blob: 6311b99c267f69fe3ac1e7cc4d72a7681e96415d [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));
Dave Chinnerc1155412010-05-07 11:05:19 +100096 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,
260 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
262 xfs_buf_t *bp;
263 xfs_buf_log_item_t *bip;
264 int error;
265
Dave Chinner7ca790a2012-04-23 15:58:55 +1000266 *bpp = NULL;
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000267 if (!tp) {
268 bp = xfs_buf_read_map(target, map, nmaps, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 if (!bp)
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000270 return (flags & XBF_TRYLOCK) ?
David Chinnera3f74ff2008-03-06 13:43:42 +1100271 EAGAIN : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000273 if (bp->b_error) {
274 error = bp->b_error;
Christoph Hellwig901796a2011-10-10 16:52:49 +0000275 xfs_buf_ioerror_alert(bp, __func__);
Dave Chinner7ca790a2012-04-23 15:58:55 +1000276 XFS_BUF_UNDONE(bp);
277 xfs_buf_stale(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 xfs_buf_relse(bp);
279 return error;
280 }
281#ifdef DEBUG
Julia Lawalla0f7bfd2009-07-27 18:15:25 +0200282 if (xfs_do_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 if (xfs_error_target == target) {
284 if (((xfs_req_num++) % xfs_error_mod) == 0) {
285 xfs_buf_relse(bp);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100286 xfs_debug(mp, "Returning error!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return XFS_ERROR(EIO);
288 }
289 }
290 }
291#endif
292 if (XFS_FORCED_SHUTDOWN(mp))
293 goto shutdown_abort;
294 *bpp = bp;
295 return 0;
296 }
297
298 /*
299 * If we find the buffer in the cache with this transaction
300 * pointer in its b_fsprivate2 field, then we know we already
301 * have it locked. If it is already read in we just increment
302 * the lock recursion count and return the buffer to the caller.
303 * If the buffer is not yet read in, then we read it in, increment
304 * the lock recursion count, and return it to the caller.
305 */
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000306 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 if (bp != NULL) {
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200308 ASSERT(xfs_buf_islocked(bp));
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200309 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200310 ASSERT(bp->b_fspriv != NULL);
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000311 ASSERT(!bp->b_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 if (!(XFS_BUF_ISDONE(bp))) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000313 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 ASSERT(!XFS_BUF_ISASYNC(bp));
315 XFS_BUF_READ(bp);
316 xfsbdstrat(tp->t_mountp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000317 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000318 if (error) {
Christoph Hellwig901796a2011-10-10 16:52:49 +0000319 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 xfs_buf_relse(bp);
321 /*
David Chinnerd64e31a2008-04-10 12:22:17 +1000322 * We can gracefully recover from most read
323 * errors. Ones we can't are those that happen
324 * after the transaction's already dirty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 */
326 if (tp->t_flags & XFS_TRANS_DIRTY)
327 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000328 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 return error;
330 }
331 }
332 /*
333 * We never locked this buf ourselves, so we shouldn't
334 * brelse it either. Just get out.
335 */
336 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000337 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 *bpp = NULL;
339 return XFS_ERROR(EIO);
340 }
341
342
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200343 bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 bip->bli_recur++;
345
346 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000347 trace_xfs_trans_read_buf_recur(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 *bpp = bp;
349 return 0;
350 }
351
Dave Chinnerde2a4f52012-06-22 18:50:11 +1000352 bp = xfs_buf_read_map(target, map, nmaps, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 if (bp == NULL) {
354 *bpp = NULL;
Dave Chinner7401aaf2011-03-26 09:14:44 +1100355 return (flags & XBF_TRYLOCK) ?
356 0 : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 }
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000358 if (bp->b_error) {
359 error = bp->b_error;
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000360 xfs_buf_stale(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000361 XFS_BUF_DONE(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +0000362 xfs_buf_ioerror_alert(bp, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (tp->t_flags & XFS_TRANS_DIRTY)
Nathan Scott7d04a332006-06-09 14:58:38 +1000364 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 xfs_buf_relse(bp);
366 return error;
367 }
368#ifdef DEBUG
369 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
370 if (xfs_error_target == target) {
371 if (((xfs_req_num++) % xfs_error_mod) == 0) {
372 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000373 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 xfs_buf_relse(bp);
Dave Chinner0b932cc2011-03-07 10:08:35 +1100375 xfs_debug(mp, "Returning trans error!");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return XFS_ERROR(EIO);
377 }
378 }
379 }
380#endif
381 if (XFS_FORCED_SHUTDOWN(mp))
382 goto shutdown_abort;
383
Christoph Hellwigd7e84f42010-02-15 23:35:09 +0000384 _xfs_trans_bjoin(tp, bp, 1);
385 trace_xfs_trans_read_buf(bp->b_fspriv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 *bpp = bp;
388 return 0;
389
390shutdown_abort:
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000391 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 xfs_buf_relse(bp);
393 *bpp = NULL;
394 return XFS_ERROR(EIO);
395}
396
397
398/*
399 * Release the buffer bp which was previously acquired with one of the
400 * xfs_trans_... buffer allocation routines if the buffer has not
401 * been modified within this transaction. If the buffer is modified
402 * within this transaction, do decrement the recursion count but do
403 * not release the buffer even if the count goes to 0. If the buffer is not
404 * modified within the transaction, decrement the recursion count and
405 * release the buffer if the recursion count goes to 0.
406 *
407 * If the buffer is to be released and it was not modified before
408 * this transaction began, then free the buf_log_item associated with it.
409 *
410 * If the transaction pointer is NULL, make this just a normal
411 * brelse() call.
412 */
413void
414xfs_trans_brelse(xfs_trans_t *tp,
415 xfs_buf_t *bp)
416{
417 xfs_buf_log_item_t *bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 /*
420 * Default to a normal brelse() call if the tp is NULL.
421 */
422 if (tp == NULL) {
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200423 ASSERT(bp->b_transp == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 xfs_buf_relse(bp);
425 return;
426 }
427
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200428 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200429 bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
431 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Dave Chinnerc1155412010-05-07 11:05:19 +1000432 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 ASSERT(atomic_read(&bip->bli_refcount) > 0);
434
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000435 trace_xfs_trans_brelse(bip);
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 /*
438 * If the release is just for a recursive lock,
439 * then decrement the count and return.
440 */
441 if (bip->bli_recur > 0) {
442 bip->bli_recur--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 return;
444 }
445
446 /*
447 * If the buffer is dirty within this transaction, we can't
448 * release it until we commit.
449 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000450 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 /*
454 * If the buffer has been invalidated, then we can't release
455 * it until the transaction commits to disk unless it is re-dirtied
456 * as part of this transaction. This prevents us from pulling
457 * the item from the AIL before we should.
458 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000459 if (bip->bli_flags & XFS_BLI_STALE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 /*
465 * Free up the log item descriptor tracking the released item.
466 */
Christoph Hellwige98c4142010-06-23 18:11:15 +1000467 xfs_trans_del_item(&bip->bli_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 /*
470 * Clear the hold flag in the buf log item if it is set.
471 * We wouldn't want the next user of the buffer to
472 * get confused.
473 */
474 if (bip->bli_flags & XFS_BLI_HOLD) {
475 bip->bli_flags &= ~XFS_BLI_HOLD;
476 }
477
478 /*
479 * Drop our reference to the buf log item.
480 */
481 atomic_dec(&bip->bli_refcount);
482
483 /*
484 * If the buf item is not tracking data in the log, then
485 * we must free it before releasing the buffer back to the
486 * free pool. Before releasing the buffer to the free pool,
487 * clear the transaction pointer in b_fsprivate2 to dissolve
488 * its relation to this transaction.
489 */
490 if (!xfs_buf_item_dirty(bip)) {
491/***
492 ASSERT(bp->b_pincount == 0);
493***/
494 ASSERT(atomic_read(&bip->bli_refcount) == 0);
495 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
496 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
497 xfs_buf_item_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 }
Christoph Hellwig5b03ff12012-02-20 02:31:22 +0000499
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200500 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
503
504/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 * Mark the buffer as not needing to be unlocked when the buf item's
506 * IOP_UNLOCK() routine is called. The buffer must already be locked
507 * and associated with the given transaction.
508 */
509/* ARGSUSED */
510void
511xfs_trans_bhold(xfs_trans_t *tp,
512 xfs_buf_t *bp)
513{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200514 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200516 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200517 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Dave Chinnerc1155412010-05-07 11:05:19 +1000519 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 bip->bli_flags |= XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000523 trace_xfs_trans_bhold(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526/*
Tim Shimminefa092f2005-09-05 08:29:01 +1000527 * Cancel the previous buffer hold request made on this buffer
528 * for this transaction.
529 */
530void
531xfs_trans_bhold_release(xfs_trans_t *tp,
532 xfs_buf_t *bp)
533{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200534 xfs_buf_log_item_t *bip = bp->b_fspriv;
Tim Shimminefa092f2005-09-05 08:29:01 +1000535
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200536 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200537 ASSERT(bip != NULL);
Tim Shimminefa092f2005-09-05 08:29:01 +1000538 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Dave Chinnerc1155412010-05-07 11:05:19 +1000539 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
Tim Shimminefa092f2005-09-05 08:29:01 +1000540 ASSERT(atomic_read(&bip->bli_refcount) > 0);
541 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000542
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200543 bip->bli_flags &= ~XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000544 trace_xfs_trans_bhold_release(bip);
Tim Shimminefa092f2005-09-05 08:29:01 +1000545}
546
547/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 * This is called to mark bytes first through last inclusive of the given
549 * buffer as needing to be logged when the transaction is committed.
550 * The buffer must already be associated with the given transaction.
551 *
552 * First and last are numbers relative to the beginning of this buffer,
553 * so the first byte in the buffer is numbered 0 regardless of the
554 * value of b_blkno.
555 */
556void
557xfs_trans_log_buf(xfs_trans_t *tp,
558 xfs_buf_t *bp,
559 uint first,
560 uint last)
561{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200562 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200564 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200565 ASSERT(bip != NULL);
Dave Chinneraa0e8832012-04-23 15:58:52 +1000566 ASSERT(first <= last && last < BBTOB(bp->b_length));
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200567 ASSERT(bp->b_iodone == NULL ||
568 bp->b_iodone == xfs_buf_iodone_callbacks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 /*
571 * Mark the buffer as needing to be written out eventually,
572 * and set its iodone function to remove the buffer's buf log
573 * item from the AIL and free it when the buffer is flushed
574 * to disk. See xfs_buf_attach_iodone() for more details
575 * on li_cb and xfs_buf_iodone_callbacks().
576 * If we end up aborting this transaction, we trap this buffer
577 * inside the b_bdstrat callback so that this won't get written to
578 * disk.
579 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 XFS_BUF_DONE(bp);
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200583 bp->b_iodone = xfs_buf_iodone_callbacks;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000584 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000586 trace_xfs_trans_log_buf(bip);
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 /*
589 * If we invalidated the buffer within this transaction, then
590 * cancel the invalidation now that we're dirtying the buffer
591 * again. There are no races with the code in xfs_buf_item_unpin(),
592 * because we have a reference to the buffer this entire time.
593 */
594 if (bip->bli_flags & XFS_BLI_STALE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 bip->bli_flags &= ~XFS_BLI_STALE;
596 ASSERT(XFS_BUF_ISSTALE(bp));
597 XFS_BUF_UNSTALE(bp);
Dave Chinnerc1155412010-05-07 11:05:19 +1000598 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 tp->t_flags |= XFS_TRANS_DIRTY;
Christoph Hellwige98c4142010-06-23 18:11:15 +1000602 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 bip->bli_flags |= XFS_BLI_LOGGED;
604 xfs_buf_item_log(bip, first, last);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
607
608/*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000609 * Invalidate a buffer that is being used within a transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 *
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000611 * Typically this is because the blocks in the buffer are being freed, so we
612 * need to prevent it from being written out when we're done. Allowing it
613 * to be written again might overwrite data in the free blocks if they are
614 * reallocated to a file.
615 *
616 * We prevent the buffer from being written out by marking it stale. We can't
617 * get rid of the buf log item at this point because the buffer may still be
618 * pinned by another transaction. If that is the case, then we'll wait until
619 * the buffer is committed to disk for the last time (we can tell by the ref
620 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
621 * keep the buffer locked so that the buffer and buf log item are not reused.
622 *
623 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
624 * the buf item. This will be used at recovery time to determine that copies
625 * of the buffer in the log before this should not be replayed.
626 *
627 * We mark the item descriptor and the transaction dirty so that we'll hold
628 * the buffer until after the commit.
629 *
630 * Since we're invalidating the buffer, we also clear the state about which
631 * parts of the buffer have been logged. We also clear the flag indicating
632 * that this is an inode buffer since the data in the buffer will no longer
633 * be valid.
634 *
635 * We set the stale bit in the buffer as well since we're getting rid of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 */
637void
638xfs_trans_binval(
639 xfs_trans_t *tp,
640 xfs_buf_t *bp)
641{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200642 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200644 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200645 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 ASSERT(atomic_read(&bip->bli_refcount) > 0);
647
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000648 trace_xfs_trans_binval(bip);
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 if (bip->bli_flags & XFS_BLI_STALE) {
651 /*
652 * If the buffer is already invalidated, then
653 * just return.
654 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 ASSERT(XFS_BUF_ISSTALE(bp));
656 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
Dave Chinnerc1155412010-05-07 11:05:19 +1000657 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
658 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwige98c4142010-06-23 18:11:15 +1000659 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 return;
662 }
663
Christoph Hellwigc867cb62011-10-10 16:52:46 +0000664 xfs_buf_stale(bp);
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 bip->bli_flags |= XFS_BLI_STALE;
Dave Chinnerccf7c232010-05-20 23:19:42 +1000667 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
Dave Chinnerc1155412010-05-07 11:05:19 +1000668 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
669 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 memset((char *)(bip->bli_format.blf_data_map), 0,
671 (bip->bli_format.blf_map_size * sizeof(uint)));
Christoph Hellwige98c4142010-06-23 18:11:15 +1000672 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 tp->t_flags |= XFS_TRANS_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676/*
Dave Chinnerccf7c232010-05-20 23:19:42 +1000677 * This call is used to indicate that the buffer contains on-disk inodes which
678 * must be handled specially during recovery. They require special handling
679 * because only the di_next_unlinked from the inodes in the buffer should be
680 * recovered. The rest of the data in the buffer is logged via the inodes
681 * themselves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 *
Dave Chinnerccf7c232010-05-20 23:19:42 +1000683 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
684 * transferred to the buffer's log format structure so that we'll know what to
685 * do at recovery time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687void
688xfs_trans_inode_buf(
689 xfs_trans_t *tp,
690 xfs_buf_t *bp)
691{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200692 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200694 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200695 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 ASSERT(atomic_read(&bip->bli_refcount) > 0);
697
Dave Chinnerccf7c232010-05-20 23:19:42 +1000698 bip->bli_flags |= XFS_BLI_INODE_BUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699}
700
701/*
702 * This call is used to indicate that the buffer is going to
703 * be staled and was an inode buffer. This means it gets
704 * special processing during unpin - where any inodes
705 * associated with the buffer should be removed from ail.
706 * There is also special processing during recovery,
707 * any replay of the inodes in the buffer needs to be
708 * prevented as the buffer may have been reused.
709 */
710void
711xfs_trans_stale_inode_buf(
712 xfs_trans_t *tp,
713 xfs_buf_t *bp)
714{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200715 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200717 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200718 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 ASSERT(atomic_read(&bip->bli_refcount) > 0);
720
721 bip->bli_flags |= XFS_BLI_STALE_INODE;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000722 bip->bli_item.li_cb = xfs_buf_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723}
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725/*
726 * Mark the buffer as being one which contains newly allocated
727 * inodes. We need to make sure that even if this buffer is
728 * relogged as an 'inode buf' we still recover all of the inode
729 * images in the face of a crash. This works in coordination with
730 * xfs_buf_item_committed() to ensure that the buffer remains in the
731 * AIL at its original location even after it has been relogged.
732 */
733/* ARGSUSED */
734void
735xfs_trans_inode_alloc_buf(
736 xfs_trans_t *tp,
737 xfs_buf_t *bp)
738{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200739 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200741 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200742 ASSERT(bip != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 ASSERT(atomic_read(&bip->bli_refcount) > 0);
744
745 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
746}
747
748
749/*
750 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
751 * dquots. However, unlike in inode buffer recovery, dquot buffers get
752 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
753 * The only thing that makes dquot buffers different from regular
754 * buffers is that we must not replay dquot bufs when recovering
755 * if a _corresponding_ quotaoff has happened. We also have to distinguish
756 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
757 * can be turned off independently.
758 */
759/* ARGSUSED */
760void
761xfs_trans_dquot_buf(
762 xfs_trans_t *tp,
763 xfs_buf_t *bp,
764 uint type)
765{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200766 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200768 ASSERT(bp->b_transp == tp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200769 ASSERT(bip != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +1000770 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
771 type == XFS_BLF_PDQUOT_BUF ||
772 type == XFS_BLF_GDQUOT_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 ASSERT(atomic_read(&bip->bli_refcount) > 0);
774
775 bip->bli_format.blf_flags |= type;
776}