blob: f7510bf68284a296daca3821213a421a39018929 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_log.h"
22#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_sb.h"
24#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110026#include "xfs_bmap_btree.h"
27#include "xfs_alloc_btree.h"
28#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_dinode.h"
30#include "xfs_inode.h"
31#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_trans_priv.h"
33#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000034#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Christoph Hellwig4a5224d2010-04-18 00:10:45 +000036/*
37 * Check to see if a buffer matching the given parameters is already
38 * a part of the given transaction.
39 */
40STATIC struct xfs_buf *
41xfs_trans_buf_item_match(
42 struct xfs_trans *tp,
43 struct xfs_buftarg *target,
Dave Chinnerde2a4f52012-06-22 18:50:11 +100044 struct xfs_buf_map *map,
45 int nmaps)
Christoph Hellwig4a5224d2010-04-18 00:10:45 +000046{
Christoph Hellwige98c4142010-06-23 18:11:15 +100047 struct xfs_log_item_desc *lidp;
48 struct xfs_buf_log_item *blip;
Dave Chinnerde2a4f52012-06-22 18:50:11 +100049 int len = 0;
50 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Dave Chinnerde2a4f52012-06-22 18:50:11 +100052 for (i = 0; i < nmaps; i++)
53 len += map[i].bm_len;
54
Christoph Hellwige98c4142010-06-23 18:11:15 +100055 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
56 blip = (struct xfs_buf_log_item *)lidp->lid_item;
57 if (blip->bli_item.li_type == XFS_LI_BUF &&
Chandra Seetharaman49074c02011-07-22 23:40:40 +000058 blip->bli_buf->b_target == target &&
Dave Chinnerde2a4f52012-06-22 18:50:11 +100059 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
60 blip->bli_buf->b_length == len) {
61 ASSERT(blip->bli_buf->b_map_count == nmaps);
Christoph Hellwige98c4142010-06-23 18:11:15 +100062 return blip->bli_buf;
Dave Chinnerde2a4f52012-06-22 18:50:11 +100063 }
Christoph Hellwig4a5224d2010-04-18 00:10:45 +000064 }
65
66 return NULL;
67}
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000069/*
70 * Add the locked buffer to the transaction.
71 *
72 * The buffer must be locked, and it cannot be associated with any
73 * transaction.
74 *
75 * If the buffer does not yet have a buf log item associated with it,
76 * then allocate one for it. Then add the buf item to the transaction.
77 */
78STATIC void
79_xfs_trans_bjoin(
80 struct xfs_trans *tp,
81 struct xfs_buf *bp,
82 int reset_recur)
83{
84 struct xfs_buf_log_item *bip;
85
Christoph Hellwigbf9d9012011-07-13 13:43:49 +020086 ASSERT(bp->b_transp == NULL);
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000087
88 /*
89 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
90 * it doesn't have one yet, then allocate one and initialize it.
91 * The checks to see if one is there are in xfs_buf_item_init().
92 */
93 xfs_buf_item_init(bp, tp->t_mountp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +020094 bip = bp->b_fspriv;
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000095 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Mark Tinguely0f22f9d2012-12-04 17:18:03 -060096 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
Christoph Hellwigd7e84f42010-02-15 23:35:09 +000097 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
98 if (reset_recur)
99 bip->bli_recur = 0;
100
101 /*
102 * Take a reference for this transaction on the buf item.
103 */
104 atomic_inc(&bip->bli_refcount);
105
106 /*
107 * Get a log_item_desc to point at the new item.
108 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000109 xfs_trans_add_item(tp, &bip->bli_item);
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000110
111 /*
112 * Initialize b_fsprivate2 so we can find it with incore_match()
113 * in xfs_trans_get_buf() and friends above.
114 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200115 bp->b_transp = tp;
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000116
117}
118
119void
120xfs_trans_bjoin(
121 struct xfs_trans *tp,
122 struct xfs_buf *bp)
123{
124 _xfs_trans_bjoin(tp, bp, 0);
125 trace_xfs_trans_bjoin(bp->b_fspriv);
126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
128/*
129 * Get and lock the buffer for the caller if it is not already
130 * locked within the given transaction. If it is already locked
131 * within the transaction, just increment its lock recursion count
132 * and return a pointer to it.
133 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * If the transaction pointer is NULL, make this just a normal
135 * get_buf() call.
136 */
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000137struct xfs_buf *
138xfs_trans_get_buf_map(
139 struct xfs_trans *tp,
140 struct xfs_buftarg *target,
141 struct xfs_buf_map *map,
142 int nmaps,
143 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 xfs_buf_t *bp;
146 xfs_buf_log_item_t *bip;
147
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000148 if (!tp)
149 return xfs_buf_get_map(target, map, nmaps, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151 /*
152 * If we find the buffer in the cache with this transaction
153 * pointer in its b_fsprivate2 field, then we know we already
154 * have it locked. In this case we just increment the lock
155 * recursion count and return the buffer to the caller.
156 */
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000157 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 if (bp != NULL) {
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200159 ASSERT(xfs_buf_islocked(bp));
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000160 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
161 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000162 XFS_BUF_DONE(bp);
163 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000164
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200165 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200166 bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 ASSERT(bip != NULL);
168 ASSERT(atomic_read(&bip->bli_refcount) > 0);
169 bip->bli_recur++;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000170 trace_xfs_trans_get_buf_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 return (bp);
172 }
173
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000174 bp = xfs_buf_get_map(target, map, nmaps, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 if (bp == NULL) {
176 return NULL;
177 }
178
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000179 ASSERT(!bp->b_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000181 _xfs_trans_bjoin(tp, bp, 1);
182 trace_xfs_trans_get_buf(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 return (bp);
184}
185
186/*
187 * Get and lock the superblock buffer of this file system for the
188 * given transaction.
189 *
190 * We don't need to use incore_match() here, because the superblock
191 * buffer is a private buffer which we keep a pointer to in the
192 * mount structure.
193 */
194xfs_buf_t *
195xfs_trans_getsb(xfs_trans_t *tp,
196 struct xfs_mount *mp,
197 int flags)
198{
199 xfs_buf_t *bp;
200 xfs_buf_log_item_t *bip;
201
202 /*
203 * Default to just trying to lock the superblock buffer
204 * if tp is NULL.
205 */
206 if (tp == NULL) {
207 return (xfs_getsb(mp, flags));
208 }
209
210 /*
211 * If the superblock buffer already has this transaction
212 * pointer in its b_fsprivate2 field, then we know we already
213 * have it locked. In this case we just increment the lock
214 * recursion count and return the buffer to the caller.
215 */
216 bp = mp->m_sb_bp;
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200217 if (bp->b_transp == tp) {
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200218 bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 ASSERT(bip != NULL);
220 ASSERT(atomic_read(&bip->bli_refcount) > 0);
221 bip->bli_recur++;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000222 trace_xfs_trans_getsb_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 return (bp);
224 }
225
226 bp = xfs_getsb(mp, flags);
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000227 if (bp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000230 _xfs_trans_bjoin(tp, bp, 1);
231 trace_xfs_trans_getsb(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return (bp);
233}
234
235#ifdef DEBUG
236xfs_buftarg_t *xfs_error_target;
237int xfs_do_error;
238int xfs_req_num;
239int xfs_error_mod = 33;
240#endif
241
242/*
243 * Get and lock the buffer for the caller if it is not already
244 * locked within the given transaction. If it has not yet been
245 * read in, read it from disk. If it is already locked
246 * within the transaction and already read in, just increment its
247 * lock recursion count and return a pointer to it.
248 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 * If the transaction pointer is NULL, make this just a normal
250 * read_buf() call.
251 */
252int
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000253xfs_trans_read_buf_map(
254 struct xfs_mount *mp,
255 struct xfs_trans *tp,
256 struct xfs_buftarg *target,
257 struct xfs_buf_map *map,
258 int nmaps,
259 xfs_buf_flags_t flags,
Dave Chinnerc3f8fc72012-11-12 22:54:01 +1100260 struct xfs_buf **bpp,
Dave Chinner1813dd62012-11-14 17:54:40 +1100261 const struct xfs_buf_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
263 xfs_buf_t *bp;
264 xfs_buf_log_item_t *bip;
265 int error;
266
Dave Chinner7ca790a2012-04-23 15:58:55 +1000267 *bpp = NULL;
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000268 if (!tp) {
Dave Chinner1813dd62012-11-14 17:54:40 +1100269 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 if (!bp)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000271 return (flags & XBF_TRYLOCK) ?
David Chinnera3f74ff2008-03-06 13:43:42 +1100272 EAGAIN : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000274 if (bp->b_error) {
275 error = bp->b_error;
Christoph Hellwig901796a2011-10-10 16:52:49 +0000276 xfs_buf_ioerror_alert(bp, __func__);
Dave Chinner7ca790a2012-04-23 15:58:55 +1000277 XFS_BUF_UNDONE(bp);
278 xfs_buf_stale(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 xfs_buf_relse(bp);
280 return error;
281 }
282#ifdef DEBUG
Julia Lawalla0f7bfd2009-07-27 18:15:25 +0200283 if (xfs_do_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 if (xfs_error_target == target) {
285 if (((xfs_req_num++) % xfs_error_mod) == 0) {
286 xfs_buf_relse(bp);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100287 xfs_debug(mp, "Returning error!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return XFS_ERROR(EIO);
289 }
290 }
291 }
292#endif
293 if (XFS_FORCED_SHUTDOWN(mp))
294 goto shutdown_abort;
295 *bpp = bp;
296 return 0;
297 }
298
299 /*
300 * If we find the buffer in the cache with this transaction
301 * pointer in its b_fsprivate2 field, then we know we already
302 * have it locked. If it is already read in we just increment
303 * the lock recursion count and return the buffer to the caller.
304 * If the buffer is not yet read in, then we read it in, increment
305 * the lock recursion count, and return it to the caller.
306 */
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000307 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 if (bp != NULL) {
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200309 ASSERT(xfs_buf_islocked(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200310 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200311 ASSERT(bp->b_fspriv != NULL);
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000312 ASSERT(!bp->b_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 if (!(XFS_BUF_ISDONE(bp))) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000314 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 ASSERT(!XFS_BUF_ISASYNC(bp));
Dave Chinnerc3f8fc72012-11-12 22:54:01 +1100316 ASSERT(bp->b_iodone == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 XFS_BUF_READ(bp);
Dave Chinner1813dd62012-11-14 17:54:40 +1100318 bp->b_ops = ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 xfsbdstrat(tp->t_mountp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000320 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000321 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +0000322 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 xfs_buf_relse(bp);
324 /*
David Chinnerd64e31a2008-04-10 12:22:17 +1000325 * We can gracefully recover from most read
326 * errors. Ones we can't are those that happen
327 * after the transaction's already dirty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
329 if (tp->t_flags & XFS_TRANS_DIRTY)
330 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000331 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return error;
333 }
334 }
335 /*
336 * We never locked this buf ourselves, so we shouldn't
337 * brelse it either. Just get out.
338 */
339 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000340 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 *bpp = NULL;
342 return XFS_ERROR(EIO);
343 }
344
345
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200346 bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 bip->bli_recur++;
348
349 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000350 trace_xfs_trans_read_buf_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 *bpp = bp;
352 return 0;
353 }
354
Dave Chinner1813dd62012-11-14 17:54:40 +1100355 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 if (bp == NULL) {
357 *bpp = NULL;
Dave Chinner7401aaf2011-03-26 09:14:44 +1100358 return (flags & XBF_TRYLOCK) ?
359 0 : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000361 if (bp->b_error) {
362 error = bp->b_error;
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000363 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000364 XFS_BUF_DONE(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000365 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 if (tp->t_flags & XFS_TRANS_DIRTY)
Nathan Scott7d04a332006-06-09 14:58:38 +1000367 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 xfs_buf_relse(bp);
369 return error;
370 }
371#ifdef DEBUG
372 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
373 if (xfs_error_target == target) {
374 if (((xfs_req_num++) % xfs_error_mod) == 0) {
375 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000376 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 xfs_buf_relse(bp);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100378 xfs_debug(mp, "Returning trans error!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 return XFS_ERROR(EIO);
380 }
381 }
382 }
383#endif
384 if (XFS_FORCED_SHUTDOWN(mp))
385 goto shutdown_abort;
386
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000387 _xfs_trans_bjoin(tp, bp, 1);
388 trace_xfs_trans_read_buf(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 *bpp = bp;
391 return 0;
392
393shutdown_abort:
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000394 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 xfs_buf_relse(bp);
396 *bpp = NULL;
397 return XFS_ERROR(EIO);
398}
399
400
401/*
402 * Release the buffer bp which was previously acquired with one of the
403 * xfs_trans_... buffer allocation routines if the buffer has not
404 * been modified within this transaction. If the buffer is modified
405 * within this transaction, do decrement the recursion count but do
406 * not release the buffer even if the count goes to 0. If the buffer is not
407 * modified within the transaction, decrement the recursion count and
408 * release the buffer if the recursion count goes to 0.
409 *
410 * If the buffer is to be released and it was not modified before
411 * this transaction began, then free the buf_log_item associated with it.
412 *
413 * If the transaction pointer is NULL, make this just a normal
414 * brelse() call.
415 */
416void
417xfs_trans_brelse(xfs_trans_t *tp,
418 xfs_buf_t *bp)
419{
420 xfs_buf_log_item_t *bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422 /*
423 * Default to a normal brelse() call if the tp is NULL.
424 */
425 if (tp == NULL) {
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200426 ASSERT(bp->b_transp == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 xfs_buf_relse(bp);
428 return;
429 }
430
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200431 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200432 bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
434 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600435 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 ASSERT(atomic_read(&bip->bli_refcount) > 0);
437
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000438 trace_xfs_trans_brelse(bip);
439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 /*
441 * If the release is just for a recursive lock,
442 * then decrement the count and return.
443 */
444 if (bip->bli_recur > 0) {
445 bip->bli_recur--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 return;
447 }
448
449 /*
450 * If the buffer is dirty within this transaction, we can't
451 * release it until we commit.
452 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000453 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456 /*
457 * If the buffer has been invalidated, then we can't release
458 * it until the transaction commits to disk unless it is re-dirtied
459 * as part of this transaction. This prevents us from pulling
460 * the item from the AIL before we should.
461 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000462 if (bip->bli_flags & XFS_BLI_STALE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467 /*
468 * Free up the log item descriptor tracking the released item.
469 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000470 xfs_trans_del_item(&bip->bli_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 /*
473 * Clear the hold flag in the buf log item if it is set.
474 * We wouldn't want the next user of the buffer to
475 * get confused.
476 */
477 if (bip->bli_flags & XFS_BLI_HOLD) {
478 bip->bli_flags &= ~XFS_BLI_HOLD;
479 }
480
481 /*
482 * Drop our reference to the buf log item.
483 */
484 atomic_dec(&bip->bli_refcount);
485
486 /*
487 * If the buf item is not tracking data in the log, then
488 * we must free it before releasing the buffer back to the
489 * free pool. Before releasing the buffer to the free pool,
490 * clear the transaction pointer in b_fsprivate2 to dissolve
491 * its relation to this transaction.
492 */
493 if (!xfs_buf_item_dirty(bip)) {
494/***
495 ASSERT(bp->b_pincount == 0);
496***/
497 ASSERT(atomic_read(&bip->bli_refcount) == 0);
498 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
499 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
500 xfs_buf_item_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 }
Christoph Hellwig5b03ff12012-02-20 02:31:22 +0000502
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200503 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
507/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 * Mark the buffer as not needing to be unlocked when the buf item's
509 * IOP_UNLOCK() routine is called. The buffer must already be locked
510 * and associated with the given transaction.
511 */
512/* ARGSUSED */
513void
514xfs_trans_bhold(xfs_trans_t *tp,
515 xfs_buf_t *bp)
516{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200517 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200519 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200520 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600522 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 bip->bli_flags |= XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000526 trace_xfs_trans_bhold(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527}
528
529/*
Tim Shimminefa092f2005-09-05 08:29:01 +1000530 * Cancel the previous buffer hold request made on this buffer
531 * for this transaction.
532 */
533void
534xfs_trans_bhold_release(xfs_trans_t *tp,
535 xfs_buf_t *bp)
536{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200537 xfs_buf_log_item_t *bip = bp->b_fspriv;
Tim Shimminefa092f2005-09-05 08:29:01 +1000538
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200539 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200540 ASSERT(bip != NULL);
Tim Shimminefa092f2005-09-05 08:29:01 +1000541 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600542 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
Tim Shimminefa092f2005-09-05 08:29:01 +1000543 ASSERT(atomic_read(&bip->bli_refcount) > 0);
544 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000545
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200546 bip->bli_flags &= ~XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000547 trace_xfs_trans_bhold_release(bip);
Tim Shimminefa092f2005-09-05 08:29:01 +1000548}
549
550/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 * This is called to mark bytes first through last inclusive of the given
552 * buffer as needing to be logged when the transaction is committed.
553 * The buffer must already be associated with the given transaction.
554 *
555 * First and last are numbers relative to the beginning of this buffer,
556 * so the first byte in the buffer is numbered 0 regardless of the
557 * value of b_blkno.
558 */
559void
560xfs_trans_log_buf(xfs_trans_t *tp,
561 xfs_buf_t *bp,
562 uint first,
563 uint last)
564{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200565 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200567 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200568 ASSERT(bip != NULL);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000569 ASSERT(first <= last && last < BBTOB(bp->b_length));
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200570 ASSERT(bp->b_iodone == NULL ||
571 bp->b_iodone == xfs_buf_iodone_callbacks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 /*
574 * Mark the buffer as needing to be written out eventually,
575 * and set its iodone function to remove the buffer's buf log
576 * item from the AIL and free it when the buffer is flushed
577 * to disk. See xfs_buf_attach_iodone() for more details
578 * on li_cb and xfs_buf_iodone_callbacks().
579 * If we end up aborting this transaction, we trap this buffer
580 * inside the b_bdstrat callback so that this won't get written to
581 * disk.
582 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 XFS_BUF_DONE(bp);
584
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200586 bp->b_iodone = xfs_buf_iodone_callbacks;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000587 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000589 trace_xfs_trans_log_buf(bip);
590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 /*
592 * If we invalidated the buffer within this transaction, then
593 * cancel the invalidation now that we're dirtying the buffer
594 * again. There are no races with the code in xfs_buf_item_unpin(),
595 * because we have a reference to the buffer this entire time.
596 */
597 if (bip->bli_flags & XFS_BLI_STALE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 bip->bli_flags &= ~XFS_BLI_STALE;
599 ASSERT(XFS_BUF_ISSTALE(bp));
600 XFS_BUF_UNSTALE(bp);
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600601 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 }
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 tp->t_flags |= XFS_TRANS_DIRTY;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000605 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 bip->bli_flags |= XFS_BLI_LOGGED;
607 xfs_buf_item_log(bip, first, last);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
609
610
611/*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000612 * Invalidate a buffer that is being used within a transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 *
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000614 * Typically this is because the blocks in the buffer are being freed, so we
615 * need to prevent it from being written out when we're done. Allowing it
616 * to be written again might overwrite data in the free blocks if they are
617 * reallocated to a file.
618 *
619 * We prevent the buffer from being written out by marking it stale. We can't
620 * get rid of the buf log item at this point because the buffer may still be
621 * pinned by another transaction. If that is the case, then we'll wait until
622 * the buffer is committed to disk for the last time (we can tell by the ref
623 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
624 * keep the buffer locked so that the buffer and buf log item are not reused.
625 *
626 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
627 * the buf item. This will be used at recovery time to determine that copies
628 * of the buffer in the log before this should not be replayed.
629 *
630 * We mark the item descriptor and the transaction dirty so that we'll hold
631 * the buffer until after the commit.
632 *
633 * Since we're invalidating the buffer, we also clear the state about which
634 * parts of the buffer have been logged. We also clear the flag indicating
635 * that this is an inode buffer since the data in the buffer will no longer
636 * be valid.
637 *
638 * We set the stale bit in the buffer as well since we're getting rid of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 */
640void
641xfs_trans_binval(
642 xfs_trans_t *tp,
643 xfs_buf_t *bp)
644{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200645 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200647 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200648 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 ASSERT(atomic_read(&bip->bli_refcount) > 0);
650
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000651 trace_xfs_trans_binval(bip);
652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (bip->bli_flags & XFS_BLI_STALE) {
654 /*
655 * If the buffer is already invalidated, then
656 * just return.
657 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 ASSERT(XFS_BUF_ISSTALE(bp));
659 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600660 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
661 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwige98c4142010-06-23 18:11:15 +1000662 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return;
665 }
666
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000667 xfs_buf_stale(bp);
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 bip->bli_flags |= XFS_BLI_STALE;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000670 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600671 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
672 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
673 memset((char *)(bip->__bli_format.blf_data_map), 0,
674 (bip->__bli_format.blf_map_size * sizeof(uint)));
Christoph Hellwige98c4142010-06-23 18:11:15 +1000675 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 tp->t_flags |= XFS_TRANS_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677}
678
679/*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000680 * This call is used to indicate that the buffer contains on-disk inodes which
681 * must be handled specially during recovery. They require special handling
682 * because only the di_next_unlinked from the inodes in the buffer should be
683 * recovered. The rest of the data in the buffer is logged via the inodes
684 * themselves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 *
Dave Chinnerccf7c232010-05-20 23:19:42 +1000686 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
687 * transferred to the buffer's log format structure so that we'll know what to
688 * do at recovery time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690void
691xfs_trans_inode_buf(
692 xfs_trans_t *tp,
693 xfs_buf_t *bp)
694{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200695 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200697 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200698 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 ASSERT(atomic_read(&bip->bli_refcount) > 0);
700
Dave Chinnerccf7c232010-05-20 23:19:42 +1000701 bip->bli_flags |= XFS_BLI_INODE_BUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
704/*
705 * This call is used to indicate that the buffer is going to
706 * be staled and was an inode buffer. This means it gets
707 * special processing during unpin - where any inodes
708 * associated with the buffer should be removed from ail.
709 * There is also special processing during recovery,
710 * any replay of the inodes in the buffer needs to be
711 * prevented as the buffer may have been reused.
712 */
713void
714xfs_trans_stale_inode_buf(
715 xfs_trans_t *tp,
716 xfs_buf_t *bp)
717{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200718 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200720 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200721 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 ASSERT(atomic_read(&bip->bli_refcount) > 0);
723
724 bip->bli_flags |= XFS_BLI_STALE_INODE;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000725 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728/*
729 * Mark the buffer as being one which contains newly allocated
730 * inodes. We need to make sure that even if this buffer is
731 * relogged as an 'inode buf' we still recover all of the inode
732 * images in the face of a crash. This works in coordination with
733 * xfs_buf_item_committed() to ensure that the buffer remains in the
734 * AIL at its original location even after it has been relogged.
735 */
736/* ARGSUSED */
737void
738xfs_trans_inode_alloc_buf(
739 xfs_trans_t *tp,
740 xfs_buf_t *bp)
741{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200742 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200744 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200745 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 ASSERT(atomic_read(&bip->bli_refcount) > 0);
747
748 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
749}
750
751
752/*
753 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
754 * dquots. However, unlike in inode buffer recovery, dquot buffers get
755 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
756 * The only thing that makes dquot buffers different from regular
757 * buffers is that we must not replay dquot bufs when recovering
758 * if a _corresponding_ quotaoff has happened. We also have to distinguish
759 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
760 * can be turned off independently.
761 */
762/* ARGSUSED */
763void
764xfs_trans_dquot_buf(
765 xfs_trans_t *tp,
766 xfs_buf_t *bp,
767 uint type)
768{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200769 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200771 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200772 ASSERT(bip != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +1000773 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
774 type == XFS_BLF_PDQUOT_BUF ||
775 type == XFS_BLF_GDQUOT_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 ASSERT(atomic_read(&bip->bli_refcount) > 0);
777
Mark Tinguely0f22f9d2012-12-04 17:18:03 -0600778 bip->__bli_format.blf_flags |= type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779}