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" |
1da177e4 LT |
21 | #include "xfs_log.h" |
22 | #include "xfs_trans.h" | |
1da177e4 LT |
23 | #include "xfs_sb.h" |
24 | #include "xfs_ag.h" | |
1da177e4 | 25 | #include "xfs_mount.h" |
a844f451 NS |
26 | #include "xfs_bmap_btree.h" |
27 | #include "xfs_alloc_btree.h" | |
28 | #include "xfs_ialloc_btree.h" | |
a844f451 NS |
29 | #include "xfs_dinode.h" |
30 | #include "xfs_inode.h" | |
31 | #include "xfs_buf_item.h" | |
1da177e4 LT |
32 | #include "xfs_trans_priv.h" |
33 | #include "xfs_error.h" | |
0b1b213f | 34 | #include "xfs_trace.h" |
1da177e4 | 35 | |
4a5224d7 CH |
36 | /* |
37 | * Check to see if a buffer matching the given parameters is already | |
38 | * a part of the given transaction. | |
39 | */ | |
40 | STATIC struct xfs_buf * | |
41 | xfs_trans_buf_item_match( | |
42 | struct xfs_trans *tp, | |
43 | struct xfs_buftarg *target, | |
de2a4f59 DC |
44 | struct xfs_buf_map *map, |
45 | int nmaps) | |
4a5224d7 | 46 | { |
e98c414f CH |
47 | struct xfs_log_item_desc *lidp; |
48 | struct xfs_buf_log_item *blip; | |
de2a4f59 DC |
49 | int len = 0; |
50 | int i; | |
51 | ||
52 | for (i = 0; i < nmaps; i++) | |
53 | len += map[i].bm_len; | |
1da177e4 | 54 | |
e98c414f CH |
55 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
56 | blip = (struct xfs_buf_log_item *)lidp->lid_item; | |
57 | if (blip->bli_item.li_type == XFS_LI_BUF && | |
49074c06 | 58 | blip->bli_buf->b_target == target && |
de2a4f59 DC |
59 | XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn && |
60 | blip->bli_buf->b_length == len) { | |
61 | ASSERT(blip->bli_buf->b_map_count == nmaps); | |
e98c414f | 62 | return blip->bli_buf; |
de2a4f59 | 63 | } |
4a5224d7 CH |
64 | } |
65 | ||
66 | return NULL; | |
67 | } | |
1da177e4 | 68 | |
d7e84f41 CH |
69 | /* |
70 | * Add the locked buffer to the transaction. | |
71 | * | |
72 | * The buffer must be locked, and it cannot be associated with any | |
73 | * transaction. | |
74 | * | |
75 | * If the buffer does not yet have a buf log item associated with it, | |
76 | * then allocate one for it. Then add the buf item to the transaction. | |
77 | */ | |
78 | STATIC void | |
79 | _xfs_trans_bjoin( | |
80 | struct xfs_trans *tp, | |
81 | struct xfs_buf *bp, | |
82 | int reset_recur) | |
83 | { | |
84 | struct xfs_buf_log_item *bip; | |
85 | ||
bf9d9013 | 86 | ASSERT(bp->b_transp == NULL); |
d7e84f41 CH |
87 | |
88 | /* | |
89 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If | |
90 | * it doesn't have one yet, then allocate one and initialize it. | |
91 | * The checks to see if one is there are in xfs_buf_item_init(). | |
92 | */ | |
93 | xfs_buf_item_init(bp, tp->t_mountp); | |
adadbeef | 94 | bip = bp->b_fspriv; |
d7e84f41 | 95 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 96 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
d7e84f41 CH |
97 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
98 | if (reset_recur) | |
99 | bip->bli_recur = 0; | |
100 | ||
101 | /* | |
102 | * Take a reference for this transaction on the buf item. | |
103 | */ | |
104 | atomic_inc(&bip->bli_refcount); | |
105 | ||
106 | /* | |
107 | * Get a log_item_desc to point at the new item. | |
108 | */ | |
e98c414f | 109 | xfs_trans_add_item(tp, &bip->bli_item); |
d7e84f41 CH |
110 | |
111 | /* | |
112 | * Initialize b_fsprivate2 so we can find it with incore_match() | |
113 | * in xfs_trans_get_buf() and friends above. | |
114 | */ | |
bf9d9013 | 115 | bp->b_transp = tp; |
d7e84f41 CH |
116 | |
117 | } | |
118 | ||
119 | void | |
120 | xfs_trans_bjoin( | |
121 | struct xfs_trans *tp, | |
122 | struct xfs_buf *bp) | |
123 | { | |
124 | _xfs_trans_bjoin(tp, bp, 0); | |
125 | trace_xfs_trans_bjoin(bp->b_fspriv); | |
126 | } | |
1da177e4 LT |
127 | |
128 | /* | |
129 | * Get and lock the buffer for the caller if it is not already | |
130 | * locked within the given transaction. If it is already locked | |
131 | * within the transaction, just increment its lock recursion count | |
132 | * and return a pointer to it. | |
133 | * | |
1da177e4 LT |
134 | * If the transaction pointer is NULL, make this just a normal |
135 | * get_buf() call. | |
136 | */ | |
de2a4f59 DC |
137 | struct xfs_buf * |
138 | xfs_trans_get_buf_map( | |
139 | struct xfs_trans *tp, | |
140 | struct xfs_buftarg *target, | |
141 | struct xfs_buf_map *map, | |
142 | int nmaps, | |
143 | xfs_buf_flags_t flags) | |
1da177e4 LT |
144 | { |
145 | xfs_buf_t *bp; | |
146 | xfs_buf_log_item_t *bip; | |
147 | ||
de2a4f59 DC |
148 | if (!tp) |
149 | return xfs_buf_get_map(target, map, nmaps, flags); | |
1da177e4 LT |
150 | |
151 | /* | |
152 | * If we find the buffer in the cache with this transaction | |
153 | * pointer in its b_fsprivate2 field, then we know we already | |
154 | * have it locked. In this case we just increment the lock | |
155 | * recursion count and return the buffer to the caller. | |
156 | */ | |
de2a4f59 | 157 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
1da177e4 | 158 | if (bp != NULL) { |
0c842ad4 | 159 | ASSERT(xfs_buf_islocked(bp)); |
c867cb61 CH |
160 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) { |
161 | xfs_buf_stale(bp); | |
c867cb61 CH |
162 | XFS_BUF_DONE(bp); |
163 | } | |
0b1b213f | 164 | |
bf9d9013 | 165 | ASSERT(bp->b_transp == tp); |
adadbeef | 166 | bip = bp->b_fspriv; |
1da177e4 LT |
167 | ASSERT(bip != NULL); |
168 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
169 | bip->bli_recur++; | |
0b1b213f | 170 | trace_xfs_trans_get_buf_recur(bip); |
1da177e4 LT |
171 | return (bp); |
172 | } | |
173 | ||
de2a4f59 | 174 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
1da177e4 LT |
175 | if (bp == NULL) { |
176 | return NULL; | |
177 | } | |
178 | ||
5a52c2a5 | 179 | ASSERT(!bp->b_error); |
1da177e4 | 180 | |
d7e84f41 CH |
181 | _xfs_trans_bjoin(tp, bp, 1); |
182 | trace_xfs_trans_get_buf(bp->b_fspriv); | |
1da177e4 LT |
183 | return (bp); |
184 | } | |
185 | ||
186 | /* | |
187 | * Get and lock the superblock buffer of this file system for the | |
188 | * given transaction. | |
189 | * | |
190 | * We don't need to use incore_match() here, because the superblock | |
191 | * buffer is a private buffer which we keep a pointer to in the | |
192 | * mount structure. | |
193 | */ | |
194 | xfs_buf_t * | |
195 | xfs_trans_getsb(xfs_trans_t *tp, | |
196 | struct xfs_mount *mp, | |
197 | int flags) | |
198 | { | |
199 | xfs_buf_t *bp; | |
200 | xfs_buf_log_item_t *bip; | |
201 | ||
202 | /* | |
203 | * Default to just trying to lock the superblock buffer | |
204 | * if tp is NULL. | |
205 | */ | |
206 | if (tp == NULL) { | |
207 | return (xfs_getsb(mp, flags)); | |
208 | } | |
209 | ||
210 | /* | |
211 | * If the superblock buffer already has this transaction | |
212 | * pointer in its b_fsprivate2 field, then we know we already | |
213 | * have it locked. In this case we just increment the lock | |
214 | * recursion count and return the buffer to the caller. | |
215 | */ | |
216 | bp = mp->m_sb_bp; | |
bf9d9013 | 217 | if (bp->b_transp == tp) { |
adadbeef | 218 | bip = bp->b_fspriv; |
1da177e4 LT |
219 | ASSERT(bip != NULL); |
220 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
221 | bip->bli_recur++; | |
0b1b213f | 222 | trace_xfs_trans_getsb_recur(bip); |
1da177e4 LT |
223 | return (bp); |
224 | } | |
225 | ||
226 | bp = xfs_getsb(mp, flags); | |
d7e84f41 | 227 | if (bp == NULL) |
1da177e4 | 228 | return NULL; |
1da177e4 | 229 | |
d7e84f41 CH |
230 | _xfs_trans_bjoin(tp, bp, 1); |
231 | trace_xfs_trans_getsb(bp->b_fspriv); | |
1da177e4 LT |
232 | return (bp); |
233 | } | |
234 | ||
235 | #ifdef DEBUG | |
236 | xfs_buftarg_t *xfs_error_target; | |
237 | int xfs_do_error; | |
238 | int xfs_req_num; | |
239 | int xfs_error_mod = 33; | |
240 | #endif | |
241 | ||
242 | /* | |
243 | * Get and lock the buffer for the caller if it is not already | |
244 | * locked within the given transaction. If it has not yet been | |
245 | * read in, read it from disk. If it is already locked | |
246 | * within the transaction and already read in, just increment its | |
247 | * lock recursion count and return a pointer to it. | |
248 | * | |
1da177e4 LT |
249 | * If the transaction pointer is NULL, make this just a normal |
250 | * read_buf() call. | |
251 | */ | |
252 | int | |
de2a4f59 DC |
253 | xfs_trans_read_buf_map( |
254 | struct xfs_mount *mp, | |
255 | struct xfs_trans *tp, | |
256 | struct xfs_buftarg *target, | |
257 | struct xfs_buf_map *map, | |
258 | int nmaps, | |
259 | xfs_buf_flags_t flags, | |
260 | struct xfs_buf **bpp) | |
1da177e4 LT |
261 | { |
262 | xfs_buf_t *bp; | |
263 | xfs_buf_log_item_t *bip; | |
264 | int error; | |
265 | ||
7ca790a5 | 266 | *bpp = NULL; |
de2a4f59 DC |
267 | if (!tp) { |
268 | bp = xfs_buf_read_map(target, map, nmaps, flags); | |
1da177e4 | 269 | if (!bp) |
0cadda1c | 270 | return (flags & XBF_TRYLOCK) ? |
a3f74ffb | 271 | EAGAIN : XFS_ERROR(ENOMEM); |
1da177e4 | 272 | |
5a52c2a5 CS |
273 | if (bp->b_error) { |
274 | error = bp->b_error; | |
901796af | 275 | xfs_buf_ioerror_alert(bp, __func__); |
7ca790a5 DC |
276 | XFS_BUF_UNDONE(bp); |
277 | xfs_buf_stale(bp); | |
1da177e4 LT |
278 | xfs_buf_relse(bp); |
279 | return error; | |
280 | } | |
281 | #ifdef DEBUG | |
a0f7bfd3 | 282 | if (xfs_do_error) { |
1da177e4 LT |
283 | if (xfs_error_target == target) { |
284 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
285 | xfs_buf_relse(bp); | |
0b932ccc | 286 | xfs_debug(mp, "Returning error!"); |
1da177e4 LT |
287 | return XFS_ERROR(EIO); |
288 | } | |
289 | } | |
290 | } | |
291 | #endif | |
292 | if (XFS_FORCED_SHUTDOWN(mp)) | |
293 | goto shutdown_abort; | |
294 | *bpp = bp; | |
295 | return 0; | |
296 | } | |
297 | ||
298 | /* | |
299 | * If we find the buffer in the cache with this transaction | |
300 | * pointer in its b_fsprivate2 field, then we know we already | |
301 | * have it locked. If it is already read in we just increment | |
302 | * the lock recursion count and return the buffer to the caller. | |
303 | * If the buffer is not yet read in, then we read it in, increment | |
304 | * the lock recursion count, and return it to the caller. | |
305 | */ | |
de2a4f59 | 306 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
1da177e4 | 307 | if (bp != NULL) { |
0c842ad4 | 308 | ASSERT(xfs_buf_islocked(bp)); |
bf9d9013 | 309 | ASSERT(bp->b_transp == tp); |
adadbeef | 310 | ASSERT(bp->b_fspriv != NULL); |
5a52c2a5 | 311 | ASSERT(!bp->b_error); |
1da177e4 | 312 | if (!(XFS_BUF_ISDONE(bp))) { |
0b1b213f | 313 | trace_xfs_trans_read_buf_io(bp, _RET_IP_); |
1da177e4 LT |
314 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
315 | XFS_BUF_READ(bp); | |
316 | xfsbdstrat(tp->t_mountp, bp); | |
1a1a3e97 | 317 | error = xfs_buf_iowait(bp); |
d64e31a2 | 318 | if (error) { |
901796af | 319 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
320 | xfs_buf_relse(bp); |
321 | /* | |
d64e31a2 DC |
322 | * We can gracefully recover from most read |
323 | * errors. Ones we can't are those that happen | |
324 | * after the transaction's already dirty. | |
1da177e4 LT |
325 | */ |
326 | if (tp->t_flags & XFS_TRANS_DIRTY) | |
327 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 328 | SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
329 | return error; |
330 | } | |
331 | } | |
332 | /* | |
333 | * We never locked this buf ourselves, so we shouldn't | |
334 | * brelse it either. Just get out. | |
335 | */ | |
336 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
0b1b213f | 337 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
338 | *bpp = NULL; |
339 | return XFS_ERROR(EIO); | |
340 | } | |
341 | ||
342 | ||
adadbeef | 343 | bip = bp->b_fspriv; |
1da177e4 LT |
344 | bip->bli_recur++; |
345 | ||
346 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 347 | trace_xfs_trans_read_buf_recur(bip); |
1da177e4 LT |
348 | *bpp = bp; |
349 | return 0; | |
350 | } | |
351 | ||
de2a4f59 | 352 | bp = xfs_buf_read_map(target, map, nmaps, flags); |
1da177e4 LT |
353 | if (bp == NULL) { |
354 | *bpp = NULL; | |
7401aafd DC |
355 | return (flags & XBF_TRYLOCK) ? |
356 | 0 : XFS_ERROR(ENOMEM); | |
1da177e4 | 357 | } |
5a52c2a5 CS |
358 | if (bp->b_error) { |
359 | error = bp->b_error; | |
c867cb61 | 360 | xfs_buf_stale(bp); |
c867cb61 | 361 | XFS_BUF_DONE(bp); |
901796af | 362 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 | 363 | if (tp->t_flags & XFS_TRANS_DIRTY) |
7d04a335 | 364 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
365 | xfs_buf_relse(bp); |
366 | return error; | |
367 | } | |
368 | #ifdef DEBUG | |
369 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { | |
370 | if (xfs_error_target == target) { | |
371 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
372 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 373 | SHUTDOWN_META_IO_ERROR); |
1da177e4 | 374 | xfs_buf_relse(bp); |
0b932ccc | 375 | xfs_debug(mp, "Returning trans error!"); |
1da177e4 LT |
376 | return XFS_ERROR(EIO); |
377 | } | |
378 | } | |
379 | } | |
380 | #endif | |
381 | if (XFS_FORCED_SHUTDOWN(mp)) | |
382 | goto shutdown_abort; | |
383 | ||
d7e84f41 CH |
384 | _xfs_trans_bjoin(tp, bp, 1); |
385 | trace_xfs_trans_read_buf(bp->b_fspriv); | |
1da177e4 | 386 | |
1da177e4 LT |
387 | *bpp = bp; |
388 | return 0; | |
389 | ||
390 | shutdown_abort: | |
0b1b213f | 391 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
392 | xfs_buf_relse(bp); |
393 | *bpp = NULL; | |
394 | return XFS_ERROR(EIO); | |
395 | } | |
396 | ||
397 | ||
398 | /* | |
399 | * Release the buffer bp which was previously acquired with one of the | |
400 | * xfs_trans_... buffer allocation routines if the buffer has not | |
401 | * been modified within this transaction. If the buffer is modified | |
402 | * within this transaction, do decrement the recursion count but do | |
403 | * not release the buffer even if the count goes to 0. If the buffer is not | |
404 | * modified within the transaction, decrement the recursion count and | |
405 | * release the buffer if the recursion count goes to 0. | |
406 | * | |
407 | * If the buffer is to be released and it was not modified before | |
408 | * this transaction began, then free the buf_log_item associated with it. | |
409 | * | |
410 | * If the transaction pointer is NULL, make this just a normal | |
411 | * brelse() call. | |
412 | */ | |
413 | void | |
414 | xfs_trans_brelse(xfs_trans_t *tp, | |
415 | xfs_buf_t *bp) | |
416 | { | |
417 | xfs_buf_log_item_t *bip; | |
1da177e4 LT |
418 | |
419 | /* | |
420 | * Default to a normal brelse() call if the tp is NULL. | |
421 | */ | |
422 | if (tp == NULL) { | |
bf9d9013 | 423 | ASSERT(bp->b_transp == NULL); |
1da177e4 LT |
424 | xfs_buf_relse(bp); |
425 | return; | |
426 | } | |
427 | ||
bf9d9013 | 428 | ASSERT(bp->b_transp == tp); |
adadbeef | 429 | bip = bp->b_fspriv; |
1da177e4 LT |
430 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
431 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 432 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
433 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
434 | ||
0b1b213f CH |
435 | trace_xfs_trans_brelse(bip); |
436 | ||
1da177e4 LT |
437 | /* |
438 | * If the release is just for a recursive lock, | |
439 | * then decrement the count and return. | |
440 | */ | |
441 | if (bip->bli_recur > 0) { | |
442 | bip->bli_recur--; | |
1da177e4 LT |
443 | return; |
444 | } | |
445 | ||
446 | /* | |
447 | * If the buffer is dirty within this transaction, we can't | |
448 | * release it until we commit. | |
449 | */ | |
e98c414f | 450 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
1da177e4 | 451 | return; |
1da177e4 LT |
452 | |
453 | /* | |
454 | * If the buffer has been invalidated, then we can't release | |
455 | * it until the transaction commits to disk unless it is re-dirtied | |
456 | * as part of this transaction. This prevents us from pulling | |
457 | * the item from the AIL before we should. | |
458 | */ | |
0b1b213f | 459 | if (bip->bli_flags & XFS_BLI_STALE) |
1da177e4 | 460 | return; |
1da177e4 LT |
461 | |
462 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); | |
1da177e4 LT |
463 | |
464 | /* | |
465 | * Free up the log item descriptor tracking the released item. | |
466 | */ | |
e98c414f | 467 | xfs_trans_del_item(&bip->bli_item); |
1da177e4 LT |
468 | |
469 | /* | |
470 | * Clear the hold flag in the buf log item if it is set. | |
471 | * We wouldn't want the next user of the buffer to | |
472 | * get confused. | |
473 | */ | |
474 | if (bip->bli_flags & XFS_BLI_HOLD) { | |
475 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
476 | } | |
477 | ||
478 | /* | |
479 | * Drop our reference to the buf log item. | |
480 | */ | |
481 | atomic_dec(&bip->bli_refcount); | |
482 | ||
483 | /* | |
484 | * If the buf item is not tracking data in the log, then | |
485 | * we must free it before releasing the buffer back to the | |
486 | * free pool. Before releasing the buffer to the free pool, | |
487 | * clear the transaction pointer in b_fsprivate2 to dissolve | |
488 | * its relation to this transaction. | |
489 | */ | |
490 | if (!xfs_buf_item_dirty(bip)) { | |
491 | /*** | |
492 | ASSERT(bp->b_pincount == 0); | |
493 | ***/ | |
494 | ASSERT(atomic_read(&bip->bli_refcount) == 0); | |
495 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); | |
496 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); | |
497 | xfs_buf_item_relse(bp); | |
1da177e4 LT |
498 | } |
499 | ||
5b03ff1b | 500 | bp->b_transp = NULL; |
1da177e4 | 501 | xfs_buf_relse(bp); |
1da177e4 LT |
502 | } |
503 | ||
1da177e4 LT |
504 | /* |
505 | * Mark the buffer as not needing to be unlocked when the buf item's | |
506 | * IOP_UNLOCK() routine is called. The buffer must already be locked | |
507 | * and associated with the given transaction. | |
508 | */ | |
509 | /* ARGSUSED */ | |
510 | void | |
511 | xfs_trans_bhold(xfs_trans_t *tp, | |
512 | xfs_buf_t *bp) | |
513 | { | |
adadbeef | 514 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 515 | |
bf9d9013 | 516 | ASSERT(bp->b_transp == tp); |
adadbeef | 517 | ASSERT(bip != NULL); |
1da177e4 | 518 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 519 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 | 520 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
adadbeef | 521 | |
1da177e4 | 522 | bip->bli_flags |= XFS_BLI_HOLD; |
0b1b213f | 523 | trace_xfs_trans_bhold(bip); |
1da177e4 LT |
524 | } |
525 | ||
efa092f3 TS |
526 | /* |
527 | * Cancel the previous buffer hold request made on this buffer | |
528 | * for this transaction. | |
529 | */ | |
530 | void | |
531 | xfs_trans_bhold_release(xfs_trans_t *tp, | |
532 | xfs_buf_t *bp) | |
533 | { | |
adadbeef | 534 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
efa092f3 | 535 | |
bf9d9013 | 536 | ASSERT(bp->b_transp == tp); |
adadbeef | 537 | ASSERT(bip != NULL); |
efa092f3 | 538 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 539 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
efa092f3 TS |
540 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
541 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); | |
0b1b213f | 542 | |
adadbeef | 543 | bip->bli_flags &= ~XFS_BLI_HOLD; |
0b1b213f | 544 | trace_xfs_trans_bhold_release(bip); |
efa092f3 TS |
545 | } |
546 | ||
1da177e4 LT |
547 | /* |
548 | * This is called to mark bytes first through last inclusive of the given | |
549 | * buffer as needing to be logged when the transaction is committed. | |
550 | * The buffer must already be associated with the given transaction. | |
551 | * | |
552 | * First and last are numbers relative to the beginning of this buffer, | |
553 | * so the first byte in the buffer is numbered 0 regardless of the | |
554 | * value of b_blkno. | |
555 | */ | |
556 | void | |
557 | xfs_trans_log_buf(xfs_trans_t *tp, | |
558 | xfs_buf_t *bp, | |
559 | uint first, | |
560 | uint last) | |
561 | { | |
adadbeef | 562 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 563 | |
bf9d9013 | 564 | ASSERT(bp->b_transp == tp); |
adadbeef | 565 | ASSERT(bip != NULL); |
aa0e8833 | 566 | ASSERT(first <= last && last < BBTOB(bp->b_length)); |
cb669ca5 CH |
567 | ASSERT(bp->b_iodone == NULL || |
568 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1da177e4 LT |
569 | |
570 | /* | |
571 | * Mark the buffer as needing to be written out eventually, | |
572 | * and set its iodone function to remove the buffer's buf log | |
573 | * item from the AIL and free it when the buffer is flushed | |
574 | * to disk. See xfs_buf_attach_iodone() for more details | |
575 | * on li_cb and xfs_buf_iodone_callbacks(). | |
576 | * If we end up aborting this transaction, we trap this buffer | |
577 | * inside the b_bdstrat callback so that this won't get written to | |
578 | * disk. | |
579 | */ | |
1da177e4 LT |
580 | XFS_BUF_DONE(bp); |
581 | ||
1da177e4 | 582 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
cb669ca5 | 583 | bp->b_iodone = xfs_buf_iodone_callbacks; |
ca30b2a7 | 584 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 | 585 | |
0b1b213f CH |
586 | trace_xfs_trans_log_buf(bip); |
587 | ||
1da177e4 LT |
588 | /* |
589 | * If we invalidated the buffer within this transaction, then | |
590 | * cancel the invalidation now that we're dirtying the buffer | |
591 | * again. There are no races with the code in xfs_buf_item_unpin(), | |
592 | * because we have a reference to the buffer this entire time. | |
593 | */ | |
594 | if (bip->bli_flags & XFS_BLI_STALE) { | |
1da177e4 LT |
595 | bip->bli_flags &= ~XFS_BLI_STALE; |
596 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
597 | XFS_BUF_UNSTALE(bp); | |
c1155410 | 598 | bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
1da177e4 LT |
599 | } |
600 | ||
1da177e4 | 601 | tp->t_flags |= XFS_TRANS_DIRTY; |
e98c414f | 602 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 LT |
603 | bip->bli_flags |= XFS_BLI_LOGGED; |
604 | xfs_buf_item_log(bip, first, last); | |
1da177e4 LT |
605 | } |
606 | ||
607 | ||
608 | /* | |
43ff2122 CH |
609 | * Invalidate a buffer that is being used within a transaction. |
610 | * | |
611 | * Typically this is because the blocks in the buffer are being freed, so we | |
612 | * need to prevent it from being written out when we're done. Allowing it | |
613 | * to be written again might overwrite data in the free blocks if they are | |
614 | * reallocated to a file. | |
1da177e4 | 615 | * |
43ff2122 CH |
616 | * We prevent the buffer from being written out by marking it stale. We can't |
617 | * get rid of the buf log item at this point because the buffer may still be | |
618 | * pinned by another transaction. If that is the case, then we'll wait until | |
619 | * the buffer is committed to disk for the last time (we can tell by the ref | |
620 | * count) and free it in xfs_buf_item_unpin(). Until that happens we will | |
621 | * keep the buffer locked so that the buffer and buf log item are not reused. | |
622 | * | |
623 | * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log | |
624 | * the buf item. This will be used at recovery time to determine that copies | |
625 | * of the buffer in the log before this should not be replayed. | |
626 | * | |
627 | * We mark the item descriptor and the transaction dirty so that we'll hold | |
628 | * the buffer until after the commit. | |
629 | * | |
630 | * Since we're invalidating the buffer, we also clear the state about which | |
631 | * parts of the buffer have been logged. We also clear the flag indicating | |
632 | * that this is an inode buffer since the data in the buffer will no longer | |
633 | * be valid. | |
634 | * | |
635 | * We set the stale bit in the buffer as well since we're getting rid of it. | |
1da177e4 LT |
636 | */ |
637 | void | |
638 | xfs_trans_binval( | |
639 | xfs_trans_t *tp, | |
640 | xfs_buf_t *bp) | |
641 | { | |
adadbeef | 642 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 643 | |
bf9d9013 | 644 | ASSERT(bp->b_transp == tp); |
adadbeef | 645 | ASSERT(bip != NULL); |
1da177e4 LT |
646 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
647 | ||
0b1b213f CH |
648 | trace_xfs_trans_binval(bip); |
649 | ||
1da177e4 LT |
650 | if (bip->bli_flags & XFS_BLI_STALE) { |
651 | /* | |
652 | * If the buffer is already invalidated, then | |
653 | * just return. | |
654 | */ | |
1da177e4 LT |
655 | ASSERT(XFS_BUF_ISSTALE(bp)); |
656 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); | |
c1155410 DC |
657 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
658 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); | |
e98c414f | 659 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
1da177e4 | 660 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
1da177e4 LT |
661 | return; |
662 | } | |
663 | ||
c867cb61 | 664 | xfs_buf_stale(bp); |
43ff2122 | 665 | |
1da177e4 | 666 | bip->bli_flags |= XFS_BLI_STALE; |
ccf7c23f | 667 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
c1155410 DC |
668 | bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
669 | bip->bli_format.blf_flags |= XFS_BLF_CANCEL; | |
1da177e4 LT |
670 | memset((char *)(bip->bli_format.blf_data_map), 0, |
671 | (bip->bli_format.blf_map_size * sizeof(uint))); | |
e98c414f | 672 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 | 673 | tp->t_flags |= XFS_TRANS_DIRTY; |
1da177e4 LT |
674 | } |
675 | ||
676 | /* | |
ccf7c23f DC |
677 | * This call is used to indicate that the buffer contains on-disk inodes which |
678 | * must be handled specially during recovery. They require special handling | |
679 | * because only the di_next_unlinked from the inodes in the buffer should be | |
680 | * recovered. The rest of the data in the buffer is logged via the inodes | |
681 | * themselves. | |
1da177e4 | 682 | * |
ccf7c23f DC |
683 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
684 | * transferred to the buffer's log format structure so that we'll know what to | |
685 | * do at recovery time. | |
1da177e4 | 686 | */ |
1da177e4 LT |
687 | void |
688 | xfs_trans_inode_buf( | |
689 | xfs_trans_t *tp, | |
690 | xfs_buf_t *bp) | |
691 | { | |
adadbeef | 692 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 693 | |
bf9d9013 | 694 | ASSERT(bp->b_transp == tp); |
adadbeef | 695 | ASSERT(bip != NULL); |
1da177e4 LT |
696 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
697 | ||
ccf7c23f | 698 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
1da177e4 LT |
699 | } |
700 | ||
701 | /* | |
702 | * This call is used to indicate that the buffer is going to | |
703 | * be staled and was an inode buffer. This means it gets | |
704 | * special processing during unpin - where any inodes | |
705 | * associated with the buffer should be removed from ail. | |
706 | * There is also special processing during recovery, | |
707 | * any replay of the inodes in the buffer needs to be | |
708 | * prevented as the buffer may have been reused. | |
709 | */ | |
710 | void | |
711 | xfs_trans_stale_inode_buf( | |
712 | xfs_trans_t *tp, | |
713 | xfs_buf_t *bp) | |
714 | { | |
adadbeef | 715 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 716 | |
bf9d9013 | 717 | ASSERT(bp->b_transp == tp); |
adadbeef | 718 | ASSERT(bip != NULL); |
1da177e4 LT |
719 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
720 | ||
721 | bip->bli_flags |= XFS_BLI_STALE_INODE; | |
ca30b2a7 | 722 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 LT |
723 | } |
724 | ||
1da177e4 LT |
725 | /* |
726 | * Mark the buffer as being one which contains newly allocated | |
727 | * inodes. We need to make sure that even if this buffer is | |
728 | * relogged as an 'inode buf' we still recover all of the inode | |
729 | * images in the face of a crash. This works in coordination with | |
730 | * xfs_buf_item_committed() to ensure that the buffer remains in the | |
731 | * AIL at its original location even after it has been relogged. | |
732 | */ | |
733 | /* ARGSUSED */ | |
734 | void | |
735 | xfs_trans_inode_alloc_buf( | |
736 | xfs_trans_t *tp, | |
737 | xfs_buf_t *bp) | |
738 | { | |
adadbeef | 739 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 740 | |
bf9d9013 | 741 | ASSERT(bp->b_transp == tp); |
adadbeef | 742 | ASSERT(bip != NULL); |
1da177e4 LT |
743 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
744 | ||
745 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; | |
746 | } | |
747 | ||
748 | ||
749 | /* | |
750 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of | |
751 | * dquots. However, unlike in inode buffer recovery, dquot buffers get | |
752 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). | |
753 | * The only thing that makes dquot buffers different from regular | |
754 | * buffers is that we must not replay dquot bufs when recovering | |
755 | * if a _corresponding_ quotaoff has happened. We also have to distinguish | |
756 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas | |
757 | * can be turned off independently. | |
758 | */ | |
759 | /* ARGSUSED */ | |
760 | void | |
761 | xfs_trans_dquot_buf( | |
762 | xfs_trans_t *tp, | |
763 | xfs_buf_t *bp, | |
764 | uint type) | |
765 | { | |
adadbeef | 766 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 767 | |
bf9d9013 | 768 | ASSERT(bp->b_transp == tp); |
adadbeef | 769 | ASSERT(bip != NULL); |
c1155410 DC |
770 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
771 | type == XFS_BLF_PDQUOT_BUF || | |
772 | type == XFS_BLF_GDQUOT_BUF); | |
1da177e4 LT |
773 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
774 | ||
775 | bip->bli_format.blf_flags |= type; | |
776 | } |