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 | 58 | XFS_BUF_ADDR(blip->bli_buf) == blkno && |
aa0e8833 | 59 | BBTOB(blip->bli_buf->b_length) == len) |
e98c414f | 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) | |
a8acad70 | 145 | flags = 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 | |
bf9d9013 | 168 | ASSERT(bp->b_transp == tp); |
adadbeef | 169 | bip = bp->b_fspriv; |
1da177e4 LT |
170 | ASSERT(bip != NULL); |
171 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
172 | bip->bli_recur++; | |
0b1b213f | 173 | trace_xfs_trans_get_buf_recur(bip); |
1da177e4 LT |
174 | return (bp); |
175 | } | |
176 | ||
177 | /* | |
0cadda1c CH |
178 | * We always specify the XBF_DONT_BLOCK flag within a transaction |
179 | * so that get_buf does not try to push out a delayed write buffer | |
1da177e4 LT |
180 | * which might cause another transaction to take place (if the |
181 | * buffer was delayed alloc). Such recursive transactions can | |
182 | * easily deadlock with our current transaction as well as cause | |
183 | * us to run out of stack space. | |
184 | */ | |
0cadda1c | 185 | bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 LT |
186 | if (bp == NULL) { |
187 | return NULL; | |
188 | } | |
189 | ||
5a52c2a5 | 190 | ASSERT(!bp->b_error); |
1da177e4 | 191 | |
d7e84f41 CH |
192 | _xfs_trans_bjoin(tp, bp, 1); |
193 | trace_xfs_trans_get_buf(bp->b_fspriv); | |
1da177e4 LT |
194 | return (bp); |
195 | } | |
196 | ||
197 | /* | |
198 | * Get and lock the superblock buffer of this file system for the | |
199 | * given transaction. | |
200 | * | |
201 | * We don't need to use incore_match() here, because the superblock | |
202 | * buffer is a private buffer which we keep a pointer to in the | |
203 | * mount structure. | |
204 | */ | |
205 | xfs_buf_t * | |
206 | xfs_trans_getsb(xfs_trans_t *tp, | |
207 | struct xfs_mount *mp, | |
208 | int flags) | |
209 | { | |
210 | xfs_buf_t *bp; | |
211 | xfs_buf_log_item_t *bip; | |
212 | ||
213 | /* | |
214 | * Default to just trying to lock the superblock buffer | |
215 | * if tp is NULL. | |
216 | */ | |
217 | if (tp == NULL) { | |
218 | return (xfs_getsb(mp, flags)); | |
219 | } | |
220 | ||
221 | /* | |
222 | * If the superblock buffer already has this transaction | |
223 | * pointer in its b_fsprivate2 field, then we know we already | |
224 | * have it locked. In this case we just increment the lock | |
225 | * recursion count and return the buffer to the caller. | |
226 | */ | |
227 | bp = mp->m_sb_bp; | |
bf9d9013 | 228 | if (bp->b_transp == tp) { |
adadbeef | 229 | bip = bp->b_fspriv; |
1da177e4 LT |
230 | ASSERT(bip != NULL); |
231 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
232 | bip->bli_recur++; | |
0b1b213f | 233 | trace_xfs_trans_getsb_recur(bip); |
1da177e4 LT |
234 | return (bp); |
235 | } | |
236 | ||
237 | bp = xfs_getsb(mp, flags); | |
d7e84f41 | 238 | if (bp == NULL) |
1da177e4 | 239 | return NULL; |
1da177e4 | 240 | |
d7e84f41 CH |
241 | _xfs_trans_bjoin(tp, bp, 1); |
242 | trace_xfs_trans_getsb(bp->b_fspriv); | |
1da177e4 LT |
243 | return (bp); |
244 | } | |
245 | ||
246 | #ifdef DEBUG | |
247 | xfs_buftarg_t *xfs_error_target; | |
248 | int xfs_do_error; | |
249 | int xfs_req_num; | |
250 | int xfs_error_mod = 33; | |
251 | #endif | |
252 | ||
253 | /* | |
254 | * Get and lock the buffer for the caller if it is not already | |
255 | * locked within the given transaction. If it has not yet been | |
256 | * read in, read it from disk. If it is already locked | |
257 | * within the transaction and already read in, just increment its | |
258 | * lock recursion count and return a pointer to it. | |
259 | * | |
1da177e4 LT |
260 | * If the transaction pointer is NULL, make this just a normal |
261 | * read_buf() call. | |
262 | */ | |
263 | int | |
264 | xfs_trans_read_buf( | |
265 | xfs_mount_t *mp, | |
266 | xfs_trans_t *tp, | |
267 | xfs_buftarg_t *target, | |
268 | xfs_daddr_t blkno, | |
269 | int len, | |
270 | uint flags, | |
271 | xfs_buf_t **bpp) | |
272 | { | |
273 | xfs_buf_t *bp; | |
274 | xfs_buf_log_item_t *bip; | |
275 | int error; | |
276 | ||
7ca790a5 DC |
277 | *bpp = NULL; |
278 | ||
1da177e4 | 279 | if (flags == 0) |
a8acad70 | 280 | flags = XBF_MAPPED; |
1da177e4 LT |
281 | |
282 | /* | |
283 | * Default to a normal get_buf() call if the tp is NULL. | |
284 | */ | |
285 | if (tp == NULL) { | |
0cadda1c | 286 | bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 | 287 | if (!bp) |
0cadda1c | 288 | return (flags & XBF_TRYLOCK) ? |
a3f74ffb | 289 | EAGAIN : XFS_ERROR(ENOMEM); |
1da177e4 | 290 | |
5a52c2a5 CS |
291 | if (bp->b_error) { |
292 | error = bp->b_error; | |
901796af | 293 | xfs_buf_ioerror_alert(bp, __func__); |
7ca790a5 DC |
294 | XFS_BUF_UNDONE(bp); |
295 | xfs_buf_stale(bp); | |
1da177e4 LT |
296 | xfs_buf_relse(bp); |
297 | return error; | |
298 | } | |
299 | #ifdef DEBUG | |
a0f7bfd3 | 300 | if (xfs_do_error) { |
1da177e4 LT |
301 | if (xfs_error_target == target) { |
302 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
303 | xfs_buf_relse(bp); | |
0b932ccc | 304 | xfs_debug(mp, "Returning error!"); |
1da177e4 LT |
305 | return XFS_ERROR(EIO); |
306 | } | |
307 | } | |
308 | } | |
309 | #endif | |
310 | if (XFS_FORCED_SHUTDOWN(mp)) | |
311 | goto shutdown_abort; | |
312 | *bpp = bp; | |
313 | return 0; | |
314 | } | |
315 | ||
316 | /* | |
317 | * If we find the buffer in the cache with this transaction | |
318 | * pointer in its b_fsprivate2 field, then we know we already | |
319 | * have it locked. If it is already read in we just increment | |
320 | * the lock recursion count and return the buffer to the caller. | |
321 | * If the buffer is not yet read in, then we read it in, increment | |
322 | * the lock recursion count, and return it to the caller. | |
323 | */ | |
4a5224d7 | 324 | bp = xfs_trans_buf_item_match(tp, target, blkno, len); |
1da177e4 | 325 | if (bp != NULL) { |
0c842ad4 | 326 | ASSERT(xfs_buf_islocked(bp)); |
bf9d9013 | 327 | ASSERT(bp->b_transp == tp); |
adadbeef | 328 | ASSERT(bp->b_fspriv != NULL); |
5a52c2a5 | 329 | ASSERT(!bp->b_error); |
1da177e4 | 330 | if (!(XFS_BUF_ISDONE(bp))) { |
0b1b213f | 331 | trace_xfs_trans_read_buf_io(bp, _RET_IP_); |
1da177e4 LT |
332 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
333 | XFS_BUF_READ(bp); | |
334 | xfsbdstrat(tp->t_mountp, bp); | |
1a1a3e97 | 335 | error = xfs_buf_iowait(bp); |
d64e31a2 | 336 | if (error) { |
901796af | 337 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 LT |
338 | xfs_buf_relse(bp); |
339 | /* | |
d64e31a2 DC |
340 | * We can gracefully recover from most read |
341 | * errors. Ones we can't are those that happen | |
342 | * after the transaction's already dirty. | |
1da177e4 LT |
343 | */ |
344 | if (tp->t_flags & XFS_TRANS_DIRTY) | |
345 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 346 | SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
347 | return error; |
348 | } | |
349 | } | |
350 | /* | |
351 | * We never locked this buf ourselves, so we shouldn't | |
352 | * brelse it either. Just get out. | |
353 | */ | |
354 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
0b1b213f | 355 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
356 | *bpp = NULL; |
357 | return XFS_ERROR(EIO); | |
358 | } | |
359 | ||
360 | ||
adadbeef | 361 | bip = bp->b_fspriv; |
1da177e4 LT |
362 | bip->bli_recur++; |
363 | ||
364 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 365 | trace_xfs_trans_read_buf_recur(bip); |
1da177e4 LT |
366 | *bpp = bp; |
367 | return 0; | |
368 | } | |
369 | ||
370 | /* | |
0cadda1c CH |
371 | * We always specify the XBF_DONT_BLOCK flag within a transaction |
372 | * so that get_buf does not try to push out a delayed write buffer | |
1da177e4 LT |
373 | * which might cause another transaction to take place (if the |
374 | * buffer was delayed alloc). Such recursive transactions can | |
375 | * easily deadlock with our current transaction as well as cause | |
376 | * us to run out of stack space. | |
377 | */ | |
0cadda1c | 378 | bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 LT |
379 | if (bp == NULL) { |
380 | *bpp = NULL; | |
7401aafd DC |
381 | return (flags & XBF_TRYLOCK) ? |
382 | 0 : XFS_ERROR(ENOMEM); | |
1da177e4 | 383 | } |
5a52c2a5 CS |
384 | if (bp->b_error) { |
385 | error = bp->b_error; | |
c867cb61 | 386 | xfs_buf_stale(bp); |
c867cb61 | 387 | XFS_BUF_DONE(bp); |
901796af | 388 | xfs_buf_ioerror_alert(bp, __func__); |
1da177e4 | 389 | if (tp->t_flags & XFS_TRANS_DIRTY) |
7d04a335 | 390 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
391 | xfs_buf_relse(bp); |
392 | return error; | |
393 | } | |
394 | #ifdef DEBUG | |
395 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { | |
396 | if (xfs_error_target == target) { | |
397 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
398 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 399 | SHUTDOWN_META_IO_ERROR); |
1da177e4 | 400 | xfs_buf_relse(bp); |
0b932ccc | 401 | xfs_debug(mp, "Returning trans error!"); |
1da177e4 LT |
402 | return XFS_ERROR(EIO); |
403 | } | |
404 | } | |
405 | } | |
406 | #endif | |
407 | if (XFS_FORCED_SHUTDOWN(mp)) | |
408 | goto shutdown_abort; | |
409 | ||
d7e84f41 CH |
410 | _xfs_trans_bjoin(tp, bp, 1); |
411 | trace_xfs_trans_read_buf(bp->b_fspriv); | |
1da177e4 | 412 | |
1da177e4 LT |
413 | *bpp = bp; |
414 | return 0; | |
415 | ||
416 | shutdown_abort: | |
0b1b213f | 417 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
418 | xfs_buf_relse(bp); |
419 | *bpp = NULL; | |
420 | return XFS_ERROR(EIO); | |
421 | } | |
422 | ||
423 | ||
424 | /* | |
425 | * Release the buffer bp which was previously acquired with one of the | |
426 | * xfs_trans_... buffer allocation routines if the buffer has not | |
427 | * been modified within this transaction. If the buffer is modified | |
428 | * within this transaction, do decrement the recursion count but do | |
429 | * not release the buffer even if the count goes to 0. If the buffer is not | |
430 | * modified within the transaction, decrement the recursion count and | |
431 | * release the buffer if the recursion count goes to 0. | |
432 | * | |
433 | * If the buffer is to be released and it was not modified before | |
434 | * this transaction began, then free the buf_log_item associated with it. | |
435 | * | |
436 | * If the transaction pointer is NULL, make this just a normal | |
437 | * brelse() call. | |
438 | */ | |
439 | void | |
440 | xfs_trans_brelse(xfs_trans_t *tp, | |
441 | xfs_buf_t *bp) | |
442 | { | |
443 | xfs_buf_log_item_t *bip; | |
1da177e4 LT |
444 | |
445 | /* | |
446 | * Default to a normal brelse() call if the tp is NULL. | |
447 | */ | |
448 | if (tp == NULL) { | |
bf9d9013 | 449 | ASSERT(bp->b_transp == NULL); |
1da177e4 LT |
450 | xfs_buf_relse(bp); |
451 | return; | |
452 | } | |
453 | ||
bf9d9013 | 454 | ASSERT(bp->b_transp == tp); |
adadbeef | 455 | bip = bp->b_fspriv; |
1da177e4 LT |
456 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
457 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 458 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
459 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
460 | ||
0b1b213f CH |
461 | trace_xfs_trans_brelse(bip); |
462 | ||
1da177e4 LT |
463 | /* |
464 | * If the release is just for a recursive lock, | |
465 | * then decrement the count and return. | |
466 | */ | |
467 | if (bip->bli_recur > 0) { | |
468 | bip->bli_recur--; | |
1da177e4 LT |
469 | return; |
470 | } | |
471 | ||
472 | /* | |
473 | * If the buffer is dirty within this transaction, we can't | |
474 | * release it until we commit. | |
475 | */ | |
e98c414f | 476 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
1da177e4 | 477 | return; |
1da177e4 LT |
478 | |
479 | /* | |
480 | * If the buffer has been invalidated, then we can't release | |
481 | * it until the transaction commits to disk unless it is re-dirtied | |
482 | * as part of this transaction. This prevents us from pulling | |
483 | * the item from the AIL before we should. | |
484 | */ | |
0b1b213f | 485 | if (bip->bli_flags & XFS_BLI_STALE) |
1da177e4 | 486 | return; |
1da177e4 LT |
487 | |
488 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); | |
1da177e4 LT |
489 | |
490 | /* | |
491 | * Free up the log item descriptor tracking the released item. | |
492 | */ | |
e98c414f | 493 | xfs_trans_del_item(&bip->bli_item); |
1da177e4 LT |
494 | |
495 | /* | |
496 | * Clear the hold flag in the buf log item if it is set. | |
497 | * We wouldn't want the next user of the buffer to | |
498 | * get confused. | |
499 | */ | |
500 | if (bip->bli_flags & XFS_BLI_HOLD) { | |
501 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
502 | } | |
503 | ||
504 | /* | |
505 | * Drop our reference to the buf log item. | |
506 | */ | |
507 | atomic_dec(&bip->bli_refcount); | |
508 | ||
509 | /* | |
510 | * If the buf item is not tracking data in the log, then | |
511 | * we must free it before releasing the buffer back to the | |
512 | * free pool. Before releasing the buffer to the free pool, | |
513 | * clear the transaction pointer in b_fsprivate2 to dissolve | |
514 | * its relation to this transaction. | |
515 | */ | |
516 | if (!xfs_buf_item_dirty(bip)) { | |
517 | /*** | |
518 | ASSERT(bp->b_pincount == 0); | |
519 | ***/ | |
520 | ASSERT(atomic_read(&bip->bli_refcount) == 0); | |
521 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); | |
522 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); | |
523 | xfs_buf_item_relse(bp); | |
1da177e4 LT |
524 | } |
525 | ||
5b03ff1b | 526 | bp->b_transp = NULL; |
1da177e4 | 527 | xfs_buf_relse(bp); |
1da177e4 LT |
528 | } |
529 | ||
1da177e4 LT |
530 | /* |
531 | * Mark the buffer as not needing to be unlocked when the buf item's | |
532 | * IOP_UNLOCK() routine is called. The buffer must already be locked | |
533 | * and associated with the given transaction. | |
534 | */ | |
535 | /* ARGSUSED */ | |
536 | void | |
537 | xfs_trans_bhold(xfs_trans_t *tp, | |
538 | xfs_buf_t *bp) | |
539 | { | |
adadbeef | 540 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 541 | |
bf9d9013 | 542 | ASSERT(bp->b_transp == tp); |
adadbeef | 543 | ASSERT(bip != NULL); |
1da177e4 | 544 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 545 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 | 546 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
adadbeef | 547 | |
1da177e4 | 548 | bip->bli_flags |= XFS_BLI_HOLD; |
0b1b213f | 549 | trace_xfs_trans_bhold(bip); |
1da177e4 LT |
550 | } |
551 | ||
efa092f3 TS |
552 | /* |
553 | * Cancel the previous buffer hold request made on this buffer | |
554 | * for this transaction. | |
555 | */ | |
556 | void | |
557 | xfs_trans_bhold_release(xfs_trans_t *tp, | |
558 | xfs_buf_t *bp) | |
559 | { | |
adadbeef | 560 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
efa092f3 | 561 | |
bf9d9013 | 562 | ASSERT(bp->b_transp == tp); |
adadbeef | 563 | ASSERT(bip != NULL); |
efa092f3 | 564 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
c1155410 | 565 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
efa092f3 TS |
566 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
567 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); | |
0b1b213f | 568 | |
adadbeef | 569 | bip->bli_flags &= ~XFS_BLI_HOLD; |
0b1b213f | 570 | trace_xfs_trans_bhold_release(bip); |
efa092f3 TS |
571 | } |
572 | ||
1da177e4 LT |
573 | /* |
574 | * This is called to mark bytes first through last inclusive of the given | |
575 | * buffer as needing to be logged when the transaction is committed. | |
576 | * The buffer must already be associated with the given transaction. | |
577 | * | |
578 | * First and last are numbers relative to the beginning of this buffer, | |
579 | * so the first byte in the buffer is numbered 0 regardless of the | |
580 | * value of b_blkno. | |
581 | */ | |
582 | void | |
583 | xfs_trans_log_buf(xfs_trans_t *tp, | |
584 | xfs_buf_t *bp, | |
585 | uint first, | |
586 | uint last) | |
587 | { | |
adadbeef | 588 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 589 | |
bf9d9013 | 590 | ASSERT(bp->b_transp == tp); |
adadbeef | 591 | ASSERT(bip != NULL); |
aa0e8833 | 592 | ASSERT(first <= last && last < BBTOB(bp->b_length)); |
cb669ca5 CH |
593 | ASSERT(bp->b_iodone == NULL || |
594 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1da177e4 LT |
595 | |
596 | /* | |
597 | * Mark the buffer as needing to be written out eventually, | |
598 | * and set its iodone function to remove the buffer's buf log | |
599 | * item from the AIL and free it when the buffer is flushed | |
600 | * to disk. See xfs_buf_attach_iodone() for more details | |
601 | * on li_cb and xfs_buf_iodone_callbacks(). | |
602 | * If we end up aborting this transaction, we trap this buffer | |
603 | * inside the b_bdstrat callback so that this won't get written to | |
604 | * disk. | |
605 | */ | |
1da177e4 LT |
606 | XFS_BUF_DONE(bp); |
607 | ||
1da177e4 | 608 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
cb669ca5 | 609 | bp->b_iodone = xfs_buf_iodone_callbacks; |
ca30b2a7 | 610 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 | 611 | |
0b1b213f CH |
612 | trace_xfs_trans_log_buf(bip); |
613 | ||
1da177e4 LT |
614 | /* |
615 | * If we invalidated the buffer within this transaction, then | |
616 | * cancel the invalidation now that we're dirtying the buffer | |
617 | * again. There are no races with the code in xfs_buf_item_unpin(), | |
618 | * because we have a reference to the buffer this entire time. | |
619 | */ | |
620 | if (bip->bli_flags & XFS_BLI_STALE) { | |
1da177e4 LT |
621 | bip->bli_flags &= ~XFS_BLI_STALE; |
622 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
623 | XFS_BUF_UNSTALE(bp); | |
c1155410 | 624 | bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
1da177e4 LT |
625 | } |
626 | ||
1da177e4 | 627 | tp->t_flags |= XFS_TRANS_DIRTY; |
e98c414f | 628 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 LT |
629 | bip->bli_flags |= XFS_BLI_LOGGED; |
630 | xfs_buf_item_log(bip, first, last); | |
1da177e4 LT |
631 | } |
632 | ||
633 | ||
634 | /* | |
43ff2122 CH |
635 | * Invalidate a buffer that is being used within a transaction. |
636 | * | |
637 | * Typically this is because the blocks in the buffer are being freed, so we | |
638 | * need to prevent it from being written out when we're done. Allowing it | |
639 | * to be written again might overwrite data in the free blocks if they are | |
640 | * reallocated to a file. | |
1da177e4 | 641 | * |
43ff2122 CH |
642 | * We prevent the buffer from being written out by marking it stale. We can't |
643 | * get rid of the buf log item at this point because the buffer may still be | |
644 | * pinned by another transaction. If that is the case, then we'll wait until | |
645 | * the buffer is committed to disk for the last time (we can tell by the ref | |
646 | * count) and free it in xfs_buf_item_unpin(). Until that happens we will | |
647 | * keep the buffer locked so that the buffer and buf log item are not reused. | |
648 | * | |
649 | * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log | |
650 | * the buf item. This will be used at recovery time to determine that copies | |
651 | * of the buffer in the log before this should not be replayed. | |
652 | * | |
653 | * We mark the item descriptor and the transaction dirty so that we'll hold | |
654 | * the buffer until after the commit. | |
655 | * | |
656 | * Since we're invalidating the buffer, we also clear the state about which | |
657 | * parts of the buffer have been logged. We also clear the flag indicating | |
658 | * that this is an inode buffer since the data in the buffer will no longer | |
659 | * be valid. | |
660 | * | |
661 | * We set the stale bit in the buffer as well since we're getting rid of it. | |
1da177e4 LT |
662 | */ |
663 | void | |
664 | xfs_trans_binval( | |
665 | xfs_trans_t *tp, | |
666 | xfs_buf_t *bp) | |
667 | { | |
adadbeef | 668 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 669 | |
bf9d9013 | 670 | ASSERT(bp->b_transp == tp); |
adadbeef | 671 | ASSERT(bip != NULL); |
1da177e4 LT |
672 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
673 | ||
0b1b213f CH |
674 | trace_xfs_trans_binval(bip); |
675 | ||
1da177e4 LT |
676 | if (bip->bli_flags & XFS_BLI_STALE) { |
677 | /* | |
678 | * If the buffer is already invalidated, then | |
679 | * just return. | |
680 | */ | |
1da177e4 LT |
681 | ASSERT(XFS_BUF_ISSTALE(bp)); |
682 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); | |
c1155410 DC |
683 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
684 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); | |
e98c414f | 685 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
1da177e4 | 686 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
1da177e4 LT |
687 | return; |
688 | } | |
689 | ||
c867cb61 | 690 | xfs_buf_stale(bp); |
43ff2122 | 691 | |
1da177e4 | 692 | bip->bli_flags |= XFS_BLI_STALE; |
ccf7c23f | 693 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
c1155410 DC |
694 | bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
695 | bip->bli_format.blf_flags |= XFS_BLF_CANCEL; | |
1da177e4 LT |
696 | memset((char *)(bip->bli_format.blf_data_map), 0, |
697 | (bip->bli_format.blf_map_size * sizeof(uint))); | |
e98c414f | 698 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 | 699 | tp->t_flags |= XFS_TRANS_DIRTY; |
1da177e4 LT |
700 | } |
701 | ||
702 | /* | |
ccf7c23f DC |
703 | * This call is used to indicate that the buffer contains on-disk inodes which |
704 | * must be handled specially during recovery. They require special handling | |
705 | * because only the di_next_unlinked from the inodes in the buffer should be | |
706 | * recovered. The rest of the data in the buffer is logged via the inodes | |
707 | * themselves. | |
1da177e4 | 708 | * |
ccf7c23f DC |
709 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
710 | * transferred to the buffer's log format structure so that we'll know what to | |
711 | * do at recovery time. | |
1da177e4 | 712 | */ |
1da177e4 LT |
713 | void |
714 | xfs_trans_inode_buf( | |
715 | xfs_trans_t *tp, | |
716 | xfs_buf_t *bp) | |
717 | { | |
adadbeef | 718 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 719 | |
bf9d9013 | 720 | ASSERT(bp->b_transp == tp); |
adadbeef | 721 | ASSERT(bip != NULL); |
1da177e4 LT |
722 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
723 | ||
ccf7c23f | 724 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
1da177e4 LT |
725 | } |
726 | ||
727 | /* | |
728 | * This call is used to indicate that the buffer is going to | |
729 | * be staled and was an inode buffer. This means it gets | |
730 | * special processing during unpin - where any inodes | |
731 | * associated with the buffer should be removed from ail. | |
732 | * There is also special processing during recovery, | |
733 | * any replay of the inodes in the buffer needs to be | |
734 | * prevented as the buffer may have been reused. | |
735 | */ | |
736 | void | |
737 | xfs_trans_stale_inode_buf( | |
738 | xfs_trans_t *tp, | |
739 | xfs_buf_t *bp) | |
740 | { | |
adadbeef | 741 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 742 | |
bf9d9013 | 743 | ASSERT(bp->b_transp == tp); |
adadbeef | 744 | ASSERT(bip != NULL); |
1da177e4 LT |
745 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
746 | ||
747 | bip->bli_flags |= XFS_BLI_STALE_INODE; | |
ca30b2a7 | 748 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 LT |
749 | } |
750 | ||
1da177e4 LT |
751 | /* |
752 | * Mark the buffer as being one which contains newly allocated | |
753 | * inodes. We need to make sure that even if this buffer is | |
754 | * relogged as an 'inode buf' we still recover all of the inode | |
755 | * images in the face of a crash. This works in coordination with | |
756 | * xfs_buf_item_committed() to ensure that the buffer remains in the | |
757 | * AIL at its original location even after it has been relogged. | |
758 | */ | |
759 | /* ARGSUSED */ | |
760 | void | |
761 | xfs_trans_inode_alloc_buf( | |
762 | xfs_trans_t *tp, | |
763 | xfs_buf_t *bp) | |
764 | { | |
adadbeef | 765 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 766 | |
bf9d9013 | 767 | ASSERT(bp->b_transp == tp); |
adadbeef | 768 | ASSERT(bip != NULL); |
1da177e4 LT |
769 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
770 | ||
771 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; | |
772 | } | |
773 | ||
774 | ||
775 | /* | |
776 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of | |
777 | * dquots. However, unlike in inode buffer recovery, dquot buffers get | |
778 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). | |
779 | * The only thing that makes dquot buffers different from regular | |
780 | * buffers is that we must not replay dquot bufs when recovering | |
781 | * if a _corresponding_ quotaoff has happened. We also have to distinguish | |
782 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas | |
783 | * can be turned off independently. | |
784 | */ | |
785 | /* ARGSUSED */ | |
786 | void | |
787 | xfs_trans_dquot_buf( | |
788 | xfs_trans_t *tp, | |
789 | xfs_buf_t *bp, | |
790 | uint type) | |
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); |
c1155410 DC |
796 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
797 | type == XFS_BLF_PDQUOT_BUF || | |
798 | type == XFS_BLF_GDQUOT_BUF); | |
1da177e4 LT |
799 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
800 | ||
801 | bip->bli_format.blf_flags |= type; | |
802 | } |