1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * 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.
13 *
14 * 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
17 */
18 #include <linux/log2.h>
19
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_dir2.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_attr.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
50 #include "xfs_log.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
53 #include "xfs_dir2_priv.h"
54
55 kmem_zone_t *xfs_inode_zone;
56
57 /*
58 * Used in xfs_itruncate_extents(). This is the maximum number of extents
59 * freed from a file in a single transaction.
60 */
61 #define XFS_ITRUNC_MAX_EXTENTS 2
62
63 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
64 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
65 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
66
67 /*
68 * helper function to extract extent size hint from inode
69 */
70 xfs_extlen_t
xfs_get_extsz_hint(struct xfs_inode * ip)71 xfs_get_extsz_hint(
72 struct xfs_inode *ip)
73 {
74 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
75 return ip->i_d.di_extsize;
76 if (XFS_IS_REALTIME_INODE(ip))
77 return ip->i_mount->m_sb.sb_rextsize;
78 return 0;
79 }
80
81 /*
82 * Helper function to extract CoW extent size hint from inode.
83 * Between the extent size hint and the CoW extent size hint, we
84 * return the greater of the two. If the value is zero (automatic),
85 * use the default size.
86 */
87 xfs_extlen_t
xfs_get_cowextsz_hint(struct xfs_inode * ip)88 xfs_get_cowextsz_hint(
89 struct xfs_inode *ip)
90 {
91 xfs_extlen_t a, b;
92
93 a = 0;
94 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
95 a = ip->i_d.di_cowextsize;
96 b = xfs_get_extsz_hint(ip);
97
98 a = max(a, b);
99 if (a == 0)
100 return XFS_DEFAULT_COWEXTSZ_HINT;
101 return a;
102 }
103
104 /*
105 * These two are wrapper routines around the xfs_ilock() routine used to
106 * centralize some grungy code. They are used in places that wish to lock the
107 * inode solely for reading the extents. The reason these places can't just
108 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
109 * bringing in of the extents from disk for a file in b-tree format. If the
110 * inode is in b-tree format, then we need to lock the inode exclusively until
111 * the extents are read in. Locking it exclusively all the time would limit
112 * our parallelism unnecessarily, though. What we do instead is check to see
113 * if the extents have been read in yet, and only lock the inode exclusively
114 * if they have not.
115 *
116 * The functions return a value which should be given to the corresponding
117 * xfs_iunlock() call.
118 */
119 uint
xfs_ilock_data_map_shared(struct xfs_inode * ip)120 xfs_ilock_data_map_shared(
121 struct xfs_inode *ip)
122 {
123 uint lock_mode = XFS_ILOCK_SHARED;
124
125 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
126 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
127 lock_mode = XFS_ILOCK_EXCL;
128 xfs_ilock(ip, lock_mode);
129 return lock_mode;
130 }
131
132 uint
xfs_ilock_attr_map_shared(struct xfs_inode * ip)133 xfs_ilock_attr_map_shared(
134 struct xfs_inode *ip)
135 {
136 uint lock_mode = XFS_ILOCK_SHARED;
137
138 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
139 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
140 lock_mode = XFS_ILOCK_EXCL;
141 xfs_ilock(ip, lock_mode);
142 return lock_mode;
143 }
144
145 /*
146 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
147 * the i_lock. This routine allows various combinations of the locks to be
148 * obtained.
149 *
150 * The 3 locks should always be ordered so that the IO lock is obtained first,
151 * the mmap lock second and the ilock last in order to prevent deadlock.
152 *
153 * Basic locking order:
154 *
155 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
156 *
157 * mmap_sem locking order:
158 *
159 * i_iolock -> page lock -> mmap_sem
160 * mmap_sem -> i_mmap_lock -> page_lock
161 *
162 * The difference in mmap_sem locking order mean that we cannot hold the
163 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
164 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
165 * in get_user_pages() to map the user pages into the kernel address space for
166 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
167 * page faults already hold the mmap_sem.
168 *
169 * Hence to serialise fully against both syscall and mmap based IO, we need to
170 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
171 * taken in places where we need to invalidate the page cache in a race
172 * free manner (e.g. truncate, hole punch and other extent manipulation
173 * functions).
174 */
175 void
xfs_ilock(xfs_inode_t * ip,uint lock_flags)176 xfs_ilock(
177 xfs_inode_t *ip,
178 uint lock_flags)
179 {
180 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
181
182 /*
183 * You can't set both SHARED and EXCL for the same lock,
184 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
185 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
186 */
187 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
188 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
189 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
190 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
191 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
192 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
193 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
194
195 if (lock_flags & XFS_IOLOCK_EXCL)
196 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
197 else if (lock_flags & XFS_IOLOCK_SHARED)
198 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
199
200 if (lock_flags & XFS_MMAPLOCK_EXCL)
201 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
202 else if (lock_flags & XFS_MMAPLOCK_SHARED)
203 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
204
205 if (lock_flags & XFS_ILOCK_EXCL)
206 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
207 else if (lock_flags & XFS_ILOCK_SHARED)
208 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
209 }
210
211 /*
212 * This is just like xfs_ilock(), except that the caller
213 * is guaranteed not to sleep. It returns 1 if it gets
214 * the requested locks and 0 otherwise. If the IO lock is
215 * obtained but the inode lock cannot be, then the IO lock
216 * is dropped before returning.
217 *
218 * ip -- the inode being locked
219 * lock_flags -- this parameter indicates the inode's locks to be
220 * to be locked. See the comment for xfs_ilock() for a list
221 * of valid values.
222 */
223 int
xfs_ilock_nowait(xfs_inode_t * ip,uint lock_flags)224 xfs_ilock_nowait(
225 xfs_inode_t *ip,
226 uint lock_flags)
227 {
228 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
229
230 /*
231 * You can't set both SHARED and EXCL for the same lock,
232 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
233 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
234 */
235 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
236 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
237 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
238 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
239 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
240 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
241 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
242
243 if (lock_flags & XFS_IOLOCK_EXCL) {
244 if (!mrtryupdate(&ip->i_iolock))
245 goto out;
246 } else if (lock_flags & XFS_IOLOCK_SHARED) {
247 if (!mrtryaccess(&ip->i_iolock))
248 goto out;
249 }
250
251 if (lock_flags & XFS_MMAPLOCK_EXCL) {
252 if (!mrtryupdate(&ip->i_mmaplock))
253 goto out_undo_iolock;
254 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
255 if (!mrtryaccess(&ip->i_mmaplock))
256 goto out_undo_iolock;
257 }
258
259 if (lock_flags & XFS_ILOCK_EXCL) {
260 if (!mrtryupdate(&ip->i_lock))
261 goto out_undo_mmaplock;
262 } else if (lock_flags & XFS_ILOCK_SHARED) {
263 if (!mrtryaccess(&ip->i_lock))
264 goto out_undo_mmaplock;
265 }
266 return 1;
267
268 out_undo_mmaplock:
269 if (lock_flags & XFS_MMAPLOCK_EXCL)
270 mrunlock_excl(&ip->i_mmaplock);
271 else if (lock_flags & XFS_MMAPLOCK_SHARED)
272 mrunlock_shared(&ip->i_mmaplock);
273 out_undo_iolock:
274 if (lock_flags & XFS_IOLOCK_EXCL)
275 mrunlock_excl(&ip->i_iolock);
276 else if (lock_flags & XFS_IOLOCK_SHARED)
277 mrunlock_shared(&ip->i_iolock);
278 out:
279 return 0;
280 }
281
282 /*
283 * xfs_iunlock() is used to drop the inode locks acquired with
284 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
285 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
286 * that we know which locks to drop.
287 *
288 * ip -- the inode being unlocked
289 * lock_flags -- this parameter indicates the inode's locks to be
290 * to be unlocked. See the comment for xfs_ilock() for a list
291 * of valid values for this parameter.
292 *
293 */
294 void
xfs_iunlock(xfs_inode_t * ip,uint lock_flags)295 xfs_iunlock(
296 xfs_inode_t *ip,
297 uint lock_flags)
298 {
299 /*
300 * You can't set both SHARED and EXCL for the same lock,
301 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
302 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
303 */
304 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
305 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
306 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
307 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
308 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
309 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
310 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
311 ASSERT(lock_flags != 0);
312
313 if (lock_flags & XFS_IOLOCK_EXCL)
314 mrunlock_excl(&ip->i_iolock);
315 else if (lock_flags & XFS_IOLOCK_SHARED)
316 mrunlock_shared(&ip->i_iolock);
317
318 if (lock_flags & XFS_MMAPLOCK_EXCL)
319 mrunlock_excl(&ip->i_mmaplock);
320 else if (lock_flags & XFS_MMAPLOCK_SHARED)
321 mrunlock_shared(&ip->i_mmaplock);
322
323 if (lock_flags & XFS_ILOCK_EXCL)
324 mrunlock_excl(&ip->i_lock);
325 else if (lock_flags & XFS_ILOCK_SHARED)
326 mrunlock_shared(&ip->i_lock);
327
328 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
329 }
330
331 /*
332 * give up write locks. the i/o lock cannot be held nested
333 * if it is being demoted.
334 */
335 void
xfs_ilock_demote(xfs_inode_t * ip,uint lock_flags)336 xfs_ilock_demote(
337 xfs_inode_t *ip,
338 uint lock_flags)
339 {
340 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
341 ASSERT((lock_flags &
342 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
343
344 if (lock_flags & XFS_ILOCK_EXCL)
345 mrdemote(&ip->i_lock);
346 if (lock_flags & XFS_MMAPLOCK_EXCL)
347 mrdemote(&ip->i_mmaplock);
348 if (lock_flags & XFS_IOLOCK_EXCL)
349 mrdemote(&ip->i_iolock);
350
351 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
352 }
353
354 #if defined(DEBUG) || defined(XFS_WARN)
355 int
xfs_isilocked(xfs_inode_t * ip,uint lock_flags)356 xfs_isilocked(
357 xfs_inode_t *ip,
358 uint lock_flags)
359 {
360 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
361 if (!(lock_flags & XFS_ILOCK_SHARED))
362 return !!ip->i_lock.mr_writer;
363 return rwsem_is_locked(&ip->i_lock.mr_lock);
364 }
365
366 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
367 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
368 return !!ip->i_mmaplock.mr_writer;
369 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
370 }
371
372 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
373 if (!(lock_flags & XFS_IOLOCK_SHARED))
374 return !!ip->i_iolock.mr_writer;
375 return rwsem_is_locked(&ip->i_iolock.mr_lock);
376 }
377
378 ASSERT(0);
379 return 0;
380 }
381 #endif
382
383 #ifdef DEBUG
384 int xfs_locked_n;
385 int xfs_small_retries;
386 int xfs_middle_retries;
387 int xfs_lots_retries;
388 int xfs_lock_delays;
389 #endif
390
391 /*
392 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
393 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
394 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
395 * errors and warnings.
396 */
397 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
398 static bool
xfs_lockdep_subclass_ok(int subclass)399 xfs_lockdep_subclass_ok(
400 int subclass)
401 {
402 return subclass < MAX_LOCKDEP_SUBCLASSES;
403 }
404 #else
405 #define xfs_lockdep_subclass_ok(subclass) (true)
406 #endif
407
408 /*
409 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
410 * value. This can be called for any type of inode lock combination, including
411 * parent locking. Care must be taken to ensure we don't overrun the subclass
412 * storage fields in the class mask we build.
413 */
414 static inline int
xfs_lock_inumorder(int lock_mode,int subclass)415 xfs_lock_inumorder(int lock_mode, int subclass)
416 {
417 int class = 0;
418
419 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
420 XFS_ILOCK_RTSUM)));
421 ASSERT(xfs_lockdep_subclass_ok(subclass));
422
423 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
424 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
425 ASSERT(xfs_lockdep_subclass_ok(subclass +
426 XFS_IOLOCK_PARENT_VAL));
427 class += subclass << XFS_IOLOCK_SHIFT;
428 if (lock_mode & XFS_IOLOCK_PARENT)
429 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
430 }
431
432 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
433 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
434 class += subclass << XFS_MMAPLOCK_SHIFT;
435 }
436
437 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
438 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
439 class += subclass << XFS_ILOCK_SHIFT;
440 }
441
442 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
443 }
444
445 /*
446 * The following routine will lock n inodes in exclusive mode. We assume the
447 * caller calls us with the inodes in i_ino order.
448 *
449 * We need to detect deadlock where an inode that we lock is in the AIL and we
450 * start waiting for another inode that is locked by a thread in a long running
451 * transaction (such as truncate). This can result in deadlock since the long
452 * running trans might need to wait for the inode we just locked in order to
453 * push the tail and free space in the log.
454 *
455 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
456 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
457 * lock more than one at a time, lockdep will report false positives saying we
458 * have violated locking orders.
459 */
460 static void
xfs_lock_inodes(xfs_inode_t ** ips,int inodes,uint lock_mode)461 xfs_lock_inodes(
462 xfs_inode_t **ips,
463 int inodes,
464 uint lock_mode)
465 {
466 int attempts = 0, i, j, try_lock;
467 xfs_log_item_t *lp;
468
469 /*
470 * Currently supports between 2 and 5 inodes with exclusive locking. We
471 * support an arbitrary depth of locking here, but absolute limits on
472 * inodes depend on the the type of locking and the limits placed by
473 * lockdep annotations in xfs_lock_inumorder. These are all checked by
474 * the asserts.
475 */
476 ASSERT(ips && inodes >= 2 && inodes <= 5);
477 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
478 XFS_ILOCK_EXCL));
479 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
480 XFS_ILOCK_SHARED)));
481 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
482 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
483 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
484 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
485 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
486 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
487
488 if (lock_mode & XFS_IOLOCK_EXCL) {
489 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
490 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
491 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
492
493 try_lock = 0;
494 i = 0;
495 again:
496 for (; i < inodes; i++) {
497 ASSERT(ips[i]);
498
499 if (i && (ips[i] == ips[i - 1])) /* Already locked */
500 continue;
501
502 /*
503 * If try_lock is not set yet, make sure all locked inodes are
504 * not in the AIL. If any are, set try_lock to be used later.
505 */
506 if (!try_lock) {
507 for (j = (i - 1); j >= 0 && !try_lock; j--) {
508 lp = (xfs_log_item_t *)ips[j]->i_itemp;
509 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
510 try_lock++;
511 }
512 }
513
514 /*
515 * If any of the previous locks we have locked is in the AIL,
516 * we must TRY to get the second and subsequent locks. If
517 * we can't get any, we must release all we have
518 * and try again.
519 */
520 if (!try_lock) {
521 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
522 continue;
523 }
524
525 /* try_lock means we have an inode locked that is in the AIL. */
526 ASSERT(i != 0);
527 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
528 continue;
529
530 /*
531 * Unlock all previous guys and try again. xfs_iunlock will try
532 * to push the tail if the inode is in the AIL.
533 */
534 attempts++;
535 for (j = i - 1; j >= 0; j--) {
536 /*
537 * Check to see if we've already unlocked this one. Not
538 * the first one going back, and the inode ptr is the
539 * same.
540 */
541 if (j != (i - 1) && ips[j] == ips[j + 1])
542 continue;
543
544 xfs_iunlock(ips[j], lock_mode);
545 }
546
547 if ((attempts % 5) == 0) {
548 delay(1); /* Don't just spin the CPU */
549 #ifdef DEBUG
550 xfs_lock_delays++;
551 #endif
552 }
553 i = 0;
554 try_lock = 0;
555 goto again;
556 }
557
558 #ifdef DEBUG
559 if (attempts) {
560 if (attempts < 5) xfs_small_retries++;
561 else if (attempts < 100) xfs_middle_retries++;
562 else xfs_lots_retries++;
563 } else {
564 xfs_locked_n++;
565 }
566 #endif
567 }
568
569 /*
570 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
571 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
572 * lock more than one at a time, lockdep will report false positives saying we
573 * have violated locking orders.
574 */
575 void
xfs_lock_two_inodes(xfs_inode_t * ip0,xfs_inode_t * ip1,uint lock_mode)576 xfs_lock_two_inodes(
577 xfs_inode_t *ip0,
578 xfs_inode_t *ip1,
579 uint lock_mode)
580 {
581 xfs_inode_t *temp;
582 int attempts = 0;
583 xfs_log_item_t *lp;
584
585 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
586 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
587 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
588 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
589 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
590
591 ASSERT(ip0->i_ino != ip1->i_ino);
592
593 if (ip0->i_ino > ip1->i_ino) {
594 temp = ip0;
595 ip0 = ip1;
596 ip1 = temp;
597 }
598
599 again:
600 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
601
602 /*
603 * If the first lock we have locked is in the AIL, we must TRY to get
604 * the second lock. If we can't get it, we must release the first one
605 * and try again.
606 */
607 lp = (xfs_log_item_t *)ip0->i_itemp;
608 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
609 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
610 xfs_iunlock(ip0, lock_mode);
611 if ((++attempts % 5) == 0)
612 delay(1); /* Don't just spin the CPU */
613 goto again;
614 }
615 } else {
616 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
617 }
618 }
619
620
621 void
__xfs_iflock(struct xfs_inode * ip)622 __xfs_iflock(
623 struct xfs_inode *ip)
624 {
625 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
626 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
627
628 do {
629 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
630 if (xfs_isiflocked(ip))
631 io_schedule();
632 } while (!xfs_iflock_nowait(ip));
633
634 finish_wait(wq, &wait.wait);
635 }
636
637 STATIC uint
_xfs_dic2xflags(__uint16_t di_flags,uint64_t di_flags2,bool has_attr)638 _xfs_dic2xflags(
639 __uint16_t di_flags,
640 uint64_t di_flags2,
641 bool has_attr)
642 {
643 uint flags = 0;
644
645 if (di_flags & XFS_DIFLAG_ANY) {
646 if (di_flags & XFS_DIFLAG_REALTIME)
647 flags |= FS_XFLAG_REALTIME;
648 if (di_flags & XFS_DIFLAG_PREALLOC)
649 flags |= FS_XFLAG_PREALLOC;
650 if (di_flags & XFS_DIFLAG_IMMUTABLE)
651 flags |= FS_XFLAG_IMMUTABLE;
652 if (di_flags & XFS_DIFLAG_APPEND)
653 flags |= FS_XFLAG_APPEND;
654 if (di_flags & XFS_DIFLAG_SYNC)
655 flags |= FS_XFLAG_SYNC;
656 if (di_flags & XFS_DIFLAG_NOATIME)
657 flags |= FS_XFLAG_NOATIME;
658 if (di_flags & XFS_DIFLAG_NODUMP)
659 flags |= FS_XFLAG_NODUMP;
660 if (di_flags & XFS_DIFLAG_RTINHERIT)
661 flags |= FS_XFLAG_RTINHERIT;
662 if (di_flags & XFS_DIFLAG_PROJINHERIT)
663 flags |= FS_XFLAG_PROJINHERIT;
664 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
665 flags |= FS_XFLAG_NOSYMLINKS;
666 if (di_flags & XFS_DIFLAG_EXTSIZE)
667 flags |= FS_XFLAG_EXTSIZE;
668 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
669 flags |= FS_XFLAG_EXTSZINHERIT;
670 if (di_flags & XFS_DIFLAG_NODEFRAG)
671 flags |= FS_XFLAG_NODEFRAG;
672 if (di_flags & XFS_DIFLAG_FILESTREAM)
673 flags |= FS_XFLAG_FILESTREAM;
674 }
675
676 if (di_flags2 & XFS_DIFLAG2_ANY) {
677 if (di_flags2 & XFS_DIFLAG2_DAX)
678 flags |= FS_XFLAG_DAX;
679 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
680 flags |= FS_XFLAG_COWEXTSIZE;
681 }
682
683 if (has_attr)
684 flags |= FS_XFLAG_HASATTR;
685
686 return flags;
687 }
688
689 uint
xfs_ip2xflags(struct xfs_inode * ip)690 xfs_ip2xflags(
691 struct xfs_inode *ip)
692 {
693 struct xfs_icdinode *dic = &ip->i_d;
694
695 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
696 }
697
698 /*
699 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
700 * is allowed, otherwise it has to be an exact match. If a CI match is found,
701 * ci_name->name will point to a the actual name (caller must free) or
702 * will be set to NULL if an exact match is found.
703 */
704 int
xfs_lookup(xfs_inode_t * dp,struct xfs_name * name,xfs_inode_t ** ipp,struct xfs_name * ci_name)705 xfs_lookup(
706 xfs_inode_t *dp,
707 struct xfs_name *name,
708 xfs_inode_t **ipp,
709 struct xfs_name *ci_name)
710 {
711 xfs_ino_t inum;
712 int error;
713
714 trace_xfs_lookup(dp, name);
715
716 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
717 return -EIO;
718
719 xfs_ilock(dp, XFS_IOLOCK_SHARED);
720 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
721 if (error)
722 goto out_unlock;
723
724 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
725 if (error)
726 goto out_free_name;
727
728 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
729 return 0;
730
731 out_free_name:
732 if (ci_name)
733 kmem_free(ci_name->name);
734 out_unlock:
735 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
736 *ipp = NULL;
737 return error;
738 }
739
740 /*
741 * Allocate an inode on disk and return a copy of its in-core version.
742 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
743 * appropriately within the inode. The uid and gid for the inode are
744 * set according to the contents of the given cred structure.
745 *
746 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
747 * has a free inode available, call xfs_iget() to obtain the in-core
748 * version of the allocated inode. Finally, fill in the inode and
749 * log its initial contents. In this case, ialloc_context would be
750 * set to NULL.
751 *
752 * If xfs_dialloc() does not have an available inode, it will replenish
753 * its supply by doing an allocation. Since we can only do one
754 * allocation within a transaction without deadlocks, we must commit
755 * the current transaction before returning the inode itself.
756 * In this case, therefore, we will set ialloc_context and return.
757 * The caller should then commit the current transaction, start a new
758 * transaction, and call xfs_ialloc() again to actually get the inode.
759 *
760 * To ensure that some other process does not grab the inode that
761 * was allocated during the first call to xfs_ialloc(), this routine
762 * also returns the [locked] bp pointing to the head of the freelist
763 * as ialloc_context. The caller should hold this buffer across
764 * the commit and pass it back into this routine on the second call.
765 *
766 * If we are allocating quota inodes, we do not have a parent inode
767 * to attach to or associate with (i.e. pip == NULL) because they
768 * are not linked into the directory structure - they are attached
769 * directly to the superblock - and so have no parent.
770 */
771 static int
xfs_ialloc(xfs_trans_t * tp,xfs_inode_t * pip,umode_t mode,xfs_nlink_t nlink,xfs_dev_t rdev,prid_t prid,int okalloc,xfs_buf_t ** ialloc_context,xfs_inode_t ** ipp)772 xfs_ialloc(
773 xfs_trans_t *tp,
774 xfs_inode_t *pip,
775 umode_t mode,
776 xfs_nlink_t nlink,
777 xfs_dev_t rdev,
778 prid_t prid,
779 int okalloc,
780 xfs_buf_t **ialloc_context,
781 xfs_inode_t **ipp)
782 {
783 struct xfs_mount *mp = tp->t_mountp;
784 xfs_ino_t ino;
785 xfs_inode_t *ip;
786 uint flags;
787 int error;
788 struct timespec tv;
789 struct inode *inode;
790
791 /*
792 * Call the space management code to pick
793 * the on-disk inode to be allocated.
794 */
795 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
796 ialloc_context, &ino);
797 if (error)
798 return error;
799 if (*ialloc_context || ino == NULLFSINO) {
800 *ipp = NULL;
801 return 0;
802 }
803 ASSERT(*ialloc_context == NULL);
804
805 /*
806 * Get the in-core inode with the lock held exclusively.
807 * This is because we're setting fields here we need
808 * to prevent others from looking at until we're done.
809 */
810 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
811 XFS_ILOCK_EXCL, &ip);
812 if (error)
813 return error;
814 ASSERT(ip != NULL);
815 inode = VFS_I(ip);
816
817 /*
818 * We always convert v1 inodes to v2 now - we only support filesystems
819 * with >= v2 inode capability, so there is no reason for ever leaving
820 * an inode in v1 format.
821 */
822 if (ip->i_d.di_version == 1)
823 ip->i_d.di_version = 2;
824
825 inode->i_mode = mode;
826 set_nlink(inode, nlink);
827 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
828 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
829 xfs_set_projid(ip, prid);
830
831 if (pip && XFS_INHERIT_GID(pip)) {
832 ip->i_d.di_gid = pip->i_d.di_gid;
833 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
834 inode->i_mode |= S_ISGID;
835 }
836
837 /*
838 * If the group ID of the new file does not match the effective group
839 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
840 * (and only if the irix_sgid_inherit compatibility variable is set).
841 */
842 if ((irix_sgid_inherit) &&
843 (inode->i_mode & S_ISGID) &&
844 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
845 inode->i_mode &= ~S_ISGID;
846
847 ip->i_d.di_size = 0;
848 ip->i_d.di_nextents = 0;
849 ASSERT(ip->i_d.di_nblocks == 0);
850
851 tv = current_time(inode);
852 inode->i_mtime = tv;
853 inode->i_atime = tv;
854 inode->i_ctime = tv;
855
856 ip->i_d.di_extsize = 0;
857 ip->i_d.di_dmevmask = 0;
858 ip->i_d.di_dmstate = 0;
859 ip->i_d.di_flags = 0;
860
861 if (ip->i_d.di_version == 3) {
862 inode->i_version = 1;
863 ip->i_d.di_flags2 = 0;
864 ip->i_d.di_cowextsize = 0;
865 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
866 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
867 }
868
869
870 flags = XFS_ILOG_CORE;
871 switch (mode & S_IFMT) {
872 case S_IFIFO:
873 case S_IFCHR:
874 case S_IFBLK:
875 case S_IFSOCK:
876 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
877 ip->i_df.if_u2.if_rdev = rdev;
878 ip->i_df.if_flags = 0;
879 flags |= XFS_ILOG_DEV;
880 break;
881 case S_IFREG:
882 case S_IFDIR:
883 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
884 uint di_flags = 0;
885
886 if (S_ISDIR(mode)) {
887 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
888 di_flags |= XFS_DIFLAG_RTINHERIT;
889 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
890 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
891 ip->i_d.di_extsize = pip->i_d.di_extsize;
892 }
893 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
894 di_flags |= XFS_DIFLAG_PROJINHERIT;
895 } else if (S_ISREG(mode)) {
896 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
897 di_flags |= XFS_DIFLAG_REALTIME;
898 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
899 di_flags |= XFS_DIFLAG_EXTSIZE;
900 ip->i_d.di_extsize = pip->i_d.di_extsize;
901 }
902 }
903 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
904 xfs_inherit_noatime)
905 di_flags |= XFS_DIFLAG_NOATIME;
906 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
907 xfs_inherit_nodump)
908 di_flags |= XFS_DIFLAG_NODUMP;
909 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
910 xfs_inherit_sync)
911 di_flags |= XFS_DIFLAG_SYNC;
912 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
913 xfs_inherit_nosymlinks)
914 di_flags |= XFS_DIFLAG_NOSYMLINKS;
915 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
916 xfs_inherit_nodefrag)
917 di_flags |= XFS_DIFLAG_NODEFRAG;
918 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
919 di_flags |= XFS_DIFLAG_FILESTREAM;
920
921 ip->i_d.di_flags |= di_flags;
922 }
923 if (pip &&
924 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
925 pip->i_d.di_version == 3 &&
926 ip->i_d.di_version == 3) {
927 uint64_t di_flags2 = 0;
928
929 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
930 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
931 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
932 }
933 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
934 di_flags2 |= XFS_DIFLAG2_DAX;
935
936 ip->i_d.di_flags2 |= di_flags2;
937 }
938 /* FALLTHROUGH */
939 case S_IFLNK:
940 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
941 ip->i_df.if_flags = XFS_IFEXTENTS;
942 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
943 ip->i_df.if_u1.if_extents = NULL;
944 break;
945 default:
946 ASSERT(0);
947 }
948 /*
949 * Attribute fork settings for new inode.
950 */
951 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
952 ip->i_d.di_anextents = 0;
953
954 /*
955 * Log the new values stuffed into the inode.
956 */
957 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
958 xfs_trans_log_inode(tp, ip, flags);
959
960 /* now that we have an i_mode we can setup the inode structure */
961 xfs_setup_inode(ip);
962
963 *ipp = ip;
964 return 0;
965 }
966
967 /*
968 * Allocates a new inode from disk and return a pointer to the
969 * incore copy. This routine will internally commit the current
970 * transaction and allocate a new one if the Space Manager needed
971 * to do an allocation to replenish the inode free-list.
972 *
973 * This routine is designed to be called from xfs_create and
974 * xfs_create_dir.
975 *
976 */
977 int
xfs_dir_ialloc(xfs_trans_t ** tpp,xfs_inode_t * dp,umode_t mode,xfs_nlink_t nlink,xfs_dev_t rdev,prid_t prid,int okalloc,xfs_inode_t ** ipp,int * committed)978 xfs_dir_ialloc(
979 xfs_trans_t **tpp, /* input: current transaction;
980 output: may be a new transaction. */
981 xfs_inode_t *dp, /* directory within whose allocate
982 the inode. */
983 umode_t mode,
984 xfs_nlink_t nlink,
985 xfs_dev_t rdev,
986 prid_t prid, /* project id */
987 int okalloc, /* ok to allocate new space */
988 xfs_inode_t **ipp, /* pointer to inode; it will be
989 locked. */
990 int *committed)
991
992 {
993 xfs_trans_t *tp;
994 xfs_inode_t *ip;
995 xfs_buf_t *ialloc_context = NULL;
996 int code;
997 void *dqinfo;
998 uint tflags;
999
1000 tp = *tpp;
1001 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1002
1003 /*
1004 * xfs_ialloc will return a pointer to an incore inode if
1005 * the Space Manager has an available inode on the free
1006 * list. Otherwise, it will do an allocation and replenish
1007 * the freelist. Since we can only do one allocation per
1008 * transaction without deadlocks, we will need to commit the
1009 * current transaction and start a new one. We will then
1010 * need to call xfs_ialloc again to get the inode.
1011 *
1012 * If xfs_ialloc did an allocation to replenish the freelist,
1013 * it returns the bp containing the head of the freelist as
1014 * ialloc_context. We will hold a lock on it across the
1015 * transaction commit so that no other process can steal
1016 * the inode(s) that we've just allocated.
1017 */
1018 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1019 &ialloc_context, &ip);
1020
1021 /*
1022 * Return an error if we were unable to allocate a new inode.
1023 * This should only happen if we run out of space on disk or
1024 * encounter a disk error.
1025 */
1026 if (code) {
1027 *ipp = NULL;
1028 return code;
1029 }
1030 if (!ialloc_context && !ip) {
1031 *ipp = NULL;
1032 return -ENOSPC;
1033 }
1034
1035 /*
1036 * If the AGI buffer is non-NULL, then we were unable to get an
1037 * inode in one operation. We need to commit the current
1038 * transaction and call xfs_ialloc() again. It is guaranteed
1039 * to succeed the second time.
1040 */
1041 if (ialloc_context) {
1042 /*
1043 * Normally, xfs_trans_commit releases all the locks.
1044 * We call bhold to hang on to the ialloc_context across
1045 * the commit. Holding this buffer prevents any other
1046 * processes from doing any allocations in this
1047 * allocation group.
1048 */
1049 xfs_trans_bhold(tp, ialloc_context);
1050
1051 /*
1052 * We want the quota changes to be associated with the next
1053 * transaction, NOT this one. So, detach the dqinfo from this
1054 * and attach it to the next transaction.
1055 */
1056 dqinfo = NULL;
1057 tflags = 0;
1058 if (tp->t_dqinfo) {
1059 dqinfo = (void *)tp->t_dqinfo;
1060 tp->t_dqinfo = NULL;
1061 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1062 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1063 }
1064
1065 code = xfs_trans_roll(&tp, NULL);
1066 if (committed != NULL)
1067 *committed = 1;
1068
1069 /*
1070 * Re-attach the quota info that we detached from prev trx.
1071 */
1072 if (dqinfo) {
1073 tp->t_dqinfo = dqinfo;
1074 tp->t_flags |= tflags;
1075 }
1076
1077 if (code) {
1078 xfs_buf_relse(ialloc_context);
1079 *tpp = tp;
1080 *ipp = NULL;
1081 return code;
1082 }
1083 xfs_trans_bjoin(tp, ialloc_context);
1084
1085 /*
1086 * Call ialloc again. Since we've locked out all
1087 * other allocations in this allocation group,
1088 * this call should always succeed.
1089 */
1090 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1091 okalloc, &ialloc_context, &ip);
1092
1093 /*
1094 * If we get an error at this point, return to the caller
1095 * so that the current transaction can be aborted.
1096 */
1097 if (code) {
1098 *tpp = tp;
1099 *ipp = NULL;
1100 return code;
1101 }
1102 ASSERT(!ialloc_context && ip);
1103
1104 } else {
1105 if (committed != NULL)
1106 *committed = 0;
1107 }
1108
1109 *ipp = ip;
1110 *tpp = tp;
1111
1112 return 0;
1113 }
1114
1115 /*
1116 * Decrement the link count on an inode & log the change. If this causes the
1117 * link count to go to zero, move the inode to AGI unlinked list so that it can
1118 * be freed when the last active reference goes away via xfs_inactive().
1119 */
1120 static int /* error */
xfs_droplink(xfs_trans_t * tp,xfs_inode_t * ip)1121 xfs_droplink(
1122 xfs_trans_t *tp,
1123 xfs_inode_t *ip)
1124 {
1125 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1126
1127 drop_nlink(VFS_I(ip));
1128 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1129
1130 if (VFS_I(ip)->i_nlink)
1131 return 0;
1132
1133 return xfs_iunlink(tp, ip);
1134 }
1135
1136 /*
1137 * Increment the link count on an inode & log the change.
1138 */
1139 static int
xfs_bumplink(xfs_trans_t * tp,xfs_inode_t * ip)1140 xfs_bumplink(
1141 xfs_trans_t *tp,
1142 xfs_inode_t *ip)
1143 {
1144 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1145
1146 ASSERT(ip->i_d.di_version > 1);
1147 inc_nlink(VFS_I(ip));
1148 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1149 return 0;
1150 }
1151
1152 int
xfs_create(xfs_inode_t * dp,struct xfs_name * name,umode_t mode,xfs_dev_t rdev,xfs_inode_t ** ipp)1153 xfs_create(
1154 xfs_inode_t *dp,
1155 struct xfs_name *name,
1156 umode_t mode,
1157 xfs_dev_t rdev,
1158 xfs_inode_t **ipp)
1159 {
1160 int is_dir = S_ISDIR(mode);
1161 struct xfs_mount *mp = dp->i_mount;
1162 struct xfs_inode *ip = NULL;
1163 struct xfs_trans *tp = NULL;
1164 int error;
1165 struct xfs_defer_ops dfops;
1166 xfs_fsblock_t first_block;
1167 bool unlock_dp_on_error = false;
1168 prid_t prid;
1169 struct xfs_dquot *udqp = NULL;
1170 struct xfs_dquot *gdqp = NULL;
1171 struct xfs_dquot *pdqp = NULL;
1172 struct xfs_trans_res *tres;
1173 uint resblks;
1174
1175 trace_xfs_create(dp, name);
1176
1177 if (XFS_FORCED_SHUTDOWN(mp))
1178 return -EIO;
1179
1180 prid = xfs_get_initial_prid(dp);
1181
1182 /*
1183 * Make sure that we have allocated dquot(s) on disk.
1184 */
1185 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1186 xfs_kgid_to_gid(current_fsgid()), prid,
1187 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1188 &udqp, &gdqp, &pdqp);
1189 if (error)
1190 return error;
1191
1192 if (is_dir) {
1193 rdev = 0;
1194 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1195 tres = &M_RES(mp)->tr_mkdir;
1196 } else {
1197 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1198 tres = &M_RES(mp)->tr_create;
1199 }
1200
1201 /*
1202 * Initially assume that the file does not exist and
1203 * reserve the resources for that case. If that is not
1204 * the case we'll drop the one we have and get a more
1205 * appropriate transaction later.
1206 */
1207 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1208 if (error == -ENOSPC) {
1209 /* flush outstanding delalloc blocks and retry */
1210 xfs_flush_inodes(mp);
1211 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1212 }
1213 if (error == -ENOSPC) {
1214 /* No space at all so try a "no-allocation" reservation */
1215 resblks = 0;
1216 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1217 }
1218 if (error)
1219 goto out_release_inode;
1220
1221 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1222 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
1223 unlock_dp_on_error = true;
1224
1225 xfs_defer_init(&dfops, &first_block);
1226
1227 /*
1228 * Reserve disk quota and the inode.
1229 */
1230 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1231 pdqp, resblks, 1, 0);
1232 if (error)
1233 goto out_trans_cancel;
1234
1235 if (!resblks) {
1236 error = xfs_dir_canenter(tp, dp, name);
1237 if (error)
1238 goto out_trans_cancel;
1239 }
1240
1241 /*
1242 * A newly created regular or special file just has one directory
1243 * entry pointing to them, but a directory also the "." entry
1244 * pointing to itself.
1245 */
1246 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1247 prid, resblks > 0, &ip, NULL);
1248 if (error)
1249 goto out_trans_cancel;
1250
1251 /*
1252 * Now we join the directory inode to the transaction. We do not do it
1253 * earlier because xfs_dir_ialloc might commit the previous transaction
1254 * (and release all the locks). An error from here on will result in
1255 * the transaction cancel unlocking dp so don't do it explicitly in the
1256 * error path.
1257 */
1258 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1259 unlock_dp_on_error = false;
1260
1261 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1262 &first_block, &dfops, resblks ?
1263 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1264 if (error) {
1265 ASSERT(error != -ENOSPC);
1266 goto out_trans_cancel;
1267 }
1268 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1269 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1270
1271 if (is_dir) {
1272 error = xfs_dir_init(tp, ip, dp);
1273 if (error)
1274 goto out_bmap_cancel;
1275
1276 error = xfs_bumplink(tp, dp);
1277 if (error)
1278 goto out_bmap_cancel;
1279 }
1280
1281 /*
1282 * If this is a synchronous mount, make sure that the
1283 * create transaction goes to disk before returning to
1284 * the user.
1285 */
1286 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1287 xfs_trans_set_sync(tp);
1288
1289 /*
1290 * Attach the dquot(s) to the inodes and modify them incore.
1291 * These ids of the inode couldn't have changed since the new
1292 * inode has been locked ever since it was created.
1293 */
1294 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1295
1296 error = xfs_defer_finish(&tp, &dfops, NULL);
1297 if (error)
1298 goto out_bmap_cancel;
1299
1300 error = xfs_trans_commit(tp);
1301 if (error)
1302 goto out_release_inode;
1303
1304 xfs_qm_dqrele(udqp);
1305 xfs_qm_dqrele(gdqp);
1306 xfs_qm_dqrele(pdqp);
1307
1308 *ipp = ip;
1309 return 0;
1310
1311 out_bmap_cancel:
1312 xfs_defer_cancel(&dfops);
1313 out_trans_cancel:
1314 xfs_trans_cancel(tp);
1315 out_release_inode:
1316 /*
1317 * Wait until after the current transaction is aborted to finish the
1318 * setup of the inode and release the inode. This prevents recursive
1319 * transactions and deadlocks from xfs_inactive.
1320 */
1321 if (ip) {
1322 xfs_finish_inode_setup(ip);
1323 IRELE(ip);
1324 }
1325
1326 xfs_qm_dqrele(udqp);
1327 xfs_qm_dqrele(gdqp);
1328 xfs_qm_dqrele(pdqp);
1329
1330 if (unlock_dp_on_error)
1331 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1332 return error;
1333 }
1334
1335 int
xfs_create_tmpfile(struct xfs_inode * dp,struct dentry * dentry,umode_t mode,struct xfs_inode ** ipp)1336 xfs_create_tmpfile(
1337 struct xfs_inode *dp,
1338 struct dentry *dentry,
1339 umode_t mode,
1340 struct xfs_inode **ipp)
1341 {
1342 struct xfs_mount *mp = dp->i_mount;
1343 struct xfs_inode *ip = NULL;
1344 struct xfs_trans *tp = NULL;
1345 int error;
1346 prid_t prid;
1347 struct xfs_dquot *udqp = NULL;
1348 struct xfs_dquot *gdqp = NULL;
1349 struct xfs_dquot *pdqp = NULL;
1350 struct xfs_trans_res *tres;
1351 uint resblks;
1352
1353 if (XFS_FORCED_SHUTDOWN(mp))
1354 return -EIO;
1355
1356 prid = xfs_get_initial_prid(dp);
1357
1358 /*
1359 * Make sure that we have allocated dquot(s) on disk.
1360 */
1361 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1362 xfs_kgid_to_gid(current_fsgid()), prid,
1363 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1364 &udqp, &gdqp, &pdqp);
1365 if (error)
1366 return error;
1367
1368 resblks = XFS_IALLOC_SPACE_RES(mp);
1369 tres = &M_RES(mp)->tr_create_tmpfile;
1370
1371 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1372 if (error == -ENOSPC) {
1373 /* No space at all so try a "no-allocation" reservation */
1374 resblks = 0;
1375 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1376 }
1377 if (error)
1378 goto out_release_inode;
1379
1380 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1381 pdqp, resblks, 1, 0);
1382 if (error)
1383 goto out_trans_cancel;
1384
1385 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1386 prid, resblks > 0, &ip, NULL);
1387 if (error)
1388 goto out_trans_cancel;
1389
1390 if (mp->m_flags & XFS_MOUNT_WSYNC)
1391 xfs_trans_set_sync(tp);
1392
1393 /*
1394 * Attach the dquot(s) to the inodes and modify them incore.
1395 * These ids of the inode couldn't have changed since the new
1396 * inode has been locked ever since it was created.
1397 */
1398 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1399
1400 error = xfs_iunlink(tp, ip);
1401 if (error)
1402 goto out_trans_cancel;
1403
1404 error = xfs_trans_commit(tp);
1405 if (error)
1406 goto out_release_inode;
1407
1408 xfs_qm_dqrele(udqp);
1409 xfs_qm_dqrele(gdqp);
1410 xfs_qm_dqrele(pdqp);
1411
1412 *ipp = ip;
1413 return 0;
1414
1415 out_trans_cancel:
1416 xfs_trans_cancel(tp);
1417 out_release_inode:
1418 /*
1419 * Wait until after the current transaction is aborted to finish the
1420 * setup of the inode and release the inode. This prevents recursive
1421 * transactions and deadlocks from xfs_inactive.
1422 */
1423 if (ip) {
1424 xfs_finish_inode_setup(ip);
1425 IRELE(ip);
1426 }
1427
1428 xfs_qm_dqrele(udqp);
1429 xfs_qm_dqrele(gdqp);
1430 xfs_qm_dqrele(pdqp);
1431
1432 return error;
1433 }
1434
1435 int
xfs_link(xfs_inode_t * tdp,xfs_inode_t * sip,struct xfs_name * target_name)1436 xfs_link(
1437 xfs_inode_t *tdp,
1438 xfs_inode_t *sip,
1439 struct xfs_name *target_name)
1440 {
1441 xfs_mount_t *mp = tdp->i_mount;
1442 xfs_trans_t *tp;
1443 int error;
1444 struct xfs_defer_ops dfops;
1445 xfs_fsblock_t first_block;
1446 int resblks;
1447
1448 trace_xfs_link(tdp, target_name);
1449
1450 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1451
1452 if (XFS_FORCED_SHUTDOWN(mp))
1453 return -EIO;
1454
1455 error = xfs_qm_dqattach(sip, 0);
1456 if (error)
1457 goto std_return;
1458
1459 error = xfs_qm_dqattach(tdp, 0);
1460 if (error)
1461 goto std_return;
1462
1463 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1464 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1465 if (error == -ENOSPC) {
1466 resblks = 0;
1467 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1468 }
1469 if (error)
1470 goto std_return;
1471
1472 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
1473 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1474
1475 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1476 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1477
1478 /*
1479 * If we are using project inheritance, we only allow hard link
1480 * creation in our tree when the project IDs are the same; else
1481 * the tree quota mechanism could be circumvented.
1482 */
1483 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1484 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1485 error = -EXDEV;
1486 goto error_return;
1487 }
1488
1489 if (!resblks) {
1490 error = xfs_dir_canenter(tp, tdp, target_name);
1491 if (error)
1492 goto error_return;
1493 }
1494
1495 xfs_defer_init(&dfops, &first_block);
1496
1497 /*
1498 * Handle initial link state of O_TMPFILE inode
1499 */
1500 if (VFS_I(sip)->i_nlink == 0) {
1501 error = xfs_iunlink_remove(tp, sip);
1502 if (error)
1503 goto error_return;
1504 }
1505
1506 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1507 &first_block, &dfops, resblks);
1508 if (error)
1509 goto error_return;
1510 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1511 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1512
1513 error = xfs_bumplink(tp, sip);
1514 if (error)
1515 goto error_return;
1516
1517 /*
1518 * If this is a synchronous mount, make sure that the
1519 * link transaction goes to disk before returning to
1520 * the user.
1521 */
1522 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1523 xfs_trans_set_sync(tp);
1524
1525 error = xfs_defer_finish(&tp, &dfops, NULL);
1526 if (error) {
1527 xfs_defer_cancel(&dfops);
1528 goto error_return;
1529 }
1530
1531 return xfs_trans_commit(tp);
1532
1533 error_return:
1534 xfs_trans_cancel(tp);
1535 std_return:
1536 return error;
1537 }
1538
1539 /*
1540 * Free up the underlying blocks past new_size. The new size must be smaller
1541 * than the current size. This routine can be used both for the attribute and
1542 * data fork, and does not modify the inode size, which is left to the caller.
1543 *
1544 * The transaction passed to this routine must have made a permanent log
1545 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1546 * given transaction and start new ones, so make sure everything involved in
1547 * the transaction is tidy before calling here. Some transaction will be
1548 * returned to the caller to be committed. The incoming transaction must
1549 * already include the inode, and both inode locks must be held exclusively.
1550 * The inode must also be "held" within the transaction. On return the inode
1551 * will be "held" within the returned transaction. This routine does NOT
1552 * require any disk space to be reserved for it within the transaction.
1553 *
1554 * If we get an error, we must return with the inode locked and linked into the
1555 * current transaction. This keeps things simple for the higher level code,
1556 * because it always knows that the inode is locked and held in the transaction
1557 * that returns to it whether errors occur or not. We don't mark the inode
1558 * dirty on error so that transactions can be easily aborted if possible.
1559 */
1560 int
xfs_itruncate_extents(struct xfs_trans ** tpp,struct xfs_inode * ip,int whichfork,xfs_fsize_t new_size)1561 xfs_itruncate_extents(
1562 struct xfs_trans **tpp,
1563 struct xfs_inode *ip,
1564 int whichfork,
1565 xfs_fsize_t new_size)
1566 {
1567 struct xfs_mount *mp = ip->i_mount;
1568 struct xfs_trans *tp = *tpp;
1569 struct xfs_defer_ops dfops;
1570 xfs_fsblock_t first_block;
1571 xfs_fileoff_t first_unmap_block;
1572 xfs_fileoff_t last_block;
1573 xfs_filblks_t unmap_len;
1574 int error = 0;
1575 int done = 0;
1576
1577 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1578 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1579 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1580 ASSERT(new_size <= XFS_ISIZE(ip));
1581 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1582 ASSERT(ip->i_itemp != NULL);
1583 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1584 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1585
1586 trace_xfs_itruncate_extents_start(ip, new_size);
1587
1588 /*
1589 * Since it is possible for space to become allocated beyond
1590 * the end of the file (in a crash where the space is allocated
1591 * but the inode size is not yet updated), simply remove any
1592 * blocks which show up between the new EOF and the maximum
1593 * possible file size. If the first block to be removed is
1594 * beyond the maximum file size (ie it is the same as last_block),
1595 * then there is nothing to do.
1596 */
1597 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1598 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1599 if (first_unmap_block == last_block)
1600 return 0;
1601
1602 ASSERT(first_unmap_block < last_block);
1603 unmap_len = last_block - first_unmap_block + 1;
1604 while (!done) {
1605 xfs_defer_init(&dfops, &first_block);
1606 error = xfs_bunmapi(tp, ip,
1607 first_unmap_block, unmap_len,
1608 xfs_bmapi_aflag(whichfork),
1609 XFS_ITRUNC_MAX_EXTENTS,
1610 &first_block, &dfops,
1611 &done);
1612 if (error)
1613 goto out_bmap_cancel;
1614
1615 /*
1616 * Duplicate the transaction that has the permanent
1617 * reservation and commit the old transaction.
1618 */
1619 error = xfs_defer_finish(&tp, &dfops, ip);
1620 if (error)
1621 goto out_bmap_cancel;
1622
1623 error = xfs_trans_roll(&tp, ip);
1624 if (error)
1625 goto out;
1626 }
1627
1628 /* Remove all pending CoW reservations. */
1629 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1630 last_block, true);
1631 if (error)
1632 goto out;
1633
1634 /*
1635 * Clear the reflink flag if there are no data fork blocks and
1636 * there are no extents staged in the cow fork.
1637 */
1638 if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) {
1639 if (ip->i_d.di_nblocks == 0)
1640 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1641 xfs_inode_clear_cowblocks_tag(ip);
1642 }
1643
1644 /*
1645 * Always re-log the inode so that our permanent transaction can keep
1646 * on rolling it forward in the log.
1647 */
1648 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1649
1650 trace_xfs_itruncate_extents_end(ip, new_size);
1651
1652 out:
1653 *tpp = tp;
1654 return error;
1655 out_bmap_cancel:
1656 /*
1657 * If the bunmapi call encounters an error, return to the caller where
1658 * the transaction can be properly aborted. We just need to make sure
1659 * we're not holding any resources that we were not when we came in.
1660 */
1661 xfs_defer_cancel(&dfops);
1662 goto out;
1663 }
1664
1665 int
xfs_release(xfs_inode_t * ip)1666 xfs_release(
1667 xfs_inode_t *ip)
1668 {
1669 xfs_mount_t *mp = ip->i_mount;
1670 int error;
1671
1672 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1673 return 0;
1674
1675 /* If this is a read-only mount, don't do this (would generate I/O) */
1676 if (mp->m_flags & XFS_MOUNT_RDONLY)
1677 return 0;
1678
1679 if (!XFS_FORCED_SHUTDOWN(mp)) {
1680 int truncated;
1681
1682 /*
1683 * If we previously truncated this file and removed old data
1684 * in the process, we want to initiate "early" writeout on
1685 * the last close. This is an attempt to combat the notorious
1686 * NULL files problem which is particularly noticeable from a
1687 * truncate down, buffered (re-)write (delalloc), followed by
1688 * a crash. What we are effectively doing here is
1689 * significantly reducing the time window where we'd otherwise
1690 * be exposed to that problem.
1691 */
1692 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1693 if (truncated) {
1694 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1695 if (ip->i_delayed_blks > 0) {
1696 error = filemap_flush(VFS_I(ip)->i_mapping);
1697 if (error)
1698 return error;
1699 }
1700 }
1701 }
1702
1703 if (VFS_I(ip)->i_nlink == 0)
1704 return 0;
1705
1706 if (xfs_can_free_eofblocks(ip, false)) {
1707
1708 /*
1709 * Check if the inode is being opened, written and closed
1710 * frequently and we have delayed allocation blocks outstanding
1711 * (e.g. streaming writes from the NFS server), truncating the
1712 * blocks past EOF will cause fragmentation to occur.
1713 *
1714 * In this case don't do the truncation, but we have to be
1715 * careful how we detect this case. Blocks beyond EOF show up as
1716 * i_delayed_blks even when the inode is clean, so we need to
1717 * truncate them away first before checking for a dirty release.
1718 * Hence on the first dirty close we will still remove the
1719 * speculative allocation, but after that we will leave it in
1720 * place.
1721 */
1722 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1723 return 0;
1724 /*
1725 * If we can't get the iolock just skip truncating the blocks
1726 * past EOF because we could deadlock with the mmap_sem
1727 * otherwise. We'll get another chance to drop them once the
1728 * last reference to the inode is dropped, so we'll never leak
1729 * blocks permanently.
1730 */
1731 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1732 error = xfs_free_eofblocks(ip);
1733 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1734 if (error)
1735 return error;
1736 }
1737
1738 /* delalloc blocks after truncation means it really is dirty */
1739 if (ip->i_delayed_blks)
1740 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1741 }
1742 return 0;
1743 }
1744
1745 /*
1746 * xfs_inactive_truncate
1747 *
1748 * Called to perform a truncate when an inode becomes unlinked.
1749 */
1750 STATIC int
xfs_inactive_truncate(struct xfs_inode * ip)1751 xfs_inactive_truncate(
1752 struct xfs_inode *ip)
1753 {
1754 struct xfs_mount *mp = ip->i_mount;
1755 struct xfs_trans *tp;
1756 int error;
1757
1758 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1759 if (error) {
1760 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1761 return error;
1762 }
1763
1764 xfs_ilock(ip, XFS_ILOCK_EXCL);
1765 xfs_trans_ijoin(tp, ip, 0);
1766
1767 /*
1768 * Log the inode size first to prevent stale data exposure in the event
1769 * of a system crash before the truncate completes. See the related
1770 * comment in xfs_vn_setattr_size() for details.
1771 */
1772 ip->i_d.di_size = 0;
1773 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1774
1775 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1776 if (error)
1777 goto error_trans_cancel;
1778
1779 ASSERT(ip->i_d.di_nextents == 0);
1780
1781 error = xfs_trans_commit(tp);
1782 if (error)
1783 goto error_unlock;
1784
1785 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1786 return 0;
1787
1788 error_trans_cancel:
1789 xfs_trans_cancel(tp);
1790 error_unlock:
1791 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1792 return error;
1793 }
1794
1795 /*
1796 * xfs_inactive_ifree()
1797 *
1798 * Perform the inode free when an inode is unlinked.
1799 */
1800 STATIC int
xfs_inactive_ifree(struct xfs_inode * ip)1801 xfs_inactive_ifree(
1802 struct xfs_inode *ip)
1803 {
1804 struct xfs_defer_ops dfops;
1805 xfs_fsblock_t first_block;
1806 struct xfs_mount *mp = ip->i_mount;
1807 struct xfs_trans *tp;
1808 int error;
1809
1810 /*
1811 * We try to use a per-AG reservation for any block needed by the finobt
1812 * tree, but as the finobt feature predates the per-AG reservation
1813 * support a degraded file system might not have enough space for the
1814 * reservation at mount time. In that case try to dip into the reserved
1815 * pool and pray.
1816 *
1817 * Send a warning if the reservation does happen to fail, as the inode
1818 * now remains allocated and sits on the unlinked list until the fs is
1819 * repaired.
1820 */
1821 if (unlikely(mp->m_inotbt_nores)) {
1822 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1823 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1824 &tp);
1825 } else {
1826 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1827 }
1828 if (error) {
1829 if (error == -ENOSPC) {
1830 xfs_warn_ratelimited(mp,
1831 "Failed to remove inode(s) from unlinked list. "
1832 "Please free space, unmount and run xfs_repair.");
1833 } else {
1834 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1835 }
1836 return error;
1837 }
1838
1839 xfs_ilock(ip, XFS_ILOCK_EXCL);
1840 xfs_trans_ijoin(tp, ip, 0);
1841
1842 xfs_defer_init(&dfops, &first_block);
1843 error = xfs_ifree(tp, ip, &dfops);
1844 if (error) {
1845 /*
1846 * If we fail to free the inode, shut down. The cancel
1847 * might do that, we need to make sure. Otherwise the
1848 * inode might be lost for a long time or forever.
1849 */
1850 if (!XFS_FORCED_SHUTDOWN(mp)) {
1851 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1852 __func__, error);
1853 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1854 }
1855 xfs_trans_cancel(tp);
1856 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1857 return error;
1858 }
1859
1860 /*
1861 * Credit the quota account(s). The inode is gone.
1862 */
1863 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1864
1865 /*
1866 * Just ignore errors at this point. There is nothing we can do except
1867 * to try to keep going. Make sure it's not a silent error.
1868 */
1869 error = xfs_defer_finish(&tp, &dfops, NULL);
1870 if (error) {
1871 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1872 __func__, error);
1873 xfs_defer_cancel(&dfops);
1874 }
1875 error = xfs_trans_commit(tp);
1876 if (error)
1877 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1878 __func__, error);
1879
1880 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1881 return 0;
1882 }
1883
1884 /*
1885 * xfs_inactive
1886 *
1887 * This is called when the vnode reference count for the vnode
1888 * goes to zero. If the file has been unlinked, then it must
1889 * now be truncated. Also, we clear all of the read-ahead state
1890 * kept for the inode here since the file is now closed.
1891 */
1892 void
xfs_inactive(xfs_inode_t * ip)1893 xfs_inactive(
1894 xfs_inode_t *ip)
1895 {
1896 struct xfs_mount *mp;
1897 int error;
1898 int truncate = 0;
1899
1900 /*
1901 * If the inode is already free, then there can be nothing
1902 * to clean up here.
1903 */
1904 if (VFS_I(ip)->i_mode == 0) {
1905 ASSERT(ip->i_df.if_real_bytes == 0);
1906 ASSERT(ip->i_df.if_broot_bytes == 0);
1907 return;
1908 }
1909
1910 mp = ip->i_mount;
1911 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1912
1913 /* If this is a read-only mount, don't do this (would generate I/O) */
1914 if (mp->m_flags & XFS_MOUNT_RDONLY)
1915 return;
1916
1917 if (VFS_I(ip)->i_nlink != 0) {
1918 /*
1919 * force is true because we are evicting an inode from the
1920 * cache. Post-eof blocks must be freed, lest we end up with
1921 * broken free space accounting.
1922 *
1923 * Note: don't bother with iolock here since lockdep complains
1924 * about acquiring it in reclaim context. We have the only
1925 * reference to the inode at this point anyways.
1926 */
1927 if (xfs_can_free_eofblocks(ip, true))
1928 xfs_free_eofblocks(ip);
1929
1930 return;
1931 }
1932
1933 if (S_ISREG(VFS_I(ip)->i_mode) &&
1934 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1935 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1936 truncate = 1;
1937
1938 error = xfs_qm_dqattach(ip, 0);
1939 if (error)
1940 return;
1941
1942 if (S_ISLNK(VFS_I(ip)->i_mode))
1943 error = xfs_inactive_symlink(ip);
1944 else if (truncate)
1945 error = xfs_inactive_truncate(ip);
1946 if (error)
1947 return;
1948
1949 /*
1950 * If there are attributes associated with the file then blow them away
1951 * now. The code calls a routine that recursively deconstructs the
1952 * attribute fork. If also blows away the in-core attribute fork.
1953 */
1954 if (XFS_IFORK_Q(ip)) {
1955 error = xfs_attr_inactive(ip);
1956 if (error)
1957 return;
1958 }
1959
1960 ASSERT(!ip->i_afp);
1961 ASSERT(ip->i_d.di_anextents == 0);
1962 ASSERT(ip->i_d.di_forkoff == 0);
1963
1964 /*
1965 * Free the inode.
1966 */
1967 error = xfs_inactive_ifree(ip);
1968 if (error)
1969 return;
1970
1971 /*
1972 * Release the dquots held by inode, if any.
1973 */
1974 xfs_qm_dqdetach(ip);
1975 }
1976
1977 /*
1978 * This is called when the inode's link count goes to 0 or we are creating a
1979 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1980 * set to true as the link count is dropped to zero by the VFS after we've
1981 * created the file successfully, so we have to add it to the unlinked list
1982 * while the link count is non-zero.
1983 *
1984 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1985 * list when the inode is freed.
1986 */
1987 STATIC int
xfs_iunlink(struct xfs_trans * tp,struct xfs_inode * ip)1988 xfs_iunlink(
1989 struct xfs_trans *tp,
1990 struct xfs_inode *ip)
1991 {
1992 xfs_mount_t *mp = tp->t_mountp;
1993 xfs_agi_t *agi;
1994 xfs_dinode_t *dip;
1995 xfs_buf_t *agibp;
1996 xfs_buf_t *ibp;
1997 xfs_agino_t agino;
1998 short bucket_index;
1999 int offset;
2000 int error;
2001
2002 ASSERT(VFS_I(ip)->i_mode != 0);
2003
2004 /*
2005 * Get the agi buffer first. It ensures lock ordering
2006 * on the list.
2007 */
2008 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
2009 if (error)
2010 return error;
2011 agi = XFS_BUF_TO_AGI(agibp);
2012
2013 /*
2014 * Get the index into the agi hash table for the
2015 * list this inode will go on.
2016 */
2017 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2018 ASSERT(agino != 0);
2019 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2020 ASSERT(agi->agi_unlinked[bucket_index]);
2021 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2022
2023 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2024 /*
2025 * There is already another inode in the bucket we need
2026 * to add ourselves to. Add us at the front of the list.
2027 * Here we put the head pointer into our next pointer,
2028 * and then we fall through to point the head at us.
2029 */
2030 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2031 0, 0);
2032 if (error)
2033 return error;
2034
2035 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2036 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2037 offset = ip->i_imap.im_boffset +
2038 offsetof(xfs_dinode_t, di_next_unlinked);
2039
2040 /* need to recalc the inode CRC if appropriate */
2041 xfs_dinode_calc_crc(mp, dip);
2042
2043 xfs_trans_inode_buf(tp, ibp);
2044 xfs_trans_log_buf(tp, ibp, offset,
2045 (offset + sizeof(xfs_agino_t) - 1));
2046 xfs_inobp_check(mp, ibp);
2047 }
2048
2049 /*
2050 * Point the bucket head pointer at the inode being inserted.
2051 */
2052 ASSERT(agino != 0);
2053 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2054 offset = offsetof(xfs_agi_t, agi_unlinked) +
2055 (sizeof(xfs_agino_t) * bucket_index);
2056 xfs_trans_log_buf(tp, agibp, offset,
2057 (offset + sizeof(xfs_agino_t) - 1));
2058 return 0;
2059 }
2060
2061 /*
2062 * Pull the on-disk inode from the AGI unlinked list.
2063 */
2064 STATIC int
xfs_iunlink_remove(xfs_trans_t * tp,xfs_inode_t * ip)2065 xfs_iunlink_remove(
2066 xfs_trans_t *tp,
2067 xfs_inode_t *ip)
2068 {
2069 xfs_ino_t next_ino;
2070 xfs_mount_t *mp;
2071 xfs_agi_t *agi;
2072 xfs_dinode_t *dip;
2073 xfs_buf_t *agibp;
2074 xfs_buf_t *ibp;
2075 xfs_agnumber_t agno;
2076 xfs_agino_t agino;
2077 xfs_agino_t next_agino;
2078 xfs_buf_t *last_ibp;
2079 xfs_dinode_t *last_dip = NULL;
2080 short bucket_index;
2081 int offset, last_offset = 0;
2082 int error;
2083
2084 mp = tp->t_mountp;
2085 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2086
2087 /*
2088 * Get the agi buffer first. It ensures lock ordering
2089 * on the list.
2090 */
2091 error = xfs_read_agi(mp, tp, agno, &agibp);
2092 if (error)
2093 return error;
2094
2095 agi = XFS_BUF_TO_AGI(agibp);
2096
2097 /*
2098 * Get the index into the agi hash table for the
2099 * list this inode will go on.
2100 */
2101 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2102 ASSERT(agino != 0);
2103 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2104 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2105 ASSERT(agi->agi_unlinked[bucket_index]);
2106
2107 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2108 /*
2109 * We're at the head of the list. Get the inode's on-disk
2110 * buffer to see if there is anyone after us on the list.
2111 * Only modify our next pointer if it is not already NULLAGINO.
2112 * This saves us the overhead of dealing with the buffer when
2113 * there is no need to change it.
2114 */
2115 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2116 0, 0);
2117 if (error) {
2118 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2119 __func__, error);
2120 return error;
2121 }
2122 next_agino = be32_to_cpu(dip->di_next_unlinked);
2123 ASSERT(next_agino != 0);
2124 if (next_agino != NULLAGINO) {
2125 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2126 offset = ip->i_imap.im_boffset +
2127 offsetof(xfs_dinode_t, di_next_unlinked);
2128
2129 /* need to recalc the inode CRC if appropriate */
2130 xfs_dinode_calc_crc(mp, dip);
2131
2132 xfs_trans_inode_buf(tp, ibp);
2133 xfs_trans_log_buf(tp, ibp, offset,
2134 (offset + sizeof(xfs_agino_t) - 1));
2135 xfs_inobp_check(mp, ibp);
2136 } else {
2137 xfs_trans_brelse(tp, ibp);
2138 }
2139 /*
2140 * Point the bucket head pointer at the next inode.
2141 */
2142 ASSERT(next_agino != 0);
2143 ASSERT(next_agino != agino);
2144 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2145 offset = offsetof(xfs_agi_t, agi_unlinked) +
2146 (sizeof(xfs_agino_t) * bucket_index);
2147 xfs_trans_log_buf(tp, agibp, offset,
2148 (offset + sizeof(xfs_agino_t) - 1));
2149 } else {
2150 /*
2151 * We need to search the list for the inode being freed.
2152 */
2153 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2154 last_ibp = NULL;
2155 while (next_agino != agino) {
2156 struct xfs_imap imap;
2157
2158 if (last_ibp)
2159 xfs_trans_brelse(tp, last_ibp);
2160
2161 imap.im_blkno = 0;
2162 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2163
2164 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2165 if (error) {
2166 xfs_warn(mp,
2167 "%s: xfs_imap returned error %d.",
2168 __func__, error);
2169 return error;
2170 }
2171
2172 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2173 &last_ibp, 0, 0);
2174 if (error) {
2175 xfs_warn(mp,
2176 "%s: xfs_imap_to_bp returned error %d.",
2177 __func__, error);
2178 return error;
2179 }
2180
2181 last_offset = imap.im_boffset;
2182 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2183 ASSERT(next_agino != NULLAGINO);
2184 ASSERT(next_agino != 0);
2185 }
2186
2187 /*
2188 * Now last_ibp points to the buffer previous to us on the
2189 * unlinked list. Pull us from the list.
2190 */
2191 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2192 0, 0);
2193 if (error) {
2194 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2195 __func__, error);
2196 return error;
2197 }
2198 next_agino = be32_to_cpu(dip->di_next_unlinked);
2199 ASSERT(next_agino != 0);
2200 ASSERT(next_agino != agino);
2201 if (next_agino != NULLAGINO) {
2202 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2203 offset = ip->i_imap.im_boffset +
2204 offsetof(xfs_dinode_t, di_next_unlinked);
2205
2206 /* need to recalc the inode CRC if appropriate */
2207 xfs_dinode_calc_crc(mp, dip);
2208
2209 xfs_trans_inode_buf(tp, ibp);
2210 xfs_trans_log_buf(tp, ibp, offset,
2211 (offset + sizeof(xfs_agino_t) - 1));
2212 xfs_inobp_check(mp, ibp);
2213 } else {
2214 xfs_trans_brelse(tp, ibp);
2215 }
2216 /*
2217 * Point the previous inode on the list to the next inode.
2218 */
2219 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2220 ASSERT(next_agino != 0);
2221 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2222
2223 /* need to recalc the inode CRC if appropriate */
2224 xfs_dinode_calc_crc(mp, last_dip);
2225
2226 xfs_trans_inode_buf(tp, last_ibp);
2227 xfs_trans_log_buf(tp, last_ibp, offset,
2228 (offset + sizeof(xfs_agino_t) - 1));
2229 xfs_inobp_check(mp, last_ibp);
2230 }
2231 return 0;
2232 }
2233
2234 /*
2235 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2236 * inodes that are in memory - they all must be marked stale and attached to
2237 * the cluster buffer.
2238 */
2239 STATIC int
xfs_ifree_cluster(xfs_inode_t * free_ip,xfs_trans_t * tp,struct xfs_icluster * xic)2240 xfs_ifree_cluster(
2241 xfs_inode_t *free_ip,
2242 xfs_trans_t *tp,
2243 struct xfs_icluster *xic)
2244 {
2245 xfs_mount_t *mp = free_ip->i_mount;
2246 int blks_per_cluster;
2247 int inodes_per_cluster;
2248 int nbufs;
2249 int i, j;
2250 int ioffset;
2251 xfs_daddr_t blkno;
2252 xfs_buf_t *bp;
2253 xfs_inode_t *ip;
2254 xfs_inode_log_item_t *iip;
2255 xfs_log_item_t *lip;
2256 struct xfs_perag *pag;
2257 xfs_ino_t inum;
2258
2259 inum = xic->first_ino;
2260 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2261 blks_per_cluster = xfs_icluster_size_fsb(mp);
2262 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2263 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2264
2265 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2266 /*
2267 * The allocation bitmap tells us which inodes of the chunk were
2268 * physically allocated. Skip the cluster if an inode falls into
2269 * a sparse region.
2270 */
2271 ioffset = inum - xic->first_ino;
2272 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2273 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2274 continue;
2275 }
2276
2277 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2278 XFS_INO_TO_AGBNO(mp, inum));
2279
2280 /*
2281 * We obtain and lock the backing buffer first in the process
2282 * here, as we have to ensure that any dirty inode that we
2283 * can't get the flush lock on is attached to the buffer.
2284 * If we scan the in-memory inodes first, then buffer IO can
2285 * complete before we get a lock on it, and hence we may fail
2286 * to mark all the active inodes on the buffer stale.
2287 */
2288 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2289 mp->m_bsize * blks_per_cluster,
2290 XBF_UNMAPPED);
2291
2292 if (!bp)
2293 return -ENOMEM;
2294
2295 /*
2296 * This buffer may not have been correctly initialised as we
2297 * didn't read it from disk. That's not important because we are
2298 * only using to mark the buffer as stale in the log, and to
2299 * attach stale cached inodes on it. That means it will never be
2300 * dispatched for IO. If it is, we want to know about it, and we
2301 * want it to fail. We can acheive this by adding a write
2302 * verifier to the buffer.
2303 */
2304 bp->b_ops = &xfs_inode_buf_ops;
2305
2306 /*
2307 * Walk the inodes already attached to the buffer and mark them
2308 * stale. These will all have the flush locks held, so an
2309 * in-memory inode walk can't lock them. By marking them all
2310 * stale first, we will not attempt to lock them in the loop
2311 * below as the XFS_ISTALE flag will be set.
2312 */
2313 lip = bp->b_fspriv;
2314 while (lip) {
2315 if (lip->li_type == XFS_LI_INODE) {
2316 iip = (xfs_inode_log_item_t *)lip;
2317 ASSERT(iip->ili_logged == 1);
2318 lip->li_cb = xfs_istale_done;
2319 xfs_trans_ail_copy_lsn(mp->m_ail,
2320 &iip->ili_flush_lsn,
2321 &iip->ili_item.li_lsn);
2322 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2323 }
2324 lip = lip->li_bio_list;
2325 }
2326
2327
2328 /*
2329 * For each inode in memory attempt to add it to the inode
2330 * buffer and set it up for being staled on buffer IO
2331 * completion. This is safe as we've locked out tail pushing
2332 * and flushing by locking the buffer.
2333 *
2334 * We have already marked every inode that was part of a
2335 * transaction stale above, which means there is no point in
2336 * even trying to lock them.
2337 */
2338 for (i = 0; i < inodes_per_cluster; i++) {
2339 retry:
2340 rcu_read_lock();
2341 ip = radix_tree_lookup(&pag->pag_ici_root,
2342 XFS_INO_TO_AGINO(mp, (inum + i)));
2343
2344 /* Inode not in memory, nothing to do */
2345 if (!ip) {
2346 rcu_read_unlock();
2347 continue;
2348 }
2349
2350 /*
2351 * because this is an RCU protected lookup, we could
2352 * find a recently freed or even reallocated inode
2353 * during the lookup. We need to check under the
2354 * i_flags_lock for a valid inode here. Skip it if it
2355 * is not valid, the wrong inode or stale.
2356 */
2357 spin_lock(&ip->i_flags_lock);
2358 if (ip->i_ino != inum + i ||
2359 __xfs_iflags_test(ip, XFS_ISTALE)) {
2360 spin_unlock(&ip->i_flags_lock);
2361 rcu_read_unlock();
2362 continue;
2363 }
2364 spin_unlock(&ip->i_flags_lock);
2365
2366 /*
2367 * Don't try to lock/unlock the current inode, but we
2368 * _cannot_ skip the other inodes that we did not find
2369 * in the list attached to the buffer and are not
2370 * already marked stale. If we can't lock it, back off
2371 * and retry.
2372 */
2373 if (ip != free_ip) {
2374 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2375 rcu_read_unlock();
2376 delay(1);
2377 goto retry;
2378 }
2379
2380 /*
2381 * Check the inode number again in case we're
2382 * racing with freeing in xfs_reclaim_inode().
2383 * See the comments in that function for more
2384 * information as to why the initial check is
2385 * not sufficient.
2386 */
2387 if (ip->i_ino != inum + i) {
2388 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2389 rcu_read_unlock();
2390 continue;
2391 }
2392 }
2393 rcu_read_unlock();
2394
2395 xfs_iflock(ip);
2396 xfs_iflags_set(ip, XFS_ISTALE);
2397
2398 /*
2399 * we don't need to attach clean inodes or those only
2400 * with unlogged changes (which we throw away, anyway).
2401 */
2402 iip = ip->i_itemp;
2403 if (!iip || xfs_inode_clean(ip)) {
2404 ASSERT(ip != free_ip);
2405 xfs_ifunlock(ip);
2406 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2407 continue;
2408 }
2409
2410 iip->ili_last_fields = iip->ili_fields;
2411 iip->ili_fields = 0;
2412 iip->ili_fsync_fields = 0;
2413 iip->ili_logged = 1;
2414 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2415 &iip->ili_item.li_lsn);
2416
2417 xfs_buf_attach_iodone(bp, xfs_istale_done,
2418 &iip->ili_item);
2419
2420 if (ip != free_ip)
2421 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2422 }
2423
2424 xfs_trans_stale_inode_buf(tp, bp);
2425 xfs_trans_binval(tp, bp);
2426 }
2427
2428 xfs_perag_put(pag);
2429 return 0;
2430 }
2431
2432 /*
2433 * Free any local-format buffers sitting around before we reset to
2434 * extents format.
2435 */
2436 static inline void
xfs_ifree_local_data(struct xfs_inode * ip,int whichfork)2437 xfs_ifree_local_data(
2438 struct xfs_inode *ip,
2439 int whichfork)
2440 {
2441 struct xfs_ifork *ifp;
2442
2443 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2444 return;
2445
2446 ifp = XFS_IFORK_PTR(ip, whichfork);
2447 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2448 }
2449
2450 /*
2451 * This is called to return an inode to the inode free list.
2452 * The inode should already be truncated to 0 length and have
2453 * no pages associated with it. This routine also assumes that
2454 * the inode is already a part of the transaction.
2455 *
2456 * The on-disk copy of the inode will have been added to the list
2457 * of unlinked inodes in the AGI. We need to remove the inode from
2458 * that list atomically with respect to freeing it here.
2459 */
2460 int
xfs_ifree(xfs_trans_t * tp,xfs_inode_t * ip,struct xfs_defer_ops * dfops)2461 xfs_ifree(
2462 xfs_trans_t *tp,
2463 xfs_inode_t *ip,
2464 struct xfs_defer_ops *dfops)
2465 {
2466 int error;
2467 struct xfs_icluster xic = { 0 };
2468
2469 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2470 ASSERT(VFS_I(ip)->i_nlink == 0);
2471 ASSERT(ip->i_d.di_nextents == 0);
2472 ASSERT(ip->i_d.di_anextents == 0);
2473 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2474 ASSERT(ip->i_d.di_nblocks == 0);
2475
2476 /*
2477 * Pull the on-disk inode from the AGI unlinked list.
2478 */
2479 error = xfs_iunlink_remove(tp, ip);
2480 if (error)
2481 return error;
2482
2483 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2484 if (error)
2485 return error;
2486
2487 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2488 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2489
2490 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2491 ip->i_d.di_flags = 0;
2492 ip->i_d.di_dmevmask = 0;
2493 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2494 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2495 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2496 /*
2497 * Bump the generation count so no one will be confused
2498 * by reincarnations of this inode.
2499 */
2500 VFS_I(ip)->i_generation++;
2501 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2502
2503 if (xic.deleted)
2504 error = xfs_ifree_cluster(ip, tp, &xic);
2505
2506 return error;
2507 }
2508
2509 /*
2510 * This is called to unpin an inode. The caller must have the inode locked
2511 * in at least shared mode so that the buffer cannot be subsequently pinned
2512 * once someone is waiting for it to be unpinned.
2513 */
2514 static void
xfs_iunpin(struct xfs_inode * ip)2515 xfs_iunpin(
2516 struct xfs_inode *ip)
2517 {
2518 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2519
2520 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2521
2522 /* Give the log a push to start the unpinning I/O */
2523 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2524
2525 }
2526
2527 static void
__xfs_iunpin_wait(struct xfs_inode * ip)2528 __xfs_iunpin_wait(
2529 struct xfs_inode *ip)
2530 {
2531 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2532 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2533
2534 xfs_iunpin(ip);
2535
2536 do {
2537 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2538 if (xfs_ipincount(ip))
2539 io_schedule();
2540 } while (xfs_ipincount(ip));
2541 finish_wait(wq, &wait.wait);
2542 }
2543
2544 void
xfs_iunpin_wait(struct xfs_inode * ip)2545 xfs_iunpin_wait(
2546 struct xfs_inode *ip)
2547 {
2548 if (xfs_ipincount(ip))
2549 __xfs_iunpin_wait(ip);
2550 }
2551
2552 /*
2553 * Removing an inode from the namespace involves removing the directory entry
2554 * and dropping the link count on the inode. Removing the directory entry can
2555 * result in locking an AGF (directory blocks were freed) and removing a link
2556 * count can result in placing the inode on an unlinked list which results in
2557 * locking an AGI.
2558 *
2559 * The big problem here is that we have an ordering constraint on AGF and AGI
2560 * locking - inode allocation locks the AGI, then can allocate a new extent for
2561 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2562 * removes the inode from the unlinked list, requiring that we lock the AGI
2563 * first, and then freeing the inode can result in an inode chunk being freed
2564 * and hence freeing disk space requiring that we lock an AGF.
2565 *
2566 * Hence the ordering that is imposed by other parts of the code is AGI before
2567 * AGF. This means we cannot remove the directory entry before we drop the inode
2568 * reference count and put it on the unlinked list as this results in a lock
2569 * order of AGF then AGI, and this can deadlock against inode allocation and
2570 * freeing. Therefore we must drop the link counts before we remove the
2571 * directory entry.
2572 *
2573 * This is still safe from a transactional point of view - it is not until we
2574 * get to xfs_defer_finish() that we have the possibility of multiple
2575 * transactions in this operation. Hence as long as we remove the directory
2576 * entry and drop the link count in the first transaction of the remove
2577 * operation, there are no transactional constraints on the ordering here.
2578 */
2579 int
xfs_remove(xfs_inode_t * dp,struct xfs_name * name,xfs_inode_t * ip)2580 xfs_remove(
2581 xfs_inode_t *dp,
2582 struct xfs_name *name,
2583 xfs_inode_t *ip)
2584 {
2585 xfs_mount_t *mp = dp->i_mount;
2586 xfs_trans_t *tp = NULL;
2587 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2588 int error = 0;
2589 struct xfs_defer_ops dfops;
2590 xfs_fsblock_t first_block;
2591 uint resblks;
2592
2593 trace_xfs_remove(dp, name);
2594
2595 if (XFS_FORCED_SHUTDOWN(mp))
2596 return -EIO;
2597
2598 error = xfs_qm_dqattach(dp, 0);
2599 if (error)
2600 goto std_return;
2601
2602 error = xfs_qm_dqattach(ip, 0);
2603 if (error)
2604 goto std_return;
2605
2606 /*
2607 * We try to get the real space reservation first,
2608 * allowing for directory btree deletion(s) implying
2609 * possible bmap insert(s). If we can't get the space
2610 * reservation then we use 0 instead, and avoid the bmap
2611 * btree insert(s) in the directory code by, if the bmap
2612 * insert tries to happen, instead trimming the LAST
2613 * block from the directory.
2614 */
2615 resblks = XFS_REMOVE_SPACE_RES(mp);
2616 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2617 if (error == -ENOSPC) {
2618 resblks = 0;
2619 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2620 &tp);
2621 }
2622 if (error) {
2623 ASSERT(error != -ENOSPC);
2624 goto std_return;
2625 }
2626
2627 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2628 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2629
2630 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2631 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2632
2633 /*
2634 * If we're removing a directory perform some additional validation.
2635 */
2636 if (is_dir) {
2637 ASSERT(VFS_I(ip)->i_nlink >= 2);
2638 if (VFS_I(ip)->i_nlink != 2) {
2639 error = -ENOTEMPTY;
2640 goto out_trans_cancel;
2641 }
2642 if (!xfs_dir_isempty(ip)) {
2643 error = -ENOTEMPTY;
2644 goto out_trans_cancel;
2645 }
2646
2647 /* Drop the link from ip's "..". */
2648 error = xfs_droplink(tp, dp);
2649 if (error)
2650 goto out_trans_cancel;
2651
2652 /* Drop the "." link from ip to self. */
2653 error = xfs_droplink(tp, ip);
2654 if (error)
2655 goto out_trans_cancel;
2656 } else {
2657 /*
2658 * When removing a non-directory we need to log the parent
2659 * inode here. For a directory this is done implicitly
2660 * by the xfs_droplink call for the ".." entry.
2661 */
2662 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2663 }
2664 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2665
2666 /* Drop the link from dp to ip. */
2667 error = xfs_droplink(tp, ip);
2668 if (error)
2669 goto out_trans_cancel;
2670
2671 xfs_defer_init(&dfops, &first_block);
2672 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2673 &first_block, &dfops, resblks);
2674 if (error) {
2675 ASSERT(error != -ENOENT);
2676 goto out_bmap_cancel;
2677 }
2678
2679 /*
2680 * If this is a synchronous mount, make sure that the
2681 * remove transaction goes to disk before returning to
2682 * the user.
2683 */
2684 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2685 xfs_trans_set_sync(tp);
2686
2687 error = xfs_defer_finish(&tp, &dfops, NULL);
2688 if (error)
2689 goto out_bmap_cancel;
2690
2691 error = xfs_trans_commit(tp);
2692 if (error)
2693 goto std_return;
2694
2695 if (is_dir && xfs_inode_is_filestream(ip))
2696 xfs_filestream_deassociate(ip);
2697
2698 return 0;
2699
2700 out_bmap_cancel:
2701 xfs_defer_cancel(&dfops);
2702 out_trans_cancel:
2703 xfs_trans_cancel(tp);
2704 std_return:
2705 return error;
2706 }
2707
2708 /*
2709 * Enter all inodes for a rename transaction into a sorted array.
2710 */
2711 #define __XFS_SORT_INODES 5
2712 STATIC void
xfs_sort_for_rename(struct xfs_inode * dp1,struct xfs_inode * dp2,struct xfs_inode * ip1,struct xfs_inode * ip2,struct xfs_inode * wip,struct xfs_inode ** i_tab,int * num_inodes)2713 xfs_sort_for_rename(
2714 struct xfs_inode *dp1, /* in: old (source) directory inode */
2715 struct xfs_inode *dp2, /* in: new (target) directory inode */
2716 struct xfs_inode *ip1, /* in: inode of old entry */
2717 struct xfs_inode *ip2, /* in: inode of new entry */
2718 struct xfs_inode *wip, /* in: whiteout inode */
2719 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2720 int *num_inodes) /* in/out: inodes in array */
2721 {
2722 int i, j;
2723
2724 ASSERT(*num_inodes == __XFS_SORT_INODES);
2725 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2726
2727 /*
2728 * i_tab contains a list of pointers to inodes. We initialize
2729 * the table here & we'll sort it. We will then use it to
2730 * order the acquisition of the inode locks.
2731 *
2732 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2733 */
2734 i = 0;
2735 i_tab[i++] = dp1;
2736 i_tab[i++] = dp2;
2737 i_tab[i++] = ip1;
2738 if (ip2)
2739 i_tab[i++] = ip2;
2740 if (wip)
2741 i_tab[i++] = wip;
2742 *num_inodes = i;
2743
2744 /*
2745 * Sort the elements via bubble sort. (Remember, there are at
2746 * most 5 elements to sort, so this is adequate.)
2747 */
2748 for (i = 0; i < *num_inodes; i++) {
2749 for (j = 1; j < *num_inodes; j++) {
2750 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2751 struct xfs_inode *temp = i_tab[j];
2752 i_tab[j] = i_tab[j-1];
2753 i_tab[j-1] = temp;
2754 }
2755 }
2756 }
2757 }
2758
2759 static int
xfs_finish_rename(struct xfs_trans * tp,struct xfs_defer_ops * dfops)2760 xfs_finish_rename(
2761 struct xfs_trans *tp,
2762 struct xfs_defer_ops *dfops)
2763 {
2764 int error;
2765
2766 /*
2767 * If this is a synchronous mount, make sure that the rename transaction
2768 * goes to disk before returning to the user.
2769 */
2770 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2771 xfs_trans_set_sync(tp);
2772
2773 error = xfs_defer_finish(&tp, dfops, NULL);
2774 if (error) {
2775 xfs_defer_cancel(dfops);
2776 xfs_trans_cancel(tp);
2777 return error;
2778 }
2779
2780 return xfs_trans_commit(tp);
2781 }
2782
2783 /*
2784 * xfs_cross_rename()
2785 *
2786 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2787 */
2788 STATIC int
xfs_cross_rename(struct xfs_trans * tp,struct xfs_inode * dp1,struct xfs_name * name1,struct xfs_inode * ip1,struct xfs_inode * dp2,struct xfs_name * name2,struct xfs_inode * ip2,struct xfs_defer_ops * dfops,xfs_fsblock_t * first_block,int spaceres)2789 xfs_cross_rename(
2790 struct xfs_trans *tp,
2791 struct xfs_inode *dp1,
2792 struct xfs_name *name1,
2793 struct xfs_inode *ip1,
2794 struct xfs_inode *dp2,
2795 struct xfs_name *name2,
2796 struct xfs_inode *ip2,
2797 struct xfs_defer_ops *dfops,
2798 xfs_fsblock_t *first_block,
2799 int spaceres)
2800 {
2801 int error = 0;
2802 int ip1_flags = 0;
2803 int ip2_flags = 0;
2804 int dp2_flags = 0;
2805
2806 /* Swap inode number for dirent in first parent */
2807 error = xfs_dir_replace(tp, dp1, name1,
2808 ip2->i_ino,
2809 first_block, dfops, spaceres);
2810 if (error)
2811 goto out_trans_abort;
2812
2813 /* Swap inode number for dirent in second parent */
2814 error = xfs_dir_replace(tp, dp2, name2,
2815 ip1->i_ino,
2816 first_block, dfops, spaceres);
2817 if (error)
2818 goto out_trans_abort;
2819
2820 /*
2821 * If we're renaming one or more directories across different parents,
2822 * update the respective ".." entries (and link counts) to match the new
2823 * parents.
2824 */
2825 if (dp1 != dp2) {
2826 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2827
2828 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2829 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2830 dp1->i_ino, first_block,
2831 dfops, spaceres);
2832 if (error)
2833 goto out_trans_abort;
2834
2835 /* transfer ip2 ".." reference to dp1 */
2836 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2837 error = xfs_droplink(tp, dp2);
2838 if (error)
2839 goto out_trans_abort;
2840 error = xfs_bumplink(tp, dp1);
2841 if (error)
2842 goto out_trans_abort;
2843 }
2844
2845 /*
2846 * Although ip1 isn't changed here, userspace needs
2847 * to be warned about the change, so that applications
2848 * relying on it (like backup ones), will properly
2849 * notify the change
2850 */
2851 ip1_flags |= XFS_ICHGTIME_CHG;
2852 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2853 }
2854
2855 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2856 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2857 dp2->i_ino, first_block,
2858 dfops, spaceres);
2859 if (error)
2860 goto out_trans_abort;
2861
2862 /* transfer ip1 ".." reference to dp2 */
2863 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2864 error = xfs_droplink(tp, dp1);
2865 if (error)
2866 goto out_trans_abort;
2867 error = xfs_bumplink(tp, dp2);
2868 if (error)
2869 goto out_trans_abort;
2870 }
2871
2872 /*
2873 * Although ip2 isn't changed here, userspace needs
2874 * to be warned about the change, so that applications
2875 * relying on it (like backup ones), will properly
2876 * notify the change
2877 */
2878 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2879 ip2_flags |= XFS_ICHGTIME_CHG;
2880 }
2881 }
2882
2883 if (ip1_flags) {
2884 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2885 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2886 }
2887 if (ip2_flags) {
2888 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2889 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2890 }
2891 if (dp2_flags) {
2892 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2893 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2894 }
2895 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2896 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2897 return xfs_finish_rename(tp, dfops);
2898
2899 out_trans_abort:
2900 xfs_defer_cancel(dfops);
2901 xfs_trans_cancel(tp);
2902 return error;
2903 }
2904
2905 /*
2906 * xfs_rename_alloc_whiteout()
2907 *
2908 * Return a referenced, unlinked, unlocked inode that that can be used as a
2909 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2910 * crash between allocating the inode and linking it into the rename transaction
2911 * recovery will free the inode and we won't leak it.
2912 */
2913 static int
xfs_rename_alloc_whiteout(struct xfs_inode * dp,struct xfs_inode ** wip)2914 xfs_rename_alloc_whiteout(
2915 struct xfs_inode *dp,
2916 struct xfs_inode **wip)
2917 {
2918 struct xfs_inode *tmpfile;
2919 int error;
2920
2921 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2922 if (error)
2923 return error;
2924
2925 /*
2926 * Prepare the tmpfile inode as if it were created through the VFS.
2927 * Otherwise, the link increment paths will complain about nlink 0->1.
2928 * Drop the link count as done by d_tmpfile(), complete the inode setup
2929 * and flag it as linkable.
2930 */
2931 drop_nlink(VFS_I(tmpfile));
2932 xfs_setup_iops(tmpfile);
2933 xfs_finish_inode_setup(tmpfile);
2934 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2935
2936 *wip = tmpfile;
2937 return 0;
2938 }
2939
2940 /*
2941 * xfs_rename
2942 */
2943 int
xfs_rename(struct xfs_inode * src_dp,struct xfs_name * src_name,struct xfs_inode * src_ip,struct xfs_inode * target_dp,struct xfs_name * target_name,struct xfs_inode * target_ip,unsigned int flags)2944 xfs_rename(
2945 struct xfs_inode *src_dp,
2946 struct xfs_name *src_name,
2947 struct xfs_inode *src_ip,
2948 struct xfs_inode *target_dp,
2949 struct xfs_name *target_name,
2950 struct xfs_inode *target_ip,
2951 unsigned int flags)
2952 {
2953 struct xfs_mount *mp = src_dp->i_mount;
2954 struct xfs_trans *tp;
2955 struct xfs_defer_ops dfops;
2956 xfs_fsblock_t first_block;
2957 struct xfs_inode *wip = NULL; /* whiteout inode */
2958 struct xfs_inode *inodes[__XFS_SORT_INODES];
2959 int num_inodes = __XFS_SORT_INODES;
2960 bool new_parent = (src_dp != target_dp);
2961 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2962 int spaceres;
2963 int error;
2964
2965 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2966
2967 if ((flags & RENAME_EXCHANGE) && !target_ip)
2968 return -EINVAL;
2969
2970 /*
2971 * If we are doing a whiteout operation, allocate the whiteout inode
2972 * we will be placing at the target and ensure the type is set
2973 * appropriately.
2974 */
2975 if (flags & RENAME_WHITEOUT) {
2976 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2977 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2978 if (error)
2979 return error;
2980
2981 /* setup target dirent info as whiteout */
2982 src_name->type = XFS_DIR3_FT_CHRDEV;
2983 }
2984
2985 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2986 inodes, &num_inodes);
2987
2988 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2989 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2990 if (error == -ENOSPC) {
2991 spaceres = 0;
2992 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2993 &tp);
2994 }
2995 if (error)
2996 goto out_release_wip;
2997
2998 /*
2999 * Attach the dquots to the inodes
3000 */
3001 error = xfs_qm_vop_rename_dqattach(inodes);
3002 if (error)
3003 goto out_trans_cancel;
3004
3005 /*
3006 * Lock all the participating inodes. Depending upon whether
3007 * the target_name exists in the target directory, and
3008 * whether the target directory is the same as the source
3009 * directory, we can lock from 2 to 4 inodes.
3010 */
3011 if (!new_parent)
3012 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
3013 else
3014 xfs_lock_two_inodes(src_dp, target_dp,
3015 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
3016
3017 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3018
3019 /*
3020 * Join all the inodes to the transaction. From this point on,
3021 * we can rely on either trans_commit or trans_cancel to unlock
3022 * them.
3023 */
3024 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
3025 if (new_parent)
3026 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
3027 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3028 if (target_ip)
3029 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3030 if (wip)
3031 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3032
3033 /*
3034 * If we are using project inheritance, we only allow renames
3035 * into our tree when the project IDs are the same; else the
3036 * tree quota mechanism would be circumvented.
3037 */
3038 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3039 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3040 error = -EXDEV;
3041 goto out_trans_cancel;
3042 }
3043
3044 xfs_defer_init(&dfops, &first_block);
3045
3046 /* RENAME_EXCHANGE is unique from here on. */
3047 if (flags & RENAME_EXCHANGE)
3048 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3049 target_dp, target_name, target_ip,
3050 &dfops, &first_block, spaceres);
3051
3052 /*
3053 * Set up the target.
3054 */
3055 if (target_ip == NULL) {
3056 /*
3057 * If there's no space reservation, check the entry will
3058 * fit before actually inserting it.
3059 */
3060 if (!spaceres) {
3061 error = xfs_dir_canenter(tp, target_dp, target_name);
3062 if (error)
3063 goto out_trans_cancel;
3064 }
3065 /*
3066 * If target does not exist and the rename crosses
3067 * directories, adjust the target directory link count
3068 * to account for the ".." reference from the new entry.
3069 */
3070 error = xfs_dir_createname(tp, target_dp, target_name,
3071 src_ip->i_ino, &first_block,
3072 &dfops, spaceres);
3073 if (error)
3074 goto out_bmap_cancel;
3075
3076 xfs_trans_ichgtime(tp, target_dp,
3077 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3078
3079 if (new_parent && src_is_directory) {
3080 error = xfs_bumplink(tp, target_dp);
3081 if (error)
3082 goto out_bmap_cancel;
3083 }
3084 } else { /* target_ip != NULL */
3085 /*
3086 * If target exists and it's a directory, check that both
3087 * target and source are directories and that target can be
3088 * destroyed, or that neither is a directory.
3089 */
3090 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3091 /*
3092 * Make sure target dir is empty.
3093 */
3094 if (!(xfs_dir_isempty(target_ip)) ||
3095 (VFS_I(target_ip)->i_nlink > 2)) {
3096 error = -EEXIST;
3097 goto out_trans_cancel;
3098 }
3099 }
3100
3101 /*
3102 * Link the source inode under the target name.
3103 * If the source inode is a directory and we are moving
3104 * it across directories, its ".." entry will be
3105 * inconsistent until we replace that down below.
3106 *
3107 * In case there is already an entry with the same
3108 * name at the destination directory, remove it first.
3109 */
3110 error = xfs_dir_replace(tp, target_dp, target_name,
3111 src_ip->i_ino,
3112 &first_block, &dfops, spaceres);
3113 if (error)
3114 goto out_bmap_cancel;
3115
3116 xfs_trans_ichgtime(tp, target_dp,
3117 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3118
3119 /*
3120 * Decrement the link count on the target since the target
3121 * dir no longer points to it.
3122 */
3123 error = xfs_droplink(tp, target_ip);
3124 if (error)
3125 goto out_bmap_cancel;
3126
3127 if (src_is_directory) {
3128 /*
3129 * Drop the link from the old "." entry.
3130 */
3131 error = xfs_droplink(tp, target_ip);
3132 if (error)
3133 goto out_bmap_cancel;
3134 }
3135 } /* target_ip != NULL */
3136
3137 /*
3138 * Remove the source.
3139 */
3140 if (new_parent && src_is_directory) {
3141 /*
3142 * Rewrite the ".." entry to point to the new
3143 * directory.
3144 */
3145 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3146 target_dp->i_ino,
3147 &first_block, &dfops, spaceres);
3148 ASSERT(error != -EEXIST);
3149 if (error)
3150 goto out_bmap_cancel;
3151 }
3152
3153 /*
3154 * We always want to hit the ctime on the source inode.
3155 *
3156 * This isn't strictly required by the standards since the source
3157 * inode isn't really being changed, but old unix file systems did
3158 * it and some incremental backup programs won't work without it.
3159 */
3160 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3161 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3162
3163 /*
3164 * Adjust the link count on src_dp. This is necessary when
3165 * renaming a directory, either within one parent when
3166 * the target existed, or across two parent directories.
3167 */
3168 if (src_is_directory && (new_parent || target_ip != NULL)) {
3169
3170 /*
3171 * Decrement link count on src_directory since the
3172 * entry that's moved no longer points to it.
3173 */
3174 error = xfs_droplink(tp, src_dp);
3175 if (error)
3176 goto out_bmap_cancel;
3177 }
3178
3179 /*
3180 * For whiteouts, we only need to update the source dirent with the
3181 * inode number of the whiteout inode rather than removing it
3182 * altogether.
3183 */
3184 if (wip) {
3185 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3186 &first_block, &dfops, spaceres);
3187 } else
3188 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3189 &first_block, &dfops, spaceres);
3190 if (error)
3191 goto out_bmap_cancel;
3192
3193 /*
3194 * For whiteouts, we need to bump the link count on the whiteout inode.
3195 * This means that failures all the way up to this point leave the inode
3196 * on the unlinked list and so cleanup is a simple matter of dropping
3197 * the remaining reference to it. If we fail here after bumping the link
3198 * count, we're shutting down the filesystem so we'll never see the
3199 * intermediate state on disk.
3200 */
3201 if (wip) {
3202 ASSERT(VFS_I(wip)->i_nlink == 0);
3203 error = xfs_bumplink(tp, wip);
3204 if (error)
3205 goto out_bmap_cancel;
3206 error = xfs_iunlink_remove(tp, wip);
3207 if (error)
3208 goto out_bmap_cancel;
3209 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3210
3211 /*
3212 * Now we have a real link, clear the "I'm a tmpfile" state
3213 * flag from the inode so it doesn't accidentally get misused in
3214 * future.
3215 */
3216 VFS_I(wip)->i_state &= ~I_LINKABLE;
3217 }
3218
3219 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3220 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3221 if (new_parent)
3222 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3223
3224 error = xfs_finish_rename(tp, &dfops);
3225 if (wip)
3226 IRELE(wip);
3227 return error;
3228
3229 out_bmap_cancel:
3230 xfs_defer_cancel(&dfops);
3231 out_trans_cancel:
3232 xfs_trans_cancel(tp);
3233 out_release_wip:
3234 if (wip)
3235 IRELE(wip);
3236 return error;
3237 }
3238
3239 STATIC int
xfs_iflush_cluster(struct xfs_inode * ip,struct xfs_buf * bp)3240 xfs_iflush_cluster(
3241 struct xfs_inode *ip,
3242 struct xfs_buf *bp)
3243 {
3244 struct xfs_mount *mp = ip->i_mount;
3245 struct xfs_perag *pag;
3246 unsigned long first_index, mask;
3247 unsigned long inodes_per_cluster;
3248 int cilist_size;
3249 struct xfs_inode **cilist;
3250 struct xfs_inode *cip;
3251 int nr_found;
3252 int clcount = 0;
3253 int bufwasdelwri;
3254 int i;
3255
3256 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3257
3258 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3259 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3260 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3261 if (!cilist)
3262 goto out_put;
3263
3264 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3265 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3266 rcu_read_lock();
3267 /* really need a gang lookup range call here */
3268 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3269 first_index, inodes_per_cluster);
3270 if (nr_found == 0)
3271 goto out_free;
3272
3273 for (i = 0; i < nr_found; i++) {
3274 cip = cilist[i];
3275 if (cip == ip)
3276 continue;
3277
3278 /*
3279 * because this is an RCU protected lookup, we could find a
3280 * recently freed or even reallocated inode during the lookup.
3281 * We need to check under the i_flags_lock for a valid inode
3282 * here. Skip it if it is not valid or the wrong inode.
3283 */
3284 spin_lock(&cip->i_flags_lock);
3285 if (!cip->i_ino ||
3286 __xfs_iflags_test(cip, XFS_ISTALE)) {
3287 spin_unlock(&cip->i_flags_lock);
3288 continue;
3289 }
3290
3291 /*
3292 * Once we fall off the end of the cluster, no point checking
3293 * any more inodes in the list because they will also all be
3294 * outside the cluster.
3295 */
3296 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3297 spin_unlock(&cip->i_flags_lock);
3298 break;
3299 }
3300 spin_unlock(&cip->i_flags_lock);
3301
3302 /*
3303 * Do an un-protected check to see if the inode is dirty and
3304 * is a candidate for flushing. These checks will be repeated
3305 * later after the appropriate locks are acquired.
3306 */
3307 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3308 continue;
3309
3310 /*
3311 * Try to get locks. If any are unavailable or it is pinned,
3312 * then this inode cannot be flushed and is skipped.
3313 */
3314
3315 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3316 continue;
3317 if (!xfs_iflock_nowait(cip)) {
3318 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3319 continue;
3320 }
3321 if (xfs_ipincount(cip)) {
3322 xfs_ifunlock(cip);
3323 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3324 continue;
3325 }
3326
3327
3328 /*
3329 * Check the inode number again, just to be certain we are not
3330 * racing with freeing in xfs_reclaim_inode(). See the comments
3331 * in that function for more information as to why the initial
3332 * check is not sufficient.
3333 */
3334 if (!cip->i_ino) {
3335 xfs_ifunlock(cip);
3336 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3337 continue;
3338 }
3339
3340 /*
3341 * arriving here means that this inode can be flushed. First
3342 * re-check that it's dirty before flushing.
3343 */
3344 if (!xfs_inode_clean(cip)) {
3345 int error;
3346 error = xfs_iflush_int(cip, bp);
3347 if (error) {
3348 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3349 goto cluster_corrupt_out;
3350 }
3351 clcount++;
3352 } else {
3353 xfs_ifunlock(cip);
3354 }
3355 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3356 }
3357
3358 if (clcount) {
3359 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3360 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3361 }
3362
3363 out_free:
3364 rcu_read_unlock();
3365 kmem_free(cilist);
3366 out_put:
3367 xfs_perag_put(pag);
3368 return 0;
3369
3370
3371 cluster_corrupt_out:
3372 /*
3373 * Corruption detected in the clustering loop. Invalidate the
3374 * inode buffer and shut down the filesystem.
3375 */
3376 rcu_read_unlock();
3377 /*
3378 * Clean up the buffer. If it was delwri, just release it --
3379 * brelse can handle it with no problems. If not, shut down the
3380 * filesystem before releasing the buffer.
3381 */
3382 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3383 if (bufwasdelwri)
3384 xfs_buf_relse(bp);
3385
3386 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3387
3388 if (!bufwasdelwri) {
3389 /*
3390 * Just like incore_relse: if we have b_iodone functions,
3391 * mark the buffer as an error and call them. Otherwise
3392 * mark it as stale and brelse.
3393 */
3394 if (bp->b_iodone) {
3395 bp->b_flags &= ~XBF_DONE;
3396 xfs_buf_stale(bp);
3397 xfs_buf_ioerror(bp, -EIO);
3398 xfs_buf_ioend(bp);
3399 } else {
3400 xfs_buf_stale(bp);
3401 xfs_buf_relse(bp);
3402 }
3403 }
3404
3405 /*
3406 * Unlocks the flush lock
3407 */
3408 xfs_iflush_abort(cip, false);
3409 kmem_free(cilist);
3410 xfs_perag_put(pag);
3411 return -EFSCORRUPTED;
3412 }
3413
3414 /*
3415 * Flush dirty inode metadata into the backing buffer.
3416 *
3417 * The caller must have the inode lock and the inode flush lock held. The
3418 * inode lock will still be held upon return to the caller, and the inode
3419 * flush lock will be released after the inode has reached the disk.
3420 *
3421 * The caller must write out the buffer returned in *bpp and release it.
3422 */
3423 int
xfs_iflush(struct xfs_inode * ip,struct xfs_buf ** bpp)3424 xfs_iflush(
3425 struct xfs_inode *ip,
3426 struct xfs_buf **bpp)
3427 {
3428 struct xfs_mount *mp = ip->i_mount;
3429 struct xfs_buf *bp = NULL;
3430 struct xfs_dinode *dip;
3431 int error;
3432
3433 XFS_STATS_INC(mp, xs_iflush_count);
3434
3435 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3436 ASSERT(xfs_isiflocked(ip));
3437 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3438 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3439
3440 *bpp = NULL;
3441
3442 xfs_iunpin_wait(ip);
3443
3444 /*
3445 * For stale inodes we cannot rely on the backing buffer remaining
3446 * stale in cache for the remaining life of the stale inode and so
3447 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3448 * inodes below. We have to check this after ensuring the inode is
3449 * unpinned so that it is safe to reclaim the stale inode after the
3450 * flush call.
3451 */
3452 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3453 xfs_ifunlock(ip);
3454 return 0;
3455 }
3456
3457 /*
3458 * This may have been unpinned because the filesystem is shutting
3459 * down forcibly. If that's the case we must not write this inode
3460 * to disk, because the log record didn't make it to disk.
3461 *
3462 * We also have to remove the log item from the AIL in this case,
3463 * as we wait for an empty AIL as part of the unmount process.
3464 */
3465 if (XFS_FORCED_SHUTDOWN(mp)) {
3466 error = -EIO;
3467 goto abort_out;
3468 }
3469
3470 /*
3471 * Get the buffer containing the on-disk inode. We are doing a try-lock
3472 * operation here, so we may get an EAGAIN error. In that case, we
3473 * simply want to return with the inode still dirty.
3474 *
3475 * If we get any other error, we effectively have a corruption situation
3476 * and we cannot flush the inode, so we treat it the same as failing
3477 * xfs_iflush_int().
3478 */
3479 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3480 0);
3481 if (error == -EAGAIN) {
3482 xfs_ifunlock(ip);
3483 return error;
3484 }
3485 if (error)
3486 goto corrupt_out;
3487
3488 /*
3489 * First flush out the inode that xfs_iflush was called with.
3490 */
3491 error = xfs_iflush_int(ip, bp);
3492 if (error)
3493 goto corrupt_out;
3494
3495 /*
3496 * If the buffer is pinned then push on the log now so we won't
3497 * get stuck waiting in the write for too long.
3498 */
3499 if (xfs_buf_ispinned(bp))
3500 xfs_log_force(mp, 0);
3501
3502 /*
3503 * inode clustering:
3504 * see if other inodes can be gathered into this write
3505 */
3506 error = xfs_iflush_cluster(ip, bp);
3507 if (error)
3508 goto cluster_corrupt_out;
3509
3510 *bpp = bp;
3511 return 0;
3512
3513 corrupt_out:
3514 if (bp)
3515 xfs_buf_relse(bp);
3516 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3517 cluster_corrupt_out:
3518 error = -EFSCORRUPTED;
3519 abort_out:
3520 /*
3521 * Unlocks the flush lock
3522 */
3523 xfs_iflush_abort(ip, false);
3524 return error;
3525 }
3526
3527 STATIC int
xfs_iflush_int(struct xfs_inode * ip,struct xfs_buf * bp)3528 xfs_iflush_int(
3529 struct xfs_inode *ip,
3530 struct xfs_buf *bp)
3531 {
3532 struct xfs_inode_log_item *iip = ip->i_itemp;
3533 struct xfs_dinode *dip;
3534 struct xfs_mount *mp = ip->i_mount;
3535
3536 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3537 ASSERT(xfs_isiflocked(ip));
3538 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3539 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3540 ASSERT(iip != NULL && iip->ili_fields != 0);
3541 ASSERT(ip->i_d.di_version > 1);
3542
3543 /* set *dip = inode's place in the buffer */
3544 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3545
3546 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3547 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3548 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3549 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3550 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3551 goto corrupt_out;
3552 }
3553 if (S_ISREG(VFS_I(ip)->i_mode)) {
3554 if (XFS_TEST_ERROR(
3555 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3556 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3557 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3558 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3559 "%s: Bad regular inode %Lu, ptr 0x%p",
3560 __func__, ip->i_ino, ip);
3561 goto corrupt_out;
3562 }
3563 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3564 if (XFS_TEST_ERROR(
3565 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3566 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3567 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3568 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3569 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3570 "%s: Bad directory inode %Lu, ptr 0x%p",
3571 __func__, ip->i_ino, ip);
3572 goto corrupt_out;
3573 }
3574 }
3575 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3576 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3577 XFS_RANDOM_IFLUSH_5)) {
3578 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3579 "%s: detected corrupt incore inode %Lu, "
3580 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3581 __func__, ip->i_ino,
3582 ip->i_d.di_nextents + ip->i_d.di_anextents,
3583 ip->i_d.di_nblocks, ip);
3584 goto corrupt_out;
3585 }
3586 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3587 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3588 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3589 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3590 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3591 goto corrupt_out;
3592 }
3593
3594 /*
3595 * Inode item log recovery for v2 inodes are dependent on the
3596 * di_flushiter count for correct sequencing. We bump the flush
3597 * iteration count so we can detect flushes which postdate a log record
3598 * during recovery. This is redundant as we now log every change and
3599 * hence this can't happen but we need to still do it to ensure
3600 * backwards compatibility with old kernels that predate logging all
3601 * inode changes.
3602 */
3603 if (ip->i_d.di_version < 3)
3604 ip->i_d.di_flushiter++;
3605
3606 /* Check the inline directory data. */
3607 if (S_ISDIR(VFS_I(ip)->i_mode) &&
3608 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3609 xfs_dir2_sf_verify(ip))
3610 goto corrupt_out;
3611
3612 /*
3613 * Copy the dirty parts of the inode into the on-disk inode. We always
3614 * copy out the core of the inode, because if the inode is dirty at all
3615 * the core must be.
3616 */
3617 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3618
3619 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3620 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3621 ip->i_d.di_flushiter = 0;
3622
3623 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3624 if (XFS_IFORK_Q(ip))
3625 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3626 xfs_inobp_check(mp, bp);
3627
3628 /*
3629 * We've recorded everything logged in the inode, so we'd like to clear
3630 * the ili_fields bits so we don't log and flush things unnecessarily.
3631 * However, we can't stop logging all this information until the data
3632 * we've copied into the disk buffer is written to disk. If we did we
3633 * might overwrite the copy of the inode in the log with all the data
3634 * after re-logging only part of it, and in the face of a crash we
3635 * wouldn't have all the data we need to recover.
3636 *
3637 * What we do is move the bits to the ili_last_fields field. When
3638 * logging the inode, these bits are moved back to the ili_fields field.
3639 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3640 * know that the information those bits represent is permanently on
3641 * disk. As long as the flush completes before the inode is logged
3642 * again, then both ili_fields and ili_last_fields will be cleared.
3643 *
3644 * We can play with the ili_fields bits here, because the inode lock
3645 * must be held exclusively in order to set bits there and the flush
3646 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3647 * done routine can tell whether or not to look in the AIL. Also, store
3648 * the current LSN of the inode so that we can tell whether the item has
3649 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3650 * need the AIL lock, because it is a 64 bit value that cannot be read
3651 * atomically.
3652 */
3653 iip->ili_last_fields = iip->ili_fields;
3654 iip->ili_fields = 0;
3655 iip->ili_fsync_fields = 0;
3656 iip->ili_logged = 1;
3657
3658 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3659 &iip->ili_item.li_lsn);
3660
3661 /*
3662 * Attach the function xfs_iflush_done to the inode's
3663 * buffer. This will remove the inode from the AIL
3664 * and unlock the inode's flush lock when the inode is
3665 * completely written to disk.
3666 */
3667 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3668
3669 /* generate the checksum. */
3670 xfs_dinode_calc_crc(mp, dip);
3671
3672 ASSERT(bp->b_fspriv != NULL);
3673 ASSERT(bp->b_iodone != NULL);
3674 return 0;
3675
3676 corrupt_out:
3677 return -EFSCORRUPTED;
3678 }
3679