 | * | |
509 | * This call ignores the recursion count. It is only called | |
510 | * when the buffer should REALLY be unlocked, regardless | |
511 | * of the recursion count. | |
512 | * | |
513 | * If the XFS_BLI_HOLD flag is set in the buf log item, then | |
514 | * free the log item if necessary but do not unlock the buffer. | |
515 | * This is for support of xfs_trans_bhold(). Make sure the | |
516 | * XFS_BLI_HOLD field is cleared if we don't free the item. | |
517 | */ | |
ba0f32d4 | 518 | STATIC void |
1da177e4 LT |
519 | xfs_buf_item_unlock( |
520 | xfs_buf_log_item_t *bip) | |
521 | { | |
522 | int aborted; | |
523 | xfs_buf_t *bp; | |
524 | uint hold; | |
525 | ||
526 | bp = bip->bli_buf; | |
527 | xfs_buftrace("XFS_UNLOCK", bp); | |
528 | ||
529 | /* | |
530 | * Clear the buffer's association with this transaction. | |
531 | */ | |
532 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); | |
533 | ||
534 | /* | |
535 | * If this is a transaction abort, don't return early. | |
536 | * Instead, allow the brelse to happen. | |
537 | * Normally it would be done for stale (cancelled) buffers | |
538 | * at unpin time, but we'll never go through the pin/unpin | |
539 | * cycle if we abort inside commit. | |
540 | */ | |
541 | aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0; | |
542 | ||
543 | /* | |
544 | * If the buf item is marked stale, then don't do anything. | |
545 | * We'll unlock the buffer and free the buf item when the | |
546 | * buffer is unpinned for the last time. | |
547 | */ | |
548 | if (bip->bli_flags & XFS_BLI_STALE) { | |
549 | bip->bli_flags &= ~XFS_BLI_LOGGED; | |
550 | xfs_buf_item_trace("UNLOCK STALE", bip); | |
551 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); | |
552 | if (!aborted) | |
553 | return; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Drop the transaction's reference to the log item if | |
558 | * it was not logged as part of the transaction. Otherwise | |
559 | * we'll drop the reference in xfs_buf_item_unpin() when | |
560 | * the transaction is really through with the buffer. | |
561 | */ | |
562 | if (!(bip->bli_flags & XFS_BLI_LOGGED)) { | |
563 | atomic_dec(&bip->bli_refcount); | |
564 | } else { | |
565 | /* | |
566 | * Clear the logged flag since this is per | |
567 | * transaction state. | |
568 | */ | |
569 | bip->bli_flags &= ~XFS_BLI_LOGGED; | |
570 | } | |
571 | ||
572 | /* | |
573 | * Before possibly freeing the buf item, determine if we should | |
574 | * release the buffer at the end of this routine. | |
575 | */ | |
576 | hold = bip->bli_flags & XFS_BLI_HOLD; | |
577 | xfs_buf_item_trace("UNLOCK", bip); | |
578 | ||
579 | /* | |
580 | * If the buf item isn't tracking any data, free it. | |
581 | * Otherwise, if XFS_BLI_HOLD is set clear it. | |
582 | */ | |
24ad33ff ES |
583 | if (xfs_bitmap_empty(bip->bli_format.blf_data_map, |
584 | bip->bli_format.blf_map_size)) { | |
1da177e4 LT |
585 | xfs_buf_item_relse(bp); |
586 | } else if (hold) { | |
587 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
588 | } | |
589 | ||
590 | /* | |
591 | * Release the buffer if XFS_BLI_HOLD was not set. | |
592 | */ | |
593 | if (!hold) { | |
594 | xfs_buf_relse(bp); | |
595 | } | |
596 | } | |
597 | ||
598 | /* | |
599 | * This is called to find out where the oldest active copy of the | |
600 | * buf log item in the on disk log resides now that the last log | |
601 | * write of it completed at the given lsn. | |
602 | * We always re-log all the dirty data in a buffer, so usually the | |
603 | * latest copy in the on disk log is the only one that matters. For | |
604 | * those cases we simply return the given lsn. | |
605 | * | |
606 | * The one exception to this is for buffers full of newly allocated | |
607 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
608 | * flag set, indicating that only the di_next_unlinked fields from the | |
609 | * inodes in the buffers will be replayed during recovery. If the | |
610 | * original newly allocated inode images have not yet been flushed | |
611 | * when the buffer is so relogged, then we need to make sure that we | |
612 | * keep the old images in the 'active' portion of the log. We do this | |
613 | * by returning the original lsn of that transaction here rather than | |
614 | * the current one. | |
615 | */ | |
ba0f32d4 | 616 | STATIC xfs_lsn_t |
1da177e4 LT |
617 | xfs_buf_item_committed( |
618 | xfs_buf_log_item_t *bip, | |
619 | xfs_lsn_t lsn) | |
620 | { | |
621 | xfs_buf_item_trace("COMMITTED", bip); | |
622 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
623 | (bip->bli_item.li_lsn != 0)) { | |
624 | return bip->bli_item.li_lsn; | |
625 | } | |
626 | return (lsn); | |
627 | } | |
628 | ||
1da177e4 LT |
629 | /* |
630 | * This is called to asynchronously write the buffer associated with this | |
631 | * buf log item out to disk. The buffer will already have been locked by | |
632 | * a successful call to xfs_buf_item_trylock(). If the buffer still has | |
633 | * B_DELWRI set, then get it going out to disk with a call to bawrite(). | |
634 | * If not, then just release the buffer. | |
635 | */ | |
ba0f32d4 | 636 | STATIC void |
1da177e4 LT |
637 | xfs_buf_item_push( |
638 | xfs_buf_log_item_t *bip) | |
639 | { | |
640 | xfs_buf_t *bp; | |
641 | ||
642 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
643 | xfs_buf_item_trace("PUSH", bip); | |
644 | ||
645 | bp = bip->bli_buf; | |
646 | ||
647 | if (XFS_BUF_ISDELAYWRITE(bp)) { | |
db7a19f2 DC |
648 | int error; |
649 | error = xfs_bawrite(bip->bli_item.li_mountp, bp); | |
650 | if (error) | |
651 | xfs_fs_cmn_err(CE_WARN, bip->bli_item.li_mountp, | |
652 | "xfs_buf_item_push: pushbuf error %d on bip %p, bp %p", | |
653 | error, bip, bp); | |
1da177e4 LT |
654 | } else { |
655 | xfs_buf_relse(bp); | |
656 | } | |
657 | } | |
658 | ||
659 | /* ARGSUSED */ | |
ba0f32d4 | 660 | STATIC void |
1da177e4 LT |
661 | xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn) |
662 | { | |
663 | } | |
664 | ||
665 | /* | |
666 | * This is the ops vector shared by all buf log items. | |
667 | */ | |
7989cb8e | 668 | static struct xfs_item_ops xfs_buf_item_ops = { |
1da177e4 LT |
669 | .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size, |
670 | .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*)) | |
671 | xfs_buf_item_format, | |
672 | .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin, | |
673 | .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin, | |
674 | .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *)) | |
675 | xfs_buf_item_unpin_remove, | |
676 | .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock, | |
677 | .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock, | |
678 | .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t)) | |
679 | xfs_buf_item_committed, | |
680 | .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push, | |
1da177e4 LT |
681 | .iop_pushbuf = NULL, |
682 | .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t)) | |
683 | xfs_buf_item_committing | |
684 | }; | |
685 | ||
686 | ||
687 | /* | |
688 | * Allocate a new buf log item to go with the given buffer. | |
689 | * Set the buffer's b_fsprivate field to point to the new | |
690 | * buf log item. If there are other item's attached to the | |
691 | * buffer (see xfs_buf_attach_iodone() below), then put the | |
692 | * buf log item at the front. | |
693 | */ | |
694 | void | |
695 | xfs_buf_item_init( | |
696 | xfs_buf_t *bp, | |
697 | xfs_mount_t *mp) | |
698 | { | |
699 | xfs_log_item_t *lip; | |
700 | xfs_buf_log_item_t *bip; | |
701 | int chunks; | |
702 | int map_size; | |
703 | ||
704 | /* | |
705 | * Check to see if there is already a buf log item for | |
706 | * this buffer. If there is, it is guaranteed to be | |
707 | * the first. If we do already have one, there is | |
708 | * nothing to do here so return. | |
709 | */ | |
710 | if (XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *) != mp) | |
711 | XFS_BUF_SET_FSPRIVATE3(bp, mp); | |
712 | XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb); | |
713 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
714 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
715 | if (lip->li_type == XFS_LI_BUF) { | |
716 | return; | |
717 | } | |
718 | } | |
719 | ||
720 | /* | |
721 | * chunks is the number of XFS_BLI_CHUNK size pieces | |
722 | * the buffer can be divided into. Make sure not to | |
723 | * truncate any pieces. map_size is the size of the | |
724 | * bitmap needed to describe the chunks of the buffer. | |
725 | */ | |
726 | chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT); | |
727 | map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT); | |
728 | ||
729 | bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone, | |
730 | KM_SLEEP); | |
731 | bip->bli_item.li_type = XFS_LI_BUF; | |
732 | bip->bli_item.li_ops = &xfs_buf_item_ops; | |
733 | bip->bli_item.li_mountp = mp; | |
734 | bip->bli_buf = bp; | |
e1f5dbd7 | 735 | xfs_buf_hold(bp); |
1da177e4 LT |
736 | bip->bli_format.blf_type = XFS_LI_BUF; |
737 | bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp); | |
738 | bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp)); | |
739 | bip->bli_format.blf_map_size = map_size; | |
740 | #ifdef XFS_BLI_TRACE | |
5695ef46 | 741 | bip->bli_trace = ktrace_alloc(XFS_BLI_TRACE_SIZE, KM_NOFS); |
1da177e4 LT |
742 | #endif |
743 | ||
744 | #ifdef XFS_TRANS_DEBUG | |
745 | /* | |
746 | * Allocate the arrays for tracking what needs to be logged | |
747 | * and what our callers request to be logged. bli_orig | |
748 | * holds a copy of the original, clean buffer for comparison | |
749 | * against, and bli_logged keeps a 1 bit flag per byte in | |
750 | * the buffer to indicate which bytes the callers have asked | |
751 | * to have logged. | |
752 | */ | |
753 | bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP); | |
754 | memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp)); | |
755 | bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP); | |
756 | #endif | |
757 | ||
758 | /* | |
759 | * Put the buf item into the list of items attached to the | |
760 | * buffer at the front. | |
761 | */ | |
762 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
763 | bip->bli_item.li_bio_list = | |
764 | XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
765 | } | |
766 | XFS_BUF_SET_FSPRIVATE(bp, bip); | |
767 | } | |
768 | ||
769 | ||
770 | /* | |
771 | * Mark bytes first through last inclusive as dirty in the buf | |
772 | * item's bitmap. | |
773 | */ | |
774 | void | |
775 | xfs_buf_item_log( | |
776 | xfs_buf_log_item_t *bip, | |
777 | uint first, | |
778 | uint last) | |
779 | { | |
780 | uint first_bit; | |
781 | uint last_bit; | |
782 | uint bits_to_set; | |
783 | uint bits_set; | |
784 | uint word_num; | |
785 | uint *wordp; | |
786 | uint bit; | |
787 | uint end_bit; | |
788 | uint mask; | |
789 | ||
790 | /* | |
791 | * Mark the item as having some dirty data for | |
792 | * quick reference in xfs_buf_item_dirty. | |
793 | */ | |
794 | bip->bli_flags |= XFS_BLI_DIRTY; | |
795 | ||
796 | /* | |
797 | * Convert byte offsets to bit numbers. | |
798 | */ | |
799 | first_bit = first >> XFS_BLI_SHIFT; | |
800 | last_bit = last >> XFS_BLI_SHIFT; | |
801 | ||
802 | /* | |
803 | * Calculate the total number of bits to be set. | |
804 | */ | |
805 | bits_to_set = last_bit - first_bit + 1; | |
806 | ||
807 | /* | |
808 | * Get a pointer to the first word in the bitmap | |
809 | * to set a bit in. | |
810 | */ | |
811 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
812 | wordp = &(bip->bli_format.blf_data_map[word_num]); | |
813 | ||
814 | /* | |
815 | * Calculate the starting bit in the first word. | |
816 | */ | |
817 | bit = first_bit & (uint)(NBWORD - 1); | |
818 | ||
819 | /* | |
820 | * First set any bits in the first word of our range. | |
821 | * If it starts at bit 0 of the word, it will be | |
822 | * set below rather than here. That is what the variable | |
823 | * bit tells us. The variable bits_set tracks the number | |
824 | * of bits that have been set so far. End_bit is the number | |
825 | * of the last bit to be set in this word plus one. | |
826 | */ | |
827 | if (bit) { | |
828 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); | |
829 | mask = ((1 << (end_bit - bit)) - 1) << bit; | |
830 | *wordp |= mask; | |
831 | wordp++; | |
832 | bits_set = end_bit - bit; | |
833 | } else { | |
834 | bits_set = 0; | |
835 | } | |
836 | ||
837 | /* | |
838 | * Now set bits a whole word at a time that are between | |
839 | * first_bit and last_bit. | |
840 | */ | |
841 | while ((bits_to_set - bits_set) >= NBWORD) { | |
842 | *wordp |= 0xffffffff; | |
843 | bits_set += NBWORD; | |
844 | wordp++; | |
845 | } | |
846 | ||
847 | /* | |
848 | * Finally, set any bits left to be set in one last partial word. | |
849 | */ | |
850 | end_bit = bits_to_set - bits_set; | |
851 | if (end_bit) { | |
852 | mask = (1 << end_bit) - 1; | |
853 | *wordp |= mask; | |
854 | } | |
855 | ||
856 | xfs_buf_item_log_debug(bip, first, last); | |
857 | } | |
858 | ||
859 | ||
860 | /* | |
861 | * Return 1 if the buffer has some data that has been logged (at any | |
862 | * point, not just the current transaction) and 0 if not. | |
863 | */ | |
864 | uint | |
865 | xfs_buf_item_dirty( | |
866 | xfs_buf_log_item_t *bip) | |
867 | { | |
868 | return (bip->bli_flags & XFS_BLI_DIRTY); | |
869 | } | |
870 | ||
e1f5dbd7 LM |
871 | STATIC void |
872 | xfs_buf_item_free( | |
873 | xfs_buf_log_item_t *bip) | |
874 | { | |
875 | #ifdef XFS_TRANS_DEBUG | |
876 | kmem_free(bip->bli_orig); | |
877 | kmem_free(bip->bli_logged); | |
878 | #endif /* XFS_TRANS_DEBUG */ | |
879 | ||
880 | #ifdef XFS_BLI_TRACE | |
881 | ktrace_free(bip->bli_trace); | |
882 | #endif | |
883 | kmem_zone_free(xfs_buf_item_zone, bip); | |
884 | } | |
885 | ||
1da177e4 LT |
886 | /* |
887 | * This is called when the buf log item is no longer needed. It should | |
888 | * free the buf log item associated with the given buffer and clear | |
889 | * the buffer's pointer to the buf log item. If there are no more | |
890 | * items in the list, clear the b_iodone field of the buffer (see | |
891 | * xfs_buf_attach_iodone() below). | |
892 | */ | |
893 | void | |
894 | xfs_buf_item_relse( | |
895 | xfs_buf_t *bp) | |
896 | { | |
897 | xfs_buf_log_item_t *bip; | |
898 | ||
899 | xfs_buftrace("XFS_RELSE", bp); | |
900 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); | |
901 | XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list); | |
902 | if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) && | |
903 | (XFS_BUF_IODONE_FUNC(bp) != NULL)) { | |
1da177e4 LT |
904 | XFS_BUF_CLR_IODONE_FUNC(bp); |
905 | } | |
e1f5dbd7 LM |
906 | xfs_buf_rele(bp); |
907 | xfs_buf_item_free(bip); | |
1da177e4 LT |
908 | } |
909 | ||
910 | ||
911 | /* | |
912 | * Add the given log item with its callback to the list of callbacks | |
913 | * to be called when the buffer's I/O completes. If it is not set | |
914 | * already, set the buffer's b_iodone() routine to be | |
915 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
916 | * items rooted at b_fsprivate. Items are always added as the second | |
917 | * entry in the list if there is a first, because the buf item code | |
918 | * assumes that the buf log item is first. | |
919 | */ | |
920 | void | |
921 | xfs_buf_attach_iodone( | |
922 | xfs_buf_t *bp, | |
923 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), | |
924 | xfs_log_item_t *lip) | |
925 | { | |
926 | xfs_log_item_t *head_lip; | |
927 | ||
928 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
929 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
930 | ||
931 | lip->li_cb = cb; | |
932 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
933 | head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
934 | lip->li_bio_list = head_lip->li_bio_list; | |
935 | head_lip->li_bio_list = lip; | |
936 | } else { | |
937 | XFS_BUF_SET_FSPRIVATE(bp, lip); | |
938 | } | |
939 | ||
940 | ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) || | |
941 | (XFS_BUF_IODONE_FUNC(bp) == NULL)); | |
942 | XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks); | |
943 | } | |
944 | ||
945 | STATIC void | |
946 | xfs_buf_do_callbacks( | |
947 | xfs_buf_t *bp, | |
948 | xfs_log_item_t *lip) | |
949 | { | |
950 | xfs_log_item_t *nlip; | |
951 | ||
952 | while (lip != NULL) { | |
953 | nlip = lip->li_bio_list; | |
954 | ASSERT(lip->li_cb != NULL); | |
955 | /* | |
956 | * Clear the next pointer so we don't have any | |
957 | * confusion if the item is added to another buf. | |
958 | * Don't touch the log item after calling its | |
959 | * callback, because it could have freed itself. | |
960 | */ | |
961 | lip->li_bio_list = NULL; | |
962 | lip->li_cb(bp, lip); | |
963 | lip = nlip; | |
964 | } | |
965 | } | |
966 | ||
967 | /* | |
968 | * This is the iodone() function for buffers which have had callbacks | |
969 | * attached to them by xfs_buf_attach_iodone(). It should remove each | |
970 | * log item from the buffer's list and call the callback of each in turn. | |
971 | * When done, the buffer's fsprivate field is set to NULL and the buffer | |
972 | * is unlocked with a call to iodone(). | |
973 | */ | |
974 | void | |
975 | xfs_buf_iodone_callbacks( | |
976 | xfs_buf_t *bp) | |
977 | { | |
978 | xfs_log_item_t *lip; | |
979 | static ulong lasttime; | |
980 | static xfs_buftarg_t *lasttarg; | |
981 | xfs_mount_t *mp; | |
982 | ||
983 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
984 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
985 | ||
986 | if (XFS_BUF_GETERROR(bp) != 0) { | |
987 | /* | |
988 | * If we've already decided to shutdown the filesystem | |
989 | * because of IO errors, there's no point in giving this | |
990 | * a retry. | |
991 | */ | |
992 | mp = lip->li_mountp; | |
993 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
994 | ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp); | |
995 | XFS_BUF_SUPER_STALE(bp); | |
996 | xfs_buftrace("BUF_IODONE_CB", bp); | |
997 | xfs_buf_do_callbacks(bp, lip); | |
998 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
999 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
1000 | ||
1001 | /* | |
1002 | * XFS_SHUT flag gets set when we go thru the | |
1003 | * entire buffer cache and deliberately start | |
1004 | * throwing away delayed write buffers. | |
1005 | * Since there's no biowait done on those, | |
1006 | * we should just brelse them. | |
1007 | */ | |
1008 | if (XFS_BUF_ISSHUT(bp)) { | |
1009 | XFS_BUF_UNSHUT(bp); | |
1010 | xfs_buf_relse(bp); | |
1011 | } else { | |
1012 | xfs_biodone(bp); | |
1013 | } | |
1014 | ||
1015 | return; | |
1016 | } | |
1017 | ||
1018 | if ((XFS_BUF_TARGET(bp) != lasttarg) || | |
1019 | (time_after(jiffies, (lasttime + 5*HZ)))) { | |
1020 | lasttime = jiffies; | |
b6574520 NS |
1021 | cmn_err(CE_ALERT, "Device %s, XFS metadata write error" |
1022 | " block 0x%llx in %s", | |
1023 | XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)), | |
1da177e4 LT |
1024 | (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname); |
1025 | } | |
1026 | lasttarg = XFS_BUF_TARGET(bp); | |
1027 | ||
1028 | if (XFS_BUF_ISASYNC(bp)) { | |
1029 | /* | |
1030 | * If the write was asynchronous then noone will be | |
1031 | * looking for the error. Clear the error state | |
1032 | * and write the buffer out again delayed write. | |
1033 | * | |
1034 | * XXXsup This is OK, so long as we catch these | |
1035 | * before we start the umount; we don't want these | |
1036 | * DELWRI metadata bufs to be hanging around. | |
1037 | */ | |
1038 | XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */ | |
1039 | ||
1040 | if (!(XFS_BUF_ISSTALE(bp))) { | |
1041 | XFS_BUF_DELAYWRITE(bp); | |
1042 | XFS_BUF_DONE(bp); | |
1043 | XFS_BUF_SET_START(bp); | |
1044 | } | |
1045 | ASSERT(XFS_BUF_IODONE_FUNC(bp)); | |
1046 | xfs_buftrace("BUF_IODONE ASYNC", bp); | |
1047 | xfs_buf_relse(bp); | |
1048 | } else { | |
1049 | /* | |
1050 | * If the write of the buffer was not asynchronous, | |
1051 | * then we want to make sure to return the error | |
1052 | * to the caller of bwrite(). Because of this we | |
1053 | * cannot clear the B_ERROR state at this point. | |
1054 | * Instead we install a callback function that | |
1055 | * will be called when the buffer is released, and | |
1056 | * that routine will clear the error state and | |
1057 | * set the buffer to be written out again after | |
1058 | * some delay. | |
1059 | */ | |
1060 | /* We actually overwrite the existing b-relse | |
1061 | function at times, but we're gonna be shutting down | |
1062 | anyway. */ | |
1063 | XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse); | |
1064 | XFS_BUF_DONE(bp); | |
b4dd330b | 1065 | XFS_BUF_FINISH_IOWAIT(bp); |
1da177e4 LT |
1066 | } |
1067 | return; | |
1068 | } | |
1069 | #ifdef XFSERRORDEBUG | |
1070 | xfs_buftrace("XFS BUFCB NOERR", bp); | |
1071 | #endif | |
1072 | xfs_buf_do_callbacks(bp, lip); | |
1073 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
1074 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
1075 | xfs_biodone(bp); | |
1076 | } | |
1077 | ||
1078 | /* | |
1079 | * This is a callback routine attached to a buffer which gets an error | |
1080 | * when being written out synchronously. | |
1081 | */ | |
1082 | STATIC void | |
1083 | xfs_buf_error_relse( | |
1084 | xfs_buf_t *bp) | |
1085 | { | |
1086 | xfs_log_item_t *lip; | |
1087 | xfs_mount_t *mp; | |
1088 | ||
1089 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
1090 | mp = (xfs_mount_t *)lip->li_mountp; | |
1091 | ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp); | |
1092 | ||
1093 | XFS_BUF_STALE(bp); | |
1094 | XFS_BUF_DONE(bp); | |
1095 | XFS_BUF_UNDELAYWRITE(bp); | |
1096 | XFS_BUF_ERROR(bp,0); | |
1097 | xfs_buftrace("BUF_ERROR_RELSE", bp); | |
1098 | if (! XFS_FORCED_SHUTDOWN(mp)) | |
7d04a335 | 1099 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1100 | /* |
1101 | * We have to unpin the pinned buffers so do the | |
1102 | * callbacks. | |
1103 | */ | |
1104 | xfs_buf_do_callbacks(bp, lip); | |
1105 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
1106 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
1107 | XFS_BUF_SET_BRELSE_FUNC(bp,NULL); | |
1108 | xfs_buf_relse(bp); | |
1109 | } | |
1110 | ||
1111 | ||
1112 | /* | |
1113 | * This is the iodone() function for buffers which have been | |
1114 | * logged. It is called when they are eventually flushed out. | |
1115 | * It should remove the buf item from the AIL, and free the buf item. | |
1116 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1117 | * care of cleaning up the buffer itself. | |
1118 | */ | |
1119 | /* ARGSUSED */ | |
1120 | void | |
1121 | xfs_buf_iodone( | |
1122 | xfs_buf_t *bp, | |
1123 | xfs_buf_log_item_t *bip) | |
1124 | { | |
1125 | struct xfs_mount *mp; | |
1da177e4 LT |
1126 | |
1127 | ASSERT(bip->bli_buf == bp); | |
1128 | ||
e1f5dbd7 | 1129 | xfs_buf_rele(bp); |
1da177e4 LT |
1130 | mp = bip->bli_item.li_mountp; |
1131 | ||
1132 | /* | |
1133 | * If we are forcibly shutting down, this may well be | |
1134 | * off the AIL already. That's because we simulate the | |
1135 | * log-committed callbacks to unpin these buffers. Or we may never | |
1136 | * have put this item on AIL because of the transaction was | |
1137 | * aborted forcibly. xfs_trans_delete_ail() takes care of these. | |
1138 | * | |
1139 | * Either way, AIL is useless if we're forcing a shutdown. | |
1140 | */ | |
287f3dad | 1141 | spin_lock(&mp->m_ail_lock); |
1da177e4 LT |
1142 | /* |
1143 | * xfs_trans_delete_ail() drops the AIL lock. | |
1144 | */ | |
287f3dad | 1145 | xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip); |
e1f5dbd7 | 1146 | xfs_buf_item_free(bip); |
1da177e4 LT |
1147 | } |
1148 | ||
1149 | #if defined(XFS_BLI_TRACE) | |
1150 | void | |
1151 | xfs_buf_item_trace( | |
1152 | char *id, | |
1153 | xfs_buf_log_item_t *bip) | |
1154 | { | |
1155 | xfs_buf_t *bp; | |
1156 | ASSERT(bip->bli_trace != NULL); | |
1157 | ||
1158 | bp = bip->bli_buf; | |
1159 | ktrace_enter(bip->bli_trace, | |
1160 | (void *)id, | |
1161 | (void *)bip->bli_buf, | |
1162 | (void *)((unsigned long)bip->bli_flags), | |
1163 | (void *)((unsigned long)bip->bli_recur), | |
1164 | (void *)((unsigned long)atomic_read(&bip->bli_refcount)), | |
1165 | (void *)((unsigned long) | |
1166 | (0xFFFFFFFF & XFS_BUF_ADDR(bp) >> 32)), | |
1167 | (void *)((unsigned long)(0xFFFFFFFF & XFS_BUF_ADDR(bp))), | |
1168 | (void *)((unsigned long)XFS_BUF_COUNT(bp)), | |
1169 | (void *)((unsigned long)XFS_BUF_BFLAGS(bp)), | |
1170 | XFS_BUF_FSPRIVATE(bp, void *), | |
1171 | XFS_BUF_FSPRIVATE2(bp, void *), | |
1172 | (void *)(unsigned long)XFS_BUF_ISPINNED(bp), | |
1173 | (void *)XFS_BUF_IODONE_FUNC(bp), | |
1174 | (void *)((unsigned long)(XFS_BUF_VALUSEMA(bp))), | |
1175 | (void *)bip->bli_item.li_desc, | |
1176 | (void *)((unsigned long)bip->bli_item.li_flags)); | |
1177 | } | |
1178 | #endif /* XFS_BLI_TRACE */ |