blob: abaafdbb3e658e9c484fb82dcfd5b63afade43e9 [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"
Nathan Scotta844f452005-11-02 14:38:42 +110026#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_inode_item.h"
David Chinnerdb7a19f2008-04-10 12:22:24 +100033#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000034#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36
37kmem_zone_t *xfs_ili_zone; /* inode log item zone */
38
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100039static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
40{
41 return container_of(lip, struct xfs_inode_log_item, ili_item);
42}
43
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*
46 * This returns the number of iovecs needed to log the given inode item.
47 *
48 * We need one iovec for the inode log format structure, one for the
49 * inode core, and possibly one for the inode data/extents/b-tree root
50 * and one for the inode attribute data/extents/b-tree root.
51 */
52STATIC uint
53xfs_inode_item_size(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100054 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -070055{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100056 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
57 struct xfs_inode *ip = iip->ili_inode;
58 uint nvecs = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60 /*
61 * Only log the data/extents/b-tree root if there is something
62 * left to log.
63 */
64 iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
65
66 switch (ip->i_d.di_format) {
67 case XFS_DINODE_FMT_EXTENTS:
68 iip->ili_format.ilf_fields &=
69 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
70 XFS_ILOG_DEV | XFS_ILOG_UUID);
71 if ((iip->ili_format.ilf_fields & XFS_ILOG_DEXT) &&
72 (ip->i_d.di_nextents > 0) &&
73 (ip->i_df.if_bytes > 0)) {
74 ASSERT(ip->i_df.if_u1.if_extents != NULL);
75 nvecs++;
76 } else {
77 iip->ili_format.ilf_fields &= ~XFS_ILOG_DEXT;
78 }
79 break;
80
81 case XFS_DINODE_FMT_BTREE:
82 ASSERT(ip->i_df.if_ext_max ==
83 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
84 iip->ili_format.ilf_fields &=
85 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
86 XFS_ILOG_DEV | XFS_ILOG_UUID);
87 if ((iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) &&
88 (ip->i_df.if_broot_bytes > 0)) {
89 ASSERT(ip->i_df.if_broot != NULL);
90 nvecs++;
91 } else {
92 ASSERT(!(iip->ili_format.ilf_fields &
93 XFS_ILOG_DBROOT));
94#ifdef XFS_TRANS_DEBUG
95 if (iip->ili_root_size > 0) {
96 ASSERT(iip->ili_root_size ==
97 ip->i_df.if_broot_bytes);
98 ASSERT(memcmp(iip->ili_orig_root,
99 ip->i_df.if_broot,
100 iip->ili_root_size) == 0);
101 } else {
102 ASSERT(ip->i_df.if_broot_bytes == 0);
103 }
104#endif
105 iip->ili_format.ilf_fields &= ~XFS_ILOG_DBROOT;
106 }
107 break;
108
109 case XFS_DINODE_FMT_LOCAL:
110 iip->ili_format.ilf_fields &=
111 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
112 XFS_ILOG_DEV | XFS_ILOG_UUID);
113 if ((iip->ili_format.ilf_fields & XFS_ILOG_DDATA) &&
114 (ip->i_df.if_bytes > 0)) {
115 ASSERT(ip->i_df.if_u1.if_data != NULL);
116 ASSERT(ip->i_d.di_size > 0);
117 nvecs++;
118 } else {
119 iip->ili_format.ilf_fields &= ~XFS_ILOG_DDATA;
120 }
121 break;
122
123 case XFS_DINODE_FMT_DEV:
124 iip->ili_format.ilf_fields &=
125 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
126 XFS_ILOG_DEXT | XFS_ILOG_UUID);
127 break;
128
129 case XFS_DINODE_FMT_UUID:
130 iip->ili_format.ilf_fields &=
131 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
132 XFS_ILOG_DEXT | XFS_ILOG_DEV);
133 break;
134
135 default:
136 ASSERT(0);
137 break;
138 }
139
140 /*
141 * If there are no attributes associated with this file,
142 * then there cannot be anything more to log.
143 * Clear all attribute-related log flags.
144 */
145 if (!XFS_IFORK_Q(ip)) {
146 iip->ili_format.ilf_fields &=
147 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
148 return nvecs;
149 }
150
151 /*
152 * Log any necessary attribute data.
153 */
154 switch (ip->i_d.di_aformat) {
155 case XFS_DINODE_FMT_EXTENTS:
156 iip->ili_format.ilf_fields &=
157 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
158 if ((iip->ili_format.ilf_fields & XFS_ILOG_AEXT) &&
159 (ip->i_d.di_anextents > 0) &&
160 (ip->i_afp->if_bytes > 0)) {
161 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
162 nvecs++;
163 } else {
164 iip->ili_format.ilf_fields &= ~XFS_ILOG_AEXT;
165 }
166 break;
167
168 case XFS_DINODE_FMT_BTREE:
169 iip->ili_format.ilf_fields &=
170 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
171 if ((iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) &&
172 (ip->i_afp->if_broot_bytes > 0)) {
173 ASSERT(ip->i_afp->if_broot != NULL);
174 nvecs++;
175 } else {
176 iip->ili_format.ilf_fields &= ~XFS_ILOG_ABROOT;
177 }
178 break;
179
180 case XFS_DINODE_FMT_LOCAL:
181 iip->ili_format.ilf_fields &=
182 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
183 if ((iip->ili_format.ilf_fields & XFS_ILOG_ADATA) &&
184 (ip->i_afp->if_bytes > 0)) {
185 ASSERT(ip->i_afp->if_u1.if_data != NULL);
186 nvecs++;
187 } else {
188 iip->ili_format.ilf_fields &= ~XFS_ILOG_ADATA;
189 }
190 break;
191
192 default:
193 ASSERT(0);
194 break;
195 }
196
197 return nvecs;
198}
199
200/*
Dave Chinnere8287762011-04-08 12:45:07 +1000201 * xfs_inode_item_format_extents - convert in-core extents to on-disk form
202 *
203 * For either the data or attr fork in extent format, we need to endian convert
204 * the in-core extent as we place them into the on-disk inode. In this case, we
205 * need to do this conversion before we write the extents into the log. Because
206 * we don't have the disk inode to write into here, we allocate a buffer and
207 * format the extents into it via xfs_iextents_copy(). We free the buffer in
208 * the unlock routine after the copy for the log has been made.
209 *
210 * In the case of the data fork, the in-core and on-disk fork sizes can be
211 * different due to delayed allocation extents. We only log on-disk extents
212 * here, so always use the physical fork size to determine the size of the
213 * buffer we need to allocate.
214 */
215STATIC void
216xfs_inode_item_format_extents(
217 struct xfs_inode *ip,
218 struct xfs_log_iovec *vecp,
219 int whichfork,
220 int type)
221{
222 xfs_bmbt_rec_t *ext_buffer;
223
224 ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP);
225 if (whichfork == XFS_DATA_FORK)
226 ip->i_itemp->ili_extents_buf = ext_buffer;
227 else
228 ip->i_itemp->ili_aextents_buf = ext_buffer;
229
230 vecp->i_addr = ext_buffer;
231 vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork);
232 vecp->i_type = type;
233}
234
235/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 * This is called to fill in the vector of log iovecs for the
237 * given inode log item. It fills the first item with an inode
238 * log format structure, the second with the on-disk inode structure,
239 * and a possible third and/or fourth with the inode data/extents/b-tree
240 * root and inode attributes data/extents/b-tree root.
241 */
242STATIC void
243xfs_inode_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000244 struct xfs_log_item *lip,
245 struct xfs_log_iovec *vecp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000247 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
248 struct xfs_inode *ip = iip->ili_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 uint nvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 size_t data_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 xfs_mount_t *mp;
252
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000253 vecp->i_addr = &iip->ili_format;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 vecp->i_len = sizeof(xfs_inode_log_format_t);
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000255 vecp->i_type = XLOG_REG_TYPE_IFORMAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 vecp++;
257 nvecs = 1;
258
259 /*
260 * Clear i_update_core if the timestamps (or any other
261 * non-transactional modification) need flushing/logging
262 * and we're about to log them with the rest of the core.
263 *
264 * This is the same logic as xfs_iflush() but this code can't
265 * run at the same time as xfs_iflush because we're in commit
266 * processing here and so we have the inode lock held in
267 * exclusive mode. Although it doesn't really matter
268 * for the timestamps if both routines were to grab the
269 * timestamps or not. That would be ok.
270 *
271 * We clear i_update_core before copying out the data.
272 * This is for coordination with our timestamp updates
273 * that don't hold the inode lock. They will always
274 * update the timestamps BEFORE setting i_update_core,
275 * so if we clear i_update_core after they set it we
276 * are guaranteed to see their updates to the timestamps
277 * either here. Likewise, if they set it after we clear it
278 * here, we'll see it either on the next commit of this
279 * inode or the next time the inode gets flushed via
280 * xfs_iflush(). This depends on strongly ordered memory
281 * semantics, but we have that. We use the SYNCHRONIZE
282 * macro to make sure that the compiler does not reorder
283 * the i_update_core access below the data copy below.
284 */
285 if (ip->i_update_core) {
286 ip->i_update_core = 0;
287 SYNCHRONIZE();
288 }
289
290 /*
Christoph Hellwigf9581b12009-10-06 20:29:26 +0000291 * Make sure to get the latest timestamps from the Linux inode.
Christoph Hellwig42fe2b12006-01-11 15:35:17 +1100292 */
Christoph Hellwigf9581b12009-10-06 20:29:26 +0000293 xfs_synchronize_times(ip);
David Chinner5d51eff2007-11-23 16:29:18 +1100294
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000295 vecp->i_addr = &ip->i_d;
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100296 vecp->i_len = sizeof(struct xfs_icdinode);
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000297 vecp->i_type = XLOG_REG_TYPE_ICORE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 vecp++;
299 nvecs++;
300 iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
301
302 /*
303 * If this is really an old format inode, then we need to
304 * log it as such. This means that we have to copy the link
305 * count from the new field to the old. We don't have to worry
306 * about the new fields, because nothing trusts them as long as
307 * the old inode version number is there. If the superblock already
308 * has a new version number, then we don't bother converting back.
309 */
310 mp = ip->i_mount;
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100311 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
312 if (ip->i_d.di_version == 1) {
Eric Sandeen62118702008-03-06 13:44:28 +1100313 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 /*
315 * Convert it back.
316 */
317 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
318 ip->i_d.di_onlink = ip->i_d.di_nlink;
319 } else {
320 /*
321 * The superblock version has already been bumped,
322 * so just make the conversion to the new inode
323 * format permanent.
324 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100325 ip->i_d.di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 ip->i_d.di_onlink = 0;
327 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
328 }
329 }
330
331 switch (ip->i_d.di_format) {
332 case XFS_DINODE_FMT_EXTENTS:
333 ASSERT(!(iip->ili_format.ilf_fields &
334 (XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
335 XFS_ILOG_DEV | XFS_ILOG_UUID)));
336 if (iip->ili_format.ilf_fields & XFS_ILOG_DEXT) {
337 ASSERT(ip->i_df.if_bytes > 0);
338 ASSERT(ip->i_df.if_u1.if_extents != NULL);
339 ASSERT(ip->i_d.di_nextents > 0);
340 ASSERT(iip->ili_extents_buf == NULL);
Christoph Hellwig73523a22010-07-20 17:54:45 +1000341 ASSERT((ip->i_df.if_bytes /
342 (uint)sizeof(xfs_bmbt_rec_t)) > 0);
Nathan Scottf016bad2005-09-08 15:30:05 +1000343#ifdef XFS_NATIVE_HOST
Dave Chinner696123f2010-07-26 13:51:46 -0500344 if (ip->i_d.di_nextents == ip->i_df.if_bytes /
345 (uint)sizeof(xfs_bmbt_rec_t)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 /*
347 * There are no delayed allocation
348 * extents, so just point to the
349 * real extents array.
350 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000351 vecp->i_addr = ip->i_df.if_u1.if_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 vecp->i_len = ip->i_df.if_bytes;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000353 vecp->i_type = XLOG_REG_TYPE_IEXT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 } else
355#endif
356 {
Dave Chinnere8287762011-04-08 12:45:07 +1000357 xfs_inode_item_format_extents(ip, vecp,
358 XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
360 ASSERT(vecp->i_len <= ip->i_df.if_bytes);
361 iip->ili_format.ilf_dsize = vecp->i_len;
362 vecp++;
363 nvecs++;
364 }
365 break;
366
367 case XFS_DINODE_FMT_BTREE:
368 ASSERT(!(iip->ili_format.ilf_fields &
369 (XFS_ILOG_DDATA | XFS_ILOG_DEXT |
370 XFS_ILOG_DEV | XFS_ILOG_UUID)));
371 if (iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) {
372 ASSERT(ip->i_df.if_broot_bytes > 0);
373 ASSERT(ip->i_df.if_broot != NULL);
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000374 vecp->i_addr = ip->i_df.if_broot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 vecp->i_len = ip->i_df.if_broot_bytes;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000376 vecp->i_type = XLOG_REG_TYPE_IBROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 vecp++;
378 nvecs++;
379 iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
380 }
381 break;
382
383 case XFS_DINODE_FMT_LOCAL:
384 ASSERT(!(iip->ili_format.ilf_fields &
385 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
386 XFS_ILOG_DEV | XFS_ILOG_UUID)));
387 if (iip->ili_format.ilf_fields & XFS_ILOG_DDATA) {
388 ASSERT(ip->i_df.if_bytes > 0);
389 ASSERT(ip->i_df.if_u1.if_data != NULL);
390 ASSERT(ip->i_d.di_size > 0);
391
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000392 vecp->i_addr = ip->i_df.if_u1.if_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 /*
394 * Round i_bytes up to a word boundary.
395 * The underlying memory is guaranteed to
396 * to be there by xfs_idata_realloc().
397 */
398 data_bytes = roundup(ip->i_df.if_bytes, 4);
399 ASSERT((ip->i_df.if_real_bytes == 0) ||
400 (ip->i_df.if_real_bytes == data_bytes));
401 vecp->i_len = (int)data_bytes;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000402 vecp->i_type = XLOG_REG_TYPE_ILOCAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 vecp++;
404 nvecs++;
405 iip->ili_format.ilf_dsize = (unsigned)data_bytes;
406 }
407 break;
408
409 case XFS_DINODE_FMT_DEV:
410 ASSERT(!(iip->ili_format.ilf_fields &
411 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
412 XFS_ILOG_DDATA | XFS_ILOG_UUID)));
413 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
414 iip->ili_format.ilf_u.ilfu_rdev =
415 ip->i_df.if_u2.if_rdev;
416 }
417 break;
418
419 case XFS_DINODE_FMT_UUID:
420 ASSERT(!(iip->ili_format.ilf_fields &
421 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
422 XFS_ILOG_DDATA | XFS_ILOG_DEV)));
423 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
424 iip->ili_format.ilf_u.ilfu_uuid =
425 ip->i_df.if_u2.if_uuid;
426 }
427 break;
428
429 default:
430 ASSERT(0);
431 break;
432 }
433
434 /*
435 * If there are no attributes associated with the file,
436 * then we're done.
437 * Assert that no attribute-related log flags are set.
438 */
439 if (!XFS_IFORK_Q(ip)) {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000440 ASSERT(nvecs == lip->li_desc->lid_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 iip->ili_format.ilf_size = nvecs;
442 ASSERT(!(iip->ili_format.ilf_fields &
443 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
444 return;
445 }
446
447 switch (ip->i_d.di_aformat) {
448 case XFS_DINODE_FMT_EXTENTS:
449 ASSERT(!(iip->ili_format.ilf_fields &
450 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT)));
451 if (iip->ili_format.ilf_fields & XFS_ILOG_AEXT) {
Christoph Hellwig73523a22010-07-20 17:54:45 +1000452#ifdef DEBUG
453 int nrecs = ip->i_afp->if_bytes /
454 (uint)sizeof(xfs_bmbt_rec_t);
455 ASSERT(nrecs > 0);
456 ASSERT(nrecs == ip->i_d.di_anextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 ASSERT(ip->i_afp->if_bytes > 0);
458 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
459 ASSERT(ip->i_d.di_anextents > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460#endif
Nathan Scottf016bad2005-09-08 15:30:05 +1000461#ifdef XFS_NATIVE_HOST
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 /*
463 * There are not delayed allocation extents
464 * for attributes, so just point at the array.
465 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000466 vecp->i_addr = ip->i_afp->if_u1.if_extents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 vecp->i_len = ip->i_afp->if_bytes;
Dave Chinnere8287762011-04-08 12:45:07 +1000468 vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469#else
470 ASSERT(iip->ili_aextents_buf == NULL);
Dave Chinnere8287762011-04-08 12:45:07 +1000471 xfs_inode_item_format_extents(ip, vecp,
472 XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473#endif
474 iip->ili_format.ilf_asize = vecp->i_len;
475 vecp++;
476 nvecs++;
477 }
478 break;
479
480 case XFS_DINODE_FMT_BTREE:
481 ASSERT(!(iip->ili_format.ilf_fields &
482 (XFS_ILOG_ADATA | XFS_ILOG_AEXT)));
483 if (iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) {
484 ASSERT(ip->i_afp->if_broot_bytes > 0);
485 ASSERT(ip->i_afp->if_broot != NULL);
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000486 vecp->i_addr = ip->i_afp->if_broot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 vecp->i_len = ip->i_afp->if_broot_bytes;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000488 vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 vecp++;
490 nvecs++;
491 iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
492 }
493 break;
494
495 case XFS_DINODE_FMT_LOCAL:
496 ASSERT(!(iip->ili_format.ilf_fields &
497 (XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
498 if (iip->ili_format.ilf_fields & XFS_ILOG_ADATA) {
499 ASSERT(ip->i_afp->if_bytes > 0);
500 ASSERT(ip->i_afp->if_u1.if_data != NULL);
501
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000502 vecp->i_addr = ip->i_afp->if_u1.if_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 /*
504 * Round i_bytes up to a word boundary.
505 * The underlying memory is guaranteed to
506 * to be there by xfs_idata_realloc().
507 */
508 data_bytes = roundup(ip->i_afp->if_bytes, 4);
509 ASSERT((ip->i_afp->if_real_bytes == 0) ||
510 (ip->i_afp->if_real_bytes == data_bytes));
511 vecp->i_len = (int)data_bytes;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000512 vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 vecp++;
514 nvecs++;
515 iip->ili_format.ilf_asize = (unsigned)data_bytes;
516 }
517 break;
518
519 default:
520 ASSERT(0);
521 break;
522 }
523
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000524 ASSERT(nvecs == lip->li_desc->lid_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 iip->ili_format.ilf_size = nvecs;
526}
527
528
529/*
530 * This is called to pin the inode associated with the inode log
Christoph Hellwiga14a5ab2010-02-18 12:43:22 +0000531 * item in memory so it cannot be written out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 */
533STATIC void
534xfs_inode_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000535 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000537 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
Christoph Hellwiga14a5ab2010-02-18 12:43:22 +0000538
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000539 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
540
541 trace_xfs_inode_pin(ip, _RET_IP_);
542 atomic_inc(&ip->i_pincount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543}
544
545
546/*
547 * This is called to unpin the inode associated with the inode log
548 * item which was previously pinned with a call to xfs_inode_item_pin().
Christoph Hellwiga14a5ab2010-02-18 12:43:22 +0000549 *
550 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552STATIC void
553xfs_inode_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000554 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000555 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000557 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
Christoph Hellwiga14a5ab2010-02-18 12:43:22 +0000558
Dave Chinner4aaf15d2010-03-08 11:24:07 +1100559 trace_xfs_inode_unpin(ip, _RET_IP_);
Christoph Hellwiga14a5ab2010-02-18 12:43:22 +0000560 ASSERT(atomic_read(&ip->i_pincount) > 0);
561 if (atomic_dec_and_test(&ip->i_pincount))
562 wake_up(&ip->i_ipin_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563}
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565/*
566 * This is called to attempt to lock the inode associated with this
567 * inode log item, in preparation for the push routine which does the actual
568 * iflush. Don't sleep on the inode lock or the flush lock.
569 *
570 * If the flush lock is already held, indicating that the inode has
571 * been or is in the process of being flushed, then (ideally) we'd like to
572 * see if the inode's buffer is still incore, and if so give it a nudge.
573 * We delay doing so until the pushbuf routine, though, to avoid holding
Nathan Scottc41564b2006-03-29 08:55:14 +1000574 * the AIL lock across a call to the blackhole which is the buffer cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 * Also we don't want to sleep in any device strategy routines, which can happen
576 * if we do the subsequent bawrite in here.
577 */
578STATIC uint
579xfs_inode_item_trylock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000580 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000582 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
583 struct xfs_inode *ip = iip->ili_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000585 if (xfs_ipincount(ip) > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return XFS_ITEM_PINNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000588 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return XFS_ITEM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 if (!xfs_iflock_nowait(ip)) {
592 /*
Dave Chinnerd808f612010-02-02 10:13:42 +1100593 * inode has already been flushed to the backing buffer,
594 * leave it locked in shared mode, pushbuf routine will
595 * unlock it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 */
Dave Chinnerd808f612010-02-02 10:13:42 +1100597 return XFS_ITEM_PUSHBUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
599
600 /* Stale items should force out the iclog */
601 if (ip->i_flags & XFS_ISTALE) {
602 xfs_ifunlock(ip);
Dave Chinnerd808f612010-02-02 10:13:42 +1100603 /*
604 * we hold the AIL lock - notify the unlock routine of this
605 * so it doesn't try to get the lock again.
606 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
608 return XFS_ITEM_PINNED;
609 }
610
611#ifdef DEBUG
612 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
613 ASSERT(iip->ili_format.ilf_fields != 0);
614 ASSERT(iip->ili_logged == 0);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000615 ASSERT(lip->li_flags & XFS_LI_IN_AIL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
617#endif
618 return XFS_ITEM_SUCCESS;
619}
620
621/*
622 * Unlock the inode associated with the inode log item.
623 * Clear the fields of the inode and inode log item that
624 * are specific to the current transaction. If the
625 * hold flags is set, do not unlock the inode.
626 */
627STATIC void
628xfs_inode_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000629 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000631 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
632 struct xfs_inode *ip = iip->ili_inode;
Christoph Hellwig898621d2010-06-24 11:36:58 +1000633 unsigned short lock_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Christoph Hellwigf3ca87382011-07-08 14:34:47 +0200635 ASSERT(ip->i_itemp != NULL);
636 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
638 /*
639 * If the inode needed a separate buffer with which to log
640 * its extents, then free it now.
641 */
642 if (iip->ili_extents_buf != NULL) {
643 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
644 ASSERT(ip->i_d.di_nextents > 0);
645 ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_DEXT);
646 ASSERT(ip->i_df.if_bytes > 0);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000647 kmem_free(iip->ili_extents_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 iip->ili_extents_buf = NULL;
649 }
650 if (iip->ili_aextents_buf != NULL) {
651 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
652 ASSERT(ip->i_d.di_anextents > 0);
653 ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_AEXT);
654 ASSERT(ip->i_afp->if_bytes > 0);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000655 kmem_free(iip->ili_aextents_buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 iip->ili_aextents_buf = NULL;
657 }
658
Christoph Hellwig898621d2010-06-24 11:36:58 +1000659 lock_flags = iip->ili_lock_flags;
660 iip->ili_lock_flags = 0;
Christoph Hellwigddc34152011-09-19 15:00:54 +0000661 if (lock_flags)
Christoph Hellwigf3ca87382011-07-08 14:34:47 +0200662 xfs_iunlock(ip, lock_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/*
Dave Chinnerde25c182010-11-30 15:15:46 +1100666 * This is called to find out where the oldest active copy of the inode log
667 * item in the on disk log resides now that the last log write of it completed
668 * at the given lsn. Since we always re-log all dirty data in an inode, the
669 * latest copy in the on disk log is the only one that matters. Therefore,
670 * simply return the given lsn.
671 *
672 * If the inode has been marked stale because the cluster is being freed, we
673 * don't want to (re-)insert this inode into the AIL. There is a race condition
674 * where the cluster buffer may be unpinned before the inode is inserted into
675 * the AIL during transaction committed processing. If the buffer is unpinned
676 * before the inode item has been committed and inserted, then it is possible
Dave Chinner1316d4d2011-07-04 05:27:36 +0000677 * for the buffer to be written and IO completes before the inode is inserted
Dave Chinnerde25c182010-11-30 15:15:46 +1100678 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
679 * AIL which will never get removed. It will, however, get reclaimed which
680 * triggers an assert in xfs_inode_free() complaining about freein an inode
681 * still in the AIL.
682 *
Dave Chinner1316d4d2011-07-04 05:27:36 +0000683 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
684 * transaction committed code knows that it does not need to do any further
685 * processing on the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687STATIC xfs_lsn_t
688xfs_inode_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000689 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 xfs_lsn_t lsn)
691{
Dave Chinnerde25c182010-11-30 15:15:46 +1100692 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
693 struct xfs_inode *ip = iip->ili_inode;
694
Dave Chinner1316d4d2011-07-04 05:27:36 +0000695 if (xfs_iflags_test(ip, XFS_ISTALE)) {
696 xfs_inode_item_unpin(lip, 0);
697 return -1;
698 }
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000699 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
702/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
704 * failed to get the inode flush lock but did get the inode locked SHARED.
705 * Here we're trying to see if the inode buffer is incore, and if so whether it's
Dave Chinnerd808f612010-02-02 10:13:42 +1100706 * marked delayed write. If that's the case, we'll promote it and that will
707 * allow the caller to write the buffer by triggering the xfsbufd to run.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 */
Christoph Hellwig17b38472011-10-11 15:14:09 +0000709STATIC bool
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710xfs_inode_item_pushbuf(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000711 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000713 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
714 struct xfs_inode *ip = iip->ili_inode;
715 struct xfs_buf *bp;
Christoph Hellwig17b38472011-10-11 15:14:09 +0000716 bool ret = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000718 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720 /*
David Chinnerc63942d2008-08-13 16:41:16 +1000721 * If a flush is not in progress anymore, chances are that the
722 * inode was taken off the AIL. So, just get out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 */
David Chinnerc63942d2008-08-13 16:41:16 +1000724 if (completion_done(&ip->i_flush) ||
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000725 !(lip->li_flags & XFS_LI_IN_AIL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000727 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
729
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000730 bp = xfs_incore(ip->i_mount->m_ddev_targp, iip->ili_format.ilf_blkno,
731 iip->ili_format.ilf_len, XBF_TRYLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Dave Chinnerd808f612010-02-02 10:13:42 +1100734 if (!bp)
Christoph Hellwig17b38472011-10-11 15:14:09 +0000735 return true;
Dave Chinnerd808f612010-02-02 10:13:42 +1100736 if (XFS_BUF_ISDELAYWRITE(bp))
737 xfs_buf_delwri_promote(bp);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000738 if (xfs_buf_ispinned(bp))
739 ret = false;
Dave Chinnerd808f612010-02-02 10:13:42 +1100740 xfs_buf_relse(bp);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000741 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744/*
745 * This is called to asynchronously write the inode associated with this
746 * inode log item out to disk. The inode will already have been locked by
747 * a successful call to xfs_inode_item_trylock().
748 */
749STATIC void
750xfs_inode_item_push(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000751 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000753 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
754 struct xfs_inode *ip = iip->ili_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000756 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +1000757 ASSERT(!completion_done(&ip->i_flush));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 /*
760 * Since we were able to lock the inode's flush lock and
761 * we found it on the AIL, the inode must be dirty. This
762 * is because the inode is removed from the AIL while still
763 * holding the flush lock in xfs_iflush_done(). Thus, if
764 * we found it in the AIL and were able to obtain the flush
765 * lock without sleeping, then there must not have been
766 * anyone in the process of flushing the inode.
767 */
768 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
769 iip->ili_format.ilf_fields != 0);
770
771 /*
Dave Chinnerc8543632010-02-06 12:39:36 +1100772 * Push the inode to it's backing buffer. This will not remove the
773 * inode from the AIL - a further push will be required to trigger a
774 * buffer push. However, this allows all the dirty inodes to be pushed
Dave Chinner1bfd8d02011-03-26 09:13:55 +1100775 * to the buffer before it is pushed to disk. The buffer IO completion
776 * will pull the inode from the AIL, mark it clean and unlock the flush
Dave Chinnerc8543632010-02-06 12:39:36 +1100777 * lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Dave Chinner1bfd8d02011-03-26 09:13:55 +1100779 (void) xfs_iflush(ip, SYNC_TRYLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
783/*
784 * XXX rcc - this one really has to do something. Probably needs
785 * to stamp in a new field in the incore inode.
786 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787STATIC void
788xfs_inode_item_committing(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000789 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 xfs_lsn_t lsn)
791{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000792 INODE_ITEM(lip)->ili_last_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
795/*
796 * This is the ops vector shared by all buf log items.
797 */
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000798static const struct xfs_item_ops xfs_inode_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000799 .iop_size = xfs_inode_item_size,
800 .iop_format = xfs_inode_item_format,
801 .iop_pin = xfs_inode_item_pin,
802 .iop_unpin = xfs_inode_item_unpin,
803 .iop_trylock = xfs_inode_item_trylock,
804 .iop_unlock = xfs_inode_item_unlock,
805 .iop_committed = xfs_inode_item_committed,
806 .iop_push = xfs_inode_item_push,
807 .iop_pushbuf = xfs_inode_item_pushbuf,
808 .iop_committing = xfs_inode_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809};
810
811
812/*
813 * Initialize the inode log item for a newly allocated (in-core) inode.
814 */
815void
816xfs_inode_item_init(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000817 struct xfs_inode *ip,
818 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000820 struct xfs_inode_log_item *iip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 ASSERT(ip->i_itemp == NULL);
823 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 iip->ili_inode = ip;
Dave Chinner43f5efc2010-03-23 10:10:00 +1100826 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
827 &xfs_inode_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 iip->ili_format.ilf_type = XFS_LI_INODE;
829 iip->ili_format.ilf_ino = ip->i_ino;
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100830 iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
831 iip->ili_format.ilf_len = ip->i_imap.im_len;
832 iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
835/*
836 * Free the inode log item and any memory hanging off of it.
837 */
838void
839xfs_inode_item_destroy(
840 xfs_inode_t *ip)
841{
842#ifdef XFS_TRANS_DEBUG
843 if (ip->i_itemp->ili_root_size != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000844 kmem_free(ip->i_itemp->ili_orig_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 }
846#endif
847 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
848}
849
850
851/*
852 * This is the inode flushing I/O completion routine. It is called
853 * from interrupt level when the buffer containing the inode is
854 * flushed to disk. It is responsible for removing the inode item
855 * from the AIL if it has not been re-logged, and unlocking the inode's
856 * flush lock.
Dave Chinner30136832010-12-20 12:03:17 +1100857 *
858 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
859 * list for other inodes that will run this function. We remove them from the
860 * buffer list so we can process all the inode IO completions in one AIL lock
861 * traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863void
864xfs_iflush_done(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000865 struct xfs_buf *bp,
866 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867{
Dave Chinner30136832010-12-20 12:03:17 +1100868 struct xfs_inode_log_item *iip;
869 struct xfs_log_item *blip;
870 struct xfs_log_item *next;
871 struct xfs_log_item *prev;
Christoph Hellwigca30b2a2010-06-23 18:11:15 +1000872 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner30136832010-12-20 12:03:17 +1100873 int need_ail = 0;
874
875 /*
876 * Scan the buffer IO completions for other inodes being completed and
877 * attach them to the current inode log item.
878 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200879 blip = bp->b_fspriv;
Dave Chinner30136832010-12-20 12:03:17 +1100880 prev = NULL;
881 while (blip != NULL) {
882 if (lip->li_cb != xfs_iflush_done) {
883 prev = blip;
884 blip = blip->li_bio_list;
885 continue;
886 }
887
888 /* remove from list */
889 next = blip->li_bio_list;
890 if (!prev) {
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200891 bp->b_fspriv = next;
Dave Chinner30136832010-12-20 12:03:17 +1100892 } else {
893 prev->li_bio_list = next;
894 }
895
896 /* add to current list */
897 blip->li_bio_list = lip->li_bio_list;
898 lip->li_bio_list = blip;
899
900 /*
901 * while we have the item, do the unlocked check for needing
902 * the AIL lock.
903 */
904 iip = INODE_ITEM(blip);
905 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
906 need_ail++;
907
908 blip = next;
909 }
910
911 /* make sure we capture the state of the initial inode. */
912 iip = INODE_ITEM(lip);
913 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
914 need_ail++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 /*
917 * We only want to pull the item from the AIL if it is
918 * actually there and its location in the log has not
919 * changed since we started the flush. Thus, we only bother
920 * if the ili_logged flag is set and the inode's lsn has not
921 * changed. First we check the lsn outside
922 * the lock since it's cheaper, and then we recheck while
923 * holding the lock before removing the inode from the AIL.
924 */
Dave Chinner30136832010-12-20 12:03:17 +1100925 if (need_ail) {
926 struct xfs_log_item *log_items[need_ail];
927 int i = 0;
David Chinner783a2f62008-10-30 17:39:58 +1100928 spin_lock(&ailp->xa_lock);
Dave Chinner30136832010-12-20 12:03:17 +1100929 for (blip = lip; blip; blip = blip->li_bio_list) {
930 iip = INODE_ITEM(blip);
931 if (iip->ili_logged &&
932 blip->li_lsn == iip->ili_flush_lsn) {
933 log_items[i++] = blip;
934 }
935 ASSERT(i <= need_ail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 }
Dave Chinner30136832010-12-20 12:03:17 +1100937 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
938 xfs_trans_ail_delete_bulk(ailp, log_items, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
942 /*
Dave Chinner30136832010-12-20 12:03:17 +1100943 * clean up and unlock the flush lock now we are done. We can clear the
944 * ili_last_fields bits now that we know that the data corresponding to
945 * them is safely on disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 */
Dave Chinner30136832010-12-20 12:03:17 +1100947 for (blip = lip; blip; blip = next) {
948 next = blip->li_bio_list;
949 blip->li_bio_list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Dave Chinner30136832010-12-20 12:03:17 +1100951 iip = INODE_ITEM(blip);
952 iip->ili_logged = 0;
953 iip->ili_last_fields = 0;
954 xfs_ifunlock(iip->ili_inode);
955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956}
957
958/*
959 * This is the inode flushing abort routine. It is called
960 * from xfs_iflush when the filesystem is shutting down to clean
961 * up the inode state.
962 * It is responsible for removing the inode item
963 * from the AIL if it has not been re-logged, and unlocking the inode's
964 * flush lock.
965 */
966void
967xfs_iflush_abort(
968 xfs_inode_t *ip)
969{
David Chinner783a2f62008-10-30 17:39:58 +1100970 xfs_inode_log_item_t *iip = ip->i_itemp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (iip) {
David Chinner783a2f62008-10-30 17:39:58 +1100973 struct xfs_ail *ailp = iip->ili_item.li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
David Chinner783a2f62008-10-30 17:39:58 +1100975 spin_lock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
David Chinner783a2f62008-10-30 17:39:58 +1100977 /* xfs_trans_ail_delete() drops the AIL lock. */
978 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)iip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 } else
David Chinner783a2f62008-10-30 17:39:58 +1100980 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 }
982 iip->ili_logged = 0;
983 /*
984 * Clear the ili_last_fields bits now that we know that the
985 * data corresponding to them is safely on disk.
986 */
987 iip->ili_last_fields = 0;
988 /*
989 * Clear the inode logging fields so no more flushes are
990 * attempted.
991 */
992 iip->ili_format.ilf_fields = 0;
993 }
994 /*
995 * Release the inode's flush lock since we're done with it.
996 */
997 xfs_ifunlock(ip);
998}
999
1000void
1001xfs_istale_done(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001002 struct xfs_buf *bp,
1003 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001005 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
Tim Shimmin6d192a92006-06-09 14:55:38 +10001007
1008/*
1009 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
1010 * (which can have different field alignments) to the native version
1011 */
1012int
1013xfs_inode_item_format_convert(
1014 xfs_log_iovec_t *buf,
1015 xfs_inode_log_format_t *in_f)
1016{
1017 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001018 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10001019
Tim Shimmin6d192a92006-06-09 14:55:38 +10001020 in_f->ilf_type = in_f32->ilf_type;
1021 in_f->ilf_size = in_f32->ilf_size;
1022 in_f->ilf_fields = in_f32->ilf_fields;
1023 in_f->ilf_asize = in_f32->ilf_asize;
1024 in_f->ilf_dsize = in_f32->ilf_dsize;
1025 in_f->ilf_ino = in_f32->ilf_ino;
1026 /* copy biggest field of ilf_u */
1027 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1028 in_f32->ilf_u.ilfu_uuid.__u_bits,
1029 sizeof(uuid_t));
1030 in_f->ilf_blkno = in_f32->ilf_blkno;
1031 in_f->ilf_len = in_f32->ilf_len;
1032 in_f->ilf_boffset = in_f32->ilf_boffset;
1033 return 0;
1034 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001035 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10001036
Tim Shimmin6d192a92006-06-09 14:55:38 +10001037 in_f->ilf_type = in_f64->ilf_type;
1038 in_f->ilf_size = in_f64->ilf_size;
1039 in_f->ilf_fields = in_f64->ilf_fields;
1040 in_f->ilf_asize = in_f64->ilf_asize;
1041 in_f->ilf_dsize = in_f64->ilf_dsize;
1042 in_f->ilf_ino = in_f64->ilf_ino;
1043 /* copy biggest field of ilf_u */
1044 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1045 in_f64->ilf_u.ilfu_uuid.__u_bits,
1046 sizeof(uuid_t));
1047 in_f->ilf_blkno = in_f64->ilf_blkno;
1048 in_f->ilf_len = in_f64->ilf_len;
1049 in_f->ilf_boffset = in_f64->ilf_boffset;
1050 return 0;
1051 }
1052 return EFSCORRUPTED;
1053}