]>
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 | 24 | #include "xfs_mount.h" |
a844f451 | 25 | #include "xfs_inode.h" |
239880ef | 26 | #include "xfs_trans.h" |
a844f451 | 27 | #include "xfs_buf_item.h" |
1da177e4 LT |
28 | #include "xfs_trans_priv.h" |
29 | #include "xfs_error.h" | |
0b1b213f | 30 | #include "xfs_trace.h" |
1da177e4 | 31 | |
4a5224d7 CH |
32 | /* |
33 | * Check to see if a buffer matching the given parameters is already | |
34 | * a part of the given transaction. | |
35 | */ | |
36 | STATIC struct xfs_buf * | |
37 | xfs_trans_buf_item_match( | |
38 | struct xfs_trans *tp, | |
39 | struct xfs_buftarg *target, | |
de2a4f59 DC |
40 | struct xfs_buf_map *map, |
41 | int nmaps) | |
4a5224d7 | 42 | { |
e98c414f CH |
43 | struct xfs_log_item_desc *lidp; |
44 | struct xfs_buf_log_item *blip; | |
de2a4f59 DC |
45 | int len = 0; |
46 | int i; | |
47 | ||
48 | for (i = 0; i < nmaps; i++) | |
49 | len += map[i].bm_len; | |
1da177e4 | 50 | |
e98c414f CH |
51 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
52 | blip = (struct xfs_buf_log_item *)lidp->lid_item; | |
53 | if (blip->bli_item.li_type == XFS_LI_BUF && | |
49074c06 | 54 | blip->bli_buf->b_target == target && |
de2a4f59 DC |
55 | XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn && |
56 | blip->bli_buf->b_length == len) { | |
57 | ASSERT(blip->bli_buf->b_map_count == nmaps); | |
e98c414f | 58 | return blip->bli_buf; |
de2a4f59 | 59 | } |
4a5224d7 CH |
60 | } |
61 | ||
62 | return NULL; | |
63 | } | |
1da177e4 | 64 | |
d7e84f41 CH |
65 | /* |
66 | * Add the locked buffer to the transaction. | |
67 | * | |
68 | * The buffer must be locked, and it cannot be associated with any | |
69 | * transaction. | |
70 | * | |
71 | * If the buffer does not yet have a buf log item associated with it, | |
72 | * then allocate one for it. Then add the buf item to the transaction. | |
73 | */ | |
74 | STATIC void | |
75 | _xfs_trans_bjoin( | |
76 | struct xfs_trans *tp, | |
77 | struct xfs_buf *bp, | |
78 | int reset_recur) | |
79 | { | |
80 | struct xfs_buf_log_item *bip; | |
81 | ||
bf9d9013 | 82 | ASSERT(bp->b_transp == NULL); |
d7e84f41 CH |
83 | |
84 | /* | |
85 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If | |
86 | * it doesn't have one yet, then allocate one and initialize it. | |
87 | * The checks to see if one is there are in xfs_buf_item_init(). | |
88 | */ | |
89 | xfs_buf_item_init(bp, tp->t_mountp); | |
adadbeef | 90 | bip = bp->b_fspriv; |
d7e84f41 | 91 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
0f22f9d0 | 92 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
d7e84f41 CH |
93 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
94 | if (reset_recur) | |
95 | bip->bli_recur = 0; | |
96 | ||
97 | /* | |
98 | * Take a reference for this transaction on the buf item. | |
99 | */ | |
100 | atomic_inc(&bip->bli_refcount); | |
101 | ||
102 | /* | |
103 | * Get a log_item_desc to point at the new item. | |
104 | */ | |
e98c414f | 105 | xfs_trans_add_item(tp, &bip->bli_item); |
d7e84f41 CH |
106 | |
107 | /* | |
108 | * Initialize b_fsprivate2 so we can find it with incore_match() | |
109 | * in xfs_trans_get_buf() and friends above. | |
110 | */ | |
bf9d9013 | 111 | bp->b_transp = tp; |
d7e84f41 CH |
112 | |
113 | } | |
114 | ||
115 | void | |
116 | xfs_trans_bjoin( | |
117 | struct xfs_trans *tp, | |
118 | struct xfs_buf *bp) | |
119 | { | |
120 | _xfs_trans_bjoin(tp, bp, 0); | |
121 | trace_xfs_trans_bjoin(bp->b_fspriv); | |
122 | } | |
1da177e4 LT |
123 | |
124 | /* | |
125 | * Get and lock the buffer for the caller if it is not already | |
126 | * locked within the given transaction. If it is already locked | |
127 | * within the transaction, just increment its lock recursion count | |
128 | * and return a pointer to it. | |
129 | * | |
1da177e4 LT |
130 | * If the transaction pointer is NULL, make this just a normal |
131 | * get_buf() call. | |
132 | */ | |
de2a4f59 DC |
133 | struct xfs_buf * |
134 | xfs_trans_get_buf_map( | |
135 | struct xfs_trans *tp, | |
136 | struct xfs_buftarg *target, | |
137 | struct xfs_buf_map *map, | |
138 | int nmaps, | |
139 | xfs_buf_flags_t flags) | |
1da177e4 LT |
140 | { |
141 | xfs_buf_t *bp; | |
142 | xfs_buf_log_item_t *bip; | |
143 | ||
de2a4f59 DC |
144 | if (!tp) |
145 | return xfs_buf_get_map(target, map, nmaps, flags); | |
1da177e4 LT |
146 | |
147 | /* | |
148 | * If we find the buffer in the cache with this transaction | |
149 | * pointer in its b_fsprivate2 field, then we know we already | |
150 | * have it locked. In this case we just increment the lock | |
151 | * recursion count and return the buffer to the caller. | |
152 | */ | |
de2a4f59 | 153 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
1da177e4 | 154 | if (bp != NULL) { |
0c842ad4 | 155 | ASSERT(xfs_buf_islocked(bp)); |
c867cb61 CH |
156 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) { |
157 | xfs_buf_stale(bp); | |
b0388bf1 | 158 | bp->b_flags |= XBF_DONE; |
c867cb61 | 159 | } |
0b1b213f | 160 | |
bf9d9013 | 161 | ASSERT(bp->b_transp == tp); |
adadbeef | 162 | bip = bp->b_fspriv; |
1da177e4 LT |
163 | ASSERT(bip != NULL); |
164 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
165 | bip->bli_recur++; | |
0b1b213f | 166 | trace_xfs_trans_get_buf_recur(bip); |
d99831ff | 167 | return bp; |
1da177e4 LT |
168 | } |
169 | ||
de2a4f59 | 170 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
1da177e4 LT |
171 | if (bp == NULL) { |
172 | return NULL; | |
173 | } | |
174 | ||
5a52c2a5 | 175 | ASSERT(!bp->b_error); |
1da177e4 | 176 | |
d7e84f41 CH |
177 | _xfs_trans_bjoin(tp, bp, 1); |
178 | trace_xfs_trans_get_buf(bp->b_fspriv); | |
d99831ff | 179 | return bp; |
1da177e4 LT |
180 | } |
181 | ||
182 | /* | |
183 | * Get and lock the superblock buffer of this file system for the | |
184 | * given transaction. | |
185 | * | |
186 | * We don't need to use incore_match() here, because the superblock | |
187 | * buffer is a private buffer which we keep a pointer to in the | |
188 | * mount structure. | |
189 | */ | |
190 | xfs_buf_t * | |
191 | xfs_trans_getsb(xfs_trans_t *tp, | |
192 | struct xfs_mount *mp, | |
193 | int flags) | |
194 | { | |
195 | xfs_buf_t *bp; | |
196 | xfs_buf_log_item_t *bip; | |
197 | ||
198 | /* | |
199 | * Default to just trying to lock the superblock buffer | |
200 | * if tp is NULL. | |
201 | */ | |
d99831ff ES |
202 | if (tp == NULL) |
203 | return xfs_getsb(mp, flags); | |
1da177e4 LT |
204 | |
205 | /* | |
206 | * If the superblock buffer already has this transaction | |
207 | * pointer in its b_fsprivate2 field, then we know we already | |
208 | * have it locked. In this case we just increment the lock | |
209 | * recursion count and return the buffer to the caller. | |
210 | */ | |
211 | bp = mp->m_sb_bp; | |
bf9d9013 | 212 | if (bp->b_transp == tp) { |
adadbeef | 213 | bip = bp->b_fspriv; |
1da177e4 LT |
214 | ASSERT(bip != NULL); |
215 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
216 | bip->bli_recur++; | |
0b1b213f | 217 | trace_xfs_trans_getsb_recur(bip); |
d99831ff | 218 | return bp; |
1da177e4 LT |
219 | } |
220 | ||
221 | bp = xfs_getsb(mp, flags); | |
d7e84f41 | 222 | if (bp == NULL) |
1da177e4 | 223 | return NULL; |
1da177e4 | 224 | |
d7e84f41 CH |
225 | _xfs_trans_bjoin(tp, bp, 1); |
226 | trace_xfs_trans_getsb(bp->b_fspriv); | |
d99831ff | 227 | return bp; |
1da177e4 LT |
228 | } |
229 | ||
1da177e4 LT |
230 | /* |
231 | * Get and lock the buffer for the caller if it is not already | |
232 | * locked within the given transaction. If it has not yet been | |
233 | * read in, read it from disk. If it is already locked | |
234 | * within the transaction and already read in, just increment its | |
235 | * lock recursion count and return a pointer to it. | |
236 | * | |
1da177e4 LT |
237 | * If the transaction pointer is NULL, make this just a normal |
238 | * read_buf() call. | |
239 | */ | |
240 | int | |
de2a4f59 DC |
241 | xfs_trans_read_buf_map( |
242 | struct xfs_mount *mp, | |
243 | struct xfs_trans *tp, | |
244 | struct xfs_buftarg *target, | |
245 | struct xfs_buf_map *map, | |
246 | int nmaps, | |
247 | xfs_buf_flags_t flags, | |
c3f8fc73 | 248 | struct xfs_buf **bpp, |
1813dd64 | 249 | const struct xfs_buf_ops *ops) |
1da177e4 | 250 | { |
2d3d0c53 DC |
251 | struct xfs_buf *bp = NULL; |
252 | struct xfs_buf_log_item *bip; | |
1da177e4 LT |
253 | int error; |
254 | ||
7ca790a5 | 255 | *bpp = NULL; |
1da177e4 LT |
256 | /* |
257 | * If we find the buffer in the cache with this transaction | |
258 | * pointer in its b_fsprivate2 field, then we know we already | |
259 | * have it locked. If it is already read in we just increment | |
260 | * the lock recursion count and return the buffer to the caller. | |
261 | * If the buffer is not yet read in, then we read it in, increment | |
262 | * the lock recursion count, and return it to the caller. | |
263 | */ | |
2d3d0c53 DC |
264 | if (tp) |
265 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); | |
266 | if (bp) { | |
0c842ad4 | 267 | ASSERT(xfs_buf_islocked(bp)); |
bf9d9013 | 268 | ASSERT(bp->b_transp == tp); |
adadbeef | 269 | ASSERT(bp->b_fspriv != NULL); |
5a52c2a5 | 270 | ASSERT(!bp->b_error); |
2d3d0c53 DC |
271 | ASSERT(bp->b_flags & XBF_DONE); |
272 | ||
1da177e4 LT |
273 | /* |
274 | * We never locked this buf ourselves, so we shouldn't | |
275 | * brelse it either. Just get out. | |
276 | */ | |
277 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
0b1b213f | 278 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
2451337d | 279 | return -EIO; |
1da177e4 LT |
280 | } |
281 | ||
adadbeef | 282 | bip = bp->b_fspriv; |
1da177e4 LT |
283 | bip->bli_recur++; |
284 | ||
285 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
0b1b213f | 286 | trace_xfs_trans_read_buf_recur(bip); |
1da177e4 LT |
287 | *bpp = bp; |
288 | return 0; | |
289 | } | |
290 | ||
1813dd64 | 291 | bp = xfs_buf_read_map(target, map, nmaps, flags, ops); |
2d3d0c53 DC |
292 | if (!bp) { |
293 | if (!(flags & XBF_TRYLOCK)) | |
294 | return -ENOMEM; | |
295 | return tp ? 0 : -EAGAIN; | |
1da177e4 | 296 | } |
2d3d0c53 DC |
297 | |
298 | /* | |
299 | * If we've had a read error, then the contents of the buffer are | |
300 | * invalid and should not be used. To ensure that a followup read tries | |
301 | * to pull the buffer from disk again, we clear the XBF_DONE flag and | |
302 | * mark the buffer stale. This ensures that anyone who has a current | |
303 | * reference to the buffer will interpret it's contents correctly and | |
304 | * future cache lookups will also treat it as an empty, uninitialised | |
305 | * buffer. | |
306 | */ | |
5a52c2a5 CS |
307 | if (bp->b_error) { |
308 | error = bp->b_error; | |
2d3d0c53 DC |
309 | if (!XFS_FORCED_SHUTDOWN(mp)) |
310 | xfs_buf_ioerror_alert(bp, __func__); | |
311 | bp->b_flags &= ~XBF_DONE; | |
c867cb61 | 312 | xfs_buf_stale(bp); |
2d3d0c53 DC |
313 | |
314 | if (tp && (tp->t_flags & XFS_TRANS_DIRTY)) | |
7d04a335 | 315 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
1da177e4 | 316 | xfs_buf_relse(bp); |
ac75a1f7 DC |
317 | |
318 | /* bad CRC means corrupted metadata */ | |
2451337d DC |
319 | if (error == -EFSBADCRC) |
320 | error = -EFSCORRUPTED; | |
1da177e4 LT |
321 | return error; |
322 | } | |
2d3d0c53 DC |
323 | |
324 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
325 | xfs_buf_relse(bp); | |
326 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); | |
327 | return -EIO; | |
1da177e4 | 328 | } |
1da177e4 | 329 | |
e9892d3c | 330 | if (tp) { |
2d3d0c53 | 331 | _xfs_trans_bjoin(tp, bp, 1); |
e9892d3c DC |
332 | trace_xfs_trans_read_buf(bp->b_fspriv); |
333 | } | |
1da177e4 LT |
334 | *bpp = bp; |
335 | return 0; | |
336 | ||
1da177e4 LT |
337 | } |
338 | ||
1da177e4 LT |
339 | /* |
340 | * Release the buffer bp which was previously acquired with one of the | |
341 | * xfs_trans_... buffer allocation routines if the buffer has not | |
342 | * been modified within this transaction. If the buffer is modified | |
343 | * within this transaction, do decrement the recursion count but do | |
344 | * not release the buffer even if the count goes to 0. If the buffer is not | |
345 | * modified within the transaction, decrement the recursion count and | |
346 | * release the buffer if the recursion count goes to 0. | |
347 | * | |
348 | * If the buffer is to be released and it was not modified before | |
349 | * this transaction began, then free the buf_log_item associated with it. | |
350 | * | |
351 | * If the transaction pointer is NULL, make this just a normal | |
352 | * brelse() call. | |
353 | */ | |
354 | void | |
355 | xfs_trans_brelse(xfs_trans_t *tp, | |
356 | xfs_buf_t *bp) | |
357 | { | |
358 | xfs_buf_log_item_t *bip; | |
79e641ce | 359 | int freed; |
1da177e4 LT |
360 | |
361 | /* | |
362 | * Default to a normal brelse() call if the tp is NULL. | |
363 | */ | |
364 | if (tp == NULL) { | |
bf9d9013 | 365 | ASSERT(bp->b_transp == NULL); |
1da177e4 LT |
366 | xfs_buf_relse(bp); |
367 | return; | |
368 | } | |
369 | ||
bf9d9013 | 370 | ASSERT(bp->b_transp == tp); |
adadbeef | 371 | bip = bp->b_fspriv; |
1da177e4 LT |
372 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
373 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); | |
0f22f9d0 | 374 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 LT |
375 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
376 | ||
0b1b213f CH |
377 | trace_xfs_trans_brelse(bip); |
378 | ||
1da177e4 LT |
379 | /* |
380 | * If the release is just for a recursive lock, | |
381 | * then decrement the count and return. | |
382 | */ | |
383 | if (bip->bli_recur > 0) { | |
384 | bip->bli_recur--; | |
1da177e4 LT |
385 | return; |
386 | } | |
387 | ||
388 | /* | |
389 | * If the buffer is dirty within this transaction, we can't | |
390 | * release it until we commit. | |
391 | */ | |
e98c414f | 392 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
1da177e4 | 393 | return; |
1da177e4 LT |
394 | |
395 | /* | |
396 | * If the buffer has been invalidated, then we can't release | |
397 | * it until the transaction commits to disk unless it is re-dirtied | |
398 | * as part of this transaction. This prevents us from pulling | |
399 | * the item from the AIL before we should. | |
400 | */ | |
0b1b213f | 401 | if (bip->bli_flags & XFS_BLI_STALE) |
1da177e4 | 402 | return; |
1da177e4 LT |
403 | |
404 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); | |
1da177e4 LT |
405 | |
406 | /* | |
407 | * Free up the log item descriptor tracking the released item. | |
408 | */ | |
e98c414f | 409 | xfs_trans_del_item(&bip->bli_item); |
1da177e4 LT |
410 | |
411 | /* | |
412 | * Clear the hold flag in the buf log item if it is set. | |
413 | * We wouldn't want the next user of the buffer to | |
414 | * get confused. | |
415 | */ | |
416 | if (bip->bli_flags & XFS_BLI_HOLD) { | |
417 | bip->bli_flags &= ~XFS_BLI_HOLD; | |
418 | } | |
419 | ||
420 | /* | |
421 | * Drop our reference to the buf log item. | |
422 | */ | |
79e641ce | 423 | freed = atomic_dec_and_test(&bip->bli_refcount); |
1da177e4 LT |
424 | |
425 | /* | |
79e641ce BF |
426 | * If the buf item is not tracking data in the log, then we must free it |
427 | * before releasing the buffer back to the free pool. | |
428 | * | |
429 | * If the fs has shutdown and we dropped the last reference, it may fall | |
430 | * on us to release a (possibly dirty) bli if it never made it to the | |
431 | * AIL (e.g., the aborted unpin already happened and didn't release it | |
432 | * due to our reference). Since we're already shutdown and need xa_lock, | |
433 | * just force remove from the AIL and release the bli here. | |
1da177e4 | 434 | */ |
79e641ce BF |
435 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp) && freed) { |
436 | xfs_trans_ail_remove(&bip->bli_item, SHUTDOWN_LOG_IO_ERROR); | |
437 | xfs_buf_item_relse(bp); | |
a4f6cf6b | 438 | } else if (!(bip->bli_flags & XFS_BLI_DIRTY)) { |
1da177e4 LT |
439 | /*** |
440 | ASSERT(bp->b_pincount == 0); | |
441 | ***/ | |
442 | ASSERT(atomic_read(&bip->bli_refcount) == 0); | |
443 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); | |
444 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); | |
445 | xfs_buf_item_relse(bp); | |
1da177e4 LT |
446 | } |
447 | ||
5b03ff1b | 448 | bp->b_transp = NULL; |
1da177e4 | 449 | xfs_buf_relse(bp); |
1da177e4 LT |
450 | } |
451 | ||
1da177e4 LT |
452 | /* |
453 | * Mark the buffer as not needing to be unlocked when the buf item's | |
904c17e6 | 454 | * iop_unlock() routine is called. The buffer must already be locked |
1da177e4 LT |
455 | * and associated with the given transaction. |
456 | */ | |
457 | /* ARGSUSED */ | |
458 | void | |
459 | xfs_trans_bhold(xfs_trans_t *tp, | |
460 | xfs_buf_t *bp) | |
461 | { | |
adadbeef | 462 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 463 | |
bf9d9013 | 464 | ASSERT(bp->b_transp == tp); |
adadbeef | 465 | ASSERT(bip != NULL); |
1da177e4 | 466 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
0f22f9d0 | 467 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
1da177e4 | 468 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
adadbeef | 469 | |
1da177e4 | 470 | bip->bli_flags |= XFS_BLI_HOLD; |
0b1b213f | 471 | trace_xfs_trans_bhold(bip); |
1da177e4 LT |
472 | } |
473 | ||
efa092f3 TS |
474 | /* |
475 | * Cancel the previous buffer hold request made on this buffer | |
476 | * for this transaction. | |
477 | */ | |
478 | void | |
479 | xfs_trans_bhold_release(xfs_trans_t *tp, | |
480 | xfs_buf_t *bp) | |
481 | { | |
adadbeef | 482 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
efa092f3 | 483 | |
bf9d9013 | 484 | ASSERT(bp->b_transp == tp); |
adadbeef | 485 | ASSERT(bip != NULL); |
efa092f3 | 486 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
0f22f9d0 | 487 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
efa092f3 TS |
488 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
489 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); | |
0b1b213f | 490 | |
adadbeef | 491 | bip->bli_flags &= ~XFS_BLI_HOLD; |
0b1b213f | 492 | trace_xfs_trans_bhold_release(bip); |
efa092f3 TS |
493 | } |
494 | ||
1da177e4 | 495 | /* |
9684010d | 496 | * Mark a buffer dirty in the transaction. |
1da177e4 LT |
497 | */ |
498 | void | |
9684010d BF |
499 | xfs_trans_dirty_buf( |
500 | struct xfs_trans *tp, | |
501 | struct xfs_buf *bp) | |
1da177e4 | 502 | { |
9684010d | 503 | struct xfs_buf_log_item *bip = bp->b_fspriv; |
1da177e4 | 504 | |
bf9d9013 | 505 | ASSERT(bp->b_transp == tp); |
adadbeef | 506 | ASSERT(bip != NULL); |
cb669ca5 CH |
507 | ASSERT(bp->b_iodone == NULL || |
508 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1da177e4 LT |
509 | |
510 | /* | |
511 | * Mark the buffer as needing to be written out eventually, | |
512 | * and set its iodone function to remove the buffer's buf log | |
513 | * item from the AIL and free it when the buffer is flushed | |
514 | * to disk. See xfs_buf_attach_iodone() for more details | |
515 | * on li_cb and xfs_buf_iodone_callbacks(). | |
516 | * If we end up aborting this transaction, we trap this buffer | |
517 | * inside the b_bdstrat callback so that this won't get written to | |
518 | * disk. | |
519 | */ | |
b0388bf1 | 520 | bp->b_flags |= XBF_DONE; |
1da177e4 | 521 | |
1da177e4 | 522 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
cb669ca5 | 523 | bp->b_iodone = xfs_buf_iodone_callbacks; |
ca30b2a7 | 524 | bip->bli_item.li_cb = xfs_buf_iodone; |
1da177e4 LT |
525 | |
526 | /* | |
527 | * If we invalidated the buffer within this transaction, then | |
528 | * cancel the invalidation now that we're dirtying the buffer | |
529 | * again. There are no races with the code in xfs_buf_item_unpin(), | |
530 | * because we have a reference to the buffer this entire time. | |
531 | */ | |
532 | if (bip->bli_flags & XFS_BLI_STALE) { | |
1da177e4 | 533 | bip->bli_flags &= ~XFS_BLI_STALE; |
5cfd28b6 DC |
534 | ASSERT(bp->b_flags & XBF_STALE); |
535 | bp->b_flags &= ~XBF_STALE; | |
0f22f9d0 | 536 | bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
1da177e4 | 537 | } |
9684010d | 538 | bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED; |
1da177e4 | 539 | |
1da177e4 | 540 | tp->t_flags |= XFS_TRANS_DIRTY; |
e98c414f | 541 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
9684010d BF |
542 | } |
543 | ||
544 | /* | |
545 | * This is called to mark bytes first through last inclusive of the given | |
546 | * buffer as needing to be logged when the transaction is committed. | |
547 | * The buffer must already be associated with the given transaction. | |
548 | * | |
549 | * First and last are numbers relative to the beginning of this buffer, | |
550 | * so the first byte in the buffer is numbered 0 regardless of the | |
551 | * value of b_blkno. | |
552 | */ | |
553 | void | |
554 | xfs_trans_log_buf( | |
555 | struct xfs_trans *tp, | |
556 | struct xfs_buf *bp, | |
557 | uint first, | |
558 | uint last) | |
559 | { | |
560 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
561 | ||
562 | ASSERT(first <= last && last < BBTOB(bp->b_length)); | |
8dc518df | 563 | ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED)); |
9684010d BF |
564 | |
565 | xfs_trans_dirty_buf(tp, bp); | |
5f6bed76 | 566 | |
9684010d | 567 | trace_xfs_trans_log_buf(bip); |
8dc518df | 568 | xfs_buf_item_log(bip, first, last); |
1da177e4 LT |
569 | } |
570 | ||
571 | ||
572 | /* | |
43ff2122 CH |
573 | * Invalidate a buffer that is being used within a transaction. |
574 | * | |
575 | * Typically this is because the blocks in the buffer are being freed, so we | |
576 | * need to prevent it from being written out when we're done. Allowing it | |
577 | * to be written again might overwrite data in the free blocks if they are | |
578 | * reallocated to a file. | |
1da177e4 | 579 | * |
43ff2122 CH |
580 | * We prevent the buffer from being written out by marking it stale. We can't |
581 | * get rid of the buf log item at this point because the buffer may still be | |
582 | * pinned by another transaction. If that is the case, then we'll wait until | |
583 | * the buffer is committed to disk for the last time (we can tell by the ref | |
584 | * count) and free it in xfs_buf_item_unpin(). Until that happens we will | |
585 | * keep the buffer locked so that the buffer and buf log item are not reused. | |
586 | * | |
587 | * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log | |
588 | * the buf item. This will be used at recovery time to determine that copies | |
589 | * of the buffer in the log before this should not be replayed. | |
590 | * | |
591 | * We mark the item descriptor and the transaction dirty so that we'll hold | |
592 | * the buffer until after the commit. | |
593 | * | |
594 | * Since we're invalidating the buffer, we also clear the state about which | |
595 | * parts of the buffer have been logged. We also clear the flag indicating | |
596 | * that this is an inode buffer since the data in the buffer will no longer | |
597 | * be valid. | |
598 | * | |
599 | * We set the stale bit in the buffer as well since we're getting rid of it. | |
1da177e4 LT |
600 | */ |
601 | void | |
602 | xfs_trans_binval( | |
603 | xfs_trans_t *tp, | |
604 | xfs_buf_t *bp) | |
605 | { | |
adadbeef | 606 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
91e4bac0 | 607 | int i; |
1da177e4 | 608 | |
bf9d9013 | 609 | ASSERT(bp->b_transp == tp); |
adadbeef | 610 | ASSERT(bip != NULL); |
1da177e4 LT |
611 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
612 | ||
0b1b213f CH |
613 | trace_xfs_trans_binval(bip); |
614 | ||
1da177e4 LT |
615 | if (bip->bli_flags & XFS_BLI_STALE) { |
616 | /* | |
617 | * If the buffer is already invalidated, then | |
618 | * just return. | |
619 | */ | |
5cfd28b6 | 620 | ASSERT(bp->b_flags & XBF_STALE); |
1da177e4 | 621 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); |
0f22f9d0 | 622 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
61fe135c | 623 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK)); |
0f22f9d0 | 624 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
e98c414f | 625 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
1da177e4 | 626 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
1da177e4 LT |
627 | return; |
628 | } | |
629 | ||
c867cb61 | 630 | xfs_buf_stale(bp); |
43ff2122 | 631 | |
1da177e4 | 632 | bip->bli_flags |= XFS_BLI_STALE; |
ccf7c23f | 633 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
0f22f9d0 MT |
634 | bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
635 | bip->__bli_format.blf_flags |= XFS_BLF_CANCEL; | |
61fe135c | 636 | bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK; |
91e4bac0 MT |
637 | for (i = 0; i < bip->bli_format_count; i++) { |
638 | memset(bip->bli_formats[i].blf_data_map, 0, | |
639 | (bip->bli_formats[i].blf_map_size * sizeof(uint))); | |
640 | } | |
e98c414f | 641 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
1da177e4 | 642 | tp->t_flags |= XFS_TRANS_DIRTY; |
1da177e4 LT |
643 | } |
644 | ||
645 | /* | |
ccf7c23f DC |
646 | * This call is used to indicate that the buffer contains on-disk inodes which |
647 | * must be handled specially during recovery. They require special handling | |
648 | * because only the di_next_unlinked from the inodes in the buffer should be | |
649 | * recovered. The rest of the data in the buffer is logged via the inodes | |
650 | * themselves. | |
1da177e4 | 651 | * |
ccf7c23f DC |
652 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
653 | * transferred to the buffer's log format structure so that we'll know what to | |
654 | * do at recovery time. | |
1da177e4 | 655 | */ |
1da177e4 LT |
656 | void |
657 | xfs_trans_inode_buf( | |
658 | xfs_trans_t *tp, | |
659 | xfs_buf_t *bp) | |
660 | { | |
adadbeef | 661 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 662 | |
bf9d9013 | 663 | ASSERT(bp->b_transp == tp); |
adadbeef | 664 | ASSERT(bip != NULL); |
1da177e4 LT |
665 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
666 | ||
ccf7c23f | 667 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
61fe135c | 668 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF); |
1da177e4 LT |
669 | } |
670 | ||
671 | /* | |
672 | * This call is used to indicate that the buffer is going to | |
673 | * be staled and was an inode buffer. This means it gets | |
93848a99 | 674 | * special processing during unpin - where any inodes |
1da177e4 LT |
675 | * associated with the buffer should be removed from ail. |
676 | * There is also special processing during recovery, | |
677 | * any replay of the inodes in the buffer needs to be | |
678 | * prevented as the buffer may have been reused. | |
679 | */ | |
680 | void | |
681 | xfs_trans_stale_inode_buf( | |
682 | xfs_trans_t *tp, | |
683 | xfs_buf_t *bp) | |
684 | { | |
adadbeef | 685 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 686 | |
bf9d9013 | 687 | ASSERT(bp->b_transp == tp); |
adadbeef | 688 | ASSERT(bip != NULL); |
1da177e4 LT |
689 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
690 | ||
691 | bip->bli_flags |= XFS_BLI_STALE_INODE; | |
ca30b2a7 | 692 | bip->bli_item.li_cb = xfs_buf_iodone; |
61fe135c | 693 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF); |
1da177e4 LT |
694 | } |
695 | ||
1da177e4 LT |
696 | /* |
697 | * Mark the buffer as being one which contains newly allocated | |
698 | * inodes. We need to make sure that even if this buffer is | |
699 | * relogged as an 'inode buf' we still recover all of the inode | |
700 | * images in the face of a crash. This works in coordination with | |
701 | * xfs_buf_item_committed() to ensure that the buffer remains in the | |
702 | * AIL at its original location even after it has been relogged. | |
703 | */ | |
704 | /* ARGSUSED */ | |
705 | void | |
706 | xfs_trans_inode_alloc_buf( | |
707 | xfs_trans_t *tp, | |
708 | xfs_buf_t *bp) | |
709 | { | |
adadbeef | 710 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 711 | |
bf9d9013 | 712 | ASSERT(bp->b_transp == tp); |
adadbeef | 713 | ASSERT(bip != NULL); |
1da177e4 LT |
714 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
715 | ||
716 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; | |
61fe135c | 717 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF); |
1da177e4 LT |
718 | } |
719 | ||
5f6bed76 | 720 | /* |
8dc518df BF |
721 | * Mark the buffer as ordered for this transaction. This means that the contents |
722 | * of the buffer are not recorded in the transaction but it is tracked in the | |
723 | * AIL as though it was. This allows us to record logical changes in | |
724 | * transactions rather than the physical changes we make to the buffer without | |
725 | * changing writeback ordering constraints of metadata buffers. | |
5f6bed76 | 726 | */ |
a5814bce | 727 | bool |
5f6bed76 DC |
728 | xfs_trans_ordered_buf( |
729 | struct xfs_trans *tp, | |
730 | struct xfs_buf *bp) | |
731 | { | |
732 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
733 | ||
734 | ASSERT(bp->b_transp == tp); | |
735 | ASSERT(bip != NULL); | |
736 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
a5814bce BF |
737 | |
738 | if (xfs_buf_item_dirty_format(bip)) | |
739 | return false; | |
5f6bed76 DC |
740 | |
741 | bip->bli_flags |= XFS_BLI_ORDERED; | |
742 | trace_xfs_buf_item_ordered(bip); | |
8dc518df BF |
743 | |
744 | /* | |
745 | * We don't log a dirty range of an ordered buffer but it still needs | |
746 | * to be marked dirty and that it has been logged. | |
747 | */ | |
748 | xfs_trans_dirty_buf(tp, bp); | |
a5814bce | 749 | return true; |
5f6bed76 DC |
750 | } |
751 | ||
ee1a47ab CH |
752 | /* |
753 | * Set the type of the buffer for log recovery so that it can correctly identify | |
754 | * and hence attach the correct buffer ops to the buffer after replay. | |
755 | */ | |
756 | void | |
757 | xfs_trans_buf_set_type( | |
758 | struct xfs_trans *tp, | |
759 | struct xfs_buf *bp, | |
61fe135c | 760 | enum xfs_blft type) |
ee1a47ab CH |
761 | { |
762 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
763 | ||
d75afeb3 DC |
764 | if (!tp) |
765 | return; | |
766 | ||
ee1a47ab CH |
767 | ASSERT(bp->b_transp == tp); |
768 | ASSERT(bip != NULL); | |
769 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
ee1a47ab | 770 | |
61fe135c | 771 | xfs_blft_to_flags(&bip->__bli_format, type); |
ee1a47ab | 772 | } |
1da177e4 | 773 | |
d75afeb3 DC |
774 | void |
775 | xfs_trans_buf_copy_type( | |
776 | struct xfs_buf *dst_bp, | |
777 | struct xfs_buf *src_bp) | |
778 | { | |
779 | struct xfs_buf_log_item *sbip = src_bp->b_fspriv; | |
780 | struct xfs_buf_log_item *dbip = dst_bp->b_fspriv; | |
61fe135c | 781 | enum xfs_blft type; |
d75afeb3 | 782 | |
61fe135c DC |
783 | type = xfs_blft_from_flags(&sbip->__bli_format); |
784 | xfs_blft_to_flags(&dbip->__bli_format, type); | |
d75afeb3 DC |
785 | } |
786 | ||
1da177e4 LT |
787 | /* |
788 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of | |
789 | * dquots. However, unlike in inode buffer recovery, dquot buffers get | |
790 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). | |
791 | * The only thing that makes dquot buffers different from regular | |
792 | * buffers is that we must not replay dquot bufs when recovering | |
793 | * if a _corresponding_ quotaoff has happened. We also have to distinguish | |
794 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas | |
795 | * can be turned off independently. | |
796 | */ | |
797 | /* ARGSUSED */ | |
798 | void | |
799 | xfs_trans_dquot_buf( | |
800 | xfs_trans_t *tp, | |
801 | xfs_buf_t *bp, | |
802 | uint type) | |
803 | { | |
61fe135c DC |
804 | struct xfs_buf_log_item *bip = bp->b_fspriv; |
805 | ||
c1155410 DC |
806 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
807 | type == XFS_BLF_PDQUOT_BUF || | |
808 | type == XFS_BLF_GDQUOT_BUF); | |
1da177e4 | 809 | |
61fe135c DC |
810 | bip->__bli_format.blf_flags |= type; |
811 | ||
812 | switch (type) { | |
813 | case XFS_BLF_UDQUOT_BUF: | |
814 | type = XFS_BLFT_UDQUOT_BUF; | |
815 | break; | |
816 | case XFS_BLF_PDQUOT_BUF: | |
817 | type = XFS_BLFT_PDQUOT_BUF; | |
818 | break; | |
819 | case XFS_BLF_GDQUOT_BUF: | |
820 | type = XFS_BLFT_GDQUOT_BUF; | |
821 | break; | |
822 | default: | |
823 | type = XFS_BLFT_UNKNOWN_BUF; | |
824 | break; | |
825 | } | |
826 | ||
ee1a47ab | 827 | xfs_trans_buf_set_type(tp, bp, type); |
1da177e4 | 828 | } |