]>
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 | 31 | #include "xfs_error.h" |
0b1b213f | 32 | #include "xfs_trace.h" |
1da177e4 LT |
33 | |
34 | ||
35 | kmem_zone_t *xfs_buf_item_zone; | |
36 | ||
37 | #ifdef XFS_TRANS_DEBUG | |
38 | /* | |
39 | * This function uses an alternate strategy for tracking the bytes | |
40 | * that the user requests to be logged. This can then be used | |
41 | * in conjunction with the bli_orig array in the buf log item to | |
42 | * catch bugs in our callers' code. | |
43 | * | |
44 | * We also double check the bits set in xfs_buf_item_log using a | |
45 | * simple algorithm to check that every byte is accounted for. | |
46 | */ | |
47 | STATIC void | |
48 | xfs_buf_item_log_debug( | |
49 | xfs_buf_log_item_t *bip, | |
50 | uint first, | |
51 | uint last) | |
52 | { | |
53 | uint x; | |
54 | uint byte; | |
55 | uint nbytes; | |
56 | uint chunk_num; | |
57 | uint word_num; | |
58 | uint bit_num; | |
59 | uint bit_set; | |
60 | uint *wordp; | |
61 | ||
62 | ASSERT(bip->bli_logged != NULL); | |
63 | byte = first; | |
64 | nbytes = last - first + 1; | |
65 | bfset(bip->bli_logged, first, nbytes); | |
66 | for (x = 0; x < nbytes; x++) { | |
c1155410 | 67 | chunk_num = byte >> XFS_BLF_SHIFT; |
1da177e4 LT |
68 | word_num = chunk_num >> BIT_TO_WORD_SHIFT; |
69 | bit_num = chunk_num & (NBWORD - 1); | |
70 | wordp = &(bip->bli_format.blf_data_map[word_num]); | |
71 | bit_set = *wordp & (1 << bit_num); | |
72 | ASSERT(bit_set); | |
73 | byte++; | |
74 | } | |
75 | } | |
76 | ||
77 | /* | |
78 | * This function is called when we flush something into a buffer without | |
79 | * logging it. This happens for things like inodes which are logged | |
80 | * separately from the buffer. | |
81 | */ | |
82 | void | |
83 | xfs_buf_item_flush_log_debug( | |
84 | xfs_buf_t *bp, | |
85 | uint first, | |
86 | uint last) | |
87 | { | |
88 | xfs_buf_log_item_t *bip; | |
89 | uint nbytes; | |
90 | ||
91 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); | |
92 | if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) { | |
93 | return; | |
94 | } | |
95 | ||
96 | ASSERT(bip->bli_logged != NULL); | |
97 | nbytes = last - first + 1; | |
98 | bfset(bip->bli_logged, first, nbytes); | |
99 | } | |
100 | ||
101 | /* | |
c41564b5 | 102 | * This function is called to verify that our callers have logged |
1da177e4 LT |
103 | * all the bytes that they changed. |
104 | * | |
105 | * It does this by comparing the original copy of the buffer stored in | |
106 | * the buf log item's bli_orig array to the current copy of the buffer | |
c41564b5 | 107 | * and ensuring that all bytes which mismatch are set in the bli_logged |
1da177e4 LT |
108 | * array of the buf log item. |
109 | */ | |
110 | STATIC void | |
111 | xfs_buf_item_log_check( | |
112 | xfs_buf_log_item_t *bip) | |
113 | { | |
114 | char *orig; | |
115 | char *buffer; | |
116 | int x; | |
117 | xfs_buf_t *bp; | |
118 | ||
119 | ASSERT(bip->bli_orig != NULL); | |
120 | ASSERT(bip->bli_logged != NULL); | |
121 | ||
122 | bp = bip->bli_buf; | |
123 | ASSERT(XFS_BUF_COUNT(bp) > 0); | |
124 | ASSERT(XFS_BUF_PTR(bp) != NULL); | |
125 | orig = bip->bli_orig; | |
126 | buffer = XFS_BUF_PTR(bp); | |
127 | for (x = 0; x < XFS_BUF_COUNT(bp); x++) { | |
128 | if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) | |
129 | cmn_err(CE_PANIC, | |
130 | "xfs_buf_item_log_check bip %x buffer %x orig %x index %d", | |
131 | bip, bp, orig, x); | |
132 | } | |
133 | } | |
134 | #else | |
135 | #define xfs_buf_item_log_debug(x,y,z) | |
136 | #define xfs_buf_item_log_check(x) | |
137 | #endif | |
138 | ||
139 | STATIC void xfs_buf_error_relse(xfs_buf_t *bp); | |
140 | STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip); | |
141 | ||
142 | /* | |
143 | * This returns the number of log iovecs needed to log the | |
144 | * given buf log item. | |
145 | * | |
146 | * It calculates this as 1 iovec for the buf log format structure | |
147 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
148 | * Contiguous chunks are logged in a single iovec. | |
149 | * | |
150 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
151 | */ | |
ba0f32d4 | 152 | STATIC uint |
1da177e4 LT |
153 | xfs_buf_item_size( |
154 | xfs_buf_log_item_t *bip) | |
155 | { | |
156 | uint nvecs; | |
157 | int next_bit; | |
158 | int last_bit; | |
159 | xfs_buf_t *bp; | |
160 | ||
161 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
162 | if (bip->bli_flags & XFS_BLI_STALE) { | |
163 | /* | |
164 | * The buffer is stale, so all we need to log | |
165 | * is the buf log format structure with the | |
166 | * cancel flag in it. | |
167 | */ | |
0b1b213f | 168 | trace_xfs_buf_item_size_stale(bip); |
c1155410 | 169 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
1da177e4 LT |
170 | return 1; |
171 | } | |
172 | ||
173 | bp = bip->bli_buf; | |
174 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
175 | nvecs = 1; | |
176 | last_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
177 | bip->bli_format.blf_map_size, 0); | |
178 | ASSERT(last_bit != -1); | |
179 | nvecs++; | |
180 | while (last_bit != -1) { | |
181 | /* | |
182 | * This takes the bit number to start looking from and | |
183 | * returns the next set bit from there. It returns -1 | |
184 | * if there are no more bits set or the start bit is | |
185 | * beyond the end of the bitmap. | |
186 | */ | |
187 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
188 | bip->bli_format.blf_map_size, | |
189 | last_bit + 1); | |
190 | /* | |
191 | * If we run out of bits, leave the loop, | |
192 | * else if we find a new set of bits bump the number of vecs, | |
193 | * else keep scanning the current set of bits. | |
194 | */ | |
195 | if (next_bit == -1) { | |
196 | last_bit = -1; | |
197 | } else if (next_bit != last_bit + 1) { | |
198 | last_bit = next_bit; | |
199 | nvecs++; | |
c1155410 DC |
200 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
201 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
202 | XFS_BLF_CHUNK)) { | |
1da177e4 LT |
203 | last_bit = next_bit; |
204 | nvecs++; | |
205 | } else { | |
206 | last_bit++; | |
207 | } | |
208 | } | |
209 | ||
0b1b213f | 210 | trace_xfs_buf_item_size(bip); |
1da177e4 LT |
211 | return nvecs; |
212 | } | |
213 | ||
214 | /* | |
215 | * This is called to fill in the vector of log iovecs for the | |
216 | * given log buf item. It fills the first entry with a buf log | |
217 | * format structure, and the rest point to contiguous chunks | |
218 | * within the buffer. | |
219 | */ | |
ba0f32d4 | 220 | STATIC void |
1da177e4 LT |
221 | xfs_buf_item_format( |
222 | xfs_buf_log_item_t *bip, | |
223 | xfs_log_iovec_t *log_vector) | |
224 | { | |
225 | uint base_size; | |
226 | uint nvecs; | |
227 | xfs_log_iovec_t *vecp; | |
228 | xfs_buf_t *bp; | |
229 | int first_bit; | |
230 | int last_bit; | |
231 | int next_bit; | |
232 | uint nbits; | |
233 | uint buffer_offset; | |
234 | ||
235 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
236 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
237 | (bip->bli_flags & XFS_BLI_STALE)); | |
238 | bp = bip->bli_buf; | |
1da177e4 LT |
239 | vecp = log_vector; |
240 | ||
241 | /* | |
242 | * The size of the base structure is the size of the | |
243 | * declared structure plus the space for the extra words | |
244 | * of the bitmap. We subtract one from the map size, because | |
245 | * the first element of the bitmap is accounted for in the | |
246 | * size of the base structure. | |
247 | */ | |
248 | base_size = | |
249 | (uint)(sizeof(xfs_buf_log_format_t) + | |
250 | ((bip->bli_format.blf_map_size - 1) * sizeof(uint))); | |
251 | vecp->i_addr = (xfs_caddr_t)&bip->bli_format; | |
252 | vecp->i_len = base_size; | |
4139b3b3 | 253 | vecp->i_type = XLOG_REG_TYPE_BFORMAT; |
1da177e4 LT |
254 | vecp++; |
255 | nvecs = 1; | |
256 | ||
ccf7c23f DC |
257 | /* |
258 | * If it is an inode buffer, transfer the in-memory state to the | |
259 | * format flags and clear the in-memory state. We do not transfer | |
260 | * this state if the inode buffer allocation has not yet been committed | |
261 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
262 | * correct replay of the inode allocation. | |
263 | */ | |
264 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
265 | if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
266 | xfs_log_item_in_current_chkpt(&bip->bli_item))) | |
267 | bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF; | |
268 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; | |
269 | } | |
270 | ||
1da177e4 LT |
271 | if (bip->bli_flags & XFS_BLI_STALE) { |
272 | /* | |
273 | * The buffer is stale, so all we need to log | |
274 | * is the buf log format structure with the | |
275 | * cancel flag in it. | |
276 | */ | |
0b1b213f | 277 | trace_xfs_buf_item_format_stale(bip); |
c1155410 | 278 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
1da177e4 LT |
279 | bip->bli_format.blf_size = nvecs; |
280 | return; | |
281 | } | |
282 | ||
283 | /* | |
284 | * Fill in an iovec for each set of contiguous chunks. | |
285 | */ | |
286 | first_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
287 | bip->bli_format.blf_map_size, 0); | |
288 | ASSERT(first_bit != -1); | |
289 | last_bit = first_bit; | |
290 | nbits = 1; | |
291 | for (;;) { | |
292 | /* | |
293 | * This takes the bit number to start looking from and | |
294 | * returns the next set bit from there. It returns -1 | |
295 | * if there are no more bits set or the start bit is | |
296 | * beyond the end of the bitmap. | |
297 | */ | |
298 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, | |
299 | bip->bli_format.blf_map_size, | |
300 | (uint)last_bit + 1); | |
301 | /* | |
302 | * If we run out of bits fill in the last iovec and get | |
303 | * out of the loop. | |
304 | * Else if we start a new set of bits then fill in the | |
305 | * iovec for the series we were looking at and start | |
306 | * counting the bits in the new one. | |
307 | * Else we're still in the same set of bits so just | |
308 | * keep counting and scanning. | |
309 | */ | |
310 | if (next_bit == -1) { | |
c1155410 | 311 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
1da177e4 | 312 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 313 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 314 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
315 | nvecs++; |
316 | break; | |
317 | } else if (next_bit != last_bit + 1) { | |
c1155410 | 318 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
1da177e4 | 319 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 320 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 321 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
322 | nvecs++; |
323 | vecp++; | |
324 | first_bit = next_bit; | |
325 | last_bit = next_bit; | |
326 | nbits = 1; | |
c1155410 DC |
327 | } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) != |
328 | (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) + | |
329 | XFS_BLF_CHUNK)) { | |
330 | buffer_offset = first_bit * XFS_BLF_CHUNK; | |
1da177e4 | 331 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 332 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 333 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
334 | /* You would think we need to bump the nvecs here too, but we do not |
335 | * this number is used by recovery, and it gets confused by the boundary | |
336 | * split here | |
337 | * nvecs++; | |
338 | */ | |
339 | vecp++; | |
340 | first_bit = next_bit; | |
341 | last_bit = next_bit; | |
342 | nbits = 1; | |
343 | } else { | |
344 | last_bit++; | |
345 | nbits++; | |
346 | } | |
347 | } | |
348 | bip->bli_format.blf_size = nvecs; | |
349 | ||
350 | /* | |
351 | * Check to make sure everything is consistent. | |
352 | */ | |
0b1b213f | 353 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
354 | xfs_buf_item_log_check(bip); |
355 | } | |
356 | ||
357 | /* | |
64fc35de DC |
358 | * This is called to pin the buffer associated with the buf log item in memory |
359 | * so it cannot be written out. Simply call bpin() on the buffer to do this. | |
360 | * | |
361 | * We also always take a reference to the buffer log item here so that the bli | |
362 | * is held while the item is pinned in memory. This means that we can | |
363 | * unconditionally drop the reference count a transaction holds when the | |
364 | * transaction is completed. | |
1da177e4 | 365 | */ |
64fc35de | 366 | |
ba0f32d4 | 367 | STATIC void |
1da177e4 LT |
368 | xfs_buf_item_pin( |
369 | xfs_buf_log_item_t *bip) | |
370 | { | |
371 | xfs_buf_t *bp; | |
372 | ||
373 | bp = bip->bli_buf; | |
374 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
375 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
376 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
377 | (bip->bli_flags & XFS_BLI_STALE)); | |
64fc35de | 378 | atomic_inc(&bip->bli_refcount); |
0b1b213f | 379 | trace_xfs_buf_item_pin(bip); |
1da177e4 LT |
380 | xfs_bpin(bp); |
381 | } | |
382 | ||
383 | ||
384 | /* | |
385 | * This is called to unpin the buffer associated with the buf log | |
386 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
387 | * Just call bunpin() on the buffer to do this. | |
388 | * | |
389 | * Also drop the reference to the buf item for the current transaction. | |
390 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
391 | * then free up the buf log item and unlock the buffer. | |
392 | */ | |
ba0f32d4 | 393 | STATIC void |
1da177e4 | 394 | xfs_buf_item_unpin( |
8e123850 | 395 | xfs_buf_log_item_t *bip) |
1da177e4 | 396 | { |
783a2f65 | 397 | struct xfs_ail *ailp; |
1da177e4 LT |
398 | xfs_buf_t *bp; |
399 | int freed; | |
8e123850 | 400 | int stale = bip->bli_flags & XFS_BLI_STALE; |
1da177e4 LT |
401 | |
402 | bp = bip->bli_buf; | |
403 | ASSERT(bp != NULL); | |
404 | ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip); | |
405 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 406 | trace_xfs_buf_item_unpin(bip); |
1da177e4 LT |
407 | |
408 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
783a2f65 | 409 | ailp = bip->bli_item.li_ailp; |
1da177e4 LT |
410 | xfs_bunpin(bp); |
411 | if (freed && stale) { | |
412 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
413 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
414 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); | |
415 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
c1155410 | 416 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
0b1b213f CH |
417 | trace_xfs_buf_item_unpin_stale(bip); |
418 | ||
1da177e4 LT |
419 | /* |
420 | * If we get called here because of an IO error, we may | |
783a2f65 | 421 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
1da177e4 | 422 | * will take care of that situation. |
783a2f65 | 423 | * xfs_trans_ail_delete() drops the AIL lock. |
1da177e4 LT |
424 | */ |
425 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
426 | xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip); | |
427 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
428 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
429 | } else { | |
783a2f65 DC |
430 | spin_lock(&ailp->xa_lock); |
431 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip); | |
1da177e4 LT |
432 | xfs_buf_item_relse(bp); |
433 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL); | |
434 | } | |
435 | xfs_buf_relse(bp); | |
436 | } | |
437 | } | |
438 | ||
439 | /* | |
440 | * this is called from uncommit in the forced-shutdown path. | |
441 | * we need to check to see if the reference count on the log item | |
442 | * is going to drop to zero. If so, unpin will free the log item | |
443 | * so we need to free the item's descriptor (that points to the item) | |
444 | * in the transaction. | |
445 | */ | |
ba0f32d4 | 446 | STATIC void |
1da177e4 LT |
447 | xfs_buf_item_unpin_remove( |
448 | xfs_buf_log_item_t *bip, | |
449 | xfs_trans_t *tp) | |
450 | { | |
8e123850 | 451 | /* will xfs_buf_item_unpin() call xfs_buf_item_relse()? */ |
1da177e4 LT |
452 | if ((atomic_read(&bip->bli_refcount) == 1) && |
453 | (bip->bli_flags & XFS_BLI_STALE)) { | |
8e123850 DC |
454 | /* |
455 | * yes -- We can safely do some work here and then call | |
456 | * buf_item_unpin to do the rest because we are | |
457 | * are holding the buffer locked so no one else will be | |
458 | * able to bump up the refcount. We have to remove the | |
459 | * log item from the transaction as we are about to release | |
460 | * our reference to the buffer. If we don't, the unlock that | |
461 | * occurs later in the xfs_trans_uncommit() will try to | |
462 | * reference the buffer which we no longer have a hold on. | |
463 | */ | |
464 | struct xfs_log_item_desc *lidp; | |
465 | ||
1da177e4 | 466 | ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0); |
0b1b213f CH |
467 | trace_xfs_buf_item_unpin_stale(bip); |
468 | ||
8e123850 | 469 | lidp = xfs_trans_find_item(tp, (xfs_log_item_t *)bip); |
1da177e4 | 470 | xfs_trans_free_item(tp, lidp); |
8e123850 | 471 | |
1da177e4 | 472 | /* |
8e123850 DC |
473 | * Since the transaction no longer refers to the buffer, the |
474 | * buffer should no longer refer to the transaction. | |
1da177e4 | 475 | */ |
8e123850 | 476 | XFS_BUF_SET_FSPRIVATE2(bip->bli_buf, NULL); |
1da177e4 | 477 | } |
8e123850 | 478 | xfs_buf_item_unpin(bip); |
1da177e4 LT |
479 | } |
480 | ||
481 | /* | |
482 | * This is called to attempt to lock the buffer associated with this | |
483 | * buf log item. Don't sleep on the buffer lock. If we can't get | |
d808f617 DC |
484 | * the lock right away, return 0. If we can get the lock, take a |
485 | * reference to the buffer. If this is a delayed write buffer that | |
486 | * needs AIL help to be written back, invoke the pushbuf routine | |
487 | * rather than the normal success path. | |
1da177e4 | 488 | */ |
ba0f32d4 | 489 | STATIC uint |
1da177e4 LT |
490 | xfs_buf_item_trylock( |
491 | xfs_buf_log_item_t *bip) | |
492 | { | |
493 | xfs_buf_t *bp; | |
494 | ||
495 | bp = bip->bli_buf; | |
d808f617 | 496 | if (XFS_BUF_ISPINNED(bp)) |
1da177e4 | 497 | return XFS_ITEM_PINNED; |
d808f617 | 498 | if (!XFS_BUF_CPSEMA(bp)) |
1da177e4 | 499 | return XFS_ITEM_LOCKED; |
1da177e4 | 500 | |
d808f617 | 501 | /* take a reference to the buffer. */ |
1da177e4 LT |
502 | XFS_BUF_HOLD(bp); |
503 | ||
504 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
0b1b213f | 505 | trace_xfs_buf_item_trylock(bip); |
d808f617 DC |
506 | if (XFS_BUF_ISDELAYWRITE(bp)) |
507 | return XFS_ITEM_PUSHBUF; | |
1da177e4 LT |
508 | return XFS_ITEM_SUCCESS; |
509 | } | |
510 | ||
511 | /* | |
64fc35de DC |
512 | * Release the buffer associated with the buf log item. If there is no dirty |
513 | * logged data associated with the buffer recorded in the buf log item, then | |
514 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 515 | * |
64fc35de DC |
516 | * This call ignores the recursion count. It is only called when the buffer |
517 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 518 | * |
64fc35de DC |
519 | * We unconditionally drop the transaction's reference to the log item. If the |
520 | * item was logged, then another reference was taken when it was pinned, so we | |
521 | * can safely drop the transaction reference now. This also allows us to avoid | |
522 | * potential races with the unpin code freeing the bli by not referencing the | |
523 | * bli after we've dropped the reference count. | |
524 | * | |
525 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
526 | * if necessary but do not unlock the buffer. This is for support of | |
527 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
528 | * free the item. | |
1da177e4 | 529 | */ |
ba0f32d4 | 530 | STATIC void |
1da177e4 LT |
531 | xfs_buf_item_unlock( |
532 | xfs_buf_log_item_t *bip) | |
533 | { | |
534 | int aborted; | |
535 | xfs_buf_t *bp; | |
536 | uint hold; | |
537 | ||
538 | bp = bip->bli_buf; | |
1da177e4 | 539 | |
64fc35de | 540 | /* Clear the buffer's association with this transaction. */ |
1da177e4 LT |
541 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); |
542 | ||
543 | /* | |
64fc35de DC |
544 | * If this is a transaction abort, don't return early. Instead, allow |
545 | * the brelse to happen. Normally it would be done for stale | |
546 | * (cancelled) buffers at unpin time, but we'll never go through the | |
547 | * pin/unpin cycle if we abort inside commit. | |
1da177e4 LT |
548 | */ |
549 | aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0; | |
550 | ||
551 | /* | |
64fc35de DC |
552 | * Before possibly freeing the buf item, determine if we should |
553 | * release the buffer at the end of this routine. | |
554 | */ | |
555 | hold = bip->bli_flags & XFS_BLI_HOLD; | |
556 | ||
557 | /* Clear the per transaction state. */ | |
558 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD); | |
559 | ||
560 | /* | |
561 | * If the buf item is marked stale, then don't do anything. We'll | |
562 | * unlock the buffer and free the buf item when the buffer is unpinned | |
563 | * for the last time. | |
1da177e4 LT |
564 | */ |
565 | if (bip->bli_flags & XFS_BLI_STALE) { | |
0b1b213f | 566 | trace_xfs_buf_item_unlock_stale(bip); |
c1155410 | 567 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
64fc35de DC |
568 | if (!aborted) { |
569 | atomic_dec(&bip->bli_refcount); | |
1da177e4 | 570 | return; |
64fc35de | 571 | } |
1da177e4 LT |
572 | } |
573 | ||
0b1b213f | 574 | trace_xfs_buf_item_unlock(bip); |
1da177e4 LT |
575 | |
576 | /* | |
64fc35de DC |
577 | * If the buf item isn't tracking any data, free it, otherwise drop the |
578 | * reference we hold to it. | |
1da177e4 | 579 | */ |
24ad33ff | 580 | if (xfs_bitmap_empty(bip->bli_format.blf_data_map, |
64fc35de | 581 | bip->bli_format.blf_map_size)) |
1da177e4 | 582 | xfs_buf_item_relse(bp); |
64fc35de DC |
583 | else |
584 | atomic_dec(&bip->bli_refcount); | |
1da177e4 | 585 | |
64fc35de | 586 | if (!hold) |
1da177e4 | 587 | xfs_buf_relse(bp); |
1da177e4 LT |
588 | } |
589 | ||
590 | /* | |
591 | * This is called to find out where the oldest active copy of the | |
592 | * buf log item in the on disk log resides now that the last log | |
593 | * write of it completed at the given lsn. | |
594 | * We always re-log all the dirty data in a buffer, so usually the | |
595 | * latest copy in the on disk log is the only one that matters. For | |
596 | * those cases we simply return the given lsn. | |
597 | * | |
598 | * The one exception to this is for buffers full of newly allocated | |
599 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
600 | * flag set, indicating that only the di_next_unlinked fields from the | |
601 | * inodes in the buffers will be replayed during recovery. If the | |
602 | * original newly allocated inode images have not yet been flushed | |
603 | * when the buffer is so relogged, then we need to make sure that we | |
604 | * keep the old images in the 'active' portion of the log. We do this | |
605 | * by returning the original lsn of that transaction here rather than | |
606 | * the current one. | |
607 | */ | |
ba0f32d4 | 608 | STATIC xfs_lsn_t |
1da177e4 LT |
609 | xfs_buf_item_committed( |
610 | xfs_buf_log_item_t *bip, | |
611 | xfs_lsn_t lsn) | |
612 | { | |
0b1b213f CH |
613 | trace_xfs_buf_item_committed(bip); |
614 | ||
1da177e4 LT |
615 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
616 | (bip->bli_item.li_lsn != 0)) { | |
617 | return bip->bli_item.li_lsn; | |
618 | } | |
619 | return (lsn); | |
620 | } | |
621 | ||
1da177e4 | 622 | /* |
d808f617 DC |
623 | * The buffer is locked, but is not a delayed write buffer. This happens |
624 | * if we race with IO completion and hence we don't want to try to write it | |
625 | * again. Just release the buffer. | |
1da177e4 | 626 | */ |
ba0f32d4 | 627 | STATIC void |
1da177e4 LT |
628 | xfs_buf_item_push( |
629 | xfs_buf_log_item_t *bip) | |
630 | { | |
631 | xfs_buf_t *bp; | |
632 | ||
633 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
0b1b213f | 634 | trace_xfs_buf_item_push(bip); |
1da177e4 LT |
635 | |
636 | bp = bip->bli_buf; | |
d808f617 DC |
637 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
638 | xfs_buf_relse(bp); | |
639 | } | |
1da177e4 | 640 | |
d808f617 DC |
641 | /* |
642 | * The buffer is locked and is a delayed write buffer. Promote the buffer | |
643 | * in the delayed write queue as the caller knows that they must invoke | |
644 | * the xfsbufd to get this buffer written. We have to unlock the buffer | |
645 | * to allow the xfsbufd to write it, too. | |
646 | */ | |
647 | STATIC void | |
648 | xfs_buf_item_pushbuf( | |
649 | xfs_buf_log_item_t *bip) | |
650 | { | |
651 | xfs_buf_t *bp; | |
652 | ||
653 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
654 | trace_xfs_buf_item_pushbuf(bip); | |
655 | ||
656 | bp = bip->bli_buf; | |
657 | ASSERT(XFS_BUF_ISDELAYWRITE(bp)); | |
658 | xfs_buf_delwri_promote(bp); | |
659 | xfs_buf_relse(bp); | |
1da177e4 LT |
660 | } |
661 | ||
662 | /* ARGSUSED */ | |
ba0f32d4 | 663 | STATIC void |
1da177e4 LT |
664 | xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn) |
665 | { | |
666 | } | |
667 | ||
668 | /* | |
669 | * This is the ops vector shared by all buf log items. | |
670 | */ | |
7989cb8e | 671 | static struct xfs_item_ops xfs_buf_item_ops = { |
1da177e4 LT |
672 | .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size, |
673 | .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*)) | |
674 | xfs_buf_item_format, | |
675 | .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin, | |
8e123850 | 676 | .iop_unpin = (void(*)(xfs_log_item_t*))xfs_buf_item_unpin, |
1da177e4 LT |
677 | .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *)) |
678 | xfs_buf_item_unpin_remove, | |
679 | .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock, | |
680 | .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock, | |
681 | .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t)) | |
682 | xfs_buf_item_committed, | |
683 | .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push, | |
d808f617 | 684 | .iop_pushbuf = (void(*)(xfs_log_item_t*))xfs_buf_item_pushbuf, |
1da177e4 LT |
685 | .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t)) |
686 | xfs_buf_item_committing | |
687 | }; | |
688 | ||
689 | ||
690 | /* | |
691 | * Allocate a new buf log item to go with the given buffer. | |
692 | * Set the buffer's b_fsprivate field to point to the new | |
693 | * buf log item. If there are other item's attached to the | |
694 | * buffer (see xfs_buf_attach_iodone() below), then put the | |
695 | * buf log item at the front. | |
696 | */ | |
697 | void | |
698 | xfs_buf_item_init( | |
699 | xfs_buf_t *bp, | |
700 | xfs_mount_t *mp) | |
701 | { | |
702 | xfs_log_item_t *lip; | |
703 | xfs_buf_log_item_t *bip; | |
704 | int chunks; | |
705 | int map_size; | |
706 | ||
707 | /* | |
708 | * Check to see if there is already a buf log item for | |
709 | * this buffer. If there is, it is guaranteed to be | |
710 | * the first. If we do already have one, there is | |
711 | * nothing to do here so return. | |
712 | */ | |
15ac08a8 CH |
713 | if (bp->b_mount != mp) |
714 | bp->b_mount = mp; | |
1da177e4 LT |
715 | XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb); |
716 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
717 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
718 | if (lip->li_type == XFS_LI_BUF) { | |
719 | return; | |
720 | } | |
721 | } | |
722 | ||
723 | /* | |
c1155410 | 724 | * chunks is the number of XFS_BLF_CHUNK size pieces |
1da177e4 LT |
725 | * the buffer can be divided into. Make sure not to |
726 | * truncate any pieces. map_size is the size of the | |
727 | * bitmap needed to describe the chunks of the buffer. | |
728 | */ | |
c1155410 | 729 | chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT); |
1da177e4 LT |
730 | map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT); |
731 | ||
732 | bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone, | |
733 | KM_SLEEP); | |
43f5efc5 | 734 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 735 | bip->bli_buf = bp; |
e1f5dbd7 | 736 | xfs_buf_hold(bp); |
1da177e4 LT |
737 | bip->bli_format.blf_type = XFS_LI_BUF; |
738 | bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp); | |
739 | bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp)); | |
740 | bip->bli_format.blf_map_size = map_size; | |
1da177e4 LT |
741 | |
742 | #ifdef XFS_TRANS_DEBUG | |
743 | /* | |
744 | * Allocate the arrays for tracking what needs to be logged | |
745 | * and what our callers request to be logged. bli_orig | |
746 | * holds a copy of the original, clean buffer for comparison | |
747 | * against, and bli_logged keeps a 1 bit flag per byte in | |
748 | * the buffer to indicate which bytes the callers have asked | |
749 | * to have logged. | |
750 | */ | |
751 | bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP); | |
752 | memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp)); | |
753 | bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP); | |
754 | #endif | |
755 | ||
756 | /* | |
757 | * Put the buf item into the list of items attached to the | |
758 | * buffer at the front. | |
759 | */ | |
760 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
761 | bip->bli_item.li_bio_list = | |
762 | XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
763 | } | |
764 | XFS_BUF_SET_FSPRIVATE(bp, bip); | |
765 | } | |
766 | ||
767 | ||
768 | /* | |
769 | * Mark bytes first through last inclusive as dirty in the buf | |
770 | * item's bitmap. | |
771 | */ | |
772 | void | |
773 | xfs_buf_item_log( | |
774 | xfs_buf_log_item_t *bip, | |
775 | uint first, | |
776 | uint last) | |
777 | { | |
778 | uint first_bit; | |
779 | uint last_bit; | |
780 | uint bits_to_set; | |
781 | uint bits_set; | |
782 | uint word_num; | |
783 | uint *wordp; | |
784 | uint bit; | |
785 | uint end_bit; | |
786 | uint mask; | |
787 | ||
788 | /* | |
789 | * Mark the item as having some dirty data for | |
790 | * quick reference in xfs_buf_item_dirty. | |
791 | */ | |
792 | bip->bli_flags |= XFS_BLI_DIRTY; | |
793 | ||
794 | /* | |
795 | * Convert byte offsets to bit numbers. | |
796 | */ | |
c1155410 DC |
797 | first_bit = first >> XFS_BLF_SHIFT; |
798 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
799 | |
800 | /* | |
801 | * Calculate the total number of bits to be set. | |
802 | */ | |
803 | bits_to_set = last_bit - first_bit + 1; | |
804 | ||
805 | /* | |
806 | * Get a pointer to the first word in the bitmap | |
807 | * to set a bit in. | |
808 | */ | |
809 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
810 | wordp = &(bip->bli_format.blf_data_map[word_num]); | |
811 | ||
812 | /* | |
813 | * Calculate the starting bit in the first word. | |
814 | */ | |
815 | bit = first_bit & (uint)(NBWORD - 1); | |
816 | ||
817 | /* | |
818 | * First set any bits in the first word of our range. | |
819 | * If it starts at bit 0 of the word, it will be | |
820 | * set below rather than here. That is what the variable | |
821 | * bit tells us. The variable bits_set tracks the number | |
822 | * of bits that have been set so far. End_bit is the number | |
823 | * of the last bit to be set in this word plus one. | |
824 | */ | |
825 | if (bit) { | |
826 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); | |
827 | mask = ((1 << (end_bit - bit)) - 1) << bit; | |
828 | *wordp |= mask; | |
829 | wordp++; | |
830 | bits_set = end_bit - bit; | |
831 | } else { | |
832 | bits_set = 0; | |
833 | } | |
834 | ||
835 | /* | |
836 | * Now set bits a whole word at a time that are between | |
837 | * first_bit and last_bit. | |
838 | */ | |
839 | while ((bits_to_set - bits_set) >= NBWORD) { | |
840 | *wordp |= 0xffffffff; | |
841 | bits_set += NBWORD; | |
842 | wordp++; | |
843 | } | |
844 | ||
845 | /* | |
846 | * Finally, set any bits left to be set in one last partial word. | |
847 | */ | |
848 | end_bit = bits_to_set - bits_set; | |
849 | if (end_bit) { | |
850 | mask = (1 << end_bit) - 1; | |
851 | *wordp |= mask; | |
852 | } | |
853 | ||
854 | xfs_buf_item_log_debug(bip, first, last); | |
855 | } | |
856 | ||
857 | ||
858 | /* | |
859 | * Return 1 if the buffer has some data that has been logged (at any | |
860 | * point, not just the current transaction) and 0 if not. | |
861 | */ | |
862 | uint | |
863 | xfs_buf_item_dirty( | |
864 | xfs_buf_log_item_t *bip) | |
865 | { | |
866 | return (bip->bli_flags & XFS_BLI_DIRTY); | |
867 | } | |
868 | ||
e1f5dbd7 LM |
869 | STATIC void |
870 | xfs_buf_item_free( | |
871 | xfs_buf_log_item_t *bip) | |
872 | { | |
873 | #ifdef XFS_TRANS_DEBUG | |
874 | kmem_free(bip->bli_orig); | |
875 | kmem_free(bip->bli_logged); | |
876 | #endif /* XFS_TRANS_DEBUG */ | |
877 | ||
e1f5dbd7 LM |
878 | kmem_zone_free(xfs_buf_item_zone, bip); |
879 | } | |
880 | ||
1da177e4 LT |
881 | /* |
882 | * This is called when the buf log item is no longer needed. It should | |
883 | * free the buf log item associated with the given buffer and clear | |
884 | * the buffer's pointer to the buf log item. If there are no more | |
885 | * items in the list, clear the b_iodone field of the buffer (see | |
886 | * xfs_buf_attach_iodone() below). | |
887 | */ | |
888 | void | |
889 | xfs_buf_item_relse( | |
890 | xfs_buf_t *bp) | |
891 | { | |
892 | xfs_buf_log_item_t *bip; | |
893 | ||
0b1b213f CH |
894 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
895 | ||
1da177e4 LT |
896 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
897 | XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list); | |
898 | if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) && | |
899 | (XFS_BUF_IODONE_FUNC(bp) != NULL)) { | |
1da177e4 LT |
900 | XFS_BUF_CLR_IODONE_FUNC(bp); |
901 | } | |
e1f5dbd7 LM |
902 | xfs_buf_rele(bp); |
903 | xfs_buf_item_free(bip); | |
1da177e4 LT |
904 | } |
905 | ||
906 | ||
907 | /* | |
908 | * Add the given log item with its callback to the list of callbacks | |
909 | * to be called when the buffer's I/O completes. If it is not set | |
910 | * already, set the buffer's b_iodone() routine to be | |
911 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
912 | * items rooted at b_fsprivate. Items are always added as the second | |
913 | * entry in the list if there is a first, because the buf item code | |
914 | * assumes that the buf log item is first. | |
915 | */ | |
916 | void | |
917 | xfs_buf_attach_iodone( | |
918 | xfs_buf_t *bp, | |
919 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), | |
920 | xfs_log_item_t *lip) | |
921 | { | |
922 | xfs_log_item_t *head_lip; | |
923 | ||
924 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
925 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
926 | ||
927 | lip->li_cb = cb; | |
928 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
929 | head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
930 | lip->li_bio_list = head_lip->li_bio_list; | |
931 | head_lip->li_bio_list = lip; | |
932 | } else { | |
933 | XFS_BUF_SET_FSPRIVATE(bp, lip); | |
934 | } | |
935 | ||
936 | ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) || | |
937 | (XFS_BUF_IODONE_FUNC(bp) == NULL)); | |
938 | XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks); | |
939 | } | |
940 | ||
941 | STATIC void | |
942 | xfs_buf_do_callbacks( | |
943 | xfs_buf_t *bp, | |
944 | xfs_log_item_t *lip) | |
945 | { | |
946 | xfs_log_item_t *nlip; | |
947 | ||
948 | while (lip != NULL) { | |
949 | nlip = lip->li_bio_list; | |
950 | ASSERT(lip->li_cb != NULL); | |
951 | /* | |
952 | * Clear the next pointer so we don't have any | |
953 | * confusion if the item is added to another buf. | |
954 | * Don't touch the log item after calling its | |
955 | * callback, because it could have freed itself. | |
956 | */ | |
957 | lip->li_bio_list = NULL; | |
958 | lip->li_cb(bp, lip); | |
959 | lip = nlip; | |
960 | } | |
961 | } | |
962 | ||
963 | /* | |
964 | * This is the iodone() function for buffers which have had callbacks | |
965 | * attached to them by xfs_buf_attach_iodone(). It should remove each | |
966 | * log item from the buffer's list and call the callback of each in turn. | |
967 | * When done, the buffer's fsprivate field is set to NULL and the buffer | |
968 | * is unlocked with a call to iodone(). | |
969 | */ | |
970 | void | |
971 | xfs_buf_iodone_callbacks( | |
972 | xfs_buf_t *bp) | |
973 | { | |
974 | xfs_log_item_t *lip; | |
975 | static ulong lasttime; | |
976 | static xfs_buftarg_t *lasttarg; | |
977 | xfs_mount_t *mp; | |
978 | ||
979 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
980 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
981 | ||
982 | if (XFS_BUF_GETERROR(bp) != 0) { | |
983 | /* | |
984 | * If we've already decided to shutdown the filesystem | |
985 | * because of IO errors, there's no point in giving this | |
986 | * a retry. | |
987 | */ | |
988 | mp = lip->li_mountp; | |
989 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
990 | ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp); | |
991 | XFS_BUF_SUPER_STALE(bp); | |
0b1b213f | 992 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
1da177e4 LT |
993 | xfs_buf_do_callbacks(bp, lip); |
994 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
995 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
4fdc7781 | 996 | xfs_biodone(bp); |
1da177e4 LT |
997 | return; |
998 | } | |
999 | ||
1000 | if ((XFS_BUF_TARGET(bp) != lasttarg) || | |
1001 | (time_after(jiffies, (lasttime + 5*HZ)))) { | |
1002 | lasttime = jiffies; | |
b6574520 NS |
1003 | cmn_err(CE_ALERT, "Device %s, XFS metadata write error" |
1004 | " block 0x%llx in %s", | |
1005 | XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)), | |
1da177e4 LT |
1006 | (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname); |
1007 | } | |
1008 | lasttarg = XFS_BUF_TARGET(bp); | |
1009 | ||
1010 | if (XFS_BUF_ISASYNC(bp)) { | |
1011 | /* | |
1012 | * If the write was asynchronous then noone will be | |
1013 | * looking for the error. Clear the error state | |
1014 | * and write the buffer out again delayed write. | |
1015 | * | |
1016 | * XXXsup This is OK, so long as we catch these | |
1017 | * before we start the umount; we don't want these | |
1018 | * DELWRI metadata bufs to be hanging around. | |
1019 | */ | |
1020 | XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */ | |
1021 | ||
1022 | if (!(XFS_BUF_ISSTALE(bp))) { | |
1023 | XFS_BUF_DELAYWRITE(bp); | |
1024 | XFS_BUF_DONE(bp); | |
1025 | XFS_BUF_SET_START(bp); | |
1026 | } | |
1027 | ASSERT(XFS_BUF_IODONE_FUNC(bp)); | |
0b1b213f | 1028 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); |
1da177e4 LT |
1029 | xfs_buf_relse(bp); |
1030 | } else { | |
1031 | /* | |
1032 | * If the write of the buffer was not asynchronous, | |
1033 | * then we want to make sure to return the error | |
1034 | * to the caller of bwrite(). Because of this we | |
1035 | * cannot clear the B_ERROR state at this point. | |
1036 | * Instead we install a callback function that | |
1037 | * will be called when the buffer is released, and | |
1038 | * that routine will clear the error state and | |
1039 | * set the buffer to be written out again after | |
1040 | * some delay. | |
1041 | */ | |
1042 | /* We actually overwrite the existing b-relse | |
1043 | function at times, but we're gonna be shutting down | |
1044 | anyway. */ | |
1045 | XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse); | |
1046 | XFS_BUF_DONE(bp); | |
b4dd330b | 1047 | XFS_BUF_FINISH_IOWAIT(bp); |
1da177e4 LT |
1048 | } |
1049 | return; | |
1050 | } | |
0b1b213f | 1051 | |
1da177e4 LT |
1052 | xfs_buf_do_callbacks(bp, lip); |
1053 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
1054 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
1055 | xfs_biodone(bp); | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * This is a callback routine attached to a buffer which gets an error | |
1060 | * when being written out synchronously. | |
1061 | */ | |
1062 | STATIC void | |
1063 | xfs_buf_error_relse( | |
1064 | xfs_buf_t *bp) | |
1065 | { | |
1066 | xfs_log_item_t *lip; | |
1067 | xfs_mount_t *mp; | |
1068 | ||
1069 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
1070 | mp = (xfs_mount_t *)lip->li_mountp; | |
1071 | ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp); | |
1072 | ||
1073 | XFS_BUF_STALE(bp); | |
1074 | XFS_BUF_DONE(bp); | |
1075 | XFS_BUF_UNDELAYWRITE(bp); | |
1076 | XFS_BUF_ERROR(bp,0); | |
0b1b213f CH |
1077 | |
1078 | trace_xfs_buf_error_relse(bp, _RET_IP_); | |
1079 | ||
1da177e4 | 1080 | if (! XFS_FORCED_SHUTDOWN(mp)) |
7d04a335 | 1081 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1082 | /* |
1083 | * We have to unpin the pinned buffers so do the | |
1084 | * callbacks. | |
1085 | */ | |
1086 | xfs_buf_do_callbacks(bp, lip); | |
1087 | XFS_BUF_SET_FSPRIVATE(bp, NULL); | |
1088 | XFS_BUF_CLR_IODONE_FUNC(bp); | |
1089 | XFS_BUF_SET_BRELSE_FUNC(bp,NULL); | |
1090 | xfs_buf_relse(bp); | |
1091 | } | |
1092 | ||
1093 | ||
1094 | /* | |
1095 | * This is the iodone() function for buffers which have been | |
1096 | * logged. It is called when they are eventually flushed out. | |
1097 | * It should remove the buf item from the AIL, and free the buf item. | |
1098 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1099 | * care of cleaning up the buffer itself. | |
1100 | */ | |
1101 | /* ARGSUSED */ | |
1102 | void | |
1103 | xfs_buf_iodone( | |
1104 | xfs_buf_t *bp, | |
1105 | xfs_buf_log_item_t *bip) | |
1106 | { | |
783a2f65 | 1107 | struct xfs_ail *ailp = bip->bli_item.li_ailp; |
1da177e4 LT |
1108 | |
1109 | ASSERT(bip->bli_buf == bp); | |
1110 | ||
e1f5dbd7 | 1111 | xfs_buf_rele(bp); |
1da177e4 LT |
1112 | |
1113 | /* | |
1114 | * If we are forcibly shutting down, this may well be | |
1115 | * off the AIL already. That's because we simulate the | |
1116 | * log-committed callbacks to unpin these buffers. Or we may never | |
1117 | * have put this item on AIL because of the transaction was | |
783a2f65 | 1118 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1da177e4 LT |
1119 | * |
1120 | * Either way, AIL is useless if we're forcing a shutdown. | |
1121 | */ | |
fc1829f3 | 1122 | spin_lock(&ailp->xa_lock); |
783a2f65 | 1123 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip); |
e1f5dbd7 | 1124 | xfs_buf_item_free(bip); |
1da177e4 | 1125 | } |