Linux 4.19-rc7
[linux-2.6/btrfs-unstable.git] / fs / xfs / xfs_inode.c
blob05db9540e4597536211446475d301eac834e972e
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6 #include <linux/log2.h>
7 #include <linux/iversion.h>
9 #include "xfs.h"
10 #include "xfs_fs.h"
11 #include "xfs_shared.h"
12 #include "xfs_format.h"
13 #include "xfs_log_format.h"
14 #include "xfs_trans_resv.h"
15 #include "xfs_sb.h"
16 #include "xfs_mount.h"
17 #include "xfs_defer.h"
18 #include "xfs_inode.h"
19 #include "xfs_da_format.h"
20 #include "xfs_da_btree.h"
21 #include "xfs_dir2.h"
22 #include "xfs_attr_sf.h"
23 #include "xfs_attr.h"
24 #include "xfs_trans_space.h"
25 #include "xfs_trans.h"
26 #include "xfs_buf_item.h"
27 #include "xfs_inode_item.h"
28 #include "xfs_ialloc.h"
29 #include "xfs_bmap.h"
30 #include "xfs_bmap_util.h"
31 #include "xfs_errortag.h"
32 #include "xfs_error.h"
33 #include "xfs_quota.h"
34 #include "xfs_filestream.h"
35 #include "xfs_cksum.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_symlink.h"
39 #include "xfs_trans_priv.h"
40 #include "xfs_log.h"
41 #include "xfs_bmap_btree.h"
42 #include "xfs_reflink.h"
43 #include "xfs_dir2_priv.h"
45 kmem_zone_t *xfs_inode_zone;
48 * Used in xfs_itruncate_extents(). This is the maximum number of extents
49 * freed from a file in a single transaction.
51 #define XFS_ITRUNC_MAX_EXTENTS 2
53 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
54 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
55 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
58 * helper function to extract extent size hint from inode
60 xfs_extlen_t
61 xfs_get_extsz_hint(
62 struct xfs_inode *ip)
64 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
65 return ip->i_d.di_extsize;
66 if (XFS_IS_REALTIME_INODE(ip))
67 return ip->i_mount->m_sb.sb_rextsize;
68 return 0;
72 * Helper function to extract CoW extent size hint from inode.
73 * Between the extent size hint and the CoW extent size hint, we
74 * return the greater of the two. If the value is zero (automatic),
75 * use the default size.
77 xfs_extlen_t
78 xfs_get_cowextsz_hint(
79 struct xfs_inode *ip)
81 xfs_extlen_t a, b;
83 a = 0;
84 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
85 a = ip->i_d.di_cowextsize;
86 b = xfs_get_extsz_hint(ip);
88 a = max(a, b);
89 if (a == 0)
90 return XFS_DEFAULT_COWEXTSZ_HINT;
91 return a;
95 * These two are wrapper routines around the xfs_ilock() routine used to
96 * centralize some grungy code. They are used in places that wish to lock the
97 * inode solely for reading the extents. The reason these places can't just
98 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
99 * bringing in of the extents from disk for a file in b-tree format. If the
100 * inode is in b-tree format, then we need to lock the inode exclusively until
101 * the extents are read in. Locking it exclusively all the time would limit
102 * our parallelism unnecessarily, though. What we do instead is check to see
103 * if the extents have been read in yet, and only lock the inode exclusively
104 * if they have not.
106 * The functions return a value which should be given to the corresponding
107 * xfs_iunlock() call.
109 uint
110 xfs_ilock_data_map_shared(
111 struct xfs_inode *ip)
113 uint lock_mode = XFS_ILOCK_SHARED;
115 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
116 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
117 lock_mode = XFS_ILOCK_EXCL;
118 xfs_ilock(ip, lock_mode);
119 return lock_mode;
122 uint
123 xfs_ilock_attr_map_shared(
124 struct xfs_inode *ip)
126 uint lock_mode = XFS_ILOCK_SHARED;
128 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
129 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
132 return lock_mode;
136 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
137 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
138 * various combinations of the locks to be obtained.
140 * The 3 locks should always be ordered so that the IO lock is obtained first,
141 * the mmap lock second and the ilock last in order to prevent deadlock.
143 * Basic locking order:
145 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
147 * mmap_sem locking order:
149 * i_rwsem -> page lock -> mmap_sem
150 * mmap_sem -> i_mmap_lock -> page_lock
152 * The difference in mmap_sem locking order mean that we cannot hold the
153 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
154 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
155 * in get_user_pages() to map the user pages into the kernel address space for
156 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
157 * page faults already hold the mmap_sem.
159 * Hence to serialise fully against both syscall and mmap based IO, we need to
160 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
161 * taken in places where we need to invalidate the page cache in a race
162 * free manner (e.g. truncate, hole punch and other extent manipulation
163 * functions).
165 void
166 xfs_ilock(
167 xfs_inode_t *ip,
168 uint lock_flags)
170 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
173 * You can't set both SHARED and EXCL for the same lock,
174 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
175 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
177 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
178 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
179 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
180 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
181 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
182 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
183 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
185 if (lock_flags & XFS_IOLOCK_EXCL) {
186 down_write_nested(&VFS_I(ip)->i_rwsem,
187 XFS_IOLOCK_DEP(lock_flags));
188 } else if (lock_flags & XFS_IOLOCK_SHARED) {
189 down_read_nested(&VFS_I(ip)->i_rwsem,
190 XFS_IOLOCK_DEP(lock_flags));
193 if (lock_flags & XFS_MMAPLOCK_EXCL)
194 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
195 else if (lock_flags & XFS_MMAPLOCK_SHARED)
196 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
198 if (lock_flags & XFS_ILOCK_EXCL)
199 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
200 else if (lock_flags & XFS_ILOCK_SHARED)
201 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
205 * This is just like xfs_ilock(), except that the caller
206 * is guaranteed not to sleep. It returns 1 if it gets
207 * the requested locks and 0 otherwise. If the IO lock is
208 * obtained but the inode lock cannot be, then the IO lock
209 * is dropped before returning.
211 * ip -- the inode being locked
212 * lock_flags -- this parameter indicates the inode's locks to be
213 * to be locked. See the comment for xfs_ilock() for a list
214 * of valid values.
217 xfs_ilock_nowait(
218 xfs_inode_t *ip,
219 uint lock_flags)
221 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
224 * You can't set both SHARED and EXCL for the same lock,
225 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
226 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
228 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
229 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
230 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
231 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
232 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
233 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
234 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
236 if (lock_flags & XFS_IOLOCK_EXCL) {
237 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
238 goto out;
239 } else if (lock_flags & XFS_IOLOCK_SHARED) {
240 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
241 goto out;
244 if (lock_flags & XFS_MMAPLOCK_EXCL) {
245 if (!mrtryupdate(&ip->i_mmaplock))
246 goto out_undo_iolock;
247 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
248 if (!mrtryaccess(&ip->i_mmaplock))
249 goto out_undo_iolock;
252 if (lock_flags & XFS_ILOCK_EXCL) {
253 if (!mrtryupdate(&ip->i_lock))
254 goto out_undo_mmaplock;
255 } else if (lock_flags & XFS_ILOCK_SHARED) {
256 if (!mrtryaccess(&ip->i_lock))
257 goto out_undo_mmaplock;
259 return 1;
261 out_undo_mmaplock:
262 if (lock_flags & XFS_MMAPLOCK_EXCL)
263 mrunlock_excl(&ip->i_mmaplock);
264 else if (lock_flags & XFS_MMAPLOCK_SHARED)
265 mrunlock_shared(&ip->i_mmaplock);
266 out_undo_iolock:
267 if (lock_flags & XFS_IOLOCK_EXCL)
268 up_write(&VFS_I(ip)->i_rwsem);
269 else if (lock_flags & XFS_IOLOCK_SHARED)
270 up_read(&VFS_I(ip)->i_rwsem);
271 out:
272 return 0;
276 * xfs_iunlock() is used to drop the inode locks acquired with
277 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
278 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
279 * that we know which locks to drop.
281 * ip -- the inode being unlocked
282 * lock_flags -- this parameter indicates the inode's locks to be
283 * to be unlocked. See the comment for xfs_ilock() for a list
284 * of valid values for this parameter.
287 void
288 xfs_iunlock(
289 xfs_inode_t *ip,
290 uint lock_flags)
293 * You can't set both SHARED and EXCL for the same lock,
294 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
295 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
297 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
298 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
299 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
300 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
301 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
302 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
303 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
304 ASSERT(lock_flags != 0);
306 if (lock_flags & XFS_IOLOCK_EXCL)
307 up_write(&VFS_I(ip)->i_rwsem);
308 else if (lock_flags & XFS_IOLOCK_SHARED)
309 up_read(&VFS_I(ip)->i_rwsem);
311 if (lock_flags & XFS_MMAPLOCK_EXCL)
312 mrunlock_excl(&ip->i_mmaplock);
313 else if (lock_flags & XFS_MMAPLOCK_SHARED)
314 mrunlock_shared(&ip->i_mmaplock);
316 if (lock_flags & XFS_ILOCK_EXCL)
317 mrunlock_excl(&ip->i_lock);
318 else if (lock_flags & XFS_ILOCK_SHARED)
319 mrunlock_shared(&ip->i_lock);
321 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
325 * give up write locks. the i/o lock cannot be held nested
326 * if it is being demoted.
328 void
329 xfs_ilock_demote(
330 xfs_inode_t *ip,
331 uint lock_flags)
333 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
334 ASSERT((lock_flags &
335 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
337 if (lock_flags & XFS_ILOCK_EXCL)
338 mrdemote(&ip->i_lock);
339 if (lock_flags & XFS_MMAPLOCK_EXCL)
340 mrdemote(&ip->i_mmaplock);
341 if (lock_flags & XFS_IOLOCK_EXCL)
342 downgrade_write(&VFS_I(ip)->i_rwsem);
344 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
347 #if defined(DEBUG) || defined(XFS_WARN)
349 xfs_isilocked(
350 xfs_inode_t *ip,
351 uint lock_flags)
353 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
354 if (!(lock_flags & XFS_ILOCK_SHARED))
355 return !!ip->i_lock.mr_writer;
356 return rwsem_is_locked(&ip->i_lock.mr_lock);
359 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
360 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
361 return !!ip->i_mmaplock.mr_writer;
362 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
365 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
366 if (!(lock_flags & XFS_IOLOCK_SHARED))
367 return !debug_locks ||
368 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
369 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
372 ASSERT(0);
373 return 0;
375 #endif
378 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
379 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
380 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
381 * errors and warnings.
383 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
384 static bool
385 xfs_lockdep_subclass_ok(
386 int subclass)
388 return subclass < MAX_LOCKDEP_SUBCLASSES;
390 #else
391 #define xfs_lockdep_subclass_ok(subclass) (true)
392 #endif
395 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
396 * value. This can be called for any type of inode lock combination, including
397 * parent locking. Care must be taken to ensure we don't overrun the subclass
398 * storage fields in the class mask we build.
400 static inline int
401 xfs_lock_inumorder(int lock_mode, int subclass)
403 int class = 0;
405 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
406 XFS_ILOCK_RTSUM)));
407 ASSERT(xfs_lockdep_subclass_ok(subclass));
409 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
410 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
411 class += subclass << XFS_IOLOCK_SHIFT;
414 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
415 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
416 class += subclass << XFS_MMAPLOCK_SHIFT;
419 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
420 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
421 class += subclass << XFS_ILOCK_SHIFT;
424 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
428 * The following routine will lock n inodes in exclusive mode. We assume the
429 * caller calls us with the inodes in i_ino order.
431 * We need to detect deadlock where an inode that we lock is in the AIL and we
432 * start waiting for another inode that is locked by a thread in a long running
433 * transaction (such as truncate). This can result in deadlock since the long
434 * running trans might need to wait for the inode we just locked in order to
435 * push the tail and free space in the log.
437 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
438 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
439 * lock more than one at a time, lockdep will report false positives saying we
440 * have violated locking orders.
442 static void
443 xfs_lock_inodes(
444 xfs_inode_t **ips,
445 int inodes,
446 uint lock_mode)
448 int attempts = 0, i, j, try_lock;
449 xfs_log_item_t *lp;
452 * Currently supports between 2 and 5 inodes with exclusive locking. We
453 * support an arbitrary depth of locking here, but absolute limits on
454 * inodes depend on the the type of locking and the limits placed by
455 * lockdep annotations in xfs_lock_inumorder. These are all checked by
456 * the asserts.
458 ASSERT(ips && inodes >= 2 && inodes <= 5);
459 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
460 XFS_ILOCK_EXCL));
461 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
462 XFS_ILOCK_SHARED)));
463 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
464 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
466 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
468 if (lock_mode & XFS_IOLOCK_EXCL) {
469 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
470 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
471 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
473 try_lock = 0;
474 i = 0;
475 again:
476 for (; i < inodes; i++) {
477 ASSERT(ips[i]);
479 if (i && (ips[i] == ips[i - 1])) /* Already locked */
480 continue;
483 * If try_lock is not set yet, make sure all locked inodes are
484 * not in the AIL. If any are, set try_lock to be used later.
486 if (!try_lock) {
487 for (j = (i - 1); j >= 0 && !try_lock; j--) {
488 lp = (xfs_log_item_t *)ips[j]->i_itemp;
489 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
490 try_lock++;
495 * If any of the previous locks we have locked is in the AIL,
496 * we must TRY to get the second and subsequent locks. If
497 * we can't get any, we must release all we have
498 * and try again.
500 if (!try_lock) {
501 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
502 continue;
505 /* try_lock means we have an inode locked that is in the AIL. */
506 ASSERT(i != 0);
507 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
508 continue;
511 * Unlock all previous guys and try again. xfs_iunlock will try
512 * to push the tail if the inode is in the AIL.
514 attempts++;
515 for (j = i - 1; j >= 0; j--) {
517 * Check to see if we've already unlocked this one. Not
518 * the first one going back, and the inode ptr is the
519 * same.
521 if (j != (i - 1) && ips[j] == ips[j + 1])
522 continue;
524 xfs_iunlock(ips[j], lock_mode);
527 if ((attempts % 5) == 0) {
528 delay(1); /* Don't just spin the CPU */
530 i = 0;
531 try_lock = 0;
532 goto again;
537 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
538 * the mmaplock or the ilock, but not more than one type at a time. If we lock
539 * more than one at a time, lockdep will report false positives saying we have
540 * violated locking orders. The iolock must be double-locked separately since
541 * we use i_rwsem for that. We now support taking one lock EXCL and the other
542 * SHARED.
544 void
545 xfs_lock_two_inodes(
546 struct xfs_inode *ip0,
547 uint ip0_mode,
548 struct xfs_inode *ip1,
549 uint ip1_mode)
551 struct xfs_inode *temp;
552 uint mode_temp;
553 int attempts = 0;
554 xfs_log_item_t *lp;
556 ASSERT(hweight32(ip0_mode) == 1);
557 ASSERT(hweight32(ip1_mode) == 1);
558 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
559 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
560 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
561 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
563 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
565 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
566 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
567 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
569 ASSERT(ip0->i_ino != ip1->i_ino);
571 if (ip0->i_ino > ip1->i_ino) {
572 temp = ip0;
573 ip0 = ip1;
574 ip1 = temp;
575 mode_temp = ip0_mode;
576 ip0_mode = ip1_mode;
577 ip1_mode = mode_temp;
580 again:
581 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
584 * If the first lock we have locked is in the AIL, we must TRY to get
585 * the second lock. If we can't get it, we must release the first one
586 * and try again.
588 lp = (xfs_log_item_t *)ip0->i_itemp;
589 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
590 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
591 xfs_iunlock(ip0, ip0_mode);
592 if ((++attempts % 5) == 0)
593 delay(1); /* Don't just spin the CPU */
594 goto again;
596 } else {
597 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
601 void
602 __xfs_iflock(
603 struct xfs_inode *ip)
605 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
606 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
608 do {
609 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
610 if (xfs_isiflocked(ip))
611 io_schedule();
612 } while (!xfs_iflock_nowait(ip));
614 finish_wait(wq, &wait.wq_entry);
617 STATIC uint
618 _xfs_dic2xflags(
619 uint16_t di_flags,
620 uint64_t di_flags2,
621 bool has_attr)
623 uint flags = 0;
625 if (di_flags & XFS_DIFLAG_ANY) {
626 if (di_flags & XFS_DIFLAG_REALTIME)
627 flags |= FS_XFLAG_REALTIME;
628 if (di_flags & XFS_DIFLAG_PREALLOC)
629 flags |= FS_XFLAG_PREALLOC;
630 if (di_flags & XFS_DIFLAG_IMMUTABLE)
631 flags |= FS_XFLAG_IMMUTABLE;
632 if (di_flags & XFS_DIFLAG_APPEND)
633 flags |= FS_XFLAG_APPEND;
634 if (di_flags & XFS_DIFLAG_SYNC)
635 flags |= FS_XFLAG_SYNC;
636 if (di_flags & XFS_DIFLAG_NOATIME)
637 flags |= FS_XFLAG_NOATIME;
638 if (di_flags & XFS_DIFLAG_NODUMP)
639 flags |= FS_XFLAG_NODUMP;
640 if (di_flags & XFS_DIFLAG_RTINHERIT)
641 flags |= FS_XFLAG_RTINHERIT;
642 if (di_flags & XFS_DIFLAG_PROJINHERIT)
643 flags |= FS_XFLAG_PROJINHERIT;
644 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
645 flags |= FS_XFLAG_NOSYMLINKS;
646 if (di_flags & XFS_DIFLAG_EXTSIZE)
647 flags |= FS_XFLAG_EXTSIZE;
648 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
649 flags |= FS_XFLAG_EXTSZINHERIT;
650 if (di_flags & XFS_DIFLAG_NODEFRAG)
651 flags |= FS_XFLAG_NODEFRAG;
652 if (di_flags & XFS_DIFLAG_FILESTREAM)
653 flags |= FS_XFLAG_FILESTREAM;
656 if (di_flags2 & XFS_DIFLAG2_ANY) {
657 if (di_flags2 & XFS_DIFLAG2_DAX)
658 flags |= FS_XFLAG_DAX;
659 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
660 flags |= FS_XFLAG_COWEXTSIZE;
663 if (has_attr)
664 flags |= FS_XFLAG_HASATTR;
666 return flags;
669 uint
670 xfs_ip2xflags(
671 struct xfs_inode *ip)
673 struct xfs_icdinode *dic = &ip->i_d;
675 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
680 * is allowed, otherwise it has to be an exact match. If a CI match is found,
681 * ci_name->name will point to a the actual name (caller must free) or
682 * will be set to NULL if an exact match is found.
685 xfs_lookup(
686 xfs_inode_t *dp,
687 struct xfs_name *name,
688 xfs_inode_t **ipp,
689 struct xfs_name *ci_name)
691 xfs_ino_t inum;
692 int error;
694 trace_xfs_lookup(dp, name);
696 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
697 return -EIO;
699 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
700 if (error)
701 goto out_unlock;
703 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
704 if (error)
705 goto out_free_name;
707 return 0;
709 out_free_name:
710 if (ci_name)
711 kmem_free(ci_name->name);
712 out_unlock:
713 *ipp = NULL;
714 return error;
718 * Allocate an inode on disk and return a copy of its in-core version.
719 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
720 * appropriately within the inode. The uid and gid for the inode are
721 * set according to the contents of the given cred structure.
723 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
724 * has a free inode available, call xfs_iget() to obtain the in-core
725 * version of the allocated inode. Finally, fill in the inode and
726 * log its initial contents. In this case, ialloc_context would be
727 * set to NULL.
729 * If xfs_dialloc() does not have an available inode, it will replenish
730 * its supply by doing an allocation. Since we can only do one
731 * allocation within a transaction without deadlocks, we must commit
732 * the current transaction before returning the inode itself.
733 * In this case, therefore, we will set ialloc_context and return.
734 * The caller should then commit the current transaction, start a new
735 * transaction, and call xfs_ialloc() again to actually get the inode.
737 * To ensure that some other process does not grab the inode that
738 * was allocated during the first call to xfs_ialloc(), this routine
739 * also returns the [locked] bp pointing to the head of the freelist
740 * as ialloc_context. The caller should hold this buffer across
741 * the commit and pass it back into this routine on the second call.
743 * If we are allocating quota inodes, we do not have a parent inode
744 * to attach to or associate with (i.e. pip == NULL) because they
745 * are not linked into the directory structure - they are attached
746 * directly to the superblock - and so have no parent.
748 static int
749 xfs_ialloc(
750 xfs_trans_t *tp,
751 xfs_inode_t *pip,
752 umode_t mode,
753 xfs_nlink_t nlink,
754 dev_t rdev,
755 prid_t prid,
756 xfs_buf_t **ialloc_context,
757 xfs_inode_t **ipp)
759 struct xfs_mount *mp = tp->t_mountp;
760 xfs_ino_t ino;
761 xfs_inode_t *ip;
762 uint flags;
763 int error;
764 struct timespec64 tv;
765 struct inode *inode;
768 * Call the space management code to pick
769 * the on-disk inode to be allocated.
771 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
772 ialloc_context, &ino);
773 if (error)
774 return error;
775 if (*ialloc_context || ino == NULLFSINO) {
776 *ipp = NULL;
777 return 0;
779 ASSERT(*ialloc_context == NULL);
782 * Protect against obviously corrupt allocation btree records. Later
783 * xfs_iget checks will catch re-allocation of other active in-memory
784 * and on-disk inodes. If we don't catch reallocating the parent inode
785 * here we will deadlock in xfs_iget() so we have to do these checks
786 * first.
788 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
789 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
790 return -EFSCORRUPTED;
794 * Get the in-core inode with the lock held exclusively.
795 * This is because we're setting fields here we need
796 * to prevent others from looking at until we're done.
798 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
799 XFS_ILOCK_EXCL, &ip);
800 if (error)
801 return error;
802 ASSERT(ip != NULL);
803 inode = VFS_I(ip);
806 * We always convert v1 inodes to v2 now - we only support filesystems
807 * with >= v2 inode capability, so there is no reason for ever leaving
808 * an inode in v1 format.
810 if (ip->i_d.di_version == 1)
811 ip->i_d.di_version = 2;
813 inode->i_mode = mode;
814 set_nlink(inode, nlink);
815 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
816 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
817 inode->i_rdev = rdev;
818 xfs_set_projid(ip, prid);
820 if (pip && XFS_INHERIT_GID(pip)) {
821 ip->i_d.di_gid = pip->i_d.di_gid;
822 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
823 inode->i_mode |= S_ISGID;
827 * If the group ID of the new file does not match the effective group
828 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
829 * (and only if the irix_sgid_inherit compatibility variable is set).
831 if ((irix_sgid_inherit) &&
832 (inode->i_mode & S_ISGID) &&
833 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
834 inode->i_mode &= ~S_ISGID;
836 ip->i_d.di_size = 0;
837 ip->i_d.di_nextents = 0;
838 ASSERT(ip->i_d.di_nblocks == 0);
840 tv = current_time(inode);
841 inode->i_mtime = tv;
842 inode->i_atime = tv;
843 inode->i_ctime = tv;
845 ip->i_d.di_extsize = 0;
846 ip->i_d.di_dmevmask = 0;
847 ip->i_d.di_dmstate = 0;
848 ip->i_d.di_flags = 0;
850 if (ip->i_d.di_version == 3) {
851 inode_set_iversion(inode, 1);
852 ip->i_d.di_flags2 = 0;
853 ip->i_d.di_cowextsize = 0;
854 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
855 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
859 flags = XFS_ILOG_CORE;
860 switch (mode & S_IFMT) {
861 case S_IFIFO:
862 case S_IFCHR:
863 case S_IFBLK:
864 case S_IFSOCK:
865 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
866 ip->i_df.if_flags = 0;
867 flags |= XFS_ILOG_DEV;
868 break;
869 case S_IFREG:
870 case S_IFDIR:
871 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
872 uint di_flags = 0;
874 if (S_ISDIR(mode)) {
875 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
876 di_flags |= XFS_DIFLAG_RTINHERIT;
877 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
878 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
879 ip->i_d.di_extsize = pip->i_d.di_extsize;
881 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
882 di_flags |= XFS_DIFLAG_PROJINHERIT;
883 } else if (S_ISREG(mode)) {
884 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
885 di_flags |= XFS_DIFLAG_REALTIME;
886 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
887 di_flags |= XFS_DIFLAG_EXTSIZE;
888 ip->i_d.di_extsize = pip->i_d.di_extsize;
891 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
892 xfs_inherit_noatime)
893 di_flags |= XFS_DIFLAG_NOATIME;
894 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
895 xfs_inherit_nodump)
896 di_flags |= XFS_DIFLAG_NODUMP;
897 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
898 xfs_inherit_sync)
899 di_flags |= XFS_DIFLAG_SYNC;
900 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
901 xfs_inherit_nosymlinks)
902 di_flags |= XFS_DIFLAG_NOSYMLINKS;
903 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
904 xfs_inherit_nodefrag)
905 di_flags |= XFS_DIFLAG_NODEFRAG;
906 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
907 di_flags |= XFS_DIFLAG_FILESTREAM;
909 ip->i_d.di_flags |= di_flags;
911 if (pip &&
912 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
913 pip->i_d.di_version == 3 &&
914 ip->i_d.di_version == 3) {
915 uint64_t di_flags2 = 0;
917 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
918 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
919 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
921 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
922 di_flags2 |= XFS_DIFLAG2_DAX;
924 ip->i_d.di_flags2 |= di_flags2;
926 /* FALLTHROUGH */
927 case S_IFLNK:
928 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
929 ip->i_df.if_flags = XFS_IFEXTENTS;
930 ip->i_df.if_bytes = 0;
931 ip->i_df.if_u1.if_root = NULL;
932 break;
933 default:
934 ASSERT(0);
937 * Attribute fork settings for new inode.
939 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
940 ip->i_d.di_anextents = 0;
943 * Log the new values stuffed into the inode.
945 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
946 xfs_trans_log_inode(tp, ip, flags);
948 /* now that we have an i_mode we can setup the inode structure */
949 xfs_setup_inode(ip);
951 *ipp = ip;
952 return 0;
956 * Allocates a new inode from disk and return a pointer to the
957 * incore copy. This routine will internally commit the current
958 * transaction and allocate a new one if the Space Manager needed
959 * to do an allocation to replenish the inode free-list.
961 * This routine is designed to be called from xfs_create and
962 * xfs_create_dir.
966 xfs_dir_ialloc(
967 xfs_trans_t **tpp, /* input: current transaction;
968 output: may be a new transaction. */
969 xfs_inode_t *dp, /* directory within whose allocate
970 the inode. */
971 umode_t mode,
972 xfs_nlink_t nlink,
973 dev_t rdev,
974 prid_t prid, /* project id */
975 xfs_inode_t **ipp) /* pointer to inode; it will be
976 locked. */
978 xfs_trans_t *tp;
979 xfs_inode_t *ip;
980 xfs_buf_t *ialloc_context = NULL;
981 int code;
982 void *dqinfo;
983 uint tflags;
985 tp = *tpp;
986 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
989 * xfs_ialloc will return a pointer to an incore inode if
990 * the Space Manager has an available inode on the free
991 * list. Otherwise, it will do an allocation and replenish
992 * the freelist. Since we can only do one allocation per
993 * transaction without deadlocks, we will need to commit the
994 * current transaction and start a new one. We will then
995 * need to call xfs_ialloc again to get the inode.
997 * If xfs_ialloc did an allocation to replenish the freelist,
998 * it returns the bp containing the head of the freelist as
999 * ialloc_context. We will hold a lock on it across the
1000 * transaction commit so that no other process can steal
1001 * the inode(s) that we've just allocated.
1003 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
1004 &ip);
1007 * Return an error if we were unable to allocate a new inode.
1008 * This should only happen if we run out of space on disk or
1009 * encounter a disk error.
1011 if (code) {
1012 *ipp = NULL;
1013 return code;
1015 if (!ialloc_context && !ip) {
1016 *ipp = NULL;
1017 return -ENOSPC;
1021 * If the AGI buffer is non-NULL, then we were unable to get an
1022 * inode in one operation. We need to commit the current
1023 * transaction and call xfs_ialloc() again. It is guaranteed
1024 * to succeed the second time.
1026 if (ialloc_context) {
1028 * Normally, xfs_trans_commit releases all the locks.
1029 * We call bhold to hang on to the ialloc_context across
1030 * the commit. Holding this buffer prevents any other
1031 * processes from doing any allocations in this
1032 * allocation group.
1034 xfs_trans_bhold(tp, ialloc_context);
1037 * We want the quota changes to be associated with the next
1038 * transaction, NOT this one. So, detach the dqinfo from this
1039 * and attach it to the next transaction.
1041 dqinfo = NULL;
1042 tflags = 0;
1043 if (tp->t_dqinfo) {
1044 dqinfo = (void *)tp->t_dqinfo;
1045 tp->t_dqinfo = NULL;
1046 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1047 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1050 code = xfs_trans_roll(&tp);
1053 * Re-attach the quota info that we detached from prev trx.
1055 if (dqinfo) {
1056 tp->t_dqinfo = dqinfo;
1057 tp->t_flags |= tflags;
1060 if (code) {
1061 xfs_buf_relse(ialloc_context);
1062 *tpp = tp;
1063 *ipp = NULL;
1064 return code;
1066 xfs_trans_bjoin(tp, ialloc_context);
1069 * Call ialloc again. Since we've locked out all
1070 * other allocations in this allocation group,
1071 * this call should always succeed.
1073 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1074 &ialloc_context, &ip);
1077 * If we get an error at this point, return to the caller
1078 * so that the current transaction can be aborted.
1080 if (code) {
1081 *tpp = tp;
1082 *ipp = NULL;
1083 return code;
1085 ASSERT(!ialloc_context && ip);
1089 *ipp = ip;
1090 *tpp = tp;
1092 return 0;
1096 * Decrement the link count on an inode & log the change. If this causes the
1097 * link count to go to zero, move the inode to AGI unlinked list so that it can
1098 * be freed when the last active reference goes away via xfs_inactive().
1100 static int /* error */
1101 xfs_droplink(
1102 xfs_trans_t *tp,
1103 xfs_inode_t *ip)
1105 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1107 drop_nlink(VFS_I(ip));
1108 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1110 if (VFS_I(ip)->i_nlink)
1111 return 0;
1113 return xfs_iunlink(tp, ip);
1117 * Increment the link count on an inode & log the change.
1119 static int
1120 xfs_bumplink(
1121 xfs_trans_t *tp,
1122 xfs_inode_t *ip)
1124 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1126 ASSERT(ip->i_d.di_version > 1);
1127 inc_nlink(VFS_I(ip));
1128 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1129 return 0;
1133 xfs_create(
1134 xfs_inode_t *dp,
1135 struct xfs_name *name,
1136 umode_t mode,
1137 dev_t rdev,
1138 xfs_inode_t **ipp)
1140 int is_dir = S_ISDIR(mode);
1141 struct xfs_mount *mp = dp->i_mount;
1142 struct xfs_inode *ip = NULL;
1143 struct xfs_trans *tp = NULL;
1144 int error;
1145 bool unlock_dp_on_error = false;
1146 prid_t prid;
1147 struct xfs_dquot *udqp = NULL;
1148 struct xfs_dquot *gdqp = NULL;
1149 struct xfs_dquot *pdqp = NULL;
1150 struct xfs_trans_res *tres;
1151 uint resblks;
1153 trace_xfs_create(dp, name);
1155 if (XFS_FORCED_SHUTDOWN(mp))
1156 return -EIO;
1158 prid = xfs_get_initial_prid(dp);
1161 * Make sure that we have allocated dquot(s) on disk.
1163 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1164 xfs_kgid_to_gid(current_fsgid()), prid,
1165 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1166 &udqp, &gdqp, &pdqp);
1167 if (error)
1168 return error;
1170 if (is_dir) {
1171 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1172 tres = &M_RES(mp)->tr_mkdir;
1173 } else {
1174 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1175 tres = &M_RES(mp)->tr_create;
1179 * Initially assume that the file does not exist and
1180 * reserve the resources for that case. If that is not
1181 * the case we'll drop the one we have and get a more
1182 * appropriate transaction later.
1184 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1185 if (error == -ENOSPC) {
1186 /* flush outstanding delalloc blocks and retry */
1187 xfs_flush_inodes(mp);
1188 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1190 if (error)
1191 goto out_release_inode;
1193 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1194 unlock_dp_on_error = true;
1197 * Reserve disk quota and the inode.
1199 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1200 pdqp, resblks, 1, 0);
1201 if (error)
1202 goto out_trans_cancel;
1205 * A newly created regular or special file just has one directory
1206 * entry pointing to them, but a directory also the "." entry
1207 * pointing to itself.
1209 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
1210 if (error)
1211 goto out_trans_cancel;
1214 * Now we join the directory inode to the transaction. We do not do it
1215 * earlier because xfs_dir_ialloc might commit the previous transaction
1216 * (and release all the locks). An error from here on will result in
1217 * the transaction cancel unlocking dp so don't do it explicitly in the
1218 * error path.
1220 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1221 unlock_dp_on_error = false;
1223 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1224 resblks ?
1225 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1226 if (error) {
1227 ASSERT(error != -ENOSPC);
1228 goto out_trans_cancel;
1230 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1231 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1233 if (is_dir) {
1234 error = xfs_dir_init(tp, ip, dp);
1235 if (error)
1236 goto out_trans_cancel;
1238 error = xfs_bumplink(tp, dp);
1239 if (error)
1240 goto out_trans_cancel;
1244 * If this is a synchronous mount, make sure that the
1245 * create transaction goes to disk before returning to
1246 * the user.
1248 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1249 xfs_trans_set_sync(tp);
1252 * Attach the dquot(s) to the inodes and modify them incore.
1253 * These ids of the inode couldn't have changed since the new
1254 * inode has been locked ever since it was created.
1256 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1258 error = xfs_trans_commit(tp);
1259 if (error)
1260 goto out_release_inode;
1262 xfs_qm_dqrele(udqp);
1263 xfs_qm_dqrele(gdqp);
1264 xfs_qm_dqrele(pdqp);
1266 *ipp = ip;
1267 return 0;
1269 out_trans_cancel:
1270 xfs_trans_cancel(tp);
1271 out_release_inode:
1273 * Wait until after the current transaction is aborted to finish the
1274 * setup of the inode and release the inode. This prevents recursive
1275 * transactions and deadlocks from xfs_inactive.
1277 if (ip) {
1278 xfs_finish_inode_setup(ip);
1279 xfs_irele(ip);
1282 xfs_qm_dqrele(udqp);
1283 xfs_qm_dqrele(gdqp);
1284 xfs_qm_dqrele(pdqp);
1286 if (unlock_dp_on_error)
1287 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1288 return error;
1292 xfs_create_tmpfile(
1293 struct xfs_inode *dp,
1294 umode_t mode,
1295 struct xfs_inode **ipp)
1297 struct xfs_mount *mp = dp->i_mount;
1298 struct xfs_inode *ip = NULL;
1299 struct xfs_trans *tp = NULL;
1300 int error;
1301 prid_t prid;
1302 struct xfs_dquot *udqp = NULL;
1303 struct xfs_dquot *gdqp = NULL;
1304 struct xfs_dquot *pdqp = NULL;
1305 struct xfs_trans_res *tres;
1306 uint resblks;
1308 if (XFS_FORCED_SHUTDOWN(mp))
1309 return -EIO;
1311 prid = xfs_get_initial_prid(dp);
1314 * Make sure that we have allocated dquot(s) on disk.
1316 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1317 xfs_kgid_to_gid(current_fsgid()), prid,
1318 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1319 &udqp, &gdqp, &pdqp);
1320 if (error)
1321 return error;
1323 resblks = XFS_IALLOC_SPACE_RES(mp);
1324 tres = &M_RES(mp)->tr_create_tmpfile;
1326 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1327 if (error)
1328 goto out_release_inode;
1330 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1331 pdqp, resblks, 1, 0);
1332 if (error)
1333 goto out_trans_cancel;
1335 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip);
1336 if (error)
1337 goto out_trans_cancel;
1339 if (mp->m_flags & XFS_MOUNT_WSYNC)
1340 xfs_trans_set_sync(tp);
1343 * Attach the dquot(s) to the inodes and modify them incore.
1344 * These ids of the inode couldn't have changed since the new
1345 * inode has been locked ever since it was created.
1347 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1349 error = xfs_iunlink(tp, ip);
1350 if (error)
1351 goto out_trans_cancel;
1353 error = xfs_trans_commit(tp);
1354 if (error)
1355 goto out_release_inode;
1357 xfs_qm_dqrele(udqp);
1358 xfs_qm_dqrele(gdqp);
1359 xfs_qm_dqrele(pdqp);
1361 *ipp = ip;
1362 return 0;
1364 out_trans_cancel:
1365 xfs_trans_cancel(tp);
1366 out_release_inode:
1368 * Wait until after the current transaction is aborted to finish the
1369 * setup of the inode and release the inode. This prevents recursive
1370 * transactions and deadlocks from xfs_inactive.
1372 if (ip) {
1373 xfs_finish_inode_setup(ip);
1374 xfs_irele(ip);
1377 xfs_qm_dqrele(udqp);
1378 xfs_qm_dqrele(gdqp);
1379 xfs_qm_dqrele(pdqp);
1381 return error;
1385 xfs_link(
1386 xfs_inode_t *tdp,
1387 xfs_inode_t *sip,
1388 struct xfs_name *target_name)
1390 xfs_mount_t *mp = tdp->i_mount;
1391 xfs_trans_t *tp;
1392 int error;
1393 int resblks;
1395 trace_xfs_link(tdp, target_name);
1397 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1399 if (XFS_FORCED_SHUTDOWN(mp))
1400 return -EIO;
1402 error = xfs_qm_dqattach(sip);
1403 if (error)
1404 goto std_return;
1406 error = xfs_qm_dqattach(tdp);
1407 if (error)
1408 goto std_return;
1410 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1411 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1412 if (error == -ENOSPC) {
1413 resblks = 0;
1414 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1416 if (error)
1417 goto std_return;
1419 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1421 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1422 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1425 * If we are using project inheritance, we only allow hard link
1426 * creation in our tree when the project IDs are the same; else
1427 * the tree quota mechanism could be circumvented.
1429 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1430 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1431 error = -EXDEV;
1432 goto error_return;
1435 if (!resblks) {
1436 error = xfs_dir_canenter(tp, tdp, target_name);
1437 if (error)
1438 goto error_return;
1442 * Handle initial link state of O_TMPFILE inode
1444 if (VFS_I(sip)->i_nlink == 0) {
1445 error = xfs_iunlink_remove(tp, sip);
1446 if (error)
1447 goto error_return;
1450 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1451 resblks);
1452 if (error)
1453 goto error_return;
1454 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1455 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1457 error = xfs_bumplink(tp, sip);
1458 if (error)
1459 goto error_return;
1462 * If this is a synchronous mount, make sure that the
1463 * link transaction goes to disk before returning to
1464 * the user.
1466 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1467 xfs_trans_set_sync(tp);
1469 return xfs_trans_commit(tp);
1471 error_return:
1472 xfs_trans_cancel(tp);
1473 std_return:
1474 return error;
1477 /* Clear the reflink flag and the cowblocks tag if possible. */
1478 static void
1479 xfs_itruncate_clear_reflink_flags(
1480 struct xfs_inode *ip)
1482 struct xfs_ifork *dfork;
1483 struct xfs_ifork *cfork;
1485 if (!xfs_is_reflink_inode(ip))
1486 return;
1487 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1488 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1489 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1490 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1491 if (cfork->if_bytes == 0)
1492 xfs_inode_clear_cowblocks_tag(ip);
1496 * Free up the underlying blocks past new_size. The new size must be smaller
1497 * than the current size. This routine can be used both for the attribute and
1498 * data fork, and does not modify the inode size, which is left to the caller.
1500 * The transaction passed to this routine must have made a permanent log
1501 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1502 * given transaction and start new ones, so make sure everything involved in
1503 * the transaction is tidy before calling here. Some transaction will be
1504 * returned to the caller to be committed. The incoming transaction must
1505 * already include the inode, and both inode locks must be held exclusively.
1506 * The inode must also be "held" within the transaction. On return the inode
1507 * will be "held" within the returned transaction. This routine does NOT
1508 * require any disk space to be reserved for it within the transaction.
1510 * If we get an error, we must return with the inode locked and linked into the
1511 * current transaction. This keeps things simple for the higher level code,
1512 * because it always knows that the inode is locked and held in the transaction
1513 * that returns to it whether errors occur or not. We don't mark the inode
1514 * dirty on error so that transactions can be easily aborted if possible.
1517 xfs_itruncate_extents_flags(
1518 struct xfs_trans **tpp,
1519 struct xfs_inode *ip,
1520 int whichfork,
1521 xfs_fsize_t new_size,
1522 int flags)
1524 struct xfs_mount *mp = ip->i_mount;
1525 struct xfs_trans *tp = *tpp;
1526 xfs_fileoff_t first_unmap_block;
1527 xfs_fileoff_t last_block;
1528 xfs_filblks_t unmap_len;
1529 int error = 0;
1530 int done = 0;
1532 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1533 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1534 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1535 ASSERT(new_size <= XFS_ISIZE(ip));
1536 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1537 ASSERT(ip->i_itemp != NULL);
1538 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1539 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1541 trace_xfs_itruncate_extents_start(ip, new_size);
1543 flags |= xfs_bmapi_aflag(whichfork);
1546 * Since it is possible for space to become allocated beyond
1547 * the end of the file (in a crash where the space is allocated
1548 * but the inode size is not yet updated), simply remove any
1549 * blocks which show up between the new EOF and the maximum
1550 * possible file size. If the first block to be removed is
1551 * beyond the maximum file size (ie it is the same as last_block),
1552 * then there is nothing to do.
1554 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1555 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1556 if (first_unmap_block == last_block)
1557 return 0;
1559 ASSERT(first_unmap_block < last_block);
1560 unmap_len = last_block - first_unmap_block + 1;
1561 while (!done) {
1562 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1563 error = xfs_bunmapi(tp, ip, first_unmap_block, unmap_len, flags,
1564 XFS_ITRUNC_MAX_EXTENTS, &done);
1565 if (error)
1566 goto out;
1569 * Duplicate the transaction that has the permanent
1570 * reservation and commit the old transaction.
1572 error = xfs_defer_finish(&tp);
1573 if (error)
1574 goto out;
1576 error = xfs_trans_roll_inode(&tp, ip);
1577 if (error)
1578 goto out;
1581 if (whichfork == XFS_DATA_FORK) {
1582 /* Remove all pending CoW reservations. */
1583 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1584 first_unmap_block, last_block, true);
1585 if (error)
1586 goto out;
1588 xfs_itruncate_clear_reflink_flags(ip);
1592 * Always re-log the inode so that our permanent transaction can keep
1593 * on rolling it forward in the log.
1595 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1597 trace_xfs_itruncate_extents_end(ip, new_size);
1599 out:
1600 *tpp = tp;
1601 return error;
1605 xfs_release(
1606 xfs_inode_t *ip)
1608 xfs_mount_t *mp = ip->i_mount;
1609 int error;
1611 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1612 return 0;
1614 /* If this is a read-only mount, don't do this (would generate I/O) */
1615 if (mp->m_flags & XFS_MOUNT_RDONLY)
1616 return 0;
1618 if (!XFS_FORCED_SHUTDOWN(mp)) {
1619 int truncated;
1622 * If we previously truncated this file and removed old data
1623 * in the process, we want to initiate "early" writeout on
1624 * the last close. This is an attempt to combat the notorious
1625 * NULL files problem which is particularly noticeable from a
1626 * truncate down, buffered (re-)write (delalloc), followed by
1627 * a crash. What we are effectively doing here is
1628 * significantly reducing the time window where we'd otherwise
1629 * be exposed to that problem.
1631 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1632 if (truncated) {
1633 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1634 if (ip->i_delayed_blks > 0) {
1635 error = filemap_flush(VFS_I(ip)->i_mapping);
1636 if (error)
1637 return error;
1642 if (VFS_I(ip)->i_nlink == 0)
1643 return 0;
1645 if (xfs_can_free_eofblocks(ip, false)) {
1648 * Check if the inode is being opened, written and closed
1649 * frequently and we have delayed allocation blocks outstanding
1650 * (e.g. streaming writes from the NFS server), truncating the
1651 * blocks past EOF will cause fragmentation to occur.
1653 * In this case don't do the truncation, but we have to be
1654 * careful how we detect this case. Blocks beyond EOF show up as
1655 * i_delayed_blks even when the inode is clean, so we need to
1656 * truncate them away first before checking for a dirty release.
1657 * Hence on the first dirty close we will still remove the
1658 * speculative allocation, but after that we will leave it in
1659 * place.
1661 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1662 return 0;
1664 * If we can't get the iolock just skip truncating the blocks
1665 * past EOF because we could deadlock with the mmap_sem
1666 * otherwise. We'll get another chance to drop them once the
1667 * last reference to the inode is dropped, so we'll never leak
1668 * blocks permanently.
1670 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1671 error = xfs_free_eofblocks(ip);
1672 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1673 if (error)
1674 return error;
1677 /* delalloc blocks after truncation means it really is dirty */
1678 if (ip->i_delayed_blks)
1679 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1681 return 0;
1685 * xfs_inactive_truncate
1687 * Called to perform a truncate when an inode becomes unlinked.
1689 STATIC int
1690 xfs_inactive_truncate(
1691 struct xfs_inode *ip)
1693 struct xfs_mount *mp = ip->i_mount;
1694 struct xfs_trans *tp;
1695 int error;
1697 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1698 if (error) {
1699 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1700 return error;
1702 xfs_ilock(ip, XFS_ILOCK_EXCL);
1703 xfs_trans_ijoin(tp, ip, 0);
1706 * Log the inode size first to prevent stale data exposure in the event
1707 * of a system crash before the truncate completes. See the related
1708 * comment in xfs_vn_setattr_size() for details.
1710 ip->i_d.di_size = 0;
1711 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1713 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1714 if (error)
1715 goto error_trans_cancel;
1717 ASSERT(ip->i_d.di_nextents == 0);
1719 error = xfs_trans_commit(tp);
1720 if (error)
1721 goto error_unlock;
1723 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1724 return 0;
1726 error_trans_cancel:
1727 xfs_trans_cancel(tp);
1728 error_unlock:
1729 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1730 return error;
1734 * xfs_inactive_ifree()
1736 * Perform the inode free when an inode is unlinked.
1738 STATIC int
1739 xfs_inactive_ifree(
1740 struct xfs_inode *ip)
1742 struct xfs_mount *mp = ip->i_mount;
1743 struct xfs_trans *tp;
1744 int error;
1747 * We try to use a per-AG reservation for any block needed by the finobt
1748 * tree, but as the finobt feature predates the per-AG reservation
1749 * support a degraded file system might not have enough space for the
1750 * reservation at mount time. In that case try to dip into the reserved
1751 * pool and pray.
1753 * Send a warning if the reservation does happen to fail, as the inode
1754 * now remains allocated and sits on the unlinked list until the fs is
1755 * repaired.
1757 if (unlikely(mp->m_inotbt_nores)) {
1758 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1759 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1760 &tp);
1761 } else {
1762 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1764 if (error) {
1765 if (error == -ENOSPC) {
1766 xfs_warn_ratelimited(mp,
1767 "Failed to remove inode(s) from unlinked list. "
1768 "Please free space, unmount and run xfs_repair.");
1769 } else {
1770 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1772 return error;
1775 xfs_ilock(ip, XFS_ILOCK_EXCL);
1776 xfs_trans_ijoin(tp, ip, 0);
1778 error = xfs_ifree(tp, ip);
1779 if (error) {
1781 * If we fail to free the inode, shut down. The cancel
1782 * might do that, we need to make sure. Otherwise the
1783 * inode might be lost for a long time or forever.
1785 if (!XFS_FORCED_SHUTDOWN(mp)) {
1786 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1787 __func__, error);
1788 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1790 xfs_trans_cancel(tp);
1791 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1792 return error;
1796 * Credit the quota account(s). The inode is gone.
1798 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1801 * Just ignore errors at this point. There is nothing we can do except
1802 * to try to keep going. Make sure it's not a silent error.
1804 error = xfs_trans_commit(tp);
1805 if (error)
1806 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1807 __func__, error);
1809 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1810 return 0;
1814 * xfs_inactive
1816 * This is called when the vnode reference count for the vnode
1817 * goes to zero. If the file has been unlinked, then it must
1818 * now be truncated. Also, we clear all of the read-ahead state
1819 * kept for the inode here since the file is now closed.
1821 void
1822 xfs_inactive(
1823 xfs_inode_t *ip)
1825 struct xfs_mount *mp;
1826 int error;
1827 int truncate = 0;
1830 * If the inode is already free, then there can be nothing
1831 * to clean up here.
1833 if (VFS_I(ip)->i_mode == 0) {
1834 ASSERT(ip->i_df.if_broot_bytes == 0);
1835 return;
1838 mp = ip->i_mount;
1839 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1841 /* If this is a read-only mount, don't do this (would generate I/O) */
1842 if (mp->m_flags & XFS_MOUNT_RDONLY)
1843 return;
1845 /* Try to clean out the cow blocks if there are any. */
1846 if (xfs_inode_has_cow_data(ip))
1847 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1849 if (VFS_I(ip)->i_nlink != 0) {
1851 * force is true because we are evicting an inode from the
1852 * cache. Post-eof blocks must be freed, lest we end up with
1853 * broken free space accounting.
1855 * Note: don't bother with iolock here since lockdep complains
1856 * about acquiring it in reclaim context. We have the only
1857 * reference to the inode at this point anyways.
1859 if (xfs_can_free_eofblocks(ip, true))
1860 xfs_free_eofblocks(ip);
1862 return;
1865 if (S_ISREG(VFS_I(ip)->i_mode) &&
1866 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1867 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1868 truncate = 1;
1870 error = xfs_qm_dqattach(ip);
1871 if (error)
1872 return;
1874 if (S_ISLNK(VFS_I(ip)->i_mode))
1875 error = xfs_inactive_symlink(ip);
1876 else if (truncate)
1877 error = xfs_inactive_truncate(ip);
1878 if (error)
1879 return;
1882 * If there are attributes associated with the file then blow them away
1883 * now. The code calls a routine that recursively deconstructs the
1884 * attribute fork. If also blows away the in-core attribute fork.
1886 if (XFS_IFORK_Q(ip)) {
1887 error = xfs_attr_inactive(ip);
1888 if (error)
1889 return;
1892 ASSERT(!ip->i_afp);
1893 ASSERT(ip->i_d.di_anextents == 0);
1894 ASSERT(ip->i_d.di_forkoff == 0);
1897 * Free the inode.
1899 error = xfs_inactive_ifree(ip);
1900 if (error)
1901 return;
1904 * Release the dquots held by inode, if any.
1906 xfs_qm_dqdetach(ip);
1910 * This is called when the inode's link count goes to 0 or we are creating a
1911 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1912 * set to true as the link count is dropped to zero by the VFS after we've
1913 * created the file successfully, so we have to add it to the unlinked list
1914 * while the link count is non-zero.
1916 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1917 * list when the inode is freed.
1919 STATIC int
1920 xfs_iunlink(
1921 struct xfs_trans *tp,
1922 struct xfs_inode *ip)
1924 xfs_mount_t *mp = tp->t_mountp;
1925 xfs_agi_t *agi;
1926 xfs_dinode_t *dip;
1927 xfs_buf_t *agibp;
1928 xfs_buf_t *ibp;
1929 xfs_agino_t agino;
1930 short bucket_index;
1931 int offset;
1932 int error;
1934 ASSERT(VFS_I(ip)->i_mode != 0);
1937 * Get the agi buffer first. It ensures lock ordering
1938 * on the list.
1940 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1941 if (error)
1942 return error;
1943 agi = XFS_BUF_TO_AGI(agibp);
1946 * Get the index into the agi hash table for the
1947 * list this inode will go on.
1949 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1950 ASSERT(agino != 0);
1951 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1952 ASSERT(agi->agi_unlinked[bucket_index]);
1953 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1955 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1957 * There is already another inode in the bucket we need
1958 * to add ourselves to. Add us at the front of the list.
1959 * Here we put the head pointer into our next pointer,
1960 * and then we fall through to point the head at us.
1962 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1963 0, 0);
1964 if (error)
1965 return error;
1967 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1968 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1969 offset = ip->i_imap.im_boffset +
1970 offsetof(xfs_dinode_t, di_next_unlinked);
1972 /* need to recalc the inode CRC if appropriate */
1973 xfs_dinode_calc_crc(mp, dip);
1975 xfs_trans_inode_buf(tp, ibp);
1976 xfs_trans_log_buf(tp, ibp, offset,
1977 (offset + sizeof(xfs_agino_t) - 1));
1978 xfs_inobp_check(mp, ibp);
1982 * Point the bucket head pointer at the inode being inserted.
1984 ASSERT(agino != 0);
1985 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1986 offset = offsetof(xfs_agi_t, agi_unlinked) +
1987 (sizeof(xfs_agino_t) * bucket_index);
1988 xfs_trans_log_buf(tp, agibp, offset,
1989 (offset + sizeof(xfs_agino_t) - 1));
1990 return 0;
1994 * Pull the on-disk inode from the AGI unlinked list.
1996 STATIC int
1997 xfs_iunlink_remove(
1998 xfs_trans_t *tp,
1999 xfs_inode_t *ip)
2001 xfs_ino_t next_ino;
2002 xfs_mount_t *mp;
2003 xfs_agi_t *agi;
2004 xfs_dinode_t *dip;
2005 xfs_buf_t *agibp;
2006 xfs_buf_t *ibp;
2007 xfs_agnumber_t agno;
2008 xfs_agino_t agino;
2009 xfs_agino_t next_agino;
2010 xfs_buf_t *last_ibp;
2011 xfs_dinode_t *last_dip = NULL;
2012 short bucket_index;
2013 int offset, last_offset = 0;
2014 int error;
2016 mp = tp->t_mountp;
2017 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2020 * Get the agi buffer first. It ensures lock ordering
2021 * on the list.
2023 error = xfs_read_agi(mp, tp, agno, &agibp);
2024 if (error)
2025 return error;
2027 agi = XFS_BUF_TO_AGI(agibp);
2030 * Get the index into the agi hash table for the
2031 * list this inode will go on.
2033 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2034 if (!xfs_verify_agino(mp, agno, agino))
2035 return -EFSCORRUPTED;
2036 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2037 if (!xfs_verify_agino(mp, agno,
2038 be32_to_cpu(agi->agi_unlinked[bucket_index]))) {
2039 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2040 agi, sizeof(*agi));
2041 return -EFSCORRUPTED;
2044 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2046 * We're at the head of the list. Get the inode's on-disk
2047 * buffer to see if there is anyone after us on the list.
2048 * Only modify our next pointer if it is not already NULLAGINO.
2049 * This saves us the overhead of dealing with the buffer when
2050 * there is no need to change it.
2052 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2053 0, 0);
2054 if (error) {
2055 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2056 __func__, error);
2057 return error;
2059 next_agino = be32_to_cpu(dip->di_next_unlinked);
2060 ASSERT(next_agino != 0);
2061 if (next_agino != NULLAGINO) {
2062 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2063 offset = ip->i_imap.im_boffset +
2064 offsetof(xfs_dinode_t, di_next_unlinked);
2066 /* need to recalc the inode CRC if appropriate */
2067 xfs_dinode_calc_crc(mp, dip);
2069 xfs_trans_inode_buf(tp, ibp);
2070 xfs_trans_log_buf(tp, ibp, offset,
2071 (offset + sizeof(xfs_agino_t) - 1));
2072 xfs_inobp_check(mp, ibp);
2073 } else {
2074 xfs_trans_brelse(tp, ibp);
2077 * Point the bucket head pointer at the next inode.
2079 ASSERT(next_agino != 0);
2080 ASSERT(next_agino != agino);
2081 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2082 offset = offsetof(xfs_agi_t, agi_unlinked) +
2083 (sizeof(xfs_agino_t) * bucket_index);
2084 xfs_trans_log_buf(tp, agibp, offset,
2085 (offset + sizeof(xfs_agino_t) - 1));
2086 } else {
2088 * We need to search the list for the inode being freed.
2090 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2091 last_ibp = NULL;
2092 while (next_agino != agino) {
2093 struct xfs_imap imap;
2095 if (last_ibp)
2096 xfs_trans_brelse(tp, last_ibp);
2098 imap.im_blkno = 0;
2099 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2101 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2102 if (error) {
2103 xfs_warn(mp,
2104 "%s: xfs_imap returned error %d.",
2105 __func__, error);
2106 return error;
2109 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2110 &last_ibp, 0, 0);
2111 if (error) {
2112 xfs_warn(mp,
2113 "%s: xfs_imap_to_bp returned error %d.",
2114 __func__, error);
2115 return error;
2118 last_offset = imap.im_boffset;
2119 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2120 if (!xfs_verify_agino(mp, agno, next_agino)) {
2121 XFS_CORRUPTION_ERROR(__func__,
2122 XFS_ERRLEVEL_LOW, mp,
2123 last_dip, sizeof(*last_dip));
2124 return -EFSCORRUPTED;
2129 * Now last_ibp points to the buffer previous to us on the
2130 * unlinked list. Pull us from the list.
2132 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2133 0, 0);
2134 if (error) {
2135 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2136 __func__, error);
2137 return error;
2139 next_agino = be32_to_cpu(dip->di_next_unlinked);
2140 ASSERT(next_agino != 0);
2141 ASSERT(next_agino != agino);
2142 if (next_agino != NULLAGINO) {
2143 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2144 offset = ip->i_imap.im_boffset +
2145 offsetof(xfs_dinode_t, di_next_unlinked);
2147 /* need to recalc the inode CRC if appropriate */
2148 xfs_dinode_calc_crc(mp, dip);
2150 xfs_trans_inode_buf(tp, ibp);
2151 xfs_trans_log_buf(tp, ibp, offset,
2152 (offset + sizeof(xfs_agino_t) - 1));
2153 xfs_inobp_check(mp, ibp);
2154 } else {
2155 xfs_trans_brelse(tp, ibp);
2158 * Point the previous inode on the list to the next inode.
2160 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2161 ASSERT(next_agino != 0);
2162 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2164 /* need to recalc the inode CRC if appropriate */
2165 xfs_dinode_calc_crc(mp, last_dip);
2167 xfs_trans_inode_buf(tp, last_ibp);
2168 xfs_trans_log_buf(tp, last_ibp, offset,
2169 (offset + sizeof(xfs_agino_t) - 1));
2170 xfs_inobp_check(mp, last_ibp);
2172 return 0;
2176 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2177 * inodes that are in memory - they all must be marked stale and attached to
2178 * the cluster buffer.
2180 STATIC int
2181 xfs_ifree_cluster(
2182 xfs_inode_t *free_ip,
2183 xfs_trans_t *tp,
2184 struct xfs_icluster *xic)
2186 xfs_mount_t *mp = free_ip->i_mount;
2187 int blks_per_cluster;
2188 int inodes_per_cluster;
2189 int nbufs;
2190 int i, j;
2191 int ioffset;
2192 xfs_daddr_t blkno;
2193 xfs_buf_t *bp;
2194 xfs_inode_t *ip;
2195 xfs_inode_log_item_t *iip;
2196 struct xfs_log_item *lip;
2197 struct xfs_perag *pag;
2198 xfs_ino_t inum;
2200 inum = xic->first_ino;
2201 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2202 blks_per_cluster = xfs_icluster_size_fsb(mp);
2203 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2204 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2206 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2208 * The allocation bitmap tells us which inodes of the chunk were
2209 * physically allocated. Skip the cluster if an inode falls into
2210 * a sparse region.
2212 ioffset = inum - xic->first_ino;
2213 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2214 ASSERT(ioffset % inodes_per_cluster == 0);
2215 continue;
2218 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2219 XFS_INO_TO_AGBNO(mp, inum));
2222 * We obtain and lock the backing buffer first in the process
2223 * here, as we have to ensure that any dirty inode that we
2224 * can't get the flush lock on is attached to the buffer.
2225 * If we scan the in-memory inodes first, then buffer IO can
2226 * complete before we get a lock on it, and hence we may fail
2227 * to mark all the active inodes on the buffer stale.
2229 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2230 mp->m_bsize * blks_per_cluster,
2231 XBF_UNMAPPED);
2233 if (!bp)
2234 return -ENOMEM;
2237 * This buffer may not have been correctly initialised as we
2238 * didn't read it from disk. That's not important because we are
2239 * only using to mark the buffer as stale in the log, and to
2240 * attach stale cached inodes on it. That means it will never be
2241 * dispatched for IO. If it is, we want to know about it, and we
2242 * want it to fail. We can acheive this by adding a write
2243 * verifier to the buffer.
2245 bp->b_ops = &xfs_inode_buf_ops;
2248 * Walk the inodes already attached to the buffer and mark them
2249 * stale. These will all have the flush locks held, so an
2250 * in-memory inode walk can't lock them. By marking them all
2251 * stale first, we will not attempt to lock them in the loop
2252 * below as the XFS_ISTALE flag will be set.
2254 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
2255 if (lip->li_type == XFS_LI_INODE) {
2256 iip = (xfs_inode_log_item_t *)lip;
2257 ASSERT(iip->ili_logged == 1);
2258 lip->li_cb = xfs_istale_done;
2259 xfs_trans_ail_copy_lsn(mp->m_ail,
2260 &iip->ili_flush_lsn,
2261 &iip->ili_item.li_lsn);
2262 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2268 * For each inode in memory attempt to add it to the inode
2269 * buffer and set it up for being staled on buffer IO
2270 * completion. This is safe as we've locked out tail pushing
2271 * and flushing by locking the buffer.
2273 * We have already marked every inode that was part of a
2274 * transaction stale above, which means there is no point in
2275 * even trying to lock them.
2277 for (i = 0; i < inodes_per_cluster; i++) {
2278 retry:
2279 rcu_read_lock();
2280 ip = radix_tree_lookup(&pag->pag_ici_root,
2281 XFS_INO_TO_AGINO(mp, (inum + i)));
2283 /* Inode not in memory, nothing to do */
2284 if (!ip) {
2285 rcu_read_unlock();
2286 continue;
2290 * because this is an RCU protected lookup, we could
2291 * find a recently freed or even reallocated inode
2292 * during the lookup. We need to check under the
2293 * i_flags_lock for a valid inode here. Skip it if it
2294 * is not valid, the wrong inode or stale.
2296 spin_lock(&ip->i_flags_lock);
2297 if (ip->i_ino != inum + i ||
2298 __xfs_iflags_test(ip, XFS_ISTALE)) {
2299 spin_unlock(&ip->i_flags_lock);
2300 rcu_read_unlock();
2301 continue;
2303 spin_unlock(&ip->i_flags_lock);
2306 * Don't try to lock/unlock the current inode, but we
2307 * _cannot_ skip the other inodes that we did not find
2308 * in the list attached to the buffer and are not
2309 * already marked stale. If we can't lock it, back off
2310 * and retry.
2312 if (ip != free_ip) {
2313 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2314 rcu_read_unlock();
2315 delay(1);
2316 goto retry;
2320 * Check the inode number again in case we're
2321 * racing with freeing in xfs_reclaim_inode().
2322 * See the comments in that function for more
2323 * information as to why the initial check is
2324 * not sufficient.
2326 if (ip->i_ino != inum + i) {
2327 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2328 rcu_read_unlock();
2329 continue;
2332 rcu_read_unlock();
2334 xfs_iflock(ip);
2335 xfs_iflags_set(ip, XFS_ISTALE);
2338 * we don't need to attach clean inodes or those only
2339 * with unlogged changes (which we throw away, anyway).
2341 iip = ip->i_itemp;
2342 if (!iip || xfs_inode_clean(ip)) {
2343 ASSERT(ip != free_ip);
2344 xfs_ifunlock(ip);
2345 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2346 continue;
2349 iip->ili_last_fields = iip->ili_fields;
2350 iip->ili_fields = 0;
2351 iip->ili_fsync_fields = 0;
2352 iip->ili_logged = 1;
2353 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2354 &iip->ili_item.li_lsn);
2356 xfs_buf_attach_iodone(bp, xfs_istale_done,
2357 &iip->ili_item);
2359 if (ip != free_ip)
2360 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2363 xfs_trans_stale_inode_buf(tp, bp);
2364 xfs_trans_binval(tp, bp);
2367 xfs_perag_put(pag);
2368 return 0;
2372 * Free any local-format buffers sitting around before we reset to
2373 * extents format.
2375 static inline void
2376 xfs_ifree_local_data(
2377 struct xfs_inode *ip,
2378 int whichfork)
2380 struct xfs_ifork *ifp;
2382 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2383 return;
2385 ifp = XFS_IFORK_PTR(ip, whichfork);
2386 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2390 * This is called to return an inode to the inode free list.
2391 * The inode should already be truncated to 0 length and have
2392 * no pages associated with it. This routine also assumes that
2393 * the inode is already a part of the transaction.
2395 * The on-disk copy of the inode will have been added to the list
2396 * of unlinked inodes in the AGI. We need to remove the inode from
2397 * that list atomically with respect to freeing it here.
2400 xfs_ifree(
2401 struct xfs_trans *tp,
2402 struct xfs_inode *ip)
2404 int error;
2405 struct xfs_icluster xic = { 0 };
2407 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2408 ASSERT(VFS_I(ip)->i_nlink == 0);
2409 ASSERT(ip->i_d.di_nextents == 0);
2410 ASSERT(ip->i_d.di_anextents == 0);
2411 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2412 ASSERT(ip->i_d.di_nblocks == 0);
2415 * Pull the on-disk inode from the AGI unlinked list.
2417 error = xfs_iunlink_remove(tp, ip);
2418 if (error)
2419 return error;
2421 error = xfs_difree(tp, ip->i_ino, &xic);
2422 if (error)
2423 return error;
2425 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2426 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2428 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2429 ip->i_d.di_flags = 0;
2430 ip->i_d.di_flags2 = 0;
2431 ip->i_d.di_dmevmask = 0;
2432 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2433 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2434 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2436 /* Don't attempt to replay owner changes for a deleted inode */
2437 ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER);
2440 * Bump the generation count so no one will be confused
2441 * by reincarnations of this inode.
2443 VFS_I(ip)->i_generation++;
2444 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2446 if (xic.deleted)
2447 error = xfs_ifree_cluster(ip, tp, &xic);
2449 return error;
2453 * This is called to unpin an inode. The caller must have the inode locked
2454 * in at least shared mode so that the buffer cannot be subsequently pinned
2455 * once someone is waiting for it to be unpinned.
2457 static void
2458 xfs_iunpin(
2459 struct xfs_inode *ip)
2461 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2463 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2465 /* Give the log a push to start the unpinning I/O */
2466 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2470 static void
2471 __xfs_iunpin_wait(
2472 struct xfs_inode *ip)
2474 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2475 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2477 xfs_iunpin(ip);
2479 do {
2480 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2481 if (xfs_ipincount(ip))
2482 io_schedule();
2483 } while (xfs_ipincount(ip));
2484 finish_wait(wq, &wait.wq_entry);
2487 void
2488 xfs_iunpin_wait(
2489 struct xfs_inode *ip)
2491 if (xfs_ipincount(ip))
2492 __xfs_iunpin_wait(ip);
2496 * Removing an inode from the namespace involves removing the directory entry
2497 * and dropping the link count on the inode. Removing the directory entry can
2498 * result in locking an AGF (directory blocks were freed) and removing a link
2499 * count can result in placing the inode on an unlinked list which results in
2500 * locking an AGI.
2502 * The big problem here is that we have an ordering constraint on AGF and AGI
2503 * locking - inode allocation locks the AGI, then can allocate a new extent for
2504 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2505 * removes the inode from the unlinked list, requiring that we lock the AGI
2506 * first, and then freeing the inode can result in an inode chunk being freed
2507 * and hence freeing disk space requiring that we lock an AGF.
2509 * Hence the ordering that is imposed by other parts of the code is AGI before
2510 * AGF. This means we cannot remove the directory entry before we drop the inode
2511 * reference count and put it on the unlinked list as this results in a lock
2512 * order of AGF then AGI, and this can deadlock against inode allocation and
2513 * freeing. Therefore we must drop the link counts before we remove the
2514 * directory entry.
2516 * This is still safe from a transactional point of view - it is not until we
2517 * get to xfs_defer_finish() that we have the possibility of multiple
2518 * transactions in this operation. Hence as long as we remove the directory
2519 * entry and drop the link count in the first transaction of the remove
2520 * operation, there are no transactional constraints on the ordering here.
2523 xfs_remove(
2524 xfs_inode_t *dp,
2525 struct xfs_name *name,
2526 xfs_inode_t *ip)
2528 xfs_mount_t *mp = dp->i_mount;
2529 xfs_trans_t *tp = NULL;
2530 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2531 int error = 0;
2532 uint resblks;
2534 trace_xfs_remove(dp, name);
2536 if (XFS_FORCED_SHUTDOWN(mp))
2537 return -EIO;
2539 error = xfs_qm_dqattach(dp);
2540 if (error)
2541 goto std_return;
2543 error = xfs_qm_dqattach(ip);
2544 if (error)
2545 goto std_return;
2548 * We try to get the real space reservation first,
2549 * allowing for directory btree deletion(s) implying
2550 * possible bmap insert(s). If we can't get the space
2551 * reservation then we use 0 instead, and avoid the bmap
2552 * btree insert(s) in the directory code by, if the bmap
2553 * insert tries to happen, instead trimming the LAST
2554 * block from the directory.
2556 resblks = XFS_REMOVE_SPACE_RES(mp);
2557 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2558 if (error == -ENOSPC) {
2559 resblks = 0;
2560 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2561 &tp);
2563 if (error) {
2564 ASSERT(error != -ENOSPC);
2565 goto std_return;
2568 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2570 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2571 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2574 * If we're removing a directory perform some additional validation.
2576 if (is_dir) {
2577 ASSERT(VFS_I(ip)->i_nlink >= 2);
2578 if (VFS_I(ip)->i_nlink != 2) {
2579 error = -ENOTEMPTY;
2580 goto out_trans_cancel;
2582 if (!xfs_dir_isempty(ip)) {
2583 error = -ENOTEMPTY;
2584 goto out_trans_cancel;
2587 /* Drop the link from ip's "..". */
2588 error = xfs_droplink(tp, dp);
2589 if (error)
2590 goto out_trans_cancel;
2592 /* Drop the "." link from ip to self. */
2593 error = xfs_droplink(tp, ip);
2594 if (error)
2595 goto out_trans_cancel;
2596 } else {
2598 * When removing a non-directory we need to log the parent
2599 * inode here. For a directory this is done implicitly
2600 * by the xfs_droplink call for the ".." entry.
2602 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2604 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2606 /* Drop the link from dp to ip. */
2607 error = xfs_droplink(tp, ip);
2608 if (error)
2609 goto out_trans_cancel;
2611 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2612 if (error) {
2613 ASSERT(error != -ENOENT);
2614 goto out_trans_cancel;
2618 * If this is a synchronous mount, make sure that the
2619 * remove transaction goes to disk before returning to
2620 * the user.
2622 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2623 xfs_trans_set_sync(tp);
2625 error = xfs_trans_commit(tp);
2626 if (error)
2627 goto std_return;
2629 if (is_dir && xfs_inode_is_filestream(ip))
2630 xfs_filestream_deassociate(ip);
2632 return 0;
2634 out_trans_cancel:
2635 xfs_trans_cancel(tp);
2636 std_return:
2637 return error;
2641 * Enter all inodes for a rename transaction into a sorted array.
2643 #define __XFS_SORT_INODES 5
2644 STATIC void
2645 xfs_sort_for_rename(
2646 struct xfs_inode *dp1, /* in: old (source) directory inode */
2647 struct xfs_inode *dp2, /* in: new (target) directory inode */
2648 struct xfs_inode *ip1, /* in: inode of old entry */
2649 struct xfs_inode *ip2, /* in: inode of new entry */
2650 struct xfs_inode *wip, /* in: whiteout inode */
2651 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2652 int *num_inodes) /* in/out: inodes in array */
2654 int i, j;
2656 ASSERT(*num_inodes == __XFS_SORT_INODES);
2657 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2660 * i_tab contains a list of pointers to inodes. We initialize
2661 * the table here & we'll sort it. We will then use it to
2662 * order the acquisition of the inode locks.
2664 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2666 i = 0;
2667 i_tab[i++] = dp1;
2668 i_tab[i++] = dp2;
2669 i_tab[i++] = ip1;
2670 if (ip2)
2671 i_tab[i++] = ip2;
2672 if (wip)
2673 i_tab[i++] = wip;
2674 *num_inodes = i;
2677 * Sort the elements via bubble sort. (Remember, there are at
2678 * most 5 elements to sort, so this is adequate.)
2680 for (i = 0; i < *num_inodes; i++) {
2681 for (j = 1; j < *num_inodes; j++) {
2682 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2683 struct xfs_inode *temp = i_tab[j];
2684 i_tab[j] = i_tab[j-1];
2685 i_tab[j-1] = temp;
2691 static int
2692 xfs_finish_rename(
2693 struct xfs_trans *tp)
2696 * If this is a synchronous mount, make sure that the rename transaction
2697 * goes to disk before returning to the user.
2699 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2700 xfs_trans_set_sync(tp);
2702 return xfs_trans_commit(tp);
2706 * xfs_cross_rename()
2708 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2710 STATIC int
2711 xfs_cross_rename(
2712 struct xfs_trans *tp,
2713 struct xfs_inode *dp1,
2714 struct xfs_name *name1,
2715 struct xfs_inode *ip1,
2716 struct xfs_inode *dp2,
2717 struct xfs_name *name2,
2718 struct xfs_inode *ip2,
2719 int spaceres)
2721 int error = 0;
2722 int ip1_flags = 0;
2723 int ip2_flags = 0;
2724 int dp2_flags = 0;
2726 /* Swap inode number for dirent in first parent */
2727 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2728 if (error)
2729 goto out_trans_abort;
2731 /* Swap inode number for dirent in second parent */
2732 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2733 if (error)
2734 goto out_trans_abort;
2737 * If we're renaming one or more directories across different parents,
2738 * update the respective ".." entries (and link counts) to match the new
2739 * parents.
2741 if (dp1 != dp2) {
2742 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2744 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2745 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2746 dp1->i_ino, spaceres);
2747 if (error)
2748 goto out_trans_abort;
2750 /* transfer ip2 ".." reference to dp1 */
2751 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2752 error = xfs_droplink(tp, dp2);
2753 if (error)
2754 goto out_trans_abort;
2755 error = xfs_bumplink(tp, dp1);
2756 if (error)
2757 goto out_trans_abort;
2761 * Although ip1 isn't changed here, userspace needs
2762 * to be warned about the change, so that applications
2763 * relying on it (like backup ones), will properly
2764 * notify the change
2766 ip1_flags |= XFS_ICHGTIME_CHG;
2767 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2770 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2771 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2772 dp2->i_ino, spaceres);
2773 if (error)
2774 goto out_trans_abort;
2776 /* transfer ip1 ".." reference to dp2 */
2777 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2778 error = xfs_droplink(tp, dp1);
2779 if (error)
2780 goto out_trans_abort;
2781 error = xfs_bumplink(tp, dp2);
2782 if (error)
2783 goto out_trans_abort;
2787 * Although ip2 isn't changed here, userspace needs
2788 * to be warned about the change, so that applications
2789 * relying on it (like backup ones), will properly
2790 * notify the change
2792 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2793 ip2_flags |= XFS_ICHGTIME_CHG;
2797 if (ip1_flags) {
2798 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2799 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2801 if (ip2_flags) {
2802 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2803 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2805 if (dp2_flags) {
2806 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2807 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2809 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2810 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2811 return xfs_finish_rename(tp);
2813 out_trans_abort:
2814 xfs_trans_cancel(tp);
2815 return error;
2819 * xfs_rename_alloc_whiteout()
2821 * Return a referenced, unlinked, unlocked inode that that can be used as a
2822 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2823 * crash between allocating the inode and linking it into the rename transaction
2824 * recovery will free the inode and we won't leak it.
2826 static int
2827 xfs_rename_alloc_whiteout(
2828 struct xfs_inode *dp,
2829 struct xfs_inode **wip)
2831 struct xfs_inode *tmpfile;
2832 int error;
2834 error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2835 if (error)
2836 return error;
2839 * Prepare the tmpfile inode as if it were created through the VFS.
2840 * Otherwise, the link increment paths will complain about nlink 0->1.
2841 * Drop the link count as done by d_tmpfile(), complete the inode setup
2842 * and flag it as linkable.
2844 drop_nlink(VFS_I(tmpfile));
2845 xfs_setup_iops(tmpfile);
2846 xfs_finish_inode_setup(tmpfile);
2847 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2849 *wip = tmpfile;
2850 return 0;
2854 * xfs_rename
2857 xfs_rename(
2858 struct xfs_inode *src_dp,
2859 struct xfs_name *src_name,
2860 struct xfs_inode *src_ip,
2861 struct xfs_inode *target_dp,
2862 struct xfs_name *target_name,
2863 struct xfs_inode *target_ip,
2864 unsigned int flags)
2866 struct xfs_mount *mp = src_dp->i_mount;
2867 struct xfs_trans *tp;
2868 struct xfs_inode *wip = NULL; /* whiteout inode */
2869 struct xfs_inode *inodes[__XFS_SORT_INODES];
2870 int num_inodes = __XFS_SORT_INODES;
2871 bool new_parent = (src_dp != target_dp);
2872 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2873 int spaceres;
2874 int error;
2876 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2878 if ((flags & RENAME_EXCHANGE) && !target_ip)
2879 return -EINVAL;
2882 * If we are doing a whiteout operation, allocate the whiteout inode
2883 * we will be placing at the target and ensure the type is set
2884 * appropriately.
2886 if (flags & RENAME_WHITEOUT) {
2887 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2888 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2889 if (error)
2890 return error;
2892 /* setup target dirent info as whiteout */
2893 src_name->type = XFS_DIR3_FT_CHRDEV;
2896 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2897 inodes, &num_inodes);
2899 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2900 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2901 if (error == -ENOSPC) {
2902 spaceres = 0;
2903 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2904 &tp);
2906 if (error)
2907 goto out_release_wip;
2910 * Attach the dquots to the inodes
2912 error = xfs_qm_vop_rename_dqattach(inodes);
2913 if (error)
2914 goto out_trans_cancel;
2917 * Lock all the participating inodes. Depending upon whether
2918 * the target_name exists in the target directory, and
2919 * whether the target directory is the same as the source
2920 * directory, we can lock from 2 to 4 inodes.
2922 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2925 * Join all the inodes to the transaction. From this point on,
2926 * we can rely on either trans_commit or trans_cancel to unlock
2927 * them.
2929 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2930 if (new_parent)
2931 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2932 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2933 if (target_ip)
2934 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2935 if (wip)
2936 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2939 * If we are using project inheritance, we only allow renames
2940 * into our tree when the project IDs are the same; else the
2941 * tree quota mechanism would be circumvented.
2943 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2944 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2945 error = -EXDEV;
2946 goto out_trans_cancel;
2949 /* RENAME_EXCHANGE is unique from here on. */
2950 if (flags & RENAME_EXCHANGE)
2951 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2952 target_dp, target_name, target_ip,
2953 spaceres);
2956 * Set up the target.
2958 if (target_ip == NULL) {
2960 * If there's no space reservation, check the entry will
2961 * fit before actually inserting it.
2963 if (!spaceres) {
2964 error = xfs_dir_canenter(tp, target_dp, target_name);
2965 if (error)
2966 goto out_trans_cancel;
2969 * If target does not exist and the rename crosses
2970 * directories, adjust the target directory link count
2971 * to account for the ".." reference from the new entry.
2973 error = xfs_dir_createname(tp, target_dp, target_name,
2974 src_ip->i_ino, spaceres);
2975 if (error)
2976 goto out_trans_cancel;
2978 xfs_trans_ichgtime(tp, target_dp,
2979 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2981 if (new_parent && src_is_directory) {
2982 error = xfs_bumplink(tp, target_dp);
2983 if (error)
2984 goto out_trans_cancel;
2986 } else { /* target_ip != NULL */
2988 * If target exists and it's a directory, check that both
2989 * target and source are directories and that target can be
2990 * destroyed, or that neither is a directory.
2992 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
2994 * Make sure target dir is empty.
2996 if (!(xfs_dir_isempty(target_ip)) ||
2997 (VFS_I(target_ip)->i_nlink > 2)) {
2998 error = -EEXIST;
2999 goto out_trans_cancel;
3004 * Link the source inode under the target name.
3005 * If the source inode is a directory and we are moving
3006 * it across directories, its ".." entry will be
3007 * inconsistent until we replace that down below.
3009 * In case there is already an entry with the same
3010 * name at the destination directory, remove it first.
3012 error = xfs_dir_replace(tp, target_dp, target_name,
3013 src_ip->i_ino, spaceres);
3014 if (error)
3015 goto out_trans_cancel;
3017 xfs_trans_ichgtime(tp, target_dp,
3018 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3021 * Decrement the link count on the target since the target
3022 * dir no longer points to it.
3024 error = xfs_droplink(tp, target_ip);
3025 if (error)
3026 goto out_trans_cancel;
3028 if (src_is_directory) {
3030 * Drop the link from the old "." entry.
3032 error = xfs_droplink(tp, target_ip);
3033 if (error)
3034 goto out_trans_cancel;
3036 } /* target_ip != NULL */
3039 * Remove the source.
3041 if (new_parent && src_is_directory) {
3043 * Rewrite the ".." entry to point to the new
3044 * directory.
3046 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3047 target_dp->i_ino, spaceres);
3048 ASSERT(error != -EEXIST);
3049 if (error)
3050 goto out_trans_cancel;
3054 * We always want to hit the ctime on the source inode.
3056 * This isn't strictly required by the standards since the source
3057 * inode isn't really being changed, but old unix file systems did
3058 * it and some incremental backup programs won't work without it.
3060 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3061 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3064 * Adjust the link count on src_dp. This is necessary when
3065 * renaming a directory, either within one parent when
3066 * the target existed, or across two parent directories.
3068 if (src_is_directory && (new_parent || target_ip != NULL)) {
3071 * Decrement link count on src_directory since the
3072 * entry that's moved no longer points to it.
3074 error = xfs_droplink(tp, src_dp);
3075 if (error)
3076 goto out_trans_cancel;
3080 * For whiteouts, we only need to update the source dirent with the
3081 * inode number of the whiteout inode rather than removing it
3082 * altogether.
3084 if (wip) {
3085 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3086 spaceres);
3087 } else
3088 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3089 spaceres);
3090 if (error)
3091 goto out_trans_cancel;
3094 * For whiteouts, we need to bump the link count on the whiteout inode.
3095 * This means that failures all the way up to this point leave the inode
3096 * on the unlinked list and so cleanup is a simple matter of dropping
3097 * the remaining reference to it. If we fail here after bumping the link
3098 * count, we're shutting down the filesystem so we'll never see the
3099 * intermediate state on disk.
3101 if (wip) {
3102 ASSERT(VFS_I(wip)->i_nlink == 0);
3103 error = xfs_bumplink(tp, wip);
3104 if (error)
3105 goto out_trans_cancel;
3106 error = xfs_iunlink_remove(tp, wip);
3107 if (error)
3108 goto out_trans_cancel;
3109 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3112 * Now we have a real link, clear the "I'm a tmpfile" state
3113 * flag from the inode so it doesn't accidentally get misused in
3114 * future.
3116 VFS_I(wip)->i_state &= ~I_LINKABLE;
3119 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3120 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3121 if (new_parent)
3122 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3124 error = xfs_finish_rename(tp);
3125 if (wip)
3126 xfs_irele(wip);
3127 return error;
3129 out_trans_cancel:
3130 xfs_trans_cancel(tp);
3131 out_release_wip:
3132 if (wip)
3133 xfs_irele(wip);
3134 return error;
3137 STATIC int
3138 xfs_iflush_cluster(
3139 struct xfs_inode *ip,
3140 struct xfs_buf *bp)
3142 struct xfs_mount *mp = ip->i_mount;
3143 struct xfs_perag *pag;
3144 unsigned long first_index, mask;
3145 unsigned long inodes_per_cluster;
3146 int cilist_size;
3147 struct xfs_inode **cilist;
3148 struct xfs_inode *cip;
3149 int nr_found;
3150 int clcount = 0;
3151 int i;
3153 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3155 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3156 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3157 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3158 if (!cilist)
3159 goto out_put;
3161 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3162 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3163 rcu_read_lock();
3164 /* really need a gang lookup range call here */
3165 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3166 first_index, inodes_per_cluster);
3167 if (nr_found == 0)
3168 goto out_free;
3170 for (i = 0; i < nr_found; i++) {
3171 cip = cilist[i];
3172 if (cip == ip)
3173 continue;
3176 * because this is an RCU protected lookup, we could find a
3177 * recently freed or even reallocated inode during the lookup.
3178 * We need to check under the i_flags_lock for a valid inode
3179 * here. Skip it if it is not valid or the wrong inode.
3181 spin_lock(&cip->i_flags_lock);
3182 if (!cip->i_ino ||
3183 __xfs_iflags_test(cip, XFS_ISTALE)) {
3184 spin_unlock(&cip->i_flags_lock);
3185 continue;
3189 * Once we fall off the end of the cluster, no point checking
3190 * any more inodes in the list because they will also all be
3191 * outside the cluster.
3193 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3194 spin_unlock(&cip->i_flags_lock);
3195 break;
3197 spin_unlock(&cip->i_flags_lock);
3200 * Do an un-protected check to see if the inode is dirty and
3201 * is a candidate for flushing. These checks will be repeated
3202 * later after the appropriate locks are acquired.
3204 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3205 continue;
3208 * Try to get locks. If any are unavailable or it is pinned,
3209 * then this inode cannot be flushed and is skipped.
3212 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3213 continue;
3214 if (!xfs_iflock_nowait(cip)) {
3215 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3216 continue;
3218 if (xfs_ipincount(cip)) {
3219 xfs_ifunlock(cip);
3220 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3221 continue;
3226 * Check the inode number again, just to be certain we are not
3227 * racing with freeing in xfs_reclaim_inode(). See the comments
3228 * in that function for more information as to why the initial
3229 * check is not sufficient.
3231 if (!cip->i_ino) {
3232 xfs_ifunlock(cip);
3233 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3234 continue;
3238 * arriving here means that this inode can be flushed. First
3239 * re-check that it's dirty before flushing.
3241 if (!xfs_inode_clean(cip)) {
3242 int error;
3243 error = xfs_iflush_int(cip, bp);
3244 if (error) {
3245 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3246 goto cluster_corrupt_out;
3248 clcount++;
3249 } else {
3250 xfs_ifunlock(cip);
3252 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3255 if (clcount) {
3256 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3257 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3260 out_free:
3261 rcu_read_unlock();
3262 kmem_free(cilist);
3263 out_put:
3264 xfs_perag_put(pag);
3265 return 0;
3268 cluster_corrupt_out:
3270 * Corruption detected in the clustering loop. Invalidate the
3271 * inode buffer and shut down the filesystem.
3273 rcu_read_unlock();
3274 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3277 * We'll always have an inode attached to the buffer for completion
3278 * process by the time we are called from xfs_iflush(). Hence we have
3279 * always need to do IO completion processing to abort the inodes
3280 * attached to the buffer. handle them just like the shutdown case in
3281 * xfs_buf_submit().
3283 ASSERT(bp->b_iodone);
3284 bp->b_flags &= ~XBF_DONE;
3285 xfs_buf_stale(bp);
3286 xfs_buf_ioerror(bp, -EIO);
3287 xfs_buf_ioend(bp);
3289 /* abort the corrupt inode, as it was not attached to the buffer */
3290 xfs_iflush_abort(cip, false);
3291 kmem_free(cilist);
3292 xfs_perag_put(pag);
3293 return -EFSCORRUPTED;
3297 * Flush dirty inode metadata into the backing buffer.
3299 * The caller must have the inode lock and the inode flush lock held. The
3300 * inode lock will still be held upon return to the caller, and the inode
3301 * flush lock will be released after the inode has reached the disk.
3303 * The caller must write out the buffer returned in *bpp and release it.
3306 xfs_iflush(
3307 struct xfs_inode *ip,
3308 struct xfs_buf **bpp)
3310 struct xfs_mount *mp = ip->i_mount;
3311 struct xfs_buf *bp = NULL;
3312 struct xfs_dinode *dip;
3313 int error;
3315 XFS_STATS_INC(mp, xs_iflush_count);
3317 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3318 ASSERT(xfs_isiflocked(ip));
3319 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3320 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3322 *bpp = NULL;
3324 xfs_iunpin_wait(ip);
3327 * For stale inodes we cannot rely on the backing buffer remaining
3328 * stale in cache for the remaining life of the stale inode and so
3329 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3330 * inodes below. We have to check this after ensuring the inode is
3331 * unpinned so that it is safe to reclaim the stale inode after the
3332 * flush call.
3334 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3335 xfs_ifunlock(ip);
3336 return 0;
3340 * This may have been unpinned because the filesystem is shutting
3341 * down forcibly. If that's the case we must not write this inode
3342 * to disk, because the log record didn't make it to disk.
3344 * We also have to remove the log item from the AIL in this case,
3345 * as we wait for an empty AIL as part of the unmount process.
3347 if (XFS_FORCED_SHUTDOWN(mp)) {
3348 error = -EIO;
3349 goto abort_out;
3353 * Get the buffer containing the on-disk inode. We are doing a try-lock
3354 * operation here, so we may get an EAGAIN error. In that case, we
3355 * simply want to return with the inode still dirty.
3357 * If we get any other error, we effectively have a corruption situation
3358 * and we cannot flush the inode, so we treat it the same as failing
3359 * xfs_iflush_int().
3361 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3363 if (error == -EAGAIN) {
3364 xfs_ifunlock(ip);
3365 return error;
3367 if (error)
3368 goto corrupt_out;
3371 * First flush out the inode that xfs_iflush was called with.
3373 error = xfs_iflush_int(ip, bp);
3374 if (error)
3375 goto corrupt_out;
3378 * If the buffer is pinned then push on the log now so we won't
3379 * get stuck waiting in the write for too long.
3381 if (xfs_buf_ispinned(bp))
3382 xfs_log_force(mp, 0);
3385 * inode clustering: try to gather other inodes into this write
3387 * Note: Any error during clustering will result in the filesystem
3388 * being shut down and completion callbacks run on the cluster buffer.
3389 * As we have already flushed and attached this inode to the buffer,
3390 * it has already been aborted and released by xfs_iflush_cluster() and
3391 * so we have no further error handling to do here.
3393 error = xfs_iflush_cluster(ip, bp);
3394 if (error)
3395 return error;
3397 *bpp = bp;
3398 return 0;
3400 corrupt_out:
3401 if (bp)
3402 xfs_buf_relse(bp);
3403 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3404 abort_out:
3405 /* abort the corrupt inode, as it was not attached to the buffer */
3406 xfs_iflush_abort(ip, false);
3407 return error;
3411 * If there are inline format data / attr forks attached to this inode,
3412 * make sure they're not corrupt.
3414 bool
3415 xfs_inode_verify_forks(
3416 struct xfs_inode *ip)
3418 struct xfs_ifork *ifp;
3419 xfs_failaddr_t fa;
3421 fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops);
3422 if (fa) {
3423 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3424 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
3425 ifp->if_u1.if_data, ifp->if_bytes, fa);
3426 return false;
3429 fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops);
3430 if (fa) {
3431 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
3432 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
3433 ifp ? ifp->if_u1.if_data : NULL,
3434 ifp ? ifp->if_bytes : 0, fa);
3435 return false;
3437 return true;
3440 STATIC int
3441 xfs_iflush_int(
3442 struct xfs_inode *ip,
3443 struct xfs_buf *bp)
3445 struct xfs_inode_log_item *iip = ip->i_itemp;
3446 struct xfs_dinode *dip;
3447 struct xfs_mount *mp = ip->i_mount;
3449 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3450 ASSERT(xfs_isiflocked(ip));
3451 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3452 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3453 ASSERT(iip != NULL && iip->ili_fields != 0);
3454 ASSERT(ip->i_d.di_version > 1);
3456 /* set *dip = inode's place in the buffer */
3457 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3459 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3460 mp, XFS_ERRTAG_IFLUSH_1)) {
3461 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3462 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3463 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3464 goto corrupt_out;
3466 if (S_ISREG(VFS_I(ip)->i_mode)) {
3467 if (XFS_TEST_ERROR(
3468 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3469 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3470 mp, XFS_ERRTAG_IFLUSH_3)) {
3471 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3472 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3473 __func__, ip->i_ino, ip);
3474 goto corrupt_out;
3476 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3477 if (XFS_TEST_ERROR(
3478 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3479 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3480 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3481 mp, XFS_ERRTAG_IFLUSH_4)) {
3482 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3483 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3484 __func__, ip->i_ino, ip);
3485 goto corrupt_out;
3488 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3489 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3490 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3491 "%s: detected corrupt incore inode %Lu, "
3492 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3493 __func__, ip->i_ino,
3494 ip->i_d.di_nextents + ip->i_d.di_anextents,
3495 ip->i_d.di_nblocks, ip);
3496 goto corrupt_out;
3498 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3499 mp, XFS_ERRTAG_IFLUSH_6)) {
3500 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3501 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3502 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3503 goto corrupt_out;
3507 * Inode item log recovery for v2 inodes are dependent on the
3508 * di_flushiter count for correct sequencing. We bump the flush
3509 * iteration count so we can detect flushes which postdate a log record
3510 * during recovery. This is redundant as we now log every change and
3511 * hence this can't happen but we need to still do it to ensure
3512 * backwards compatibility with old kernels that predate logging all
3513 * inode changes.
3515 if (ip->i_d.di_version < 3)
3516 ip->i_d.di_flushiter++;
3518 /* Check the inline fork data before we write out. */
3519 if (!xfs_inode_verify_forks(ip))
3520 goto corrupt_out;
3523 * Copy the dirty parts of the inode into the on-disk inode. We always
3524 * copy out the core of the inode, because if the inode is dirty at all
3525 * the core must be.
3527 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3529 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3530 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3531 ip->i_d.di_flushiter = 0;
3533 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3534 if (XFS_IFORK_Q(ip))
3535 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3536 xfs_inobp_check(mp, bp);
3539 * We've recorded everything logged in the inode, so we'd like to clear
3540 * the ili_fields bits so we don't log and flush things unnecessarily.
3541 * However, we can't stop logging all this information until the data
3542 * we've copied into the disk buffer is written to disk. If we did we
3543 * might overwrite the copy of the inode in the log with all the data
3544 * after re-logging only part of it, and in the face of a crash we
3545 * wouldn't have all the data we need to recover.
3547 * What we do is move the bits to the ili_last_fields field. When
3548 * logging the inode, these bits are moved back to the ili_fields field.
3549 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3550 * know that the information those bits represent is permanently on
3551 * disk. As long as the flush completes before the inode is logged
3552 * again, then both ili_fields and ili_last_fields will be cleared.
3554 * We can play with the ili_fields bits here, because the inode lock
3555 * must be held exclusively in order to set bits there and the flush
3556 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3557 * done routine can tell whether or not to look in the AIL. Also, store
3558 * the current LSN of the inode so that we can tell whether the item has
3559 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3560 * need the AIL lock, because it is a 64 bit value that cannot be read
3561 * atomically.
3563 iip->ili_last_fields = iip->ili_fields;
3564 iip->ili_fields = 0;
3565 iip->ili_fsync_fields = 0;
3566 iip->ili_logged = 1;
3568 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3569 &iip->ili_item.li_lsn);
3572 * Attach the function xfs_iflush_done to the inode's
3573 * buffer. This will remove the inode from the AIL
3574 * and unlock the inode's flush lock when the inode is
3575 * completely written to disk.
3577 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3579 /* generate the checksum. */
3580 xfs_dinode_calc_crc(mp, dip);
3582 ASSERT(!list_empty(&bp->b_li_list));
3583 ASSERT(bp->b_iodone != NULL);
3584 return 0;
3586 corrupt_out:
3587 return -EFSCORRUPTED;
3590 /* Release an inode. */
3591 void
3592 xfs_irele(
3593 struct xfs_inode *ip)
3595 trace_xfs_irele(ip, _RET_IP_);
3596 iput(VFS_I(ip));