Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 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 LT |
25 | #include "xfs_sb.h" |
26 | #include "xfs_ag.h" | |
1da177e4 | 27 | #include "xfs_mount.h" |
a844f451 NS |
28 | #include "xfs_bmap_btree.h" |
29 | #include "xfs_alloc_btree.h" | |
30 | #include "xfs_ialloc_btree.h" | |
a844f451 NS |
31 | #include "xfs_dinode.h" |
32 | #include "xfs_inode.h" | |
33 | #include "xfs_buf_item.h" | |
1da177e4 LT |
34 | #include "xfs_trans_priv.h" |
35 | #include "xfs_error.h" | |
36 | #include "xfs_rw.h" | |
0b1b213f | 37 | #include "xfs_trace.h" |
1da177e4 | 38 | |
4a5224d7 CH |
39 | /* |
40 | * Check to see if a buffer matching the given parameters is already | |
41 | * a part of the given transaction. | |
42 | */ | |
43 | STATIC struct xfs_buf * | |
44 | xfs_trans_buf_item_match( | |
45 | struct xfs_trans *tp, | |
46 | struct xfs_buftarg *target, | |
47 | xfs_daddr_t blkno, | |
48 | int len) | |
49 | { | |
e98c414f CH |
50 | struct xfs_log_item_desc *lidp; |
51 | struct xfs_buf_log_item *blip; | |
1da177e4 | 52 | |
4a5224d7 | 53 | len = BBTOB(len); |
e98c414f CH |
54 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
55 | blip = (struct xfs_buf_log_item *)lidp->lid_item; | |
56 | if (blip->bli_item.li_type == XFS_LI_BUF && | |
49074c06 | 57 | blip->bli_buf->b_target == target && |
e98c414f CH |
58 | XFS_BUF_ADDR(blip->bli_buf) == blkno && |
59 | XFS_BUF_COUNT(blip->bli_buf) == len) | |
60 | return blip->bli_buf; | |
4a5224d7 CH |
61 | } |
62 | ||
63 | return NULL; | |
64 | } | |
1da177e4 | 65 | |
d7e84f41 CH |
66 | /* |
67 | * Add the locked buffer to the transaction. | |
68 | * | |
69 | * The buffer must be locked, and it cannot be associated with any | |
70 | * transaction. | |
71 | * | |
72 | * If the buffer does not yet have a buf log item associated with it, | |
73 | * then allocate one for it. Then add the buf item to the transaction. | |
74 | */ | |
75 | STATIC void | |
76 | _xfs_trans_bjoin( | |
77 | struct xfs_trans *tp, | |
78 | struct xfs_buf *bp, | |
79 | int reset_recur) | |
80 | { | |
81 | struct xfs_buf_log_item *bip; | |
82 | ||
bf9d9013 | 83 | ASSERT(bp->b_transp == NULL); |
d7e84f41 CH |
84 | |
85 | /* | |
86 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If | |
87 | * it doesn't have one yet, then allocate one and initialize it. | |
88 | * The checks to see if one is there are in xfs_buf_item_init(). | |
89 | */ | |
90 | xfs_buf_item_init(bp, tp->t_mountp); | |
adadbeef | 91 | bip = bp->b_fspriv; |
d7e84f41 | 92 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 93 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
d7e84f41 CH |
94 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
95 | if (reset_recur) | |
96 | bip->bli_recur = 0; | |
97 | ||
98 | /* | |
99 | * Take a reference for this transaction on the buf item. | |
100 | */ | |
101 | atomic_inc(&bip->bli_refcount); | |
102 | ||
103 | /* | |
104 | * Get a log_item_desc to point at the new item. | |
105 | */ | |
e98c414f | 106 | xfs_trans_add_item(tp, &bip->bli_item); |
d7e84f41 CH |
107 | |
108 | /* | |
109 | * Initialize b_fsprivate2 so we can find it with incore_match() | |
110 | * in xfs_trans_get_buf() and friends above. | |
111 | */ | |
bf9d9013 | 112 | bp->b_transp = tp; |
d7e84f41 CH |
113 | |
114 | } | |
115 | ||
116 | void | |
117 | xfs_trans_bjoin( | |
118 | struct xfs_trans *tp, | |
119 | struct xfs_buf *bp) | |
120 | { | |
121 | _xfs_trans_bjoin(tp, bp, 0); | |
122 | trace_xfs_trans_bjoin(bp->b_fspriv); | |
123 | } | |
1da177e4 LT |
124 | |
125 | /* | |
126 | * Get and lock the buffer for the caller if it is not already | |
127 | * locked within the given transaction. If it is already locked | |
128 | * within the transaction, just increment its lock recursion count | |
129 | * and return a pointer to it. | |
130 | * | |
1da177e4 LT |
131 | * If the transaction pointer is NULL, make this just a normal |
132 | * get_buf() call. | |
133 | */ | |
134 | xfs_buf_t * | |
135 | xfs_trans_get_buf(xfs_trans_t *tp, | |
136 | xfs_buftarg_t *target_dev, | |
137 | xfs_daddr_t blkno, | |
138 | int len, | |
139 | uint flags) | |
140 | { | |
141 | xfs_buf_t *bp; | |
142 | xfs_buf_log_item_t *bip; | |
143 | ||
144 | if (flags == 0) | |
0cadda1c | 145 | flags = XBF_LOCK | XBF_MAPPED; |
1da177e4 LT |
146 | |
147 | /* | |
148 | * Default to a normal get_buf() call if the tp is NULL. | |
149 | */ | |
6ad112bf | 150 | if (tp == NULL) |
0cadda1c CH |
151 | return xfs_buf_get(target_dev, blkno, len, |
152 | flags | XBF_DONT_BLOCK); | |
1da177e4 LT |
153 | |
154 | /* | |
155 | * If we find the buffer in the cache with this transaction | |
156 | * pointer in its b_fsprivate2 field, then we know we already | |
157 | * have it locked. In this case we just increment the lock | |
158 | * recursion count and return the buffer to the caller. | |
159 | */ | |
4a5224d7 | 160 | bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len); |
1da177e4 | 161 | if (bp != NULL) { |
0c842ad4 | 162 | ASSERT(xfs_buf_islocked(bp)); |
c867cb61 CH |
163 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) { |
164 | xfs_buf_stale(bp); | |
c867cb61 CH |
165 | XFS_BUF_DONE(bp); |
166 | } | |
0b1b213f | 167 | |
1da177e4 LT |
168 | /* |
169 | * If the buffer is stale then it was binval'ed | |
170 | * since last read. This doesn't matter since the | |
171 | * caller isn't allowed to use the data anyway. | |
172 | */ | |
0b1b213f | 173 | else if (XFS_BUF_ISSTALE(bp)) |
1da177e4 | 174 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
0b1b213f | 175 | |
bf9d9013 | 176 | ASSERT(bp->b_transp == tp); |
adadbeef | 177 | bip = bp->b_fspriv; |
1da177e4 LT |
178 | ASSERT(bip != NULL); |
179 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
180 | bip->bli_recur++; | |
0b1b213f | 181 | trace_xfs_trans_get_buf_recur(bip); |
1da177e4 LT |
182 | return (bp); |
183 | } | |
184 | ||
185 | /* | |
0cadda1c CH |
186 | * We always specify the XBF_DONT_BLOCK flag within a transaction |
187 | * so that get_buf does not try to push out a delayed write buffer | |
1da177e4 LT |
188 | * which might cause another transaction to take place (if the |
189 | * buffer was delayed alloc). Such recursive transactions can | |
190 | * easily deadlock with our current transaction as well as cause | |
191 | * us to run out of stack space. | |
192 | */ | |
0cadda1c | 193 | bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 LT |
194 | if (bp == NULL) { |
195 | return NULL; | |
196 | } | |
197 | ||
5a52c2a5 | 198 | ASSERT(!bp->b_error); |
1da177e4 | 199 | |
d7e84f41 CH |
200 | _xfs_trans_bjoin(tp, bp, 1); |
201 | trace_xfs_trans_get_buf(bp->b_fspriv); | |
1da177e4 LT |
202 | return (bp); |
203 | } | |
204 | ||
205 | /* | |
206 | * Get and lock the superblock buffer of this file system for the | |
207 | * given transaction. | |
208 | * | |
209 | * We don't need to use incore_match() here, because the superblock | |
210 | * buffer is a private buffer which we keep a pointer to in the | |
211 | * mount structure. | |
212 | */ | |
213 | xfs_buf_t * | |
214 | xfs_trans_getsb(xfs_trans_t *tp, | |
215 | struct xfs_mount *mp, | |
216 | int flags) | |
217 | { | |
218 | xfs_buf_t *bp; | |
219 | xfs_buf_log_item_t *bip; | |
220 | ||
221 | /* | |
222 | * Default to just trying to lock the superblock buffer | |
223 | * if tp is NULL. | |
224 | */ | |
225 | if (tp == NULL) { | |
226 | return (xfs_getsb(mp, flags)); | |
227 | } | |
228 | ||
229 | /* | |
230 | * If the superblock buffer already has this transaction | |
231 | * pointer in its b_fsprivate2 field, then we know we already | |
232 | * have it locked. In this case we just increment the lock | |
233 | * recursion count and return the buffer to the caller. | |
234 | */ | |
235 | bp = mp->m_sb_bp; | |
bf9d9013 | 236 | if (bp->b_transp == tp) { |
adadbeef | 237 | bip = bp->b_fspriv; |
1da177e4 LT |
238 | ASSERT(bip != NULL); |
239 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
240 | bip->bli_recur++; | |
0b1b213f | 241 | trace_xfs_trans_getsb_recur(bip); |
1da177e4 LT |
242 | return (bp); |
243 | } | |
244 | ||
245 | bp = xfs_getsb(mp, flags); | |
d7e84f41 | 246 | if (bp == NULL) |
1da177e4 | 247 | return NULL; |
1da177e4 | 248 | |
d7e84f41 CH |
249 | _xfs_trans_bjoin(tp, bp, 1); |
250 | trace_xfs_trans_getsb(bp->b_fspriv); | |
1da177e4 LT |
251 | return (bp); |
252 | } | |
253 | ||
254 | #ifdef DEBUG | |
255 | xfs_buftarg_t *xfs_error_target; | |
256 | int xfs_do_error; | |
257 | int xfs_req_num; | |
258 | int xfs_error_mod = 33; | |
259 | #endif | |
260 | ||
261 | /* | |
262 | * Get and lock the buffer for the caller if it is not already | |
263 | * locked within the given transaction. If it has not yet been | |
264 | * read in, read it from disk. If it is already locked | |
265 | * within the transaction and already read in, just increment its | |
266 | * lock recursion count and return a pointer to it. | |
267 | * | |
1da177e4 LT |
268 | * If the transaction pointer is NULL, make this just a normal |
269 | * read_buf() call. | |
270 | */ | |
271 | int | |
272 | xfs_trans_read_buf( | |
273 | xfs_mount_t *mp, | |
274 | xfs_trans_t *tp, | |
275 | xfs_buftarg_t *target, | |
276 | xfs_daddr_t blkno, | |
277 | int len, | |
278 | uint flags, | |
279 | xfs_buf_t **bpp) | |
280 | { | |
281 | xfs_buf_t *bp; | |
282 | xfs_buf_log_item_t *bip; | |
283 | int error; | |
284 | ||
285 | if (flags == 0) | |
0cadda1c | 286 | flags = XBF_LOCK | XBF_MAPPED; |
1da177e4 LT |
287 | |
288 | /* | |
289 | * Default to a normal get_buf() call if the tp is NULL. | |
290 | */ | |
291 | if (tp == NULL) { | |
0cadda1c | 292 | bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 | 293 | if (!bp) |
0cadda1c | 294 | return (flags & XBF_TRYLOCK) ? |
a3f74ffb | 295 | EAGAIN : XFS_ERROR(ENOMEM); |
1da177e4 | 296 | |
5a52c2a5 CS |
297 | if (bp->b_error) { |
298 | error = bp->b_error; | |
901796af | 299 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
300 | xfs_buf_relse(bp); |
301 | return error; | |
302 | } | |
303 | #ifdef DEBUG | |
a0f7bfd3 | 304 | if (xfs_do_error) { |
1da177e4 LT |
305 | if (xfs_error_target == target) { |
306 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
307 | xfs_buf_relse(bp); | |
0b932ccc | 308 | xfs_debug(mp, "Returning error!"); |
1da177e4 LT |
309 | return XFS_ERROR(EIO); |
310 | } | |
311 | } | |
312 | } | |
313 | #endif | |
314 | if (XFS_FORCED_SHUTDOWN(mp)) | |
315 | goto shutdown_abort; | |
316 | *bpp = bp; | |
317 | return 0; | |
318 | } | |
319 | ||
320 | /* | |
321 | * If we find the buffer in the cache with this transaction | |
322 | * pointer in its b_fsprivate2 field, then we know we already | |
323 | * have it locked. If it is already read in we just increment | |
324 | * the lock recursion count and return the buffer to the caller. | |
325 | * If the buffer is not yet read in, then we read it in, increment | |
326 | * the lock recursion count, and return it to the caller. | |
327 | */ | |
4a5224d7 | 328 | bp = xfs_trans_buf_item_match(tp, target, blkno, len); |
1da177e4 | 329 | if (bp != NULL) { |
0c842ad4 | 330 | ASSERT(xfs_buf_islocked(bp)); |
bf9d9013 | 331 | ASSERT(bp->b_transp == tp); |
adadbeef | 332 | ASSERT(bp->b_fspriv != NULL); |
5a52c2a5 | 333 | ASSERT(!bp->b_error); |
1da177e4 | 334 | if (!(XFS_BUF_ISDONE(bp))) { |
0b1b213f | 335 | trace_xfs_trans_read_buf_io(bp, _RET_IP_); |
1da177e4 LT |
336 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
337 | XFS_BUF_READ(bp); | |
338 | xfsbdstrat(tp->t_mountp, bp); | |
1a1a3e97 | 339 | error = xfs_buf_iowait(bp); |
d64e31a2 | 340 | if (error) { |
901796af | 341 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
342 | xfs_buf_relse(bp); |
343 | /* | |
d64e31a2 DC |
344 | * We can gracefully recover from most read |
345 | * errors. Ones we can't are those that happen | |
346 | * after the transaction's already dirty. | |
1da177e4 LT |
347 | */ |
348 | if (tp->t_flags & XFS_TRANS_DIRTY) | |
349 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 350 | SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
351 | return error; |
352 | } | |
353 | } | |
354 | /* | |
355 | * We never locked this buf ourselves, so we shouldn't | |
356 | * brelse it either. Just get out. | |
357 | */ | |
358 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
0b1b213f | 359 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
360 | *bpp = NULL; |
361 | return XFS_ERROR(EIO); | |
362 | } | |
363 | ||
364 | ||
adadbeef | 365 | bip = bp->b_fspriv; |
1da177e4 LT |
366 | bip->bli_recur++; |
367 | ||
368 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 369 | trace_xfs_trans_read_buf_recur(bip); |
1da177e4 LT |
370 | *bpp = bp; |
371 | return 0; | |
372 | } | |
373 | ||
374 | /* | |
0cadda1c CH |
375 | * We always specify the XBF_DONT_BLOCK flag within a transaction |
376 | * so that get_buf does not try to push out a delayed write buffer | |
1da177e4 LT |
377 | * which might cause another transaction to take place (if the |
378 | * buffer was delayed alloc). Such recursive transactions can | |
379 | * easily deadlock with our current transaction as well as cause | |
380 | * us to run out of stack space. | |
381 | */ | |
0cadda1c | 382 | bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 LT |
383 | if (bp == NULL) { |
384 | *bpp = NULL; | |
7401aafd DC |
385 | return (flags & XBF_TRYLOCK) ? |
386 | 0 : XFS_ERROR(ENOMEM); | |
1da177e4 | 387 | } |
5a52c2a5 CS |
388 | if (bp->b_error) { |
389 | error = bp->b_error; | |
c867cb61 | 390 | xfs_buf_stale(bp); |
c867cb61 | 391 | XFS_BUF_DONE(bp); |
901796af | 392 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 | 393 | if (tp->t_flags & XFS_TRANS_DIRTY) |
7d04a335 | 394 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
395 | xfs_buf_relse(bp); |
396 | return error; | |
397 | } | |
398 | #ifdef DEBUG | |
399 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { | |
400 | if (xfs_error_target == target) { | |
401 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
402 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 403 | SHUTDOWN_META_IO_ERROR); |
1da177e4 | 404 | xfs_buf_relse(bp); |
0b932ccc | 405 | xfs_debug(mp, "Returning trans error!"); |
1da177e4 LT |
406 | return XFS_ERROR(EIO); |
407 | } | |
408 | } | |
409 | } | |
410 | #endif | |
411 | if (XFS_FORCED_SHUTDOWN(mp)) | |
412 | goto shutdown_abort; | |
413 | ||
d7e84f41 CH |
414 | _xfs_trans_bjoin(tp, bp, 1); |
415 | trace_xfs_trans_read_buf(bp->b_fspriv); | |
1da177e4 | 416 | |
1da177e4 LT |
417 | *bpp = bp; |
418 | return 0; | |
419 | ||
420 | shutdown_abort: | |
421 | /* | |
422 | * the theory here is that buffer is good but we're | |
423 | * bailing out because the filesystem is being forcibly | |
424 | * shut down. So we should leave the b_flags alone since | |
425 | * the buffer's not staled and just get out. | |
426 | */ | |
427 | #if defined(DEBUG) | |
428 | if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp)) | |
0b932ccc | 429 | xfs_notice(mp, "about to pop assert, bp == 0x%p", bp); |
1da177e4 | 430 | #endif |
ed43233b | 431 | ASSERT((bp->b_flags & (XBF_STALE|XBF_DELWRI)) != |
0cadda1c | 432 | (XBF_STALE|XBF_DELWRI)); |
1da177e4 | 433 | |
0b1b213f | 434 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
435 | xfs_buf_relse(bp); |
436 | *bpp = NULL; | |
437 | return XFS_ERROR(EIO); | |
438 | } | |
439 | ||
440 | ||
441 | /* | |
442 | * Release the buffer bp which was previously acquired with one of the | |
443 | * xfs_trans_... buffer allocation routines if the buffer has not | |
444 | * been modified within this transaction. If the buffer is modified | |
445 | * within this transaction, do decrement the recursion count but do | |
446 | * not release the buffer even if the count goes to 0. If the buffer is not | |
447 | * modified within the transaction, decrement the recursion count and | |
448 | * release the buffer if the recursion count goes to 0. | |
449 | * | |
450 | * If the buffer is to be released and it was not modified before | |
451 | * this transaction began, then free the buf_log_item associated with it. | |
452 | * | |
453 | * If the transaction pointer is NULL, make this just a normal | |
454 | * brelse() call. | |
455 | */ | |
456 | void | |
457 | xfs_trans_brelse(xfs_trans_t *tp, | |
458 | xfs_buf_t *bp) | |
459 | { | |
460 | xfs_buf_log_item_t *bip; | |
1da177e4 LT |
461 | |
462 | /* | |
463 | * Default to a normal brelse() call if the tp is NULL. | |
464 | */ | |
465 | if (tp == NULL) { | |
adadbeef CH |
466 | struct xfs_log_item *lip = bp->b_fspriv; |
467 | ||
bf9d9013 | 468 | ASSERT(bp->b_transp == NULL); |
adadbeef | 469 | |
1da177e4 LT |
470 | /* |
471 | * If there's a buf log item attached to the buffer, | |
472 | * then let the AIL know that the buffer is being | |
473 | * unlocked. | |
474 | */ | |
adadbeef CH |
475 | if (lip != NULL && lip->li_type == XFS_LI_BUF) { |
476 | bip = bp->b_fspriv; | |
477 | xfs_trans_unlocked_item(bip->bli_item.li_ailp, lip); | |
1da177e4 LT |
478 | } |
479 | xfs_buf_relse(bp); | |
480 | return; | |
481 | } | |
482 | ||
bf9d9013 | 483 | ASSERT(bp->b_transp == tp); |
adadbeef | 484 | bip = bp->b_fspriv; |
1da177e4 LT |
485 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
486 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 487 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
488 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
489 | ||
0b1b213f CH |
490 | trace_xfs_trans_brelse(bip); |
491 | ||
1da177e4 LT |
492 | /* |
493 | * If the release is just for a recursive lock, | |
494 | * then decrement the count and return. | |
495 | */ | |
496 | if (bip->bli_recur > 0) { | |
497 | bip->bli_recur--; | |
1da177e4 LT |
498 | return; |
499 | } | |
500 | ||
501 | /* | |
502 | * If the buffer is dirty within this transaction, we can't | |
503 | * release it until we commit. | |
504 | */ | |
e98c414f | 505 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
1da177e4 | 506 | return; |
1da177e4 LT |
507 | |
508 | /* | |
509 | * If the buffer has been invalidated, then we can't release | |
510 | * it until the transaction commits to disk unless it is re-dirtied | |
511 | * as part of this transaction. This prevents us from pulling | |
512 | * the item from the AIL before we should. | |
513 | */ | |
0b1b213f | 514 | if (bip->bli_flags & XFS_BLI_STALE) |
1da177e4 | 515 | return; |
1da177e4 LT |
516 | |
517 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); | |
1da177e4 LT |
518 | |
519 | /* | |
520 | * Free up the log item descriptor tracking the released item. | |
521 | */ | |
e98c414f | 522 | xfs_trans_del_item(&bip->bli_item); |
1da177e4 LT |
523 | |
524 | /* | |
525 | * Clear the hold flag in the buf log item if it is set. | |
526 | * We wouldn't want the next user of the buffer to | |
527 | * get confused. | |
528 | */ | |
529 | if (bip->bli_flags & XFS_BLI_HOLD) { | |
530 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
531 | } | |
532 | ||
533 | /* | |
534 | * Drop our reference to the buf log item. | |
535 | */ | |
536 | atomic_dec(&bip->bli_refcount); | |
537 | ||
538 | /* | |
539 | * If the buf item is not tracking data in the log, then | |
540 | * we must free it before releasing the buffer back to the | |
541 | * free pool. Before releasing the buffer to the free pool, | |
542 | * clear the transaction pointer in b_fsprivate2 to dissolve | |
543 | * its relation to this transaction. | |
544 | */ | |
545 | if (!xfs_buf_item_dirty(bip)) { | |
546 | /*** | |
547 | ASSERT(bp->b_pincount == 0); | |
548 | ***/ | |
549 | ASSERT(atomic_read(&bip->bli_refcount) == 0); | |
550 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); | |
551 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); | |
552 | xfs_buf_item_relse(bp); | |
553 | bip = NULL; | |
554 | } | |
bf9d9013 | 555 | bp->b_transp = NULL; |
1da177e4 LT |
556 | |
557 | /* | |
558 | * If we've still got a buf log item on the buffer, then | |
559 | * tell the AIL that the buffer is being unlocked. | |
560 | */ | |
561 | if (bip != NULL) { | |
783a2f65 | 562 | xfs_trans_unlocked_item(bip->bli_item.li_ailp, |
1da177e4 LT |
563 | (xfs_log_item_t*)bip); |
564 | } | |
565 | ||
566 | xfs_buf_relse(bp); | |
567 | return; | |
568 | } | |
569 | ||
1da177e4 LT |
570 | /* |
571 | * Mark the buffer as not needing to be unlocked when the buf item's | |
572 | * IOP_UNLOCK() routine is called. The buffer must already be locked | |
573 | * and associated with the given transaction. | |
574 | */ | |
575 | /* ARGSUSED */ | |
576 | void | |
577 | xfs_trans_bhold(xfs_trans_t *tp, | |
578 | xfs_buf_t *bp) | |
579 | { | |
adadbeef | 580 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 581 | |
bf9d9013 | 582 | ASSERT(bp->b_transp == tp); |
adadbeef | 583 | ASSERT(bip != NULL); |
1da177e4 | 584 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 585 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 | 586 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
adadbeef | 587 | |
1da177e4 | 588 | bip->bli_flags |= XFS_BLI_HOLD; |
0b1b213f | 589 | trace_xfs_trans_bhold(bip); |
1da177e4 LT |
590 | } |
591 | ||
efa092f3 TS |
592 | /* |
593 | * Cancel the previous buffer hold request made on this buffer | |
594 | * for this transaction. | |
595 | */ | |
596 | void | |
597 | xfs_trans_bhold_release(xfs_trans_t *tp, | |
598 | xfs_buf_t *bp) | |
599 | { | |
adadbeef | 600 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
efa092f3 | 601 | |
bf9d9013 | 602 | ASSERT(bp->b_transp == tp); |
adadbeef | 603 | ASSERT(bip != NULL); |
efa092f3 | 604 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 605 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
efa092f3 TS |
606 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
607 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); | |
0b1b213f | 608 | |
adadbeef | 609 | bip->bli_flags &= ~XFS_BLI_HOLD; |
0b1b213f | 610 | trace_xfs_trans_bhold_release(bip); |
efa092f3 TS |
611 | } |
612 | ||
1da177e4 LT |
613 | /* |
614 | * This is called to mark bytes first through last inclusive of the given | |
615 | * buffer as needing to be logged when the transaction is committed. | |
616 | * The buffer must already be associated with the given transaction. | |
617 | * | |
618 | * First and last are numbers relative to the beginning of this buffer, | |
619 | * so the first byte in the buffer is numbered 0 regardless of the | |
620 | * value of b_blkno. | |
621 | */ | |
622 | void | |
623 | xfs_trans_log_buf(xfs_trans_t *tp, | |
624 | xfs_buf_t *bp, | |
625 | uint first, | |
626 | uint last) | |
627 | { | |
adadbeef | 628 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 629 | |
bf9d9013 | 630 | ASSERT(bp->b_transp == tp); |
adadbeef | 631 | ASSERT(bip != NULL); |
1da177e4 | 632 | ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp))); |
cb669ca5 CH |
633 | ASSERT(bp->b_iodone == NULL || |
634 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1da177e4 LT |
635 | |
636 | /* | |
637 | * Mark the buffer as needing to be written out eventually, | |
638 | * and set its iodone function to remove the buffer's buf log | |
639 | * item from the AIL and free it when the buffer is flushed | |
640 | * to disk. See xfs_buf_attach_iodone() for more details | |
641 | * on li_cb and xfs_buf_iodone_callbacks(). | |
642 | * If we end up aborting this transaction, we trap this buffer | |
643 | * inside the b_bdstrat callback so that this won't get written to | |
644 | * disk. | |
645 | */ | |
1da177e4 LT |
646 | XFS_BUF_DONE(bp); |
647 | ||
1da177e4 | 648 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
cb669ca5 | 649 | bp->b_iodone = xfs_buf_iodone_callbacks; |
ca30b2a7 | 650 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 | 651 | |
61551f1e CH |
652 | xfs_buf_delwri_queue(bp); |
653 | ||
0b1b213f CH |
654 | trace_xfs_trans_log_buf(bip); |
655 | ||
1da177e4 LT |
656 | /* |
657 | * If we invalidated the buffer within this transaction, then | |
658 | * cancel the invalidation now that we're dirtying the buffer | |
659 | * again. There are no races with the code in xfs_buf_item_unpin(), | |
660 | * because we have a reference to the buffer this entire time. | |
661 | */ | |
662 | if (bip->bli_flags & XFS_BLI_STALE) { | |
1da177e4 LT |
663 | bip->bli_flags &= ~XFS_BLI_STALE; |
664 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
665 | XFS_BUF_UNSTALE(bp); | |
c1155410 | 666 | bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
1da177e4 LT |
667 | } |
668 | ||
1da177e4 | 669 | tp->t_flags |= XFS_TRANS_DIRTY; |
e98c414f | 670 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 LT |
671 | bip->bli_flags |= XFS_BLI_LOGGED; |
672 | xfs_buf_item_log(bip, first, last); | |
1da177e4 LT |
673 | } |
674 | ||
675 | ||
676 | /* | |
677 | * This called to invalidate a buffer that is being used within | |
678 | * a transaction. Typically this is because the blocks in the | |
679 | * buffer are being freed, so we need to prevent it from being | |
680 | * written out when we're done. Allowing it to be written again | |
681 | * might overwrite data in the free blocks if they are reallocated | |
682 | * to a file. | |
683 | * | |
684 | * We prevent the buffer from being written out by clearing the | |
685 | * B_DELWRI flag. We can't always | |
686 | * get rid of the buf log item at this point, though, because | |
687 | * the buffer may still be pinned by another transaction. If that | |
688 | * is the case, then we'll wait until the buffer is committed to | |
689 | * disk for the last time (we can tell by the ref count) and | |
690 | * free it in xfs_buf_item_unpin(). Until it is cleaned up we | |
691 | * will keep the buffer locked so that the buffer and buf log item | |
692 | * are not reused. | |
693 | */ | |
694 | void | |
695 | xfs_trans_binval( | |
696 | xfs_trans_t *tp, | |
697 | xfs_buf_t *bp) | |
698 | { | |
adadbeef | 699 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 700 | |
bf9d9013 | 701 | ASSERT(bp->b_transp == tp); |
adadbeef | 702 | ASSERT(bip != NULL); |
1da177e4 LT |
703 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
704 | ||
0b1b213f CH |
705 | trace_xfs_trans_binval(bip); |
706 | ||
1da177e4 LT |
707 | if (bip->bli_flags & XFS_BLI_STALE) { |
708 | /* | |
709 | * If the buffer is already invalidated, then | |
710 | * just return. | |
711 | */ | |
712 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); | |
713 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
714 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); | |
c1155410 DC |
715 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
716 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); | |
e98c414f | 717 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
1da177e4 | 718 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
1da177e4 LT |
719 | return; |
720 | } | |
721 | ||
722 | /* | |
723 | * Clear the dirty bit in the buffer and set the STALE flag | |
724 | * in the buf log item. The STALE flag will be used in | |
725 | * xfs_buf_item_unpin() to determine if it should clean up | |
726 | * when the last reference to the buf item is given up. | |
c1155410 | 727 | * We set the XFS_BLF_CANCEL flag in the buf log format structure |
1da177e4 LT |
728 | * and log the buf item. This will be used at recovery time |
729 | * to determine that copies of the buffer in the log before | |
730 | * this should not be replayed. | |
731 | * We mark the item descriptor and the transaction dirty so | |
732 | * that we'll hold the buffer until after the commit. | |
733 | * | |
734 | * Since we're invalidating the buffer, we also clear the state | |
735 | * about which parts of the buffer have been logged. We also | |
736 | * clear the flag indicating that this is an inode buffer since | |
737 | * the data in the buffer will no longer be valid. | |
738 | * | |
739 | * We set the stale bit in the buffer as well since we're getting | |
740 | * rid of it. | |
741 | */ | |
c867cb61 | 742 | xfs_buf_stale(bp); |
1da177e4 | 743 | bip->bli_flags |= XFS_BLI_STALE; |
ccf7c23f | 744 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
c1155410 DC |
745 | bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
746 | bip->bli_format.blf_flags |= XFS_BLF_CANCEL; | |
1da177e4 LT |
747 | memset((char *)(bip->bli_format.blf_data_map), 0, |
748 | (bip->bli_format.blf_map_size * sizeof(uint))); | |
e98c414f | 749 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 | 750 | tp->t_flags |= XFS_TRANS_DIRTY; |
1da177e4 LT |
751 | } |
752 | ||
753 | /* | |
ccf7c23f DC |
754 | * This call is used to indicate that the buffer contains on-disk inodes which |
755 | * must be handled specially during recovery. They require special handling | |
756 | * because only the di_next_unlinked from the inodes in the buffer should be | |
757 | * recovered. The rest of the data in the buffer is logged via the inodes | |
758 | * themselves. | |
1da177e4 | 759 | * |
ccf7c23f DC |
760 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
761 | * transferred to the buffer's log format structure so that we'll know what to | |
762 | * do at recovery time. | |
1da177e4 | 763 | */ |
1da177e4 LT |
764 | void |
765 | xfs_trans_inode_buf( | |
766 | xfs_trans_t *tp, | |
767 | xfs_buf_t *bp) | |
768 | { | |
adadbeef | 769 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 770 | |
bf9d9013 | 771 | ASSERT(bp->b_transp == tp); |
adadbeef | 772 | ASSERT(bip != NULL); |
1da177e4 LT |
773 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
774 | ||
ccf7c23f | 775 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
1da177e4 LT |
776 | } |
777 | ||
778 | /* | |
779 | * This call is used to indicate that the buffer is going to | |
780 | * be staled and was an inode buffer. This means it gets | |
781 | * special processing during unpin - where any inodes | |
782 | * associated with the buffer should be removed from ail. | |
783 | * There is also special processing during recovery, | |
784 | * any replay of the inodes in the buffer needs to be | |
785 | * prevented as the buffer may have been reused. | |
786 | */ | |
787 | void | |
788 | xfs_trans_stale_inode_buf( | |
789 | xfs_trans_t *tp, | |
790 | xfs_buf_t *bp) | |
791 | { | |
adadbeef | 792 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 793 | |
bf9d9013 | 794 | ASSERT(bp->b_transp == tp); |
adadbeef | 795 | ASSERT(bip != NULL); |
1da177e4 LT |
796 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
797 | ||
798 | bip->bli_flags |= XFS_BLI_STALE_INODE; | |
ca30b2a7 | 799 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 LT |
800 | } |
801 | ||
1da177e4 LT |
802 | /* |
803 | * Mark the buffer as being one which contains newly allocated | |
804 | * inodes. We need to make sure that even if this buffer is | |
805 | * relogged as an 'inode buf' we still recover all of the inode | |
806 | * images in the face of a crash. This works in coordination with | |
807 | * xfs_buf_item_committed() to ensure that the buffer remains in the | |
808 | * AIL at its original location even after it has been relogged. | |
809 | */ | |
810 | /* ARGSUSED */ | |
811 | void | |
812 | xfs_trans_inode_alloc_buf( | |
813 | xfs_trans_t *tp, | |
814 | xfs_buf_t *bp) | |
815 | { | |
adadbeef | 816 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 817 | |
bf9d9013 | 818 | ASSERT(bp->b_transp == tp); |
adadbeef | 819 | ASSERT(bip != NULL); |
1da177e4 LT |
820 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
821 | ||
822 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; | |
823 | } | |
824 | ||
825 | ||
826 | /* | |
827 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of | |
828 | * dquots. However, unlike in inode buffer recovery, dquot buffers get | |
829 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). | |
830 | * The only thing that makes dquot buffers different from regular | |
831 | * buffers is that we must not replay dquot bufs when recovering | |
832 | * if a _corresponding_ quotaoff has happened. We also have to distinguish | |
833 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas | |
834 | * can be turned off independently. | |
835 | */ | |
836 | /* ARGSUSED */ | |
837 | void | |
838 | xfs_trans_dquot_buf( | |
839 | xfs_trans_t *tp, | |
840 | xfs_buf_t *bp, | |
841 | uint type) | |
842 | { | |
adadbeef | 843 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 844 | |
bf9d9013 | 845 | ASSERT(bp->b_transp == tp); |
adadbeef | 846 | ASSERT(bip != NULL); |
c1155410 DC |
847 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
848 | type == XFS_BLF_PDQUOT_BUF || | |
849 | type == XFS_BLF_GDQUOT_BUF); | |
1da177e4 LT |
850 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
851 | ||
852 | bip->bli_format.blf_flags |= type; | |
853 | } |