blob: 09a0468ef11a87dca652d166b9ff969a084ce8b3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_bmap_btree.h"
29#include "xfs_alloc_btree.h"
30#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_dinode.h"
32#include "xfs_inode.h"
33#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_trans_priv.h"
35#include "xfs_error.h"
36#include "xfs_rw.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000037#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Christoph Hellwig4a5224d2010-04-18 00:10:45 +000039/*
40 * Check to see if a buffer matching the given parameters is already
41 * a part of the given transaction.
42 */
43STATIC struct xfs_buf *
44xfs_trans_buf_item_match(
45 struct xfs_trans *tp,
46 struct xfs_buftarg *target,
47 xfs_daddr_t blkno,
48 int len)
49{
Christoph Hellwige98c4142010-06-23 18:11:15 +100050 struct xfs_log_item_desc *lidp;
51 struct xfs_buf_log_item *blip;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Christoph Hellwig4a5224d2010-04-18 00:10:45 +000053 len = BBTOB(len);
Christoph Hellwige98c4142010-06-23 18:11:15 +100054 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;
Christoph Hellwig4a5224d2010-04-18 00:10:45 +000061 }
62
63 return NULL;
64}
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000066/*
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 */
75STATIC 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));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +020084 ASSERT(bp->b_transp == NULL);
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000085
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));
Dave Chinnerc1155412010-05-07 11:05:19 +100094 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000095 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 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000107 xfs_trans_add_item(tp, &bip->bli_item);
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000108
109 /*
110 * Initialize b_fsprivate2 so we can find it with incore_match()
111 * in xfs_trans_get_buf() and friends above.
112 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200113 bp->b_transp = tp;
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000114
115}
116
117void
118xfs_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}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 * If the transaction pointer is NULL, make this just a normal
133 * get_buf() call.
134 */
135xfs_buf_t *
136xfs_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)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000146 flags = XBF_LOCK | XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148 /*
149 * Default to a normal get_buf() call if the tp is NULL.
150 */
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000151 if (tp == NULL)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000152 return xfs_buf_get(target_dev, blkno, len,
153 flags | XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
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 */
Christoph Hellwig4a5224d2010-04-18 00:10:45 +0000161 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 if (bp != NULL) {
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200163 ASSERT(xfs_buf_islocked(bp));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000164 if (XFS_FORCED_SHUTDOWN(tp->t_mountp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 XFS_BUF_SUPER_STALE(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 /*
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 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000172 else if (XFS_BUF_ISSTALE(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000174
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200175 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 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++;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000180 trace_xfs_trans_get_buf_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return (bp);
182 }
183
184 /*
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000185 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 * 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 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000192 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (bp == NULL) {
194 return NULL;
195 }
196
197 ASSERT(!XFS_BUF_GETERROR(bp));
198
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000199 _xfs_trans_bjoin(tp, bp, 1);
200 trace_xfs_trans_get_buf(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 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 */
212xfs_buf_t *
213xfs_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;
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200235 if (bp->b_transp == tp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 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++;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000240 trace_xfs_trans_getsb_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 return (bp);
242 }
243
244 bp = xfs_getsb(mp, flags);
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000245 if (bp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000248 _xfs_trans_bjoin(tp, bp, 1);
249 trace_xfs_trans_getsb(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 return (bp);
251}
252
253#ifdef DEBUG
254xfs_buftarg_t *xfs_error_target;
255int xfs_do_error;
256int xfs_req_num;
257int 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 * If the transaction pointer is NULL, make this just a normal
268 * read_buf() call.
269 */
270int
271xfs_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)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000285 flags = XBF_LOCK | XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 /*
288 * Default to a normal get_buf() call if the tp is NULL.
289 */
290 if (tp == NULL) {
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000291 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 if (!bp)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000293 return (flags & XBF_TRYLOCK) ?
David Chinnera3f74ff2008-03-06 13:43:42 +1100294 EAGAIN : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Julia Lawalla0f7bfd2009-07-27 18:15:25 +0200296 if (XFS_BUF_GETERROR(bp) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 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
Julia Lawalla0f7bfd2009-07-27 18:15:25 +0200304 if (xfs_do_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 if (xfs_error_target == target) {
306 if (((xfs_req_num++) % xfs_error_mod) == 0) {
307 xfs_buf_relse(bp);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100308 xfs_debug(mp, "Returning error!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 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 */
Christoph Hellwig4a5224d2010-04-18 00:10:45 +0000328 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 if (bp != NULL) {
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200330 ASSERT(xfs_buf_islocked(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200331 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
333 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
334 if (!(XFS_BUF_ISDONE(bp))) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000335 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 ASSERT(!XFS_BUF_ISASYNC(bp));
337 XFS_BUF_READ(bp);
338 xfsbdstrat(tp->t_mountp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000339 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000340 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 xfs_ioerror_alert("xfs_trans_read_buf", mp,
342 bp, blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 xfs_buf_relse(bp);
344 /*
David Chinnerd64e31a2008-04-10 12:22:17 +1000345 * We can gracefully recover from most read
346 * errors. Ones we can't are those that happen
347 * after the transaction's already dirty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 */
349 if (tp->t_flags & XFS_TRANS_DIRTY)
350 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000351 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 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)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000360 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 *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);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000370 trace_xfs_trans_read_buf_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 *bpp = bp;
372 return 0;
373 }
374
375 /*
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000376 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 * 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 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000383 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 if (bp == NULL) {
385 *bpp = NULL;
Dave Chinner7401aaf2011-03-26 09:14:44 +1100386 return (flags & XBF_TRYLOCK) ?
387 0 : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 }
389 if (XFS_BUF_GETERROR(bp) != 0) {
390 XFS_BUF_SUPER_STALE(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 error = XFS_BUF_GETERROR(bp);
392
393 xfs_ioerror_alert("xfs_trans_read_buf", mp,
394 bp, blkno);
395 if (tp->t_flags & XFS_TRANS_DIRTY)
Nathan Scott7d04a332006-06-09 14:58:38 +1000396 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 xfs_buf_relse(bp);
398 return error;
399 }
400#ifdef DEBUG
401 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
402 if (xfs_error_target == target) {
403 if (((xfs_req_num++) % xfs_error_mod) == 0) {
404 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000405 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 xfs_buf_relse(bp);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100407 xfs_debug(mp, "Returning trans error!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 return XFS_ERROR(EIO);
409 }
410 }
411 }
412#endif
413 if (XFS_FORCED_SHUTDOWN(mp))
414 goto shutdown_abort;
415
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000416 _xfs_trans_bjoin(tp, bp, 1);
417 trace_xfs_trans_read_buf(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 *bpp = bp;
420 return 0;
421
422shutdown_abort:
423 /*
424 * the theory here is that buffer is good but we're
425 * bailing out because the filesystem is being forcibly
426 * shut down. So we should leave the b_flags alone since
427 * the buffer's not staled and just get out.
428 */
429#if defined(DEBUG)
430 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
Dave Chinner0b932cc2011-03-07 10:08:35 +1100431 xfs_notice(mp, "about to pop assert, bp == 0x%p", bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432#endif
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000433 ASSERT((XFS_BUF_BFLAGS(bp) & (XBF_STALE|XBF_DELWRI)) !=
434 (XBF_STALE|XBF_DELWRI));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000436 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 xfs_buf_relse(bp);
438 *bpp = NULL;
439 return XFS_ERROR(EIO);
440}
441
442
443/*
444 * Release the buffer bp which was previously acquired with one of the
445 * xfs_trans_... buffer allocation routines if the buffer has not
446 * been modified within this transaction. If the buffer is modified
447 * within this transaction, do decrement the recursion count but do
448 * not release the buffer even if the count goes to 0. If the buffer is not
449 * modified within the transaction, decrement the recursion count and
450 * release the buffer if the recursion count goes to 0.
451 *
452 * If the buffer is to be released and it was not modified before
453 * this transaction began, then free the buf_log_item associated with it.
454 *
455 * If the transaction pointer is NULL, make this just a normal
456 * brelse() call.
457 */
458void
459xfs_trans_brelse(xfs_trans_t *tp,
460 xfs_buf_t *bp)
461{
462 xfs_buf_log_item_t *bip;
463 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 /*
466 * Default to a normal brelse() call if the tp is NULL.
467 */
468 if (tp == NULL) {
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200469 ASSERT(bp->b_transp == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 /*
471 * If there's a buf log item attached to the buffer,
472 * then let the AIL know that the buffer is being
473 * unlocked.
474 */
475 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
476 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
477 if (lip->li_type == XFS_LI_BUF) {
478 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
David Chinner783a2f62008-10-30 17:39:58 +1100479 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
480 lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 }
482 }
483 xfs_buf_relse(bp);
484 return;
485 }
486
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200487 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
489 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
490 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Dave Chinnerc1155412010-05-07 11:05:19 +1000491 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 ASSERT(atomic_read(&bip->bli_refcount) > 0);
493
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000494 trace_xfs_trans_brelse(bip);
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 /*
497 * If the release is just for a recursive lock,
498 * then decrement the count and return.
499 */
500 if (bip->bli_recur > 0) {
501 bip->bli_recur--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return;
503 }
504
505 /*
506 * If the buffer is dirty within this transaction, we can't
507 * release it until we commit.
508 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000509 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512 /*
513 * If the buffer has been invalidated, then we can't release
514 * it until the transaction commits to disk unless it is re-dirtied
515 * as part of this transaction. This prevents us from pulling
516 * the item from the AIL before we should.
517 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000518 if (bip->bli_flags & XFS_BLI_STALE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
523 /*
524 * Free up the log item descriptor tracking the released item.
525 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000526 xfs_trans_del_item(&bip->bli_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 /*
529 * Clear the hold flag in the buf log item if it is set.
530 * We wouldn't want the next user of the buffer to
531 * get confused.
532 */
533 if (bip->bli_flags & XFS_BLI_HOLD) {
534 bip->bli_flags &= ~XFS_BLI_HOLD;
535 }
536
537 /*
538 * Drop our reference to the buf log item.
539 */
540 atomic_dec(&bip->bli_refcount);
541
542 /*
543 * If the buf item is not tracking data in the log, then
544 * we must free it before releasing the buffer back to the
545 * free pool. Before releasing the buffer to the free pool,
546 * clear the transaction pointer in b_fsprivate2 to dissolve
547 * its relation to this transaction.
548 */
549 if (!xfs_buf_item_dirty(bip)) {
550/***
551 ASSERT(bp->b_pincount == 0);
552***/
553 ASSERT(atomic_read(&bip->bli_refcount) == 0);
554 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
555 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
556 xfs_buf_item_relse(bp);
557 bip = NULL;
558 }
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200559 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 /*
562 * If we've still got a buf log item on the buffer, then
563 * tell the AIL that the buffer is being unlocked.
564 */
565 if (bip != NULL) {
David Chinner783a2f62008-10-30 17:39:58 +1100566 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 (xfs_log_item_t*)bip);
568 }
569
570 xfs_buf_relse(bp);
571 return;
572}
573
574/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 * Mark the buffer as not needing to be unlocked when the buf item's
576 * IOP_UNLOCK() routine is called. The buffer must already be locked
577 * and associated with the given transaction.
578 */
579/* ARGSUSED */
580void
581xfs_trans_bhold(xfs_trans_t *tp,
582 xfs_buf_t *bp)
583{
584 xfs_buf_log_item_t *bip;
585
586 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200587 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
589
590 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
591 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Dave Chinnerc1155412010-05-07 11:05:19 +1000592 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 ASSERT(atomic_read(&bip->bli_refcount) > 0);
594 bip->bli_flags |= XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000595 trace_xfs_trans_bhold(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
598/*
Tim Shimminefa092f2005-09-05 08:29:01 +1000599 * Cancel the previous buffer hold request made on this buffer
600 * for this transaction.
601 */
602void
603xfs_trans_bhold_release(xfs_trans_t *tp,
604 xfs_buf_t *bp)
605{
606 xfs_buf_log_item_t *bip;
607
608 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200609 ASSERT(bp->b_transp == tp);
Tim Shimminefa092f2005-09-05 08:29:01 +1000610 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
611
612 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
613 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Dave Chinnerc1155412010-05-07 11:05:19 +1000614 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Tim Shimminefa092f2005-09-05 08:29:01 +1000615 ASSERT(atomic_read(&bip->bli_refcount) > 0);
616 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
617 bip->bli_flags &= ~XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000618
619 trace_xfs_trans_bhold_release(bip);
Tim Shimminefa092f2005-09-05 08:29:01 +1000620}
621
622/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 * This is called to mark bytes first through last inclusive of the given
624 * buffer as needing to be logged when the transaction is committed.
625 * The buffer must already be associated with the given transaction.
626 *
627 * First and last are numbers relative to the beginning of this buffer,
628 * so the first byte in the buffer is numbered 0 regardless of the
629 * value of b_blkno.
630 */
631void
632xfs_trans_log_buf(xfs_trans_t *tp,
633 xfs_buf_t *bp,
634 uint first,
635 uint last)
636{
637 xfs_buf_log_item_t *bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200640 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
642 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
643 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
644 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
645
646 /*
647 * Mark the buffer as needing to be written out eventually,
648 * and set its iodone function to remove the buffer's buf log
649 * item from the AIL and free it when the buffer is flushed
650 * to disk. See xfs_buf_attach_iodone() for more details
651 * on li_cb and xfs_buf_iodone_callbacks().
652 * If we end up aborting this transaction, we trap this buffer
653 * inside the b_bdstrat callback so that this won't get written to
654 * disk.
655 */
656 XFS_BUF_DELAYWRITE(bp);
657 XFS_BUF_DONE(bp);
658
659 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
660 ASSERT(atomic_read(&bip->bli_refcount) > 0);
661 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000662 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000664 trace_xfs_trans_log_buf(bip);
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 /*
667 * If we invalidated the buffer within this transaction, then
668 * cancel the invalidation now that we're dirtying the buffer
669 * again. There are no races with the code in xfs_buf_item_unpin(),
670 * because we have a reference to the buffer this entire time.
671 */
672 if (bip->bli_flags & XFS_BLI_STALE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 bip->bli_flags &= ~XFS_BLI_STALE;
674 ASSERT(XFS_BUF_ISSTALE(bp));
675 XFS_BUF_UNSTALE(bp);
Dave Chinnerc1155412010-05-07 11:05:19 +1000676 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
678
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 tp->t_flags |= XFS_TRANS_DIRTY;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000680 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 bip->bli_flags |= XFS_BLI_LOGGED;
682 xfs_buf_item_log(bip, first, last);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
685
686/*
687 * This called to invalidate a buffer that is being used within
688 * a transaction. Typically this is because the blocks in the
689 * buffer are being freed, so we need to prevent it from being
690 * written out when we're done. Allowing it to be written again
691 * might overwrite data in the free blocks if they are reallocated
692 * to a file.
693 *
694 * We prevent the buffer from being written out by clearing the
695 * B_DELWRI flag. We can't always
696 * get rid of the buf log item at this point, though, because
697 * the buffer may still be pinned by another transaction. If that
698 * is the case, then we'll wait until the buffer is committed to
699 * disk for the last time (we can tell by the ref count) and
700 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
701 * will keep the buffer locked so that the buffer and buf log item
702 * are not reused.
703 */
704void
705xfs_trans_binval(
706 xfs_trans_t *tp,
707 xfs_buf_t *bp)
708{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 xfs_buf_log_item_t *bip;
710
711 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200712 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
714
715 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 ASSERT(atomic_read(&bip->bli_refcount) > 0);
717
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000718 trace_xfs_trans_binval(bip);
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 if (bip->bli_flags & XFS_BLI_STALE) {
721 /*
722 * If the buffer is already invalidated, then
723 * just return.
724 */
725 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
726 ASSERT(XFS_BUF_ISSTALE(bp));
727 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
Dave Chinnerc1155412010-05-07 11:05:19 +1000728 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
729 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwige98c4142010-06-23 18:11:15 +1000730 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 return;
733 }
734
735 /*
736 * Clear the dirty bit in the buffer and set the STALE flag
737 * in the buf log item. The STALE flag will be used in
738 * xfs_buf_item_unpin() to determine if it should clean up
739 * when the last reference to the buf item is given up.
Dave Chinnerc1155412010-05-07 11:05:19 +1000740 * We set the XFS_BLF_CANCEL flag in the buf log format structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 * and log the buf item. This will be used at recovery time
742 * to determine that copies of the buffer in the log before
743 * this should not be replayed.
744 * We mark the item descriptor and the transaction dirty so
745 * that we'll hold the buffer until after the commit.
746 *
747 * Since we're invalidating the buffer, we also clear the state
748 * about which parts of the buffer have been logged. We also
749 * clear the flag indicating that this is an inode buffer since
750 * the data in the buffer will no longer be valid.
751 *
752 * We set the stale bit in the buffer as well since we're getting
753 * rid of it.
754 */
755 XFS_BUF_UNDELAYWRITE(bp);
756 XFS_BUF_STALE(bp);
757 bip->bli_flags |= XFS_BLI_STALE;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000758 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
Dave Chinnerc1155412010-05-07 11:05:19 +1000759 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
760 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 memset((char *)(bip->bli_format.blf_data_map), 0,
762 (bip->bli_format.blf_map_size * sizeof(uint)));
Christoph Hellwige98c4142010-06-23 18:11:15 +1000763 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 tp->t_flags |= XFS_TRANS_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
766
767/*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000768 * This call is used to indicate that the buffer contains on-disk inodes which
769 * must be handled specially during recovery. They require special handling
770 * because only the di_next_unlinked from the inodes in the buffer should be
771 * recovered. The rest of the data in the buffer is logged via the inodes
772 * themselves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 *
Dave Chinnerccf7c232010-05-20 23:19:42 +1000774 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
775 * transferred to the buffer's log format structure so that we'll know what to
776 * do at recovery time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778void
779xfs_trans_inode_buf(
780 xfs_trans_t *tp,
781 xfs_buf_t *bp)
782{
783 xfs_buf_log_item_t *bip;
784
785 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200786 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
788
789 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
790 ASSERT(atomic_read(&bip->bli_refcount) > 0);
791
Dave Chinnerccf7c232010-05-20 23:19:42 +1000792 bip->bli_flags |= XFS_BLI_INODE_BUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
795/*
796 * This call is used to indicate that the buffer is going to
797 * be staled and was an inode buffer. This means it gets
798 * special processing during unpin - where any inodes
799 * associated with the buffer should be removed from ail.
800 * There is also special processing during recovery,
801 * any replay of the inodes in the buffer needs to be
802 * prevented as the buffer may have been reused.
803 */
804void
805xfs_trans_stale_inode_buf(
806 xfs_trans_t *tp,
807 xfs_buf_t *bp)
808{
809 xfs_buf_log_item_t *bip;
810
811 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200812 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
814
815 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
816 ASSERT(atomic_read(&bip->bli_refcount) > 0);
817
818 bip->bli_flags |= XFS_BLI_STALE_INODE;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000819 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822/*
823 * Mark the buffer as being one which contains newly allocated
824 * inodes. We need to make sure that even if this buffer is
825 * relogged as an 'inode buf' we still recover all of the inode
826 * images in the face of a crash. This works in coordination with
827 * xfs_buf_item_committed() to ensure that the buffer remains in the
828 * AIL at its original location even after it has been relogged.
829 */
830/* ARGSUSED */
831void
832xfs_trans_inode_alloc_buf(
833 xfs_trans_t *tp,
834 xfs_buf_t *bp)
835{
836 xfs_buf_log_item_t *bip;
837
838 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200839 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
841
842 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
843 ASSERT(atomic_read(&bip->bli_refcount) > 0);
844
845 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
846}
847
848
849/*
850 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
851 * dquots. However, unlike in inode buffer recovery, dquot buffers get
852 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
853 * The only thing that makes dquot buffers different from regular
854 * buffers is that we must not replay dquot bufs when recovering
855 * if a _corresponding_ quotaoff has happened. We also have to distinguish
856 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
857 * can be turned off independently.
858 */
859/* ARGSUSED */
860void
861xfs_trans_dquot_buf(
862 xfs_trans_t *tp,
863 xfs_buf_t *bp,
864 uint type)
865{
866 xfs_buf_log_item_t *bip;
867
868 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200869 ASSERT(bp->b_transp == tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +1000871 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
872 type == XFS_BLF_PDQUOT_BUF ||
873 type == XFS_BLF_GDQUOT_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
876 ASSERT(atomic_read(&bip->bli_refcount) > 0);
877
878 bip->bli_format.blf_flags |= type;
879}