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" |
70a9883c | 20 | #include "xfs_shared.h" |
a4fbe6ab | 21 | #include "xfs_format.h" |
239880ef DC |
22 | #include "xfs_log_format.h" |
23 | #include "xfs_trans_resv.h" | |
1da177e4 LT |
24 | #include "xfs_sb.h" |
25 | #include "xfs_ag.h" | |
1da177e4 | 26 | #include "xfs_mount.h" |
a844f451 | 27 | #include "xfs_inode.h" |
239880ef | 28 | #include "xfs_trans.h" |
a844f451 | 29 | #include "xfs_buf_item.h" |
1da177e4 LT |
30 | #include "xfs_trans_priv.h" |
31 | #include "xfs_error.h" | |
0b1b213f | 32 | #include "xfs_trace.h" |
1da177e4 | 33 | |
4a5224d7 CH |
34 | /* |
35 | * Check to see if a buffer matching the given parameters is already | |
36 | * a part of the given transaction. | |
37 | */ | |
38 | STATIC struct xfs_buf * | |
39 | xfs_trans_buf_item_match( | |
40 | struct xfs_trans *tp, | |
41 | struct xfs_buftarg *target, | |
de2a4f59 DC |
42 | struct xfs_buf_map *map, |
43 | int nmaps) | |
4a5224d7 | 44 | { |
e98c414f CH |
45 | struct xfs_log_item_desc *lidp; |
46 | struct xfs_buf_log_item *blip; | |
de2a4f59 DC |
47 | int len = 0; |
48 | int i; | |
49 | ||
50 | for (i = 0; i < nmaps; i++) | |
51 | len += map[i].bm_len; | |
1da177e4 | 52 | |
e98c414f CH |
53 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
54 | blip = (struct xfs_buf_log_item *)lidp->lid_item; | |
55 | if (blip->bli_item.li_type == XFS_LI_BUF && | |
49074c06 | 56 | blip->bli_buf->b_target == target && |
de2a4f59 DC |
57 | XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn && |
58 | blip->bli_buf->b_length == len) { | |
59 | ASSERT(blip->bli_buf->b_map_count == nmaps); | |
e98c414f | 60 | return blip->bli_buf; |
de2a4f59 | 61 | } |
4a5224d7 CH |
62 | } |
63 | ||
64 | return NULL; | |
65 | } | |
1da177e4 | 66 | |
d7e84f41 CH |
67 | /* |
68 | * Add the locked buffer to the transaction. | |
69 | * | |
70 | * The buffer must be locked, and it cannot be associated with any | |
71 | * transaction. | |
72 | * | |
73 | * If the buffer does not yet have a buf log item associated with it, | |
74 | * then allocate one for it. Then add the buf item to the transaction. | |
75 | */ | |
76 | STATIC void | |
77 | _xfs_trans_bjoin( | |
78 | struct xfs_trans *tp, | |
79 | struct xfs_buf *bp, | |
80 | int reset_recur) | |
81 | { | |
82 | struct xfs_buf_log_item *bip; | |
83 | ||
bf9d9013 | 84 | ASSERT(bp->b_transp == NULL); |
d7e84f41 CH |
85 | |
86 | /* | |
87 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If | |
88 | * it doesn't have one yet, then allocate one and initialize it. | |
89 | * The checks to see if one is there are in xfs_buf_item_init(). | |
90 | */ | |
91 | xfs_buf_item_init(bp, tp->t_mountp); | |
adadbeef | 92 | bip = bp->b_fspriv; |
d7e84f41 | 93 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
0f22f9d0 | 94 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
d7e84f41 CH |
95 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
96 | if (reset_recur) | |
97 | bip->bli_recur = 0; | |
98 | ||
99 | /* | |
100 | * Take a reference for this transaction on the buf item. | |
101 | */ | |
102 | atomic_inc(&bip->bli_refcount); | |
103 | ||
104 | /* | |
105 | * Get a log_item_desc to point at the new item. | |
106 | */ | |
e98c414f | 107 | xfs_trans_add_item(tp, &bip->bli_item); |
d7e84f41 CH |
108 | |
109 | /* | |
110 | * Initialize b_fsprivate2 so we can find it with incore_match() | |
111 | * in xfs_trans_get_buf() and friends above. | |
112 | */ | |
bf9d9013 | 113 | bp->b_transp = tp; |
d7e84f41 CH |
114 | |
115 | } | |
116 | ||
117 | void | |
118 | xfs_trans_bjoin( | |
119 | struct xfs_trans *tp, | |
120 | struct xfs_buf *bp) | |
121 | { | |
122 | _xfs_trans_bjoin(tp, bp, 0); | |
123 | trace_xfs_trans_bjoin(bp->b_fspriv); | |
124 | } | |
1da177e4 LT |
125 | |
126 | /* | |
127 | * Get and lock the buffer for the caller if it is not already | |
128 | * locked within the given transaction. If it is already locked | |
129 | * within the transaction, just increment its lock recursion count | |
130 | * and return a pointer to it. | |
131 | * | |
1da177e4 LT |
132 | * If the transaction pointer is NULL, make this just a normal |
133 | * get_buf() call. | |
134 | */ | |
de2a4f59 DC |
135 | struct xfs_buf * |
136 | xfs_trans_get_buf_map( | |
137 | struct xfs_trans *tp, | |
138 | struct xfs_buftarg *target, | |
139 | struct xfs_buf_map *map, | |
140 | int nmaps, | |
141 | xfs_buf_flags_t flags) | |
1da177e4 LT |
142 | { |
143 | xfs_buf_t *bp; | |
144 | xfs_buf_log_item_t *bip; | |
145 | ||
de2a4f59 DC |
146 | if (!tp) |
147 | return xfs_buf_get_map(target, map, nmaps, flags); | |
1da177e4 LT |
148 | |
149 | /* | |
150 | * If we find the buffer in the cache with this transaction | |
151 | * pointer in its b_fsprivate2 field, then we know we already | |
152 | * have it locked. In this case we just increment the lock | |
153 | * recursion count and return the buffer to the caller. | |
154 | */ | |
de2a4f59 | 155 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
1da177e4 | 156 | if (bp != NULL) { |
0c842ad4 | 157 | ASSERT(xfs_buf_islocked(bp)); |
c867cb61 CH |
158 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) { |
159 | xfs_buf_stale(bp); | |
c867cb61 CH |
160 | XFS_BUF_DONE(bp); |
161 | } | |
0b1b213f | 162 | |
bf9d9013 | 163 | ASSERT(bp->b_transp == tp); |
adadbeef | 164 | bip = bp->b_fspriv; |
1da177e4 LT |
165 | ASSERT(bip != NULL); |
166 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
167 | bip->bli_recur++; | |
0b1b213f | 168 | trace_xfs_trans_get_buf_recur(bip); |
1da177e4 LT |
169 | return (bp); |
170 | } | |
171 | ||
de2a4f59 | 172 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
1da177e4 LT |
173 | if (bp == NULL) { |
174 | return NULL; | |
175 | } | |
176 | ||
5a52c2a5 | 177 | ASSERT(!bp->b_error); |
1da177e4 | 178 | |
d7e84f41 CH |
179 | _xfs_trans_bjoin(tp, bp, 1); |
180 | trace_xfs_trans_get_buf(bp->b_fspriv); | |
1da177e4 LT |
181 | return (bp); |
182 | } | |
183 | ||
184 | /* | |
185 | * Get and lock the superblock buffer of this file system for the | |
186 | * given transaction. | |
187 | * | |
188 | * We don't need to use incore_match() here, because the superblock | |
189 | * buffer is a private buffer which we keep a pointer to in the | |
190 | * mount structure. | |
191 | */ | |
192 | xfs_buf_t * | |
193 | xfs_trans_getsb(xfs_trans_t *tp, | |
194 | struct xfs_mount *mp, | |
195 | int flags) | |
196 | { | |
197 | xfs_buf_t *bp; | |
198 | xfs_buf_log_item_t *bip; | |
199 | ||
200 | /* | |
201 | * Default to just trying to lock the superblock buffer | |
202 | * if tp is NULL. | |
203 | */ | |
204 | if (tp == NULL) { | |
205 | return (xfs_getsb(mp, flags)); | |
206 | } | |
207 | ||
208 | /* | |
209 | * If the superblock buffer already has this transaction | |
210 | * pointer in its b_fsprivate2 field, then we know we already | |
211 | * have it locked. In this case we just increment the lock | |
212 | * recursion count and return the buffer to the caller. | |
213 | */ | |
214 | bp = mp->m_sb_bp; | |
bf9d9013 | 215 | if (bp->b_transp == tp) { |
adadbeef | 216 | bip = bp->b_fspriv; |
1da177e4 LT |
217 | ASSERT(bip != NULL); |
218 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
219 | bip->bli_recur++; | |
0b1b213f | 220 | trace_xfs_trans_getsb_recur(bip); |
1da177e4 LT |
221 | return (bp); |
222 | } | |
223 | ||
224 | bp = xfs_getsb(mp, flags); | |
d7e84f41 | 225 | if (bp == NULL) |
1da177e4 | 226 | return NULL; |
1da177e4 | 227 | |
d7e84f41 CH |
228 | _xfs_trans_bjoin(tp, bp, 1); |
229 | trace_xfs_trans_getsb(bp->b_fspriv); | |
1da177e4 LT |
230 | return (bp); |
231 | } | |
232 | ||
233 | #ifdef DEBUG | |
234 | xfs_buftarg_t *xfs_error_target; | |
235 | int xfs_do_error; | |
236 | int xfs_req_num; | |
237 | int xfs_error_mod = 33; | |
238 | #endif | |
239 | ||
240 | /* | |
241 | * Get and lock the buffer for the caller if it is not already | |
242 | * locked within the given transaction. If it has not yet been | |
243 | * read in, read it from disk. If it is already locked | |
244 | * within the transaction and already read in, just increment its | |
245 | * lock recursion count and return a pointer to it. | |
246 | * | |
1da177e4 LT |
247 | * If the transaction pointer is NULL, make this just a normal |
248 | * read_buf() call. | |
249 | */ | |
250 | int | |
de2a4f59 DC |
251 | xfs_trans_read_buf_map( |
252 | struct xfs_mount *mp, | |
253 | struct xfs_trans *tp, | |
254 | struct xfs_buftarg *target, | |
255 | struct xfs_buf_map *map, | |
256 | int nmaps, | |
257 | xfs_buf_flags_t flags, | |
c3f8fc73 | 258 | struct xfs_buf **bpp, |
1813dd64 | 259 | const struct xfs_buf_ops *ops) |
1da177e4 LT |
260 | { |
261 | xfs_buf_t *bp; | |
262 | xfs_buf_log_item_t *bip; | |
263 | int error; | |
264 | ||
7ca790a5 | 265 | *bpp = NULL; |
de2a4f59 | 266 | if (!tp) { |
1813dd64 | 267 | bp = xfs_buf_read_map(target, map, nmaps, flags, ops); |
1da177e4 | 268 | if (!bp) |
0cadda1c | 269 | return (flags & XBF_TRYLOCK) ? |
a3f74ffb | 270 | EAGAIN : XFS_ERROR(ENOMEM); |
1da177e4 | 271 | |
5a52c2a5 CS |
272 | if (bp->b_error) { |
273 | error = bp->b_error; | |
901796af | 274 | xfs_buf_ioerror_alert(bp, __func__); |
7ca790a5 DC |
275 | XFS_BUF_UNDONE(bp); |
276 | xfs_buf_stale(bp); | |
1da177e4 | 277 | xfs_buf_relse(bp); |
ac75a1f7 DC |
278 | |
279 | /* bad CRC means corrupted metadata */ | |
280 | if (error == EFSBADCRC) | |
281 | error = EFSCORRUPTED; | |
1da177e4 LT |
282 | return error; |
283 | } | |
284 | #ifdef DEBUG | |
a0f7bfd3 | 285 | if (xfs_do_error) { |
1da177e4 LT |
286 | if (xfs_error_target == target) { |
287 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
288 | xfs_buf_relse(bp); | |
0b932ccc | 289 | xfs_debug(mp, "Returning error!"); |
1da177e4 LT |
290 | return XFS_ERROR(EIO); |
291 | } | |
292 | } | |
293 | } | |
294 | #endif | |
295 | if (XFS_FORCED_SHUTDOWN(mp)) | |
296 | goto shutdown_abort; | |
297 | *bpp = bp; | |
298 | return 0; | |
299 | } | |
300 | ||
301 | /* | |
302 | * If we find the buffer in the cache with this transaction | |
303 | * pointer in its b_fsprivate2 field, then we know we already | |
304 | * have it locked. If it is already read in we just increment | |
305 | * the lock recursion count and return the buffer to the caller. | |
306 | * If the buffer is not yet read in, then we read it in, increment | |
307 | * the lock recursion count, and return it to the caller. | |
308 | */ | |
de2a4f59 | 309 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
1da177e4 | 310 | if (bp != NULL) { |
0c842ad4 | 311 | ASSERT(xfs_buf_islocked(bp)); |
bf9d9013 | 312 | ASSERT(bp->b_transp == tp); |
adadbeef | 313 | ASSERT(bp->b_fspriv != NULL); |
5a52c2a5 | 314 | ASSERT(!bp->b_error); |
1da177e4 | 315 | if (!(XFS_BUF_ISDONE(bp))) { |
0b1b213f | 316 | trace_xfs_trans_read_buf_io(bp, _RET_IP_); |
1da177e4 | 317 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
c3f8fc73 | 318 | ASSERT(bp->b_iodone == NULL); |
1da177e4 | 319 | XFS_BUF_READ(bp); |
1813dd64 | 320 | bp->b_ops = ops; |
83a0adc3 CH |
321 | |
322 | /* | |
323 | * XXX(hch): clean up the error handling here to be less | |
324 | * of a mess.. | |
325 | */ | |
326 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
327 | trace_xfs_bdstrat_shut(bp, _RET_IP_); | |
328 | xfs_bioerror_relse(bp); | |
329 | } else { | |
330 | xfs_buf_iorequest(bp); | |
331 | } | |
332 | ||
1a1a3e97 | 333 | error = xfs_buf_iowait(bp); |
d64e31a2 | 334 | if (error) { |
901796af | 335 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
336 | xfs_buf_relse(bp); |
337 | /* | |
d64e31a2 DC |
338 | * We can gracefully recover from most read |
339 | * errors. Ones we can't are those that happen | |
340 | * after the transaction's already dirty. | |
1da177e4 LT |
341 | */ |
342 | if (tp->t_flags & XFS_TRANS_DIRTY) | |
343 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 344 | SHUTDOWN_META_IO_ERROR); |
ac75a1f7 DC |
345 | /* bad CRC means corrupted metadata */ |
346 | if (error == EFSBADCRC) | |
347 | error = EFSCORRUPTED; | |
1da177e4 LT |
348 | return error; |
349 | } | |
350 | } | |
351 | /* | |
352 | * We never locked this buf ourselves, so we shouldn't | |
353 | * brelse it either. Just get out. | |
354 | */ | |
355 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
0b1b213f | 356 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
357 | *bpp = NULL; |
358 | return XFS_ERROR(EIO); | |
359 | } | |
360 | ||
361 | ||
adadbeef | 362 | bip = bp->b_fspriv; |
1da177e4 LT |
363 | bip->bli_recur++; |
364 | ||
365 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 366 | trace_xfs_trans_read_buf_recur(bip); |
1da177e4 LT |
367 | *bpp = bp; |
368 | return 0; | |
369 | } | |
370 | ||
1813dd64 | 371 | bp = xfs_buf_read_map(target, map, nmaps, flags, ops); |
1da177e4 LT |
372 | if (bp == NULL) { |
373 | *bpp = NULL; | |
7401aafd DC |
374 | return (flags & XBF_TRYLOCK) ? |
375 | 0 : XFS_ERROR(ENOMEM); | |
1da177e4 | 376 | } |
5a52c2a5 CS |
377 | if (bp->b_error) { |
378 | error = bp->b_error; | |
c867cb61 | 379 | xfs_buf_stale(bp); |
c867cb61 | 380 | XFS_BUF_DONE(bp); |
901796af | 381 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 | 382 | if (tp->t_flags & XFS_TRANS_DIRTY) |
7d04a335 | 383 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
1da177e4 | 384 | xfs_buf_relse(bp); |
ac75a1f7 DC |
385 | |
386 | /* bad CRC means corrupted metadata */ | |
387 | if (error == EFSBADCRC) | |
388 | error = EFSCORRUPTED; | |
1da177e4 LT |
389 | return error; |
390 | } | |
391 | #ifdef DEBUG | |
392 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { | |
393 | if (xfs_error_target == target) { | |
394 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
395 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 396 | SHUTDOWN_META_IO_ERROR); |
1da177e4 | 397 | xfs_buf_relse(bp); |
0b932ccc | 398 | xfs_debug(mp, "Returning trans error!"); |
1da177e4 LT |
399 | return XFS_ERROR(EIO); |
400 | } | |
401 | } | |
402 | } | |
403 | #endif | |
404 | if (XFS_FORCED_SHUTDOWN(mp)) | |
405 | goto shutdown_abort; | |
406 | ||
d7e84f41 CH |
407 | _xfs_trans_bjoin(tp, bp, 1); |
408 | trace_xfs_trans_read_buf(bp->b_fspriv); | |
1da177e4 | 409 | |
1da177e4 LT |
410 | *bpp = bp; |
411 | return 0; | |
412 | ||
413 | shutdown_abort: | |
0b1b213f | 414 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
415 | xfs_buf_relse(bp); |
416 | *bpp = NULL; | |
417 | return XFS_ERROR(EIO); | |
418 | } | |
419 | ||
1da177e4 LT |
420 | /* |
421 | * Release the buffer bp which was previously acquired with one of the | |
422 | * xfs_trans_... buffer allocation routines if the buffer has not | |
423 | * been modified within this transaction. If the buffer is modified | |
424 | * within this transaction, do decrement the recursion count but do | |
425 | * not release the buffer even if the count goes to 0. If the buffer is not | |
426 | * modified within the transaction, decrement the recursion count and | |
427 | * release the buffer if the recursion count goes to 0. | |
428 | * | |
429 | * If the buffer is to be released and it was not modified before | |
430 | * this transaction began, then free the buf_log_item associated with it. | |
431 | * | |
432 | * If the transaction pointer is NULL, make this just a normal | |
433 | * brelse() call. | |
434 | */ | |
435 | void | |
436 | xfs_trans_brelse(xfs_trans_t *tp, | |
437 | xfs_buf_t *bp) | |
438 | { | |
439 | xfs_buf_log_item_t *bip; | |
1da177e4 LT |
440 | |
441 | /* | |
442 | * Default to a normal brelse() call if the tp is NULL. | |
443 | */ | |
444 | if (tp == NULL) { | |
bf9d9013 | 445 | ASSERT(bp->b_transp == NULL); |
1da177e4 LT |
446 | xfs_buf_relse(bp); |
447 | return; | |
448 | } | |
449 | ||
bf9d9013 | 450 | ASSERT(bp->b_transp == tp); |
adadbeef | 451 | bip = bp->b_fspriv; |
1da177e4 LT |
452 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
453 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
0f22f9d0 | 454 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
455 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
456 | ||
0b1b213f CH |
457 | trace_xfs_trans_brelse(bip); |
458 | ||
1da177e4 LT |
459 | /* |
460 | * If the release is just for a recursive lock, | |
461 | * then decrement the count and return. | |
462 | */ | |
463 | if (bip->bli_recur > 0) { | |
464 | bip->bli_recur--; | |
1da177e4 LT |
465 | return; |
466 | } | |
467 | ||
468 | /* | |
469 | * If the buffer is dirty within this transaction, we can't | |
470 | * release it until we commit. | |
471 | */ | |
e98c414f | 472 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
1da177e4 | 473 | return; |
1da177e4 LT |
474 | |
475 | /* | |
476 | * If the buffer has been invalidated, then we can't release | |
477 | * it until the transaction commits to disk unless it is re-dirtied | |
478 | * as part of this transaction. This prevents us from pulling | |
479 | * the item from the AIL before we should. | |
480 | */ | |
0b1b213f | 481 | if (bip->bli_flags & XFS_BLI_STALE) |
1da177e4 | 482 | return; |
1da177e4 LT |
483 | |
484 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); | |
1da177e4 LT |
485 | |
486 | /* | |
487 | * Free up the log item descriptor tracking the released item. | |
488 | */ | |
e98c414f | 489 | xfs_trans_del_item(&bip->bli_item); |
1da177e4 LT |
490 | |
491 | /* | |
492 | * Clear the hold flag in the buf log item if it is set. | |
493 | * We wouldn't want the next user of the buffer to | |
494 | * get confused. | |
495 | */ | |
496 | if (bip->bli_flags & XFS_BLI_HOLD) { | |
497 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
498 | } | |
499 | ||
500 | /* | |
501 | * Drop our reference to the buf log item. | |
502 | */ | |
503 | atomic_dec(&bip->bli_refcount); | |
504 | ||
505 | /* | |
506 | * If the buf item is not tracking data in the log, then | |
507 | * we must free it before releasing the buffer back to the | |
508 | * free pool. Before releasing the buffer to the free pool, | |
509 | * clear the transaction pointer in b_fsprivate2 to dissolve | |
510 | * its relation to this transaction. | |
511 | */ | |
512 | if (!xfs_buf_item_dirty(bip)) { | |
513 | /*** | |
514 | ASSERT(bp->b_pincount == 0); | |
515 | ***/ | |
516 | ASSERT(atomic_read(&bip->bli_refcount) == 0); | |
517 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); | |
518 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); | |
519 | xfs_buf_item_relse(bp); | |
1da177e4 LT |
520 | } |
521 | ||
5b03ff1b | 522 | bp->b_transp = NULL; |
1da177e4 | 523 | xfs_buf_relse(bp); |
1da177e4 LT |
524 | } |
525 | ||
1da177e4 LT |
526 | /* |
527 | * Mark the buffer as not needing to be unlocked when the buf item's | |
904c17e6 | 528 | * iop_unlock() routine is called. The buffer must already be locked |
1da177e4 LT |
529 | * and associated with the given transaction. |
530 | */ | |
531 | /* ARGSUSED */ | |
532 | void | |
533 | xfs_trans_bhold(xfs_trans_t *tp, | |
534 | xfs_buf_t *bp) | |
535 | { | |
adadbeef | 536 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 537 | |
bf9d9013 | 538 | ASSERT(bp->b_transp == tp); |
adadbeef | 539 | ASSERT(bip != NULL); |
1da177e4 | 540 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
0f22f9d0 | 541 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 | 542 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
adadbeef | 543 | |
1da177e4 | 544 | bip->bli_flags |= XFS_BLI_HOLD; |
0b1b213f | 545 | trace_xfs_trans_bhold(bip); |
1da177e4 LT |
546 | } |
547 | ||
efa092f3 TS |
548 | /* |
549 | * Cancel the previous buffer hold request made on this buffer | |
550 | * for this transaction. | |
551 | */ | |
552 | void | |
553 | xfs_trans_bhold_release(xfs_trans_t *tp, | |
554 | xfs_buf_t *bp) | |
555 | { | |
adadbeef | 556 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
efa092f3 | 557 | |
bf9d9013 | 558 | ASSERT(bp->b_transp == tp); |
adadbeef | 559 | ASSERT(bip != NULL); |
efa092f3 | 560 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
0f22f9d0 | 561 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
efa092f3 TS |
562 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
563 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); | |
0b1b213f | 564 | |
adadbeef | 565 | bip->bli_flags &= ~XFS_BLI_HOLD; |
0b1b213f | 566 | trace_xfs_trans_bhold_release(bip); |
efa092f3 TS |
567 | } |
568 | ||
1da177e4 LT |
569 | /* |
570 | * This is called to mark bytes first through last inclusive of the given | |
571 | * buffer as needing to be logged when the transaction is committed. | |
572 | * The buffer must already be associated with the given transaction. | |
573 | * | |
574 | * First and last are numbers relative to the beginning of this buffer, | |
575 | * so the first byte in the buffer is numbered 0 regardless of the | |
576 | * value of b_blkno. | |
577 | */ | |
578 | void | |
579 | xfs_trans_log_buf(xfs_trans_t *tp, | |
580 | xfs_buf_t *bp, | |
581 | uint first, | |
582 | uint last) | |
583 | { | |
adadbeef | 584 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 585 | |
bf9d9013 | 586 | ASSERT(bp->b_transp == tp); |
adadbeef | 587 | ASSERT(bip != NULL); |
aa0e8833 | 588 | ASSERT(first <= last && last < BBTOB(bp->b_length)); |
cb669ca5 CH |
589 | ASSERT(bp->b_iodone == NULL || |
590 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1da177e4 LT |
591 | |
592 | /* | |
593 | * Mark the buffer as needing to be written out eventually, | |
594 | * and set its iodone function to remove the buffer's buf log | |
595 | * item from the AIL and free it when the buffer is flushed | |
596 | * to disk. See xfs_buf_attach_iodone() for more details | |
597 | * on li_cb and xfs_buf_iodone_callbacks(). | |
598 | * If we end up aborting this transaction, we trap this buffer | |
599 | * inside the b_bdstrat callback so that this won't get written to | |
600 | * disk. | |
601 | */ | |
1da177e4 LT |
602 | XFS_BUF_DONE(bp); |
603 | ||
1da177e4 | 604 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
cb669ca5 | 605 | bp->b_iodone = xfs_buf_iodone_callbacks; |
ca30b2a7 | 606 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 | 607 | |
0b1b213f CH |
608 | trace_xfs_trans_log_buf(bip); |
609 | ||
1da177e4 LT |
610 | /* |
611 | * If we invalidated the buffer within this transaction, then | |
612 | * cancel the invalidation now that we're dirtying the buffer | |
613 | * again. There are no races with the code in xfs_buf_item_unpin(), | |
614 | * because we have a reference to the buffer this entire time. | |
615 | */ | |
616 | if (bip->bli_flags & XFS_BLI_STALE) { | |
1da177e4 LT |
617 | bip->bli_flags &= ~XFS_BLI_STALE; |
618 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
619 | XFS_BUF_UNSTALE(bp); | |
0f22f9d0 | 620 | bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
1da177e4 LT |
621 | } |
622 | ||
1da177e4 | 623 | tp->t_flags |= XFS_TRANS_DIRTY; |
e98c414f | 624 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
5f6bed76 DC |
625 | |
626 | /* | |
627 | * If we have an ordered buffer we are not logging any dirty range but | |
628 | * it still needs to be marked dirty and that it has been logged. | |
629 | */ | |
630 | bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED; | |
631 | if (!(bip->bli_flags & XFS_BLI_ORDERED)) | |
632 | xfs_buf_item_log(bip, first, last); | |
1da177e4 LT |
633 | } |
634 | ||
635 | ||
636 | /* | |
43ff2122 CH |
637 | * Invalidate a buffer that is being used within a transaction. |
638 | * | |
639 | * Typically this is because the blocks in the buffer are being freed, so we | |
640 | * need to prevent it from being written out when we're done. Allowing it | |
641 | * to be written again might overwrite data in the free blocks if they are | |
642 | * reallocated to a file. | |
1da177e4 | 643 | * |
43ff2122 CH |
644 | * We prevent the buffer from being written out by marking it stale. We can't |
645 | * get rid of the buf log item at this point because the buffer may still be | |
646 | * pinned by another transaction. If that is the case, then we'll wait until | |
647 | * the buffer is committed to disk for the last time (we can tell by the ref | |
648 | * count) and free it in xfs_buf_item_unpin(). Until that happens we will | |
649 | * keep the buffer locked so that the buffer and buf log item are not reused. | |
650 | * | |
651 | * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log | |
652 | * the buf item. This will be used at recovery time to determine that copies | |
653 | * of the buffer in the log before this should not be replayed. | |
654 | * | |
655 | * We mark the item descriptor and the transaction dirty so that we'll hold | |
656 | * the buffer until after the commit. | |
657 | * | |
658 | * Since we're invalidating the buffer, we also clear the state about which | |
659 | * parts of the buffer have been logged. We also clear the flag indicating | |
660 | * that this is an inode buffer since the data in the buffer will no longer | |
661 | * be valid. | |
662 | * | |
663 | * We set the stale bit in the buffer as well since we're getting rid of it. | |
1da177e4 LT |
664 | */ |
665 | void | |
666 | xfs_trans_binval( | |
667 | xfs_trans_t *tp, | |
668 | xfs_buf_t *bp) | |
669 | { | |
adadbeef | 670 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
91e4bac0 | 671 | int i; |
1da177e4 | 672 | |
bf9d9013 | 673 | ASSERT(bp->b_transp == tp); |
adadbeef | 674 | ASSERT(bip != NULL); |
1da177e4 LT |
675 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
676 | ||
0b1b213f CH |
677 | trace_xfs_trans_binval(bip); |
678 | ||
1da177e4 LT |
679 | if (bip->bli_flags & XFS_BLI_STALE) { |
680 | /* | |
681 | * If the buffer is already invalidated, then | |
682 | * just return. | |
683 | */ | |
1da177e4 LT |
684 | ASSERT(XFS_BUF_ISSTALE(bp)); |
685 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); | |
0f22f9d0 | 686 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
61fe135c | 687 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK)); |
0f22f9d0 | 688 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
e98c414f | 689 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
1da177e4 | 690 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
1da177e4 LT |
691 | return; |
692 | } | |
693 | ||
c867cb61 | 694 | xfs_buf_stale(bp); |
43ff2122 | 695 | |
1da177e4 | 696 | bip->bli_flags |= XFS_BLI_STALE; |
ccf7c23f | 697 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
0f22f9d0 MT |
698 | bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
699 | bip->__bli_format.blf_flags |= XFS_BLF_CANCEL; | |
61fe135c | 700 | bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK; |
91e4bac0 MT |
701 | for (i = 0; i < bip->bli_format_count; i++) { |
702 | memset(bip->bli_formats[i].blf_data_map, 0, | |
703 | (bip->bli_formats[i].blf_map_size * sizeof(uint))); | |
704 | } | |
e98c414f | 705 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 | 706 | tp->t_flags |= XFS_TRANS_DIRTY; |
1da177e4 LT |
707 | } |
708 | ||
709 | /* | |
ccf7c23f DC |
710 | * This call is used to indicate that the buffer contains on-disk inodes which |
711 | * must be handled specially during recovery. They require special handling | |
712 | * because only the di_next_unlinked from the inodes in the buffer should be | |
713 | * recovered. The rest of the data in the buffer is logged via the inodes | |
714 | * themselves. | |
1da177e4 | 715 | * |
ccf7c23f DC |
716 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
717 | * transferred to the buffer's log format structure so that we'll know what to | |
718 | * do at recovery time. | |
1da177e4 | 719 | */ |
1da177e4 LT |
720 | void |
721 | xfs_trans_inode_buf( | |
722 | xfs_trans_t *tp, | |
723 | xfs_buf_t *bp) | |
724 | { | |
adadbeef | 725 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 726 | |
bf9d9013 | 727 | ASSERT(bp->b_transp == tp); |
adadbeef | 728 | ASSERT(bip != NULL); |
1da177e4 LT |
729 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
730 | ||
ccf7c23f | 731 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
61fe135c | 732 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF); |
1da177e4 LT |
733 | } |
734 | ||
735 | /* | |
736 | * This call is used to indicate that the buffer is going to | |
737 | * be staled and was an inode buffer. This means it gets | |
93848a99 | 738 | * special processing during unpin - where any inodes |
1da177e4 LT |
739 | * associated with the buffer should be removed from ail. |
740 | * There is also special processing during recovery, | |
741 | * any replay of the inodes in the buffer needs to be | |
742 | * prevented as the buffer may have been reused. | |
743 | */ | |
744 | void | |
745 | xfs_trans_stale_inode_buf( | |
746 | xfs_trans_t *tp, | |
747 | xfs_buf_t *bp) | |
748 | { | |
adadbeef | 749 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 750 | |
bf9d9013 | 751 | ASSERT(bp->b_transp == tp); |
adadbeef | 752 | ASSERT(bip != NULL); |
1da177e4 LT |
753 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
754 | ||
755 | bip->bli_flags |= XFS_BLI_STALE_INODE; | |
ca30b2a7 | 756 | bip->bli_item.li_cb = xfs_buf_iodone; |
61fe135c | 757 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF); |
1da177e4 LT |
758 | } |
759 | ||
1da177e4 LT |
760 | /* |
761 | * Mark the buffer as being one which contains newly allocated | |
762 | * inodes. We need to make sure that even if this buffer is | |
763 | * relogged as an 'inode buf' we still recover all of the inode | |
764 | * images in the face of a crash. This works in coordination with | |
765 | * xfs_buf_item_committed() to ensure that the buffer remains in the | |
766 | * AIL at its original location even after it has been relogged. | |
767 | */ | |
768 | /* ARGSUSED */ | |
769 | void | |
770 | xfs_trans_inode_alloc_buf( | |
771 | xfs_trans_t *tp, | |
772 | xfs_buf_t *bp) | |
773 | { | |
adadbeef | 774 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 775 | |
bf9d9013 | 776 | ASSERT(bp->b_transp == tp); |
adadbeef | 777 | ASSERT(bip != NULL); |
1da177e4 LT |
778 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
779 | ||
780 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; | |
61fe135c | 781 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF); |
1da177e4 LT |
782 | } |
783 | ||
5f6bed76 DC |
784 | /* |
785 | * Mark the buffer as ordered for this transaction. This means | |
786 | * that the contents of the buffer are not recorded in the transaction | |
787 | * but it is tracked in the AIL as though it was. This allows us | |
788 | * to record logical changes in transactions rather than the physical | |
789 | * changes we make to the buffer without changing writeback ordering | |
790 | * constraints of metadata buffers. | |
791 | */ | |
792 | void | |
793 | xfs_trans_ordered_buf( | |
794 | struct xfs_trans *tp, | |
795 | struct xfs_buf *bp) | |
796 | { | |
797 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
798 | ||
799 | ASSERT(bp->b_transp == tp); | |
800 | ASSERT(bip != NULL); | |
801 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
802 | ||
803 | bip->bli_flags |= XFS_BLI_ORDERED; | |
804 | trace_xfs_buf_item_ordered(bip); | |
805 | } | |
806 | ||
ee1a47ab CH |
807 | /* |
808 | * Set the type of the buffer for log recovery so that it can correctly identify | |
809 | * and hence attach the correct buffer ops to the buffer after replay. | |
810 | */ | |
811 | void | |
812 | xfs_trans_buf_set_type( | |
813 | struct xfs_trans *tp, | |
814 | struct xfs_buf *bp, | |
61fe135c | 815 | enum xfs_blft type) |
ee1a47ab CH |
816 | { |
817 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
818 | ||
d75afeb3 DC |
819 | if (!tp) |
820 | return; | |
821 | ||
ee1a47ab CH |
822 | ASSERT(bp->b_transp == tp); |
823 | ASSERT(bip != NULL); | |
824 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
ee1a47ab | 825 | |
61fe135c | 826 | xfs_blft_to_flags(&bip->__bli_format, type); |
ee1a47ab | 827 | } |
1da177e4 | 828 | |
d75afeb3 DC |
829 | void |
830 | xfs_trans_buf_copy_type( | |
831 | struct xfs_buf *dst_bp, | |
832 | struct xfs_buf *src_bp) | |
833 | { | |
834 | struct xfs_buf_log_item *sbip = src_bp->b_fspriv; | |
835 | struct xfs_buf_log_item *dbip = dst_bp->b_fspriv; | |
61fe135c | 836 | enum xfs_blft type; |
d75afeb3 | 837 | |
61fe135c DC |
838 | type = xfs_blft_from_flags(&sbip->__bli_format); |
839 | xfs_blft_to_flags(&dbip->__bli_format, type); | |
d75afeb3 DC |
840 | } |
841 | ||
1da177e4 LT |
842 | /* |
843 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of | |
844 | * dquots. However, unlike in inode buffer recovery, dquot buffers get | |
845 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). | |
846 | * The only thing that makes dquot buffers different from regular | |
847 | * buffers is that we must not replay dquot bufs when recovering | |
848 | * if a _corresponding_ quotaoff has happened. We also have to distinguish | |
849 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas | |
850 | * can be turned off independently. | |
851 | */ | |
852 | /* ARGSUSED */ | |
853 | void | |
854 | xfs_trans_dquot_buf( | |
855 | xfs_trans_t *tp, | |
856 | xfs_buf_t *bp, | |
857 | uint type) | |
858 | { | |
61fe135c DC |
859 | struct xfs_buf_log_item *bip = bp->b_fspriv; |
860 | ||
c1155410 DC |
861 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
862 | type == XFS_BLF_PDQUOT_BUF || | |
863 | type == XFS_BLF_GDQUOT_BUF); | |
1da177e4 | 864 | |
61fe135c DC |
865 | bip->__bli_format.blf_flags |= type; |
866 | ||
867 | switch (type) { | |
868 | case XFS_BLF_UDQUOT_BUF: | |
869 | type = XFS_BLFT_UDQUOT_BUF; | |
870 | break; | |
871 | case XFS_BLF_PDQUOT_BUF: | |
872 | type = XFS_BLFT_PDQUOT_BUF; | |
873 | break; | |
874 | case XFS_BLF_GDQUOT_BUF: | |
875 | type = XFS_BLFT_GDQUOT_BUF; | |
876 | break; | |
877 | default: | |
878 | type = XFS_BLFT_UNKNOWN_BUF; | |
879 | break; | |
880 | } | |
881 | ||
ee1a47ab | 882 | xfs_trans_buf_set_type(tp, bp, type); |
1da177e4 | 883 | } |