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 && | |
57 | XFS_BUF_TARGET(blip->bli_buf) == target && | |
58 | XFS_BUF_ADDR(blip->bli_buf) == blkno && | |
59 | XFS_BUF_COUNT(blip->bli_buf) == len) | |
60 | return blip->bli_buf; | |
4a5224d7 CH |
61 | } |
62 | ||
63 | return NULL; | |
64 | } | |
1da177e4 | 65 | |
d7e84f41 CH |
66 | /* |
67 | * Add the locked buffer to the transaction. | |
68 | * | |
69 | * The buffer must be locked, and it cannot be associated with any | |
70 | * transaction. | |
71 | * | |
72 | * If the buffer does not yet have a buf log item associated with it, | |
73 | * then allocate one for it. Then add the buf item to the transaction. | |
74 | */ | |
75 | STATIC void | |
76 | _xfs_trans_bjoin( | |
77 | struct xfs_trans *tp, | |
78 | struct xfs_buf *bp, | |
79 | int reset_recur) | |
80 | { | |
81 | struct xfs_buf_log_item *bip; | |
82 | ||
83 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
84 | ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL); | |
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); | |
92 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
93 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 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 | */ | |
113 | XFS_BUF_SET_FSPRIVATE2(bp, tp); | |
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 | */ | |
135 | xfs_buf_t * | |
136 | xfs_trans_get_buf(xfs_trans_t *tp, | |
137 | xfs_buftarg_t *target_dev, | |
138 | xfs_daddr_t blkno, | |
139 | int len, | |
140 | uint flags) | |
141 | { | |
142 | xfs_buf_t *bp; | |
143 | xfs_buf_log_item_t *bip; | |
144 | ||
145 | if (flags == 0) | |
0cadda1c | 146 | flags = XBF_LOCK | XBF_MAPPED; |
1da177e4 LT |
147 | |
148 | /* | |
149 | * Default to a normal get_buf() call if the tp is NULL. | |
150 | */ | |
6ad112bf | 151 | if (tp == NULL) |
0cadda1c CH |
152 | return xfs_buf_get(target_dev, blkno, len, |
153 | flags | XBF_DONT_BLOCK); | |
1da177e4 LT |
154 | |
155 | /* | |
156 | * If we find the buffer in the cache with this transaction | |
157 | * pointer in its b_fsprivate2 field, then we know we already | |
158 | * have it locked. In this case we just increment the lock | |
159 | * recursion count and return the buffer to the caller. | |
160 | */ | |
4a5224d7 | 161 | bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len); |
1da177e4 LT |
162 | if (bp != NULL) { |
163 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
0b1b213f | 164 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) |
1da177e4 | 165 | XFS_BUF_SUPER_STALE(bp); |
0b1b213f | 166 | |
1da177e4 LT |
167 | /* |
168 | * If the buffer is stale then it was binval'ed | |
169 | * since last read. This doesn't matter since the | |
170 | * caller isn't allowed to use the data anyway. | |
171 | */ | |
0b1b213f | 172 | else if (XFS_BUF_ISSTALE(bp)) |
1da177e4 | 173 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
0b1b213f | 174 | |
1da177e4 LT |
175 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
176 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
177 | ASSERT(bip != NULL); | |
178 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
179 | bip->bli_recur++; | |
0b1b213f | 180 | trace_xfs_trans_get_buf_recur(bip); |
1da177e4 LT |
181 | return (bp); |
182 | } | |
183 | ||
184 | /* | |
0cadda1c CH |
185 | * We always specify the XBF_DONT_BLOCK flag within a transaction |
186 | * so that get_buf does not try to push out a delayed write buffer | |
1da177e4 LT |
187 | * which might cause another transaction to take place (if the |
188 | * buffer was delayed alloc). Such recursive transactions can | |
189 | * easily deadlock with our current transaction as well as cause | |
190 | * us to run out of stack space. | |
191 | */ | |
0cadda1c | 192 | bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 LT |
193 | if (bp == NULL) { |
194 | return NULL; | |
195 | } | |
196 | ||
197 | ASSERT(!XFS_BUF_GETERROR(bp)); | |
198 | ||
d7e84f41 CH |
199 | _xfs_trans_bjoin(tp, bp, 1); |
200 | trace_xfs_trans_get_buf(bp->b_fspriv); | |
1da177e4 LT |
201 | return (bp); |
202 | } | |
203 | ||
204 | /* | |
205 | * Get and lock the superblock buffer of this file system for the | |
206 | * given transaction. | |
207 | * | |
208 | * We don't need to use incore_match() here, because the superblock | |
209 | * buffer is a private buffer which we keep a pointer to in the | |
210 | * mount structure. | |
211 | */ | |
212 | xfs_buf_t * | |
213 | xfs_trans_getsb(xfs_trans_t *tp, | |
214 | struct xfs_mount *mp, | |
215 | int flags) | |
216 | { | |
217 | xfs_buf_t *bp; | |
218 | xfs_buf_log_item_t *bip; | |
219 | ||
220 | /* | |
221 | * Default to just trying to lock the superblock buffer | |
222 | * if tp is NULL. | |
223 | */ | |
224 | if (tp == NULL) { | |
225 | return (xfs_getsb(mp, flags)); | |
226 | } | |
227 | ||
228 | /* | |
229 | * If the superblock buffer already has this transaction | |
230 | * pointer in its b_fsprivate2 field, then we know we already | |
231 | * have it locked. In this case we just increment the lock | |
232 | * recursion count and return the buffer to the caller. | |
233 | */ | |
234 | bp = mp->m_sb_bp; | |
235 | if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) { | |
236 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); | |
237 | ASSERT(bip != NULL); | |
238 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
239 | bip->bli_recur++; | |
0b1b213f | 240 | trace_xfs_trans_getsb_recur(bip); |
1da177e4 LT |
241 | return (bp); |
242 | } | |
243 | ||
244 | bp = xfs_getsb(mp, flags); | |
d7e84f41 | 245 | if (bp == NULL) |
1da177e4 | 246 | return NULL; |
1da177e4 | 247 | |
d7e84f41 CH |
248 | _xfs_trans_bjoin(tp, bp, 1); |
249 | trace_xfs_trans_getsb(bp->b_fspriv); | |
1da177e4 LT |
250 | return (bp); |
251 | } | |
252 | ||
253 | #ifdef DEBUG | |
254 | xfs_buftarg_t *xfs_error_target; | |
255 | int xfs_do_error; | |
256 | int xfs_req_num; | |
257 | int xfs_error_mod = 33; | |
258 | #endif | |
259 | ||
260 | /* | |
261 | * Get and lock the buffer for the caller if it is not already | |
262 | * locked within the given transaction. If it has not yet been | |
263 | * read in, read it from disk. If it is already locked | |
264 | * within the transaction and already read in, just increment its | |
265 | * lock recursion count and return a pointer to it. | |
266 | * | |
1da177e4 LT |
267 | * If the transaction pointer is NULL, make this just a normal |
268 | * read_buf() call. | |
269 | */ | |
270 | int | |
271 | xfs_trans_read_buf( | |
272 | xfs_mount_t *mp, | |
273 | xfs_trans_t *tp, | |
274 | xfs_buftarg_t *target, | |
275 | xfs_daddr_t blkno, | |
276 | int len, | |
277 | uint flags, | |
278 | xfs_buf_t **bpp) | |
279 | { | |
280 | xfs_buf_t *bp; | |
281 | xfs_buf_log_item_t *bip; | |
282 | int error; | |
283 | ||
284 | if (flags == 0) | |
0cadda1c | 285 | flags = XBF_LOCK | XBF_MAPPED; |
1da177e4 LT |
286 | |
287 | /* | |
288 | * Default to a normal get_buf() call if the tp is NULL. | |
289 | */ | |
290 | if (tp == NULL) { | |
0cadda1c | 291 | bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 | 292 | if (!bp) |
0cadda1c | 293 | return (flags & XBF_TRYLOCK) ? |
a3f74ffb | 294 | EAGAIN : XFS_ERROR(ENOMEM); |
1da177e4 | 295 | |
a0f7bfd3 | 296 | if (XFS_BUF_GETERROR(bp) != 0) { |
1da177e4 LT |
297 | xfs_ioerror_alert("xfs_trans_read_buf", mp, |
298 | bp, blkno); | |
299 | error = XFS_BUF_GETERROR(bp); | |
300 | xfs_buf_relse(bp); | |
301 | return error; | |
302 | } | |
303 | #ifdef DEBUG | |
a0f7bfd3 | 304 | if (xfs_do_error) { |
1da177e4 LT |
305 | if (xfs_error_target == target) { |
306 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
307 | xfs_buf_relse(bp); | |
b6574520 | 308 | cmn_err(CE_DEBUG, "Returning error!\n"); |
1da177e4 LT |
309 | return XFS_ERROR(EIO); |
310 | } | |
311 | } | |
312 | } | |
313 | #endif | |
314 | if (XFS_FORCED_SHUTDOWN(mp)) | |
315 | goto shutdown_abort; | |
316 | *bpp = bp; | |
317 | return 0; | |
318 | } | |
319 | ||
320 | /* | |
321 | * If we find the buffer in the cache with this transaction | |
322 | * pointer in its b_fsprivate2 field, then we know we already | |
323 | * have it locked. If it is already read in we just increment | |
324 | * the lock recursion count and return the buffer to the caller. | |
325 | * If the buffer is not yet read in, then we read it in, increment | |
326 | * the lock recursion count, and return it to the caller. | |
327 | */ | |
4a5224d7 | 328 | bp = xfs_trans_buf_item_match(tp, target, blkno, len); |
1da177e4 LT |
329 | if (bp != NULL) { |
330 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
331 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
332 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
333 | ASSERT((XFS_BUF_ISERROR(bp)) == 0); | |
334 | if (!(XFS_BUF_ISDONE(bp))) { | |
0b1b213f | 335 | trace_xfs_trans_read_buf_io(bp, _RET_IP_); |
1da177e4 LT |
336 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
337 | XFS_BUF_READ(bp); | |
338 | xfsbdstrat(tp->t_mountp, bp); | |
1a1a3e97 | 339 | error = xfs_buf_iowait(bp); |
d64e31a2 | 340 | if (error) { |
1da177e4 LT |
341 | xfs_ioerror_alert("xfs_trans_read_buf", mp, |
342 | bp, blkno); | |
1da177e4 LT |
343 | xfs_buf_relse(bp); |
344 | /* | |
d64e31a2 DC |
345 | * We can gracefully recover from most read |
346 | * errors. Ones we can't are those that happen | |
347 | * after the transaction's already dirty. | |
1da177e4 LT |
348 | */ |
349 | if (tp->t_flags & XFS_TRANS_DIRTY) | |
350 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 351 | SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
352 | return error; |
353 | } | |
354 | } | |
355 | /* | |
356 | * We never locked this buf ourselves, so we shouldn't | |
357 | * brelse it either. Just get out. | |
358 | */ | |
359 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
0b1b213f | 360 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
361 | *bpp = NULL; |
362 | return XFS_ERROR(EIO); | |
363 | } | |
364 | ||
365 | ||
366 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); | |
367 | bip->bli_recur++; | |
368 | ||
369 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 370 | trace_xfs_trans_read_buf_recur(bip); |
1da177e4 LT |
371 | *bpp = bp; |
372 | return 0; | |
373 | } | |
374 | ||
375 | /* | |
0cadda1c CH |
376 | * We always specify the XBF_DONT_BLOCK flag within a transaction |
377 | * so that get_buf does not try to push out a delayed write buffer | |
1da177e4 LT |
378 | * which might cause another transaction to take place (if the |
379 | * buffer was delayed alloc). Such recursive transactions can | |
380 | * easily deadlock with our current transaction as well as cause | |
381 | * us to run out of stack space. | |
382 | */ | |
0cadda1c | 383 | bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK); |
1da177e4 LT |
384 | if (bp == NULL) { |
385 | *bpp = NULL; | |
386 | return 0; | |
387 | } | |
388 | if (XFS_BUF_GETERROR(bp) != 0) { | |
389 | XFS_BUF_SUPER_STALE(bp); | |
1da177e4 LT |
390 | error = XFS_BUF_GETERROR(bp); |
391 | ||
392 | xfs_ioerror_alert("xfs_trans_read_buf", mp, | |
393 | bp, blkno); | |
394 | if (tp->t_flags & XFS_TRANS_DIRTY) | |
7d04a335 | 395 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
396 | xfs_buf_relse(bp); |
397 | return error; | |
398 | } | |
399 | #ifdef DEBUG | |
400 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { | |
401 | if (xfs_error_target == target) { | |
402 | if (((xfs_req_num++) % xfs_error_mod) == 0) { | |
403 | xfs_force_shutdown(tp->t_mountp, | |
7d04a335 | 404 | SHUTDOWN_META_IO_ERROR); |
1da177e4 | 405 | xfs_buf_relse(bp); |
b6574520 | 406 | cmn_err(CE_DEBUG, "Returning trans error!\n"); |
1da177e4 LT |
407 | return XFS_ERROR(EIO); |
408 | } | |
409 | } | |
410 | } | |
411 | #endif | |
412 | if (XFS_FORCED_SHUTDOWN(mp)) | |
413 | goto shutdown_abort; | |
414 | ||
d7e84f41 CH |
415 | _xfs_trans_bjoin(tp, bp, 1); |
416 | trace_xfs_trans_read_buf(bp->b_fspriv); | |
1da177e4 | 417 | |
1da177e4 LT |
418 | *bpp = bp; |
419 | return 0; | |
420 | ||
421 | shutdown_abort: | |
422 | /* | |
423 | * the theory here is that buffer is good but we're | |
424 | * bailing out because the filesystem is being forcibly | |
425 | * shut down. So we should leave the b_flags alone since | |
426 | * the buffer's not staled and just get out. | |
427 | */ | |
428 | #if defined(DEBUG) | |
429 | if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp)) | |
430 | cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp); | |
431 | #endif | |
0cadda1c CH |
432 | ASSERT((XFS_BUF_BFLAGS(bp) & (XBF_STALE|XBF_DELWRI)) != |
433 | (XBF_STALE|XBF_DELWRI)); | |
1da177e4 | 434 | |
0b1b213f | 435 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
1da177e4 LT |
436 | xfs_buf_relse(bp); |
437 | *bpp = NULL; | |
438 | return XFS_ERROR(EIO); | |
439 | } | |
440 | ||
441 | ||
442 | /* | |
443 | * Release the buffer bp which was previously acquired with one of the | |
444 | * xfs_trans_... buffer allocation routines if the buffer has not | |
445 | * been modified within this transaction. If the buffer is modified | |
446 | * within this transaction, do decrement the recursion count but do | |
447 | * not release the buffer even if the count goes to 0. If the buffer is not | |
448 | * modified within the transaction, decrement the recursion count and | |
449 | * release the buffer if the recursion count goes to 0. | |
450 | * | |
451 | * If the buffer is to be released and it was not modified before | |
452 | * this transaction began, then free the buf_log_item associated with it. | |
453 | * | |
454 | * If the transaction pointer is NULL, make this just a normal | |
455 | * brelse() call. | |
456 | */ | |
457 | void | |
458 | xfs_trans_brelse(xfs_trans_t *tp, | |
459 | xfs_buf_t *bp) | |
460 | { | |
461 | xfs_buf_log_item_t *bip; | |
462 | xfs_log_item_t *lip; | |
1da177e4 LT |
463 | |
464 | /* | |
465 | * Default to a normal brelse() call if the tp is NULL. | |
466 | */ | |
467 | if (tp == NULL) { | |
468 | ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL); | |
469 | /* | |
470 | * If there's a buf log item attached to the buffer, | |
471 | * then let the AIL know that the buffer is being | |
472 | * unlocked. | |
473 | */ | |
474 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { | |
475 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
476 | if (lip->li_type == XFS_LI_BUF) { | |
477 | bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*); | |
783a2f65 DC |
478 | xfs_trans_unlocked_item(bip->bli_item.li_ailp, |
479 | lip); | |
1da177e4 LT |
480 | } |
481 | } | |
482 | xfs_buf_relse(bp); | |
483 | return; | |
484 | } | |
485 | ||
486 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
487 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
488 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); | |
489 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 490 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
491 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
492 | ||
0b1b213f CH |
493 | trace_xfs_trans_brelse(bip); |
494 | ||
1da177e4 LT |
495 | /* |
496 | * If the release is just for a recursive lock, | |
497 | * then decrement the count and return. | |
498 | */ | |
499 | if (bip->bli_recur > 0) { | |
500 | bip->bli_recur--; | |
1da177e4 LT |
501 | return; |
502 | } | |
503 | ||
504 | /* | |
505 | * If the buffer is dirty within this transaction, we can't | |
506 | * release it until we commit. | |
507 | */ | |
e98c414f | 508 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
1da177e4 | 509 | return; |
1da177e4 LT |
510 | |
511 | /* | |
512 | * If the buffer has been invalidated, then we can't release | |
513 | * it until the transaction commits to disk unless it is re-dirtied | |
514 | * as part of this transaction. This prevents us from pulling | |
515 | * the item from the AIL before we should. | |
516 | */ | |
0b1b213f | 517 | if (bip->bli_flags & XFS_BLI_STALE) |
1da177e4 | 518 | return; |
1da177e4 LT |
519 | |
520 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); | |
1da177e4 LT |
521 | |
522 | /* | |
523 | * Free up the log item descriptor tracking the released item. | |
524 | */ | |
e98c414f | 525 | xfs_trans_del_item(&bip->bli_item); |
1da177e4 LT |
526 | |
527 | /* | |
528 | * Clear the hold flag in the buf log item if it is set. | |
529 | * We wouldn't want the next user of the buffer to | |
530 | * get confused. | |
531 | */ | |
532 | if (bip->bli_flags & XFS_BLI_HOLD) { | |
533 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
534 | } | |
535 | ||
536 | /* | |
537 | * Drop our reference to the buf log item. | |
538 | */ | |
539 | atomic_dec(&bip->bli_refcount); | |
540 | ||
541 | /* | |
542 | * If the buf item is not tracking data in the log, then | |
543 | * we must free it before releasing the buffer back to the | |
544 | * free pool. Before releasing the buffer to the free pool, | |
545 | * clear the transaction pointer in b_fsprivate2 to dissolve | |
546 | * its relation to this transaction. | |
547 | */ | |
548 | if (!xfs_buf_item_dirty(bip)) { | |
549 | /*** | |
550 | ASSERT(bp->b_pincount == 0); | |
551 | ***/ | |
552 | ASSERT(atomic_read(&bip->bli_refcount) == 0); | |
553 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); | |
554 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); | |
555 | xfs_buf_item_relse(bp); | |
556 | bip = NULL; | |
557 | } | |
558 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); | |
559 | ||
560 | /* | |
561 | * If we've still got a buf log item on the buffer, then | |
562 | * tell the AIL that the buffer is being unlocked. | |
563 | */ | |
564 | if (bip != NULL) { | |
783a2f65 | 565 | xfs_trans_unlocked_item(bip->bli_item.li_ailp, |
1da177e4 LT |
566 | (xfs_log_item_t*)bip); |
567 | } | |
568 | ||
569 | xfs_buf_relse(bp); | |
570 | return; | |
571 | } | |
572 | ||
1da177e4 LT |
573 | /* |
574 | * Mark the buffer as not needing to be unlocked when the buf item's | |
575 | * IOP_UNLOCK() routine is called. The buffer must already be locked | |
576 | * and associated with the given transaction. | |
577 | */ | |
578 | /* ARGSUSED */ | |
579 | void | |
580 | xfs_trans_bhold(xfs_trans_t *tp, | |
581 | xfs_buf_t *bp) | |
582 | { | |
583 | xfs_buf_log_item_t *bip; | |
584 | ||
585 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
586 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
587 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
588 | ||
589 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
590 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 591 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
592 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
593 | bip->bli_flags |= XFS_BLI_HOLD; | |
0b1b213f | 594 | trace_xfs_trans_bhold(bip); |
1da177e4 LT |
595 | } |
596 | ||
efa092f3 TS |
597 | /* |
598 | * Cancel the previous buffer hold request made on this buffer | |
599 | * for this transaction. | |
600 | */ | |
601 | void | |
602 | xfs_trans_bhold_release(xfs_trans_t *tp, | |
603 | xfs_buf_t *bp) | |
604 | { | |
605 | xfs_buf_log_item_t *bip; | |
606 | ||
607 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
608 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
609 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
610 | ||
611 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
612 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
c1155410 | 613 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL)); |
efa092f3 TS |
614 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
615 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); | |
616 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
0b1b213f CH |
617 | |
618 | trace_xfs_trans_bhold_release(bip); | |
efa092f3 TS |
619 | } |
620 | ||
1da177e4 LT |
621 | /* |
622 | * This is called to mark bytes first through last inclusive of the given | |
623 | * buffer as needing to be logged when the transaction is committed. | |
624 | * The buffer must already be associated with the given transaction. | |
625 | * | |
626 | * First and last are numbers relative to the beginning of this buffer, | |
627 | * so the first byte in the buffer is numbered 0 regardless of the | |
628 | * value of b_blkno. | |
629 | */ | |
630 | void | |
631 | xfs_trans_log_buf(xfs_trans_t *tp, | |
632 | xfs_buf_t *bp, | |
633 | uint first, | |
634 | uint last) | |
635 | { | |
636 | xfs_buf_log_item_t *bip; | |
1da177e4 LT |
637 | |
638 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
639 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
640 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
641 | ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp))); | |
642 | ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) || | |
643 | (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks)); | |
644 | ||
645 | /* | |
646 | * Mark the buffer as needing to be written out eventually, | |
647 | * and set its iodone function to remove the buffer's buf log | |
648 | * item from the AIL and free it when the buffer is flushed | |
649 | * to disk. See xfs_buf_attach_iodone() for more details | |
650 | * on li_cb and xfs_buf_iodone_callbacks(). | |
651 | * If we end up aborting this transaction, we trap this buffer | |
652 | * inside the b_bdstrat callback so that this won't get written to | |
653 | * disk. | |
654 | */ | |
655 | XFS_BUF_DELAYWRITE(bp); | |
656 | XFS_BUF_DONE(bp); | |
657 | ||
658 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
659 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
660 | XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks); | |
ca30b2a7 | 661 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 | 662 | |
0b1b213f CH |
663 | trace_xfs_trans_log_buf(bip); |
664 | ||
1da177e4 LT |
665 | /* |
666 | * If we invalidated the buffer within this transaction, then | |
667 | * cancel the invalidation now that we're dirtying the buffer | |
668 | * again. There are no races with the code in xfs_buf_item_unpin(), | |
669 | * because we have a reference to the buffer this entire time. | |
670 | */ | |
671 | if (bip->bli_flags & XFS_BLI_STALE) { | |
1da177e4 LT |
672 | bip->bli_flags &= ~XFS_BLI_STALE; |
673 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
674 | XFS_BUF_UNSTALE(bp); | |
c1155410 | 675 | bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
1da177e4 LT |
676 | } |
677 | ||
1da177e4 | 678 | tp->t_flags |= XFS_TRANS_DIRTY; |
e98c414f | 679 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 LT |
680 | bip->bli_flags |= XFS_BLI_LOGGED; |
681 | xfs_buf_item_log(bip, first, last); | |
1da177e4 LT |
682 | } |
683 | ||
684 | ||
685 | /* | |
686 | * This called to invalidate a buffer that is being used within | |
687 | * a transaction. Typically this is because the blocks in the | |
688 | * buffer are being freed, so we need to prevent it from being | |
689 | * written out when we're done. Allowing it to be written again | |
690 | * might overwrite data in the free blocks if they are reallocated | |
691 | * to a file. | |
692 | * | |
693 | * We prevent the buffer from being written out by clearing the | |
694 | * B_DELWRI flag. We can't always | |
695 | * get rid of the buf log item at this point, though, because | |
696 | * the buffer may still be pinned by another transaction. If that | |
697 | * is the case, then we'll wait until the buffer is committed to | |
698 | * disk for the last time (we can tell by the ref count) and | |
699 | * free it in xfs_buf_item_unpin(). Until it is cleaned up we | |
700 | * will keep the buffer locked so that the buffer and buf log item | |
701 | * are not reused. | |
702 | */ | |
703 | void | |
704 | xfs_trans_binval( | |
705 | xfs_trans_t *tp, | |
706 | xfs_buf_t *bp) | |
707 | { | |
1da177e4 LT |
708 | xfs_buf_log_item_t *bip; |
709 | ||
710 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
711 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
712 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
713 | ||
714 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
1da177e4 LT |
715 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
716 | ||
0b1b213f CH |
717 | trace_xfs_trans_binval(bip); |
718 | ||
1da177e4 LT |
719 | if (bip->bli_flags & XFS_BLI_STALE) { |
720 | /* | |
721 | * If the buffer is already invalidated, then | |
722 | * just return. | |
723 | */ | |
724 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); | |
725 | ASSERT(XFS_BUF_ISSTALE(bp)); | |
726 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); | |
c1155410 DC |
727 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
728 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); | |
e98c414f | 729 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
1da177e4 | 730 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
1da177e4 LT |
731 | return; |
732 | } | |
733 | ||
734 | /* | |
735 | * Clear the dirty bit in the buffer and set the STALE flag | |
736 | * in the buf log item. The STALE flag will be used in | |
737 | * xfs_buf_item_unpin() to determine if it should clean up | |
738 | * when the last reference to the buf item is given up. | |
c1155410 | 739 | * We set the XFS_BLF_CANCEL flag in the buf log format structure |
1da177e4 LT |
740 | * and log the buf item. This will be used at recovery time |
741 | * to determine that copies of the buffer in the log before | |
742 | * this should not be replayed. | |
743 | * We mark the item descriptor and the transaction dirty so | |
744 | * that we'll hold the buffer until after the commit. | |
745 | * | |
746 | * Since we're invalidating the buffer, we also clear the state | |
747 | * about which parts of the buffer have been logged. We also | |
748 | * clear the flag indicating that this is an inode buffer since | |
749 | * the data in the buffer will no longer be valid. | |
750 | * | |
751 | * We set the stale bit in the buffer as well since we're getting | |
752 | * rid of it. | |
753 | */ | |
754 | XFS_BUF_UNDELAYWRITE(bp); | |
755 | XFS_BUF_STALE(bp); | |
756 | bip->bli_flags |= XFS_BLI_STALE; | |
ccf7c23f | 757 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
c1155410 DC |
758 | bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
759 | bip->bli_format.blf_flags |= XFS_BLF_CANCEL; | |
1da177e4 LT |
760 | memset((char *)(bip->bli_format.blf_data_map), 0, |
761 | (bip->bli_format.blf_map_size * sizeof(uint))); | |
e98c414f | 762 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 | 763 | tp->t_flags |= XFS_TRANS_DIRTY; |
1da177e4 LT |
764 | } |
765 | ||
766 | /* | |
ccf7c23f DC |
767 | * This call is used to indicate that the buffer contains on-disk inodes which |
768 | * must be handled specially during recovery. They require special handling | |
769 | * because only the di_next_unlinked from the inodes in the buffer should be | |
770 | * recovered. The rest of the data in the buffer is logged via the inodes | |
771 | * themselves. | |
1da177e4 | 772 | * |
ccf7c23f DC |
773 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
774 | * transferred to the buffer's log format structure so that we'll know what to | |
775 | * do at recovery time. | |
1da177e4 | 776 | */ |
1da177e4 LT |
777 | void |
778 | xfs_trans_inode_buf( | |
779 | xfs_trans_t *tp, | |
780 | xfs_buf_t *bp) | |
781 | { | |
782 | xfs_buf_log_item_t *bip; | |
783 | ||
784 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
785 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
786 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
787 | ||
788 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
789 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
790 | ||
ccf7c23f | 791 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
1da177e4 LT |
792 | } |
793 | ||
794 | /* | |
795 | * This call is used to indicate that the buffer is going to | |
796 | * be staled and was an inode buffer. This means it gets | |
797 | * special processing during unpin - where any inodes | |
798 | * associated with the buffer should be removed from ail. | |
799 | * There is also special processing during recovery, | |
800 | * any replay of the inodes in the buffer needs to be | |
801 | * prevented as the buffer may have been reused. | |
802 | */ | |
803 | void | |
804 | xfs_trans_stale_inode_buf( | |
805 | xfs_trans_t *tp, | |
806 | xfs_buf_t *bp) | |
807 | { | |
808 | xfs_buf_log_item_t *bip; | |
809 | ||
810 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
811 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
812 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
813 | ||
814 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
815 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
816 | ||
817 | bip->bli_flags |= XFS_BLI_STALE_INODE; | |
ca30b2a7 | 818 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 LT |
819 | } |
820 | ||
1da177e4 LT |
821 | /* |
822 | * Mark the buffer as being one which contains newly allocated | |
823 | * inodes. We need to make sure that even if this buffer is | |
824 | * relogged as an 'inode buf' we still recover all of the inode | |
825 | * images in the face of a crash. This works in coordination with | |
826 | * xfs_buf_item_committed() to ensure that the buffer remains in the | |
827 | * AIL at its original location even after it has been relogged. | |
828 | */ | |
829 | /* ARGSUSED */ | |
830 | void | |
831 | xfs_trans_inode_alloc_buf( | |
832 | xfs_trans_t *tp, | |
833 | xfs_buf_t *bp) | |
834 | { | |
835 | xfs_buf_log_item_t *bip; | |
836 | ||
837 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
838 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
839 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
840 | ||
841 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
842 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
843 | ||
844 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; | |
845 | } | |
846 | ||
847 | ||
848 | /* | |
849 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of | |
850 | * dquots. However, unlike in inode buffer recovery, dquot buffers get | |
851 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). | |
852 | * The only thing that makes dquot buffers different from regular | |
853 | * buffers is that we must not replay dquot bufs when recovering | |
854 | * if a _corresponding_ quotaoff has happened. We also have to distinguish | |
855 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas | |
856 | * can be turned off independently. | |
857 | */ | |
858 | /* ARGSUSED */ | |
859 | void | |
860 | xfs_trans_dquot_buf( | |
861 | xfs_trans_t *tp, | |
862 | xfs_buf_t *bp, | |
863 | uint type) | |
864 | { | |
865 | xfs_buf_log_item_t *bip; | |
866 | ||
867 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
868 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); | |
869 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
c1155410 DC |
870 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
871 | type == XFS_BLF_PDQUOT_BUF || | |
872 | type == XFS_BLF_GDQUOT_BUF); | |
1da177e4 LT |
873 | |
874 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); | |
875 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
876 | ||
877 | bip->bli_format.blf_flags |= type; | |
878 | } |