blob: 90af025e6839e69a8fb32e49e76d810b2ceb9cf9 [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));
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));
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 */
113 XFS_BUF_SET_FSPRIVATE2(bp, tp);
114
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) {
163 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 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++;
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;
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++;
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);
Nathan Scottb6574522006-06-09 15:29:40 +1000308 cmn_err(CE_DEBUG, "Returning error!\n");
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) {
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))) {
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);
David Chinnerd64e31a2008-04-10 12:22:17 +1000339 error = xfs_iowait(bp);
340 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;
386 return 0;
387 }
388 if (XFS_BUF_GETERROR(bp) != 0) {
389 XFS_BUF_SUPER_STALE(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 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)
Nathan Scott7d04a332006-06-09 14:58:38 +1000395 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 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,
Nathan Scott7d04a332006-06-09 14:58:38 +1000404 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 xfs_buf_relse(bp);
Nathan Scottb6574522006-06-09 15:29:40 +1000406 cmn_err(CE_DEBUG, "Returning trans error!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 return XFS_ERROR(EIO);
408 }
409 }
410 }
411#endif
412 if (XFS_FORCED_SHUTDOWN(mp))
413 goto shutdown_abort;
414
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000415 _xfs_trans_bjoin(tp, bp, 1);
416 trace_xfs_trans_read_buf(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 *bpp = bp;
419 return 0;
420
421shutdown_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
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000432 ASSERT((XFS_BUF_BFLAGS(bp) & (XBF_STALE|XBF_DELWRI)) !=
433 (XBF_STALE|XBF_DELWRI));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000435 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 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 */
457void
458xfs_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
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*);
David Chinner783a2f62008-10-30 17:39:58 +1100478 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
479 lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 }
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));
Dave Chinnerc1155412010-05-07 11:05:19 +1000490 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 ASSERT(atomic_read(&bip->bli_refcount) > 0);
492
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000493 trace_xfs_trans_brelse(bip);
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 /*
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--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return;
502 }
503
504 /*
505 * If the buffer is dirty within this transaction, we can't
506 * release it until we commit.
507 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000508 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
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 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000517 if (bip->bli_flags & XFS_BLI_STALE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
520 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
522 /*
523 * Free up the log item descriptor tracking the released item.
524 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000525 xfs_trans_del_item(&bip->bli_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
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) {
David Chinner783a2f62008-10-30 17:39:58 +1100565 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 (xfs_log_item_t*)bip);
567 }
568
569 xfs_buf_relse(bp);
570 return;
571}
572
573/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 * 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 */
579void
580xfs_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));
Dave Chinnerc1155412010-05-07 11:05:19 +1000591 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 ASSERT(atomic_read(&bip->bli_refcount) > 0);
593 bip->bli_flags |= XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000594 trace_xfs_trans_bhold(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
597/*
Tim Shimminefa092f2005-09-05 08:29:01 +1000598 * Cancel the previous buffer hold request made on this buffer
599 * for this transaction.
600 */
601void
602xfs_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));
Dave Chinnerc1155412010-05-07 11:05:19 +1000613 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Tim Shimminefa092f2005-09-05 08:29:01 +1000614 ASSERT(atomic_read(&bip->bli_refcount) > 0);
615 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
616 bip->bli_flags &= ~XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000617
618 trace_xfs_trans_bhold_release(bip);
Tim Shimminefa092f2005-09-05 08:29:01 +1000619}
620
621/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 * 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 */
630void
631xfs_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
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);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000661 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000663 trace_xfs_trans_log_buf(bip);
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 /*
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) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 bip->bli_flags &= ~XFS_BLI_STALE;
673 ASSERT(XFS_BUF_ISSTALE(bp));
674 XFS_BUF_UNSTALE(bp);
Dave Chinnerc1155412010-05-07 11:05:19 +1000675 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 tp->t_flags |= XFS_TRANS_DIRTY;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000679 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 bip->bli_flags |= XFS_BLI_LOGGED;
681 xfs_buf_item_log(bip, first, last);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
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 */
703void
704xfs_trans_binval(
705 xfs_trans_t *tp,
706 xfs_buf_t *bp)
707{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 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 *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 ASSERT(atomic_read(&bip->bli_refcount) > 0);
716
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000717 trace_xfs_trans_binval(bip);
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 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)));
Dave Chinnerc1155412010-05-07 11:05:19 +1000727 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
728 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwige98c4142010-06-23 18:11:15 +1000729 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 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.
Dave Chinnerc1155412010-05-07 11:05:19 +1000739 * We set the XFS_BLF_CANCEL flag in the buf log format structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 * 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;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000757 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
Dave Chinnerc1155412010-05-07 11:05:19 +1000758 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
759 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 memset((char *)(bip->bli_format.blf_data_map), 0,
761 (bip->bli_format.blf_map_size * sizeof(uint)));
Christoph Hellwige98c4142010-06-23 18:11:15 +1000762 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 tp->t_flags |= XFS_TRANS_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
766/*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000767 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 *
Dave Chinnerccf7c232010-05-20 23:19:42 +1000773 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777void
778xfs_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
Dave Chinnerccf7c232010-05-20 23:19:42 +1000791 bip->bli_flags |= XFS_BLI_INODE_BUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}
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 */
803void
804xfs_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;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000818 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821/*
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 */
830void
831xfs_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 */
859void
860xfs_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);
Dave Chinnerc1155412010-05-07 11:05:19 +1000870 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
871 type == XFS_BLF_PDQUOT_BUF ||
872 type == XFS_BLF_GDQUOT_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
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}