1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
6 #include <linux/iversion.h>
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
19 #include "xfs_trans_space.h"
20 #include "xfs_trans.h"
21 #include "xfs_buf_item.h"
22 #include "xfs_inode_item.h"
23 #include "xfs_iunlink_item.h"
24 #include "xfs_ialloc.h"
26 #include "xfs_bmap_util.h"
27 #include "xfs_errortag.h"
28 #include "xfs_error.h"
29 #include "xfs_quota.h"
30 #include "xfs_filestream.h"
31 #include "xfs_trace.h"
32 #include "xfs_icache.h"
33 #include "xfs_symlink.h"
34 #include "xfs_trans_priv.h"
36 #include "xfs_bmap_btree.h"
37 #include "xfs_reflink.h"
39 #include "xfs_log_priv.h"
40 #include "xfs_health.h"
42 struct kmem_cache *xfs_inode_cache;
44 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
45 STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
49 * helper function to extract extent size hint from inode
56 * No point in aligning allocations if we need to COW to actually
59 if (xfs_is_always_cow_inode(ip))
61 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
63 if (XFS_IS_REALTIME_INODE(ip))
64 return ip->i_mount->m_sb.sb_rextsize;
69 * Helper function to extract CoW extent size hint from inode.
70 * Between the extent size hint and the CoW extent size hint, we
71 * return the greater of the two. If the value is zero (automatic),
72 * use the default size.
75 xfs_get_cowextsz_hint(
81 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
83 b = xfs_get_extsz_hint(ip);
87 return XFS_DEFAULT_COWEXTSZ_HINT;
92 * These two are wrapper routines around the xfs_ilock() routine used to
93 * centralize some grungy code. They are used in places that wish to lock the
94 * inode solely for reading the extents. The reason these places can't just
95 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
96 * bringing in of the extents from disk for a file in b-tree format. If the
97 * inode is in b-tree format, then we need to lock the inode exclusively until
98 * the extents are read in. Locking it exclusively all the time would limit
99 * our parallelism unnecessarily, though. What we do instead is check to see
100 * if the extents have been read in yet, and only lock the inode exclusively
103 * The functions return a value which should be given to the corresponding
104 * xfs_iunlock() call.
107 xfs_ilock_data_map_shared(
108 struct xfs_inode *ip)
110 uint lock_mode = XFS_ILOCK_SHARED;
112 if (xfs_need_iread_extents(&ip->i_df))
113 lock_mode = XFS_ILOCK_EXCL;
114 xfs_ilock(ip, lock_mode);
119 xfs_ilock_attr_map_shared(
120 struct xfs_inode *ip)
122 uint lock_mode = XFS_ILOCK_SHARED;
124 if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
125 lock_mode = XFS_ILOCK_EXCL;
126 xfs_ilock(ip, lock_mode);
131 * You can't set both SHARED and EXCL for the same lock,
132 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
133 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
134 * to set in lock_flags.
137 xfs_lock_flags_assert(
140 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
141 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
142 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
143 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
144 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
145 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
146 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
147 ASSERT(lock_flags != 0);
151 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
152 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
153 * various combinations of the locks to be obtained.
155 * The 3 locks should always be ordered so that the IO lock is obtained first,
156 * the mmap lock second and the ilock last in order to prevent deadlock.
158 * Basic locking order:
160 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
162 * mmap_lock locking order:
164 * i_rwsem -> page lock -> mmap_lock
165 * mmap_lock -> invalidate_lock -> page_lock
167 * The difference in mmap_lock locking order mean that we cannot hold the
168 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
169 * can fault in pages during copy in/out (for buffered IO) or require the
170 * mmap_lock in get_user_pages() to map the user pages into the kernel address
171 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
172 * fault because page faults already hold the mmap_lock.
174 * Hence to serialise fully against both syscall and mmap based IO, we need to
175 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
176 * both taken in places where we need to invalidate the page cache in a race
177 * free manner (e.g. truncate, hole punch and other extent manipulation
185 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
187 xfs_lock_flags_assert(lock_flags);
189 if (lock_flags & XFS_IOLOCK_EXCL) {
190 down_write_nested(&VFS_I(ip)->i_rwsem,
191 XFS_IOLOCK_DEP(lock_flags));
192 } else if (lock_flags & XFS_IOLOCK_SHARED) {
193 down_read_nested(&VFS_I(ip)->i_rwsem,
194 XFS_IOLOCK_DEP(lock_flags));
197 if (lock_flags & XFS_MMAPLOCK_EXCL) {
198 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
199 XFS_MMAPLOCK_DEP(lock_flags));
200 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
201 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
202 XFS_MMAPLOCK_DEP(lock_flags));
205 if (lock_flags & XFS_ILOCK_EXCL)
206 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
207 else if (lock_flags & XFS_ILOCK_SHARED)
208 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 * This is just like xfs_ilock(), except that the caller
213 * is guaranteed not to sleep. It returns 1 if it gets
214 * the requested locks and 0 otherwise. If the IO lock is
215 * obtained but the inode lock cannot be, then the IO lock
216 * is dropped before returning.
218 * ip -- the inode being locked
219 * lock_flags -- this parameter indicates the inode's locks to be
220 * to be locked. See the comment for xfs_ilock() for a list
228 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
230 xfs_lock_flags_assert(lock_flags);
232 if (lock_flags & XFS_IOLOCK_EXCL) {
233 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
235 } else if (lock_flags & XFS_IOLOCK_SHARED) {
236 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
240 if (lock_flags & XFS_MMAPLOCK_EXCL) {
241 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
242 goto out_undo_iolock;
243 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
244 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
245 goto out_undo_iolock;
248 if (lock_flags & XFS_ILOCK_EXCL) {
249 if (!mrtryupdate(&ip->i_lock))
250 goto out_undo_mmaplock;
251 } else if (lock_flags & XFS_ILOCK_SHARED) {
252 if (!mrtryaccess(&ip->i_lock))
253 goto out_undo_mmaplock;
258 if (lock_flags & XFS_MMAPLOCK_EXCL)
259 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
260 else if (lock_flags & XFS_MMAPLOCK_SHARED)
261 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
263 if (lock_flags & XFS_IOLOCK_EXCL)
264 up_write(&VFS_I(ip)->i_rwsem);
265 else if (lock_flags & XFS_IOLOCK_SHARED)
266 up_read(&VFS_I(ip)->i_rwsem);
272 * xfs_iunlock() is used to drop the inode locks acquired with
273 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
274 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
275 * that we know which locks to drop.
277 * ip -- the inode being unlocked
278 * lock_flags -- this parameter indicates the inode's locks to be
279 * to be unlocked. See the comment for xfs_ilock() for a list
280 * of valid values for this parameter.
288 xfs_lock_flags_assert(lock_flags);
290 if (lock_flags & XFS_IOLOCK_EXCL)
291 up_write(&VFS_I(ip)->i_rwsem);
292 else if (lock_flags & XFS_IOLOCK_SHARED)
293 up_read(&VFS_I(ip)->i_rwsem);
295 if (lock_flags & XFS_MMAPLOCK_EXCL)
296 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
297 else if (lock_flags & XFS_MMAPLOCK_SHARED)
298 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
300 if (lock_flags & XFS_ILOCK_EXCL)
301 mrunlock_excl(&ip->i_lock);
302 else if (lock_flags & XFS_ILOCK_SHARED)
303 mrunlock_shared(&ip->i_lock);
305 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
309 * give up write locks. the i/o lock cannot be held nested
310 * if it is being demoted.
317 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
319 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
321 if (lock_flags & XFS_ILOCK_EXCL)
322 mrdemote(&ip->i_lock);
323 if (lock_flags & XFS_MMAPLOCK_EXCL)
324 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
325 if (lock_flags & XFS_IOLOCK_EXCL)
326 downgrade_write(&VFS_I(ip)->i_rwsem);
328 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
333 struct xfs_inode *ip,
336 if (lock_flags & XFS_ILOCK_SHARED)
337 rwsem_assert_held(&ip->i_lock.mr_lock);
338 else if (lock_flags & XFS_ILOCK_EXCL)
339 ASSERT(ip->i_lock.mr_writer);
341 if (lock_flags & XFS_MMAPLOCK_SHARED)
342 rwsem_assert_held(&VFS_I(ip)->i_mapping->invalidate_lock);
343 else if (lock_flags & XFS_MMAPLOCK_EXCL)
344 rwsem_assert_held_write(&VFS_I(ip)->i_mapping->invalidate_lock);
346 if (lock_flags & XFS_IOLOCK_SHARED)
347 rwsem_assert_held(&VFS_I(ip)->i_rwsem);
348 else if (lock_flags & XFS_IOLOCK_EXCL)
349 rwsem_assert_held_write(&VFS_I(ip)->i_rwsem);
353 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
354 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
355 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
356 * errors and warnings.
358 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
360 xfs_lockdep_subclass_ok(
363 return subclass < MAX_LOCKDEP_SUBCLASSES;
366 #define xfs_lockdep_subclass_ok(subclass) (true)
370 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
371 * value. This can be called for any type of inode lock combination, including
372 * parent locking. Care must be taken to ensure we don't overrun the subclass
373 * storage fields in the class mask we build.
382 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
384 ASSERT(xfs_lockdep_subclass_ok(subclass));
386 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
387 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
388 class += subclass << XFS_IOLOCK_SHIFT;
391 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
392 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
393 class += subclass << XFS_MMAPLOCK_SHIFT;
396 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
397 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
398 class += subclass << XFS_ILOCK_SHIFT;
401 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
405 * The following routine will lock n inodes in exclusive mode. We assume the
406 * caller calls us with the inodes in i_ino order.
408 * We need to detect deadlock where an inode that we lock is in the AIL and we
409 * start waiting for another inode that is locked by a thread in a long running
410 * transaction (such as truncate). This can result in deadlock since the long
411 * running trans might need to wait for the inode we just locked in order to
412 * push the tail and free space in the log.
414 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
415 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
416 * lock more than one at a time, lockdep will report false positives saying we
417 * have violated locking orders.
421 struct xfs_inode **ips,
429 struct xfs_log_item *lp;
432 * Currently supports between 2 and 5 inodes with exclusive locking. We
433 * support an arbitrary depth of locking here, but absolute limits on
434 * inodes depend on the type of locking and the limits placed by
435 * lockdep annotations in xfs_lock_inumorder. These are all checked by
438 ASSERT(ips && inodes >= 2 && inodes <= 5);
439 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
441 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
443 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
444 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
445 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
446 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
448 if (lock_mode & XFS_IOLOCK_EXCL) {
449 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
450 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
451 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
456 for (; i < inodes; i++) {
459 if (i && (ips[i] == ips[i - 1])) /* Already locked */
463 * If try_lock is not set yet, make sure all locked inodes are
464 * not in the AIL. If any are, set try_lock to be used later.
467 for (j = (i - 1); j >= 0 && !try_lock; j--) {
468 lp = &ips[j]->i_itemp->ili_item;
469 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
475 * If any of the previous locks we have locked is in the AIL,
476 * we must TRY to get the second and subsequent locks. If
477 * we can't get any, we must release all we have
481 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
485 /* try_lock means we have an inode locked that is in the AIL. */
487 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
491 * Unlock all previous guys and try again. xfs_iunlock will try
492 * to push the tail if the inode is in the AIL.
495 for (j = i - 1; j >= 0; j--) {
497 * Check to see if we've already unlocked this one. Not
498 * the first one going back, and the inode ptr is the
501 if (j != (i - 1) && ips[j] == ips[j + 1])
504 xfs_iunlock(ips[j], lock_mode);
507 if ((attempts % 5) == 0) {
508 delay(1); /* Don't just spin the CPU */
515 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
516 * mmaplock must be double-locked separately since we use i_rwsem and
517 * invalidate_lock for that. We now support taking one lock EXCL and the
522 struct xfs_inode *ip0,
524 struct xfs_inode *ip1,
528 struct xfs_log_item *lp;
530 ASSERT(hweight32(ip0_mode) == 1);
531 ASSERT(hweight32(ip1_mode) == 1);
532 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
533 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
534 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
535 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
536 ASSERT(ip0->i_ino != ip1->i_ino);
538 if (ip0->i_ino > ip1->i_ino) {
540 swap(ip0_mode, ip1_mode);
544 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
547 * If the first lock we have locked is in the AIL, we must TRY to get
548 * the second lock. If we can't get it, we must release the first one
551 lp = &ip0->i_itemp->ili_item;
552 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
553 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
554 xfs_iunlock(ip0, ip0_mode);
555 if ((++attempts % 5) == 0)
556 delay(1); /* Don't just spin the CPU */
560 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
566 struct xfs_inode *ip)
570 if (ip->i_diflags & XFS_DIFLAG_ANY) {
571 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
572 flags |= FS_XFLAG_REALTIME;
573 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
574 flags |= FS_XFLAG_PREALLOC;
575 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
576 flags |= FS_XFLAG_IMMUTABLE;
577 if (ip->i_diflags & XFS_DIFLAG_APPEND)
578 flags |= FS_XFLAG_APPEND;
579 if (ip->i_diflags & XFS_DIFLAG_SYNC)
580 flags |= FS_XFLAG_SYNC;
581 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
582 flags |= FS_XFLAG_NOATIME;
583 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
584 flags |= FS_XFLAG_NODUMP;
585 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
586 flags |= FS_XFLAG_RTINHERIT;
587 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
588 flags |= FS_XFLAG_PROJINHERIT;
589 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
590 flags |= FS_XFLAG_NOSYMLINKS;
591 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
592 flags |= FS_XFLAG_EXTSIZE;
593 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
594 flags |= FS_XFLAG_EXTSZINHERIT;
595 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
596 flags |= FS_XFLAG_NODEFRAG;
597 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
598 flags |= FS_XFLAG_FILESTREAM;
601 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
602 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
603 flags |= FS_XFLAG_DAX;
604 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
605 flags |= FS_XFLAG_COWEXTSIZE;
608 if (xfs_inode_has_attr_fork(ip))
609 flags |= FS_XFLAG_HASATTR;
614 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
615 * is allowed, otherwise it has to be an exact match. If a CI match is found,
616 * ci_name->name will point to a the actual name (caller must free) or
617 * will be set to NULL if an exact match is found.
621 struct xfs_inode *dp,
622 const struct xfs_name *name,
623 struct xfs_inode **ipp,
624 struct xfs_name *ci_name)
629 trace_xfs_lookup(dp, name);
631 if (xfs_is_shutdown(dp->i_mount))
633 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
636 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
640 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
648 kfree(ci_name->name);
654 /* Propagate di_flags from a parent inode to a child inode. */
656 xfs_inode_inherit_flags(
657 struct xfs_inode *ip,
658 const struct xfs_inode *pip)
660 unsigned int di_flags = 0;
661 xfs_failaddr_t failaddr;
662 umode_t mode = VFS_I(ip)->i_mode;
665 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
666 di_flags |= XFS_DIFLAG_RTINHERIT;
667 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
668 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
669 ip->i_extsize = pip->i_extsize;
671 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
672 di_flags |= XFS_DIFLAG_PROJINHERIT;
673 } else if (S_ISREG(mode)) {
674 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
675 xfs_has_realtime(ip->i_mount))
676 di_flags |= XFS_DIFLAG_REALTIME;
677 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
678 di_flags |= XFS_DIFLAG_EXTSIZE;
679 ip->i_extsize = pip->i_extsize;
682 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
684 di_flags |= XFS_DIFLAG_NOATIME;
685 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
687 di_flags |= XFS_DIFLAG_NODUMP;
688 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
690 di_flags |= XFS_DIFLAG_SYNC;
691 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
692 xfs_inherit_nosymlinks)
693 di_flags |= XFS_DIFLAG_NOSYMLINKS;
694 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
695 xfs_inherit_nodefrag)
696 di_flags |= XFS_DIFLAG_NODEFRAG;
697 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
698 di_flags |= XFS_DIFLAG_FILESTREAM;
700 ip->i_diflags |= di_flags;
703 * Inode verifiers on older kernels only check that the extent size
704 * hint is an integer multiple of the rt extent size on realtime files.
705 * They did not check the hint alignment on a directory with both
706 * rtinherit and extszinherit flags set. If the misaligned hint is
707 * propagated from a directory into a new realtime file, new file
708 * allocations will fail due to math errors in the rt allocator and/or
709 * trip the verifiers. Validate the hint settings in the new file so
710 * that we don't let broken hints propagate.
712 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
713 VFS_I(ip)->i_mode, ip->i_diflags);
715 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
716 XFS_DIFLAG_EXTSZINHERIT);
721 /* Propagate di_flags2 from a parent inode to a child inode. */
723 xfs_inode_inherit_flags2(
724 struct xfs_inode *ip,
725 const struct xfs_inode *pip)
727 xfs_failaddr_t failaddr;
729 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
730 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
731 ip->i_cowextsize = pip->i_cowextsize;
733 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
734 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
736 /* Don't let invalid cowextsize hints propagate. */
737 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
738 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
740 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
741 ip->i_cowextsize = 0;
746 * Initialise a newly allocated inode and return the in-core inode to the
747 * caller locked exclusively.
751 struct mnt_idmap *idmap,
752 struct xfs_trans *tp,
753 struct xfs_inode *pip,
760 struct xfs_inode **ipp)
762 struct inode *dir = pip ? VFS_I(pip) : NULL;
763 struct xfs_mount *mp = tp->t_mountp;
764 struct xfs_inode *ip;
767 struct timespec64 tv;
771 * Protect against obviously corrupt allocation btree records. Later
772 * xfs_iget checks will catch re-allocation of other active in-memory
773 * and on-disk inodes. If we don't catch reallocating the parent inode
774 * here we will deadlock in xfs_iget() so we have to do these checks
777 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
778 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
779 return -EFSCORRUPTED;
783 * Get the in-core inode with the lock held exclusively to prevent
784 * others from looking at until we're done.
786 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
792 set_nlink(inode, nlink);
793 inode->i_rdev = rdev;
796 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
797 inode_fsuid_set(inode, idmap);
798 inode->i_gid = dir->i_gid;
799 inode->i_mode = mode;
801 inode_init_owner(idmap, inode, dir, mode);
805 * If the group ID of the new file does not match the effective group
806 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
807 * (and only if the irix_sgid_inherit compatibility variable is set).
809 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
810 !vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode)))
811 inode->i_mode &= ~S_ISGID;
814 ip->i_df.if_nextents = 0;
815 ASSERT(ip->i_nblocks == 0);
817 tv = inode_set_ctime_current(inode);
818 inode_set_mtime_to_ts(inode, tv);
819 inode_set_atime_to_ts(inode, tv);
824 if (xfs_has_v3inodes(mp)) {
825 inode_set_iversion(inode, 1);
826 ip->i_cowextsize = 0;
830 flags = XFS_ILOG_CORE;
831 switch (mode & S_IFMT) {
836 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
837 flags |= XFS_ILOG_DEV;
841 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
842 xfs_inode_inherit_flags(ip, pip);
843 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
844 xfs_inode_inherit_flags2(ip, pip);
847 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
848 ip->i_df.if_bytes = 0;
849 ip->i_df.if_data = NULL;
856 * If we need to create attributes immediately after allocating the
857 * inode, initialise an empty attribute fork right now. We use the
858 * default fork offset for attributes here as we don't know exactly what
859 * size or how many attributes we might be adding. We can do this
860 * safely here because we know the data fork is completely empty and
861 * this saves us from needing to run a separate transaction to set the
862 * fork offset in the immediate future.
864 if (init_xattrs && xfs_has_attr(mp)) {
865 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
866 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
870 * Log the new values stuffed into the inode.
872 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
873 xfs_trans_log_inode(tp, ip, flags);
875 /* now that we have an i_mode we can setup the inode structure */
883 * Decrement the link count on an inode & log the change. If this causes the
884 * link count to go to zero, move the inode to AGI unlinked list so that it can
885 * be freed when the last active reference goes away via xfs_inactive().
887 static int /* error */
892 if (VFS_I(ip)->i_nlink == 0) {
893 xfs_alert(ip->i_mount,
894 "%s: Attempt to drop inode (%llu) with nlink zero.",
895 __func__, ip->i_ino);
896 return -EFSCORRUPTED;
899 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
901 drop_nlink(VFS_I(ip));
902 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
904 if (VFS_I(ip)->i_nlink)
907 return xfs_iunlink(tp, ip);
911 * Increment the link count on an inode & log the change.
918 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
920 inc_nlink(VFS_I(ip));
921 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
926 struct mnt_idmap *idmap,
928 struct xfs_name *name,
934 int is_dir = S_ISDIR(mode);
935 struct xfs_mount *mp = dp->i_mount;
936 struct xfs_inode *ip = NULL;
937 struct xfs_trans *tp = NULL;
939 bool unlock_dp_on_error = false;
941 struct xfs_dquot *udqp = NULL;
942 struct xfs_dquot *gdqp = NULL;
943 struct xfs_dquot *pdqp = NULL;
944 struct xfs_trans_res *tres;
948 trace_xfs_create(dp, name);
950 if (xfs_is_shutdown(mp))
952 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
955 prid = xfs_get_initial_prid(dp);
958 * Make sure that we have allocated dquot(s) on disk.
960 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
961 mapped_fsgid(idmap, &init_user_ns), prid,
962 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
963 &udqp, &gdqp, &pdqp);
968 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
969 tres = &M_RES(mp)->tr_mkdir;
971 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
972 tres = &M_RES(mp)->tr_create;
976 * Initially assume that the file does not exist and
977 * reserve the resources for that case. If that is not
978 * the case we'll drop the one we have and get a more
979 * appropriate transaction later.
981 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
983 if (error == -ENOSPC) {
984 /* flush outstanding delalloc blocks and retry */
985 xfs_flush_inodes(mp);
986 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
990 goto out_release_dquots;
992 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
993 unlock_dp_on_error = true;
996 * A newly created regular or special file just has one directory
997 * entry pointing to them, but a directory also the "." entry
998 * pointing to itself.
1000 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1002 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1003 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1005 goto out_trans_cancel;
1008 * Now we join the directory inode to the transaction. We do not do it
1009 * earlier because xfs_dialloc might commit the previous transaction
1010 * (and release all the locks). An error from here on will result in
1011 * the transaction cancel unlocking dp so don't do it explicitly in the
1014 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1015 unlock_dp_on_error = false;
1017 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1018 resblks - XFS_IALLOC_SPACE_RES(mp));
1020 ASSERT(error != -ENOSPC);
1021 goto out_trans_cancel;
1023 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1024 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1027 error = xfs_dir_init(tp, ip, dp);
1029 goto out_trans_cancel;
1031 xfs_bumplink(tp, dp);
1035 * If this is a synchronous mount, make sure that the
1036 * create transaction goes to disk before returning to
1039 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1040 xfs_trans_set_sync(tp);
1043 * Attach the dquot(s) to the inodes and modify them incore.
1044 * These ids of the inode couldn't have changed since the new
1045 * inode has been locked ever since it was created.
1047 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1049 error = xfs_trans_commit(tp);
1051 goto out_release_inode;
1053 xfs_qm_dqrele(udqp);
1054 xfs_qm_dqrele(gdqp);
1055 xfs_qm_dqrele(pdqp);
1061 xfs_trans_cancel(tp);
1064 * Wait until after the current transaction is aborted to finish the
1065 * setup of the inode and release the inode. This prevents recursive
1066 * transactions and deadlocks from xfs_inactive.
1069 xfs_finish_inode_setup(ip);
1073 xfs_qm_dqrele(udqp);
1074 xfs_qm_dqrele(gdqp);
1075 xfs_qm_dqrele(pdqp);
1077 if (unlock_dp_on_error)
1078 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1084 struct mnt_idmap *idmap,
1085 struct xfs_inode *dp,
1087 struct xfs_inode **ipp)
1089 struct xfs_mount *mp = dp->i_mount;
1090 struct xfs_inode *ip = NULL;
1091 struct xfs_trans *tp = NULL;
1094 struct xfs_dquot *udqp = NULL;
1095 struct xfs_dquot *gdqp = NULL;
1096 struct xfs_dquot *pdqp = NULL;
1097 struct xfs_trans_res *tres;
1101 if (xfs_is_shutdown(mp))
1104 prid = xfs_get_initial_prid(dp);
1107 * Make sure that we have allocated dquot(s) on disk.
1109 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
1110 mapped_fsgid(idmap, &init_user_ns), prid,
1111 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1112 &udqp, &gdqp, &pdqp);
1116 resblks = XFS_IALLOC_SPACE_RES(mp);
1117 tres = &M_RES(mp)->tr_create_tmpfile;
1119 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1122 goto out_release_dquots;
1124 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1126 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1127 0, 0, prid, false, &ip);
1129 goto out_trans_cancel;
1131 if (xfs_has_wsync(mp))
1132 xfs_trans_set_sync(tp);
1135 * Attach the dquot(s) to the inodes and modify them incore.
1136 * These ids of the inode couldn't have changed since the new
1137 * inode has been locked ever since it was created.
1139 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1141 error = xfs_iunlink(tp, ip);
1143 goto out_trans_cancel;
1145 error = xfs_trans_commit(tp);
1147 goto out_release_inode;
1149 xfs_qm_dqrele(udqp);
1150 xfs_qm_dqrele(gdqp);
1151 xfs_qm_dqrele(pdqp);
1157 xfs_trans_cancel(tp);
1160 * Wait until after the current transaction is aborted to finish the
1161 * setup of the inode and release the inode. This prevents recursive
1162 * transactions and deadlocks from xfs_inactive.
1165 xfs_finish_inode_setup(ip);
1169 xfs_qm_dqrele(udqp);
1170 xfs_qm_dqrele(gdqp);
1171 xfs_qm_dqrele(pdqp);
1180 struct xfs_name *target_name)
1182 xfs_mount_t *mp = tdp->i_mount;
1184 int error, nospace_error = 0;
1187 trace_xfs_link(tdp, target_name);
1189 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1191 if (xfs_is_shutdown(mp))
1193 if (xfs_ifork_zapped(tdp, XFS_DATA_FORK))
1196 error = xfs_qm_dqattach(sip);
1200 error = xfs_qm_dqattach(tdp);
1204 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1205 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1206 &tp, &nospace_error);
1211 * If we are using project inheritance, we only allow hard link
1212 * creation in our tree when the project IDs are the same; else
1213 * the tree quota mechanism could be circumvented.
1215 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1216 tdp->i_projid != sip->i_projid)) {
1222 error = xfs_dir_canenter(tp, tdp, target_name);
1228 * Handle initial link state of O_TMPFILE inode
1230 if (VFS_I(sip)->i_nlink == 0) {
1231 struct xfs_perag *pag;
1233 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1234 error = xfs_iunlink_remove(tp, pag, sip);
1240 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1244 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1245 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1247 xfs_bumplink(tp, sip);
1250 * If this is a synchronous mount, make sure that the
1251 * link transaction goes to disk before returning to
1254 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1255 xfs_trans_set_sync(tp);
1257 return xfs_trans_commit(tp);
1260 xfs_trans_cancel(tp);
1262 if (error == -ENOSPC && nospace_error)
1263 error = nospace_error;
1267 /* Clear the reflink flag and the cowblocks tag if possible. */
1269 xfs_itruncate_clear_reflink_flags(
1270 struct xfs_inode *ip)
1272 struct xfs_ifork *dfork;
1273 struct xfs_ifork *cfork;
1275 if (!xfs_is_reflink_inode(ip))
1277 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1278 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1279 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1280 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1281 if (cfork->if_bytes == 0)
1282 xfs_inode_clear_cowblocks_tag(ip);
1286 * Free up the underlying blocks past new_size. The new size must be smaller
1287 * than the current size. This routine can be used both for the attribute and
1288 * data fork, and does not modify the inode size, which is left to the caller.
1290 * The transaction passed to this routine must have made a permanent log
1291 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1292 * given transaction and start new ones, so make sure everything involved in
1293 * the transaction is tidy before calling here. Some transaction will be
1294 * returned to the caller to be committed. The incoming transaction must
1295 * already include the inode, and both inode locks must be held exclusively.
1296 * The inode must also be "held" within the transaction. On return the inode
1297 * will be "held" within the returned transaction. This routine does NOT
1298 * require any disk space to be reserved for it within the transaction.
1300 * If we get an error, we must return with the inode locked and linked into the
1301 * current transaction. This keeps things simple for the higher level code,
1302 * because it always knows that the inode is locked and held in the transaction
1303 * that returns to it whether errors occur or not. We don't mark the inode
1304 * dirty on error so that transactions can be easily aborted if possible.
1307 xfs_itruncate_extents_flags(
1308 struct xfs_trans **tpp,
1309 struct xfs_inode *ip,
1311 xfs_fsize_t new_size,
1314 struct xfs_mount *mp = ip->i_mount;
1315 struct xfs_trans *tp = *tpp;
1316 xfs_fileoff_t first_unmap_block;
1319 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1320 if (atomic_read(&VFS_I(ip)->i_count))
1321 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
1322 ASSERT(new_size <= XFS_ISIZE(ip));
1323 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1324 ASSERT(ip->i_itemp != NULL);
1325 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1326 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1328 trace_xfs_itruncate_extents_start(ip, new_size);
1330 flags |= xfs_bmapi_aflag(whichfork);
1333 * Since it is possible for space to become allocated beyond
1334 * the end of the file (in a crash where the space is allocated
1335 * but the inode size is not yet updated), simply remove any
1336 * blocks which show up between the new EOF and the maximum
1337 * possible file size.
1339 * We have to free all the blocks to the bmbt maximum offset, even if
1340 * the page cache can't scale that far.
1342 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1343 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1344 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1348 error = xfs_bunmapi_range(&tp, ip, flags, first_unmap_block,
1353 if (whichfork == XFS_DATA_FORK) {
1354 /* Remove all pending CoW reservations. */
1355 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1356 first_unmap_block, XFS_MAX_FILEOFF, true);
1360 xfs_itruncate_clear_reflink_flags(ip);
1364 * Always re-log the inode so that our permanent transaction can keep
1365 * on rolling it forward in the log.
1367 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1369 trace_xfs_itruncate_extents_end(ip, new_size);
1380 xfs_mount_t *mp = ip->i_mount;
1383 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1386 /* If this is a read-only mount, don't do this (would generate I/O) */
1387 if (xfs_is_readonly(mp))
1390 if (!xfs_is_shutdown(mp)) {
1394 * If we previously truncated this file and removed old data
1395 * in the process, we want to initiate "early" writeout on
1396 * the last close. This is an attempt to combat the notorious
1397 * NULL files problem which is particularly noticeable from a
1398 * truncate down, buffered (re-)write (delalloc), followed by
1399 * a crash. What we are effectively doing here is
1400 * significantly reducing the time window where we'd otherwise
1401 * be exposed to that problem.
1403 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1405 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1406 if (ip->i_delayed_blks > 0) {
1407 error = filemap_flush(VFS_I(ip)->i_mapping);
1414 if (VFS_I(ip)->i_nlink == 0)
1418 * If we can't get the iolock just skip truncating the blocks past EOF
1419 * because we could deadlock with the mmap_lock otherwise. We'll get
1420 * another chance to drop them once the last reference to the inode is
1421 * dropped, so we'll never leak blocks permanently.
1423 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1426 if (xfs_can_free_eofblocks(ip, false)) {
1428 * Check if the inode is being opened, written and closed
1429 * frequently and we have delayed allocation blocks outstanding
1430 * (e.g. streaming writes from the NFS server), truncating the
1431 * blocks past EOF will cause fragmentation to occur.
1433 * In this case don't do the truncation, but we have to be
1434 * careful how we detect this case. Blocks beyond EOF show up as
1435 * i_delayed_blks even when the inode is clean, so we need to
1436 * truncate them away first before checking for a dirty release.
1437 * Hence on the first dirty close we will still remove the
1438 * speculative allocation, but after that we will leave it in
1441 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1444 error = xfs_free_eofblocks(ip);
1448 /* delalloc blocks after truncation means it really is dirty */
1449 if (ip->i_delayed_blks)
1450 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1454 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1459 * xfs_inactive_truncate
1461 * Called to perform a truncate when an inode becomes unlinked.
1464 xfs_inactive_truncate(
1465 struct xfs_inode *ip)
1467 struct xfs_mount *mp = ip->i_mount;
1468 struct xfs_trans *tp;
1471 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1473 ASSERT(xfs_is_shutdown(mp));
1476 xfs_ilock(ip, XFS_ILOCK_EXCL);
1477 xfs_trans_ijoin(tp, ip, 0);
1480 * Log the inode size first to prevent stale data exposure in the event
1481 * of a system crash before the truncate completes. See the related
1482 * comment in xfs_vn_setattr_size() for details.
1484 ip->i_disk_size = 0;
1485 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1487 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1489 goto error_trans_cancel;
1491 ASSERT(ip->i_df.if_nextents == 0);
1493 error = xfs_trans_commit(tp);
1497 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1501 xfs_trans_cancel(tp);
1503 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1508 * xfs_inactive_ifree()
1510 * Perform the inode free when an inode is unlinked.
1514 struct xfs_inode *ip)
1516 struct xfs_mount *mp = ip->i_mount;
1517 struct xfs_trans *tp;
1521 * We try to use a per-AG reservation for any block needed by the finobt
1522 * tree, but as the finobt feature predates the per-AG reservation
1523 * support a degraded file system might not have enough space for the
1524 * reservation at mount time. In that case try to dip into the reserved
1527 * Send a warning if the reservation does happen to fail, as the inode
1528 * now remains allocated and sits on the unlinked list until the fs is
1531 if (unlikely(mp->m_finobt_nores)) {
1532 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1533 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1536 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1539 if (error == -ENOSPC) {
1540 xfs_warn_ratelimited(mp,
1541 "Failed to remove inode(s) from unlinked list. "
1542 "Please free space, unmount and run xfs_repair.");
1544 ASSERT(xfs_is_shutdown(mp));
1550 * We do not hold the inode locked across the entire rolling transaction
1551 * here. We only need to hold it for the first transaction that
1552 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1553 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1554 * here breaks the relationship between cluster buffer invalidation and
1555 * stale inode invalidation on cluster buffer item journal commit
1556 * completion, and can result in leaving dirty stale inodes hanging
1559 * We have no need for serialising this inode operation against other
1560 * operations - we freed the inode and hence reallocation is required
1561 * and that will serialise on reallocating the space the deferops need
1562 * to free. Hence we can unlock the inode on the first commit of
1563 * the transaction rather than roll it right through the deferops. This
1564 * avoids relogging the XFS_ISTALE inode.
1566 * We check that xfs_ifree() hasn't grown an internal transaction roll
1567 * by asserting that the inode is still locked when it returns.
1569 xfs_ilock(ip, XFS_ILOCK_EXCL);
1570 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1572 error = xfs_ifree(tp, ip);
1573 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1576 * If we fail to free the inode, shut down. The cancel
1577 * might do that, we need to make sure. Otherwise the
1578 * inode might be lost for a long time or forever.
1580 if (!xfs_is_shutdown(mp)) {
1581 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1583 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1585 xfs_trans_cancel(tp);
1590 * Credit the quota account(s). The inode is gone.
1592 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1594 return xfs_trans_commit(tp);
1598 * Returns true if we need to update the on-disk metadata before we can free
1599 * the memory used by this inode. Updates include freeing post-eof
1600 * preallocations; freeing COW staging extents; and marking the inode free in
1601 * the inobt if it is on the unlinked list.
1604 xfs_inode_needs_inactive(
1605 struct xfs_inode *ip)
1607 struct xfs_mount *mp = ip->i_mount;
1608 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1611 * If the inode is already free, then there can be nothing
1614 if (VFS_I(ip)->i_mode == 0)
1618 * If this is a read-only mount, don't do this (would generate I/O)
1619 * unless we're in log recovery and cleaning the iunlinked list.
1621 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1624 /* If the log isn't running, push inodes straight to reclaim. */
1625 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1628 /* Metadata inodes require explicit resource cleanup. */
1629 if (xfs_is_metadata_inode(ip))
1632 /* Want to clean out the cow blocks if there are any. */
1633 if (cow_ifp && cow_ifp->if_bytes > 0)
1636 /* Unlinked files must be freed. */
1637 if (VFS_I(ip)->i_nlink == 0)
1641 * This file isn't being freed, so check if there are post-eof blocks
1642 * to free. @force is true because we are evicting an inode from the
1643 * cache. Post-eof blocks must be freed, lest we end up with broken
1644 * free space accounting.
1646 * Note: don't bother with iolock here since lockdep complains about
1647 * acquiring it in reclaim context. We have the only reference to the
1648 * inode at this point anyways.
1650 return xfs_can_free_eofblocks(ip, true);
1656 * This is called when the vnode reference count for the vnode
1657 * goes to zero. If the file has been unlinked, then it must
1658 * now be truncated. Also, we clear all of the read-ahead state
1659 * kept for the inode here since the file is now closed.
1665 struct xfs_mount *mp;
1670 * If the inode is already free, then there can be nothing
1673 if (VFS_I(ip)->i_mode == 0) {
1674 ASSERT(ip->i_df.if_broot_bytes == 0);
1679 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1682 * If this is a read-only mount, don't do this (would generate I/O)
1683 * unless we're in log recovery and cleaning the iunlinked list.
1685 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1688 /* Metadata inodes require explicit resource cleanup. */
1689 if (xfs_is_metadata_inode(ip))
1692 /* Try to clean out the cow blocks if there are any. */
1693 if (xfs_inode_has_cow_data(ip))
1694 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1696 if (VFS_I(ip)->i_nlink != 0) {
1698 * force is true because we are evicting an inode from the
1699 * cache. Post-eof blocks must be freed, lest we end up with
1700 * broken free space accounting.
1702 * Note: don't bother with iolock here since lockdep complains
1703 * about acquiring it in reclaim context. We have the only
1704 * reference to the inode at this point anyways.
1706 if (xfs_can_free_eofblocks(ip, true))
1707 error = xfs_free_eofblocks(ip);
1712 if (S_ISREG(VFS_I(ip)->i_mode) &&
1713 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1714 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1717 if (xfs_iflags_test(ip, XFS_IQUOTAUNCHECKED)) {
1719 * If this inode is being inactivated during a quotacheck and
1720 * has not yet been scanned by quotacheck, we /must/ remove
1721 * the dquots from the inode before inactivation changes the
1722 * block and inode counts. Most probably this is a result of
1723 * reloading the incore iunlinked list to purge unrecovered
1726 xfs_qm_dqdetach(ip);
1728 error = xfs_qm_dqattach(ip);
1733 if (S_ISLNK(VFS_I(ip)->i_mode))
1734 error = xfs_inactive_symlink(ip);
1736 error = xfs_inactive_truncate(ip);
1741 * If there are attributes associated with the file then blow them away
1742 * now. The code calls a routine that recursively deconstructs the
1743 * attribute fork. If also blows away the in-core attribute fork.
1745 if (xfs_inode_has_attr_fork(ip)) {
1746 error = xfs_attr_inactive(ip);
1751 ASSERT(ip->i_forkoff == 0);
1756 error = xfs_inactive_ifree(ip);
1760 * We're done making metadata updates for this inode, so we can release
1761 * the attached dquots.
1763 xfs_qm_dqdetach(ip);
1768 * In-Core Unlinked List Lookups
1769 * =============================
1771 * Every inode is supposed to be reachable from some other piece of metadata
1772 * with the exception of the root directory. Inodes with a connection to a
1773 * file descriptor but not linked from anywhere in the on-disk directory tree
1774 * are collectively known as unlinked inodes, though the filesystem itself
1775 * maintains links to these inodes so that on-disk metadata are consistent.
1777 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1778 * header contains a number of buckets that point to an inode, and each inode
1779 * record has a pointer to the next inode in the hash chain. This
1780 * singly-linked list causes scaling problems in the iunlink remove function
1781 * because we must walk that list to find the inode that points to the inode
1782 * being removed from the unlinked hash bucket list.
1784 * Hence we keep an in-memory double linked list to link each inode on an
1785 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1786 * based lists would require having 64 list heads in the perag, one for each
1787 * list. This is expensive in terms of memory (think millions of AGs) and cache
1788 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1789 * must be referenced at the VFS level to keep them on the list and hence we
1790 * have an existence guarantee for inodes on the unlinked list.
1792 * Given we have an existence guarantee, we can use lockless inode cache lookups
1793 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1794 * for the double linked unlinked list, and we don't need any extra locking to
1795 * keep the list safe as all manipulations are done under the AGI buffer lock.
1796 * Keeping the list up to date does not require memory allocation, just finding
1797 * the XFS inode and updating the next/prev unlinked list aginos.
1801 * Find an inode on the unlinked list. This does not take references to the
1802 * inode as we have existence guarantees by holding the AGI buffer lock and that
1803 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1804 * don't find the inode in cache, then let the caller handle the situation.
1806 static struct xfs_inode *
1808 struct xfs_perag *pag,
1811 struct xfs_inode *ip;
1814 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1816 /* Caller can handle inode not being in memory. */
1822 * Inode in RCU freeing limbo should not happen. Warn about this and
1823 * let the caller handle the failure.
1825 if (WARN_ON_ONCE(!ip->i_ino)) {
1829 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1835 * Update the prev pointer of the next agino. Returns -ENOLINK if the inode
1839 xfs_iunlink_update_backref(
1840 struct xfs_perag *pag,
1841 xfs_agino_t prev_agino,
1842 xfs_agino_t next_agino)
1844 struct xfs_inode *ip;
1846 /* No update necessary if we are at the end of the list. */
1847 if (next_agino == NULLAGINO)
1850 ip = xfs_iunlink_lookup(pag, next_agino);
1854 ip->i_prev_unlinked = prev_agino;
1859 * Point the AGI unlinked bucket at an inode and log the results. The caller
1860 * is responsible for validating the old value.
1863 xfs_iunlink_update_bucket(
1864 struct xfs_trans *tp,
1865 struct xfs_perag *pag,
1866 struct xfs_buf *agibp,
1867 unsigned int bucket_index,
1868 xfs_agino_t new_agino)
1870 struct xfs_agi *agi = agibp->b_addr;
1871 xfs_agino_t old_value;
1874 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
1876 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1877 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1878 old_value, new_agino);
1881 * We should never find the head of the list already set to the value
1882 * passed in because either we're adding or removing ourselves from the
1885 if (old_value == new_agino) {
1886 xfs_buf_mark_corrupt(agibp);
1887 return -EFSCORRUPTED;
1890 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
1891 offset = offsetof(struct xfs_agi, agi_unlinked) +
1892 (sizeof(xfs_agino_t) * bucket_index);
1893 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
1898 * Load the inode @next_agino into the cache and set its prev_unlinked pointer
1899 * to @prev_agino. Caller must hold the AGI to synchronize with other changes
1900 * to the unlinked list.
1903 xfs_iunlink_reload_next(
1904 struct xfs_trans *tp,
1905 struct xfs_buf *agibp,
1906 xfs_agino_t prev_agino,
1907 xfs_agino_t next_agino)
1909 struct xfs_perag *pag = agibp->b_pag;
1910 struct xfs_mount *mp = pag->pag_mount;
1911 struct xfs_inode *next_ip = NULL;
1915 ASSERT(next_agino != NULLAGINO);
1919 next_ip = radix_tree_lookup(&pag->pag_ici_root, next_agino);
1920 ASSERT(next_ip == NULL);
1924 xfs_info_ratelimited(mp,
1925 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating recovery.",
1926 next_agino, pag->pag_agno);
1929 * Use an untrusted lookup just to be cautious in case the AGI has been
1930 * corrupted and now points at a free inode. That shouldn't happen,
1931 * but we'd rather shut down now since we're already running in a weird
1934 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, next_agino);
1935 error = xfs_iget(mp, tp, ino, XFS_IGET_UNTRUSTED, 0, &next_ip);
1939 /* If this is not an unlinked inode, something is very wrong. */
1940 if (VFS_I(next_ip)->i_nlink != 0) {
1941 error = -EFSCORRUPTED;
1945 next_ip->i_prev_unlinked = prev_agino;
1946 trace_xfs_iunlink_reload_next(next_ip);
1948 ASSERT(!(VFS_I(next_ip)->i_state & I_DONTCACHE));
1949 if (xfs_is_quotacheck_running(mp) && next_ip)
1950 xfs_iflags_set(next_ip, XFS_IQUOTAUNCHECKED);
1956 xfs_iunlink_insert_inode(
1957 struct xfs_trans *tp,
1958 struct xfs_perag *pag,
1959 struct xfs_buf *agibp,
1960 struct xfs_inode *ip)
1962 struct xfs_mount *mp = tp->t_mountp;
1963 struct xfs_agi *agi = agibp->b_addr;
1964 xfs_agino_t next_agino;
1965 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1966 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1970 * Get the index into the agi hash table for the list this inode will
1971 * go on. Make sure the pointer isn't garbage and that this inode
1972 * isn't already on the list.
1974 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1975 if (next_agino == agino ||
1976 !xfs_verify_agino_or_null(pag, next_agino)) {
1977 xfs_buf_mark_corrupt(agibp);
1978 return -EFSCORRUPTED;
1982 * Update the prev pointer in the next inode to point back to this
1985 error = xfs_iunlink_update_backref(pag, agino, next_agino);
1986 if (error == -ENOLINK)
1987 error = xfs_iunlink_reload_next(tp, agibp, agino, next_agino);
1991 if (next_agino != NULLAGINO) {
1993 * There is already another inode in the bucket, so point this
1994 * inode to the current head of the list.
1996 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
1999 ip->i_next_unlinked = next_agino;
2002 /* Point the head of the list to point to this inode. */
2003 ip->i_prev_unlinked = NULLAGINO;
2004 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
2008 * This is called when the inode's link count has gone to 0 or we are creating
2009 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2011 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2012 * list when the inode is freed.
2016 struct xfs_trans *tp,
2017 struct xfs_inode *ip)
2019 struct xfs_mount *mp = tp->t_mountp;
2020 struct xfs_perag *pag;
2021 struct xfs_buf *agibp;
2024 ASSERT(VFS_I(ip)->i_nlink == 0);
2025 ASSERT(VFS_I(ip)->i_mode != 0);
2026 trace_xfs_iunlink(ip);
2028 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2030 /* Get the agi buffer first. It ensures lock ordering on the list. */
2031 error = xfs_read_agi(pag, tp, &agibp);
2035 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
2042 xfs_iunlink_remove_inode(
2043 struct xfs_trans *tp,
2044 struct xfs_perag *pag,
2045 struct xfs_buf *agibp,
2046 struct xfs_inode *ip)
2048 struct xfs_mount *mp = tp->t_mountp;
2049 struct xfs_agi *agi = agibp->b_addr;
2050 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2051 xfs_agino_t head_agino;
2052 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2055 trace_xfs_iunlink_remove(ip);
2058 * Get the index into the agi hash table for the list this inode will
2059 * go on. Make sure the head pointer isn't garbage.
2061 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2062 if (!xfs_verify_agino(pag, head_agino)) {
2063 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2065 return -EFSCORRUPTED;
2069 * Set our inode's next_unlinked pointer to NULL and then return
2070 * the old pointer value so that we can update whatever was previous
2071 * to us in the list to point to whatever was next in the list.
2073 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2078 * Update the prev pointer in the next inode to point back to previous
2079 * inode in the chain.
2081 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2082 ip->i_next_unlinked);
2083 if (error == -ENOLINK)
2084 error = xfs_iunlink_reload_next(tp, agibp, ip->i_prev_unlinked,
2085 ip->i_next_unlinked);
2089 if (head_agino != agino) {
2090 struct xfs_inode *prev_ip;
2092 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2094 return -EFSCORRUPTED;
2096 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2097 ip->i_next_unlinked);
2098 prev_ip->i_next_unlinked = ip->i_next_unlinked;
2100 /* Point the head of the list to the next unlinked inode. */
2101 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2102 ip->i_next_unlinked);
2105 ip->i_next_unlinked = NULLAGINO;
2106 ip->i_prev_unlinked = 0;
2111 * Pull the on-disk inode from the AGI unlinked list.
2115 struct xfs_trans *tp,
2116 struct xfs_perag *pag,
2117 struct xfs_inode *ip)
2119 struct xfs_buf *agibp;
2122 trace_xfs_iunlink_remove(ip);
2124 /* Get the agi buffer first. It ensures lock ordering on the list. */
2125 error = xfs_read_agi(pag, tp, &agibp);
2129 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2133 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2134 * mark it stale. We should only find clean inodes in this lookup that aren't
2138 xfs_ifree_mark_inode_stale(
2139 struct xfs_perag *pag,
2140 struct xfs_inode *free_ip,
2143 struct xfs_mount *mp = pag->pag_mount;
2144 struct xfs_inode_log_item *iip;
2145 struct xfs_inode *ip;
2149 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2151 /* Inode not in memory, nothing to do */
2158 * because this is an RCU protected lookup, we could find a recently
2159 * freed or even reallocated inode during the lookup. We need to check
2160 * under the i_flags_lock for a valid inode here. Skip it if it is not
2161 * valid, the wrong inode or stale.
2163 spin_lock(&ip->i_flags_lock);
2164 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2165 goto out_iflags_unlock;
2168 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2169 * other inodes that we did not find in the list attached to the buffer
2170 * and are not already marked stale. If we can't lock it, back off and
2173 if (ip != free_ip) {
2174 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2175 spin_unlock(&ip->i_flags_lock);
2181 ip->i_flags |= XFS_ISTALE;
2184 * If the inode is flushing, it is already attached to the buffer. All
2185 * we needed to do here is mark the inode stale so buffer IO completion
2186 * will remove it from the AIL.
2189 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2190 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2191 ASSERT(iip->ili_last_fields);
2196 * Inodes not attached to the buffer can be released immediately.
2197 * Everything else has to go through xfs_iflush_abort() on journal
2198 * commit as the flock synchronises removal of the inode from the
2199 * cluster buffer against inode reclaim.
2201 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2204 __xfs_iflags_set(ip, XFS_IFLUSHING);
2205 spin_unlock(&ip->i_flags_lock);
2208 /* we have a dirty inode in memory that has not yet been flushed. */
2209 spin_lock(&iip->ili_lock);
2210 iip->ili_last_fields = iip->ili_fields;
2211 iip->ili_fields = 0;
2212 iip->ili_fsync_fields = 0;
2213 spin_unlock(&iip->ili_lock);
2214 ASSERT(iip->ili_last_fields);
2217 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2222 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2224 spin_unlock(&ip->i_flags_lock);
2229 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2230 * inodes that are in memory - they all must be marked stale and attached to
2231 * the cluster buffer.
2235 struct xfs_trans *tp,
2236 struct xfs_perag *pag,
2237 struct xfs_inode *free_ip,
2238 struct xfs_icluster *xic)
2240 struct xfs_mount *mp = free_ip->i_mount;
2241 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2244 xfs_ino_t inum = xic->first_ino;
2250 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2252 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2254 * The allocation bitmap tells us which inodes of the chunk were
2255 * physically allocated. Skip the cluster if an inode falls into
2258 ioffset = inum - xic->first_ino;
2259 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2260 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2264 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2265 XFS_INO_TO_AGBNO(mp, inum));
2268 * We obtain and lock the backing buffer first in the process
2269 * here to ensure dirty inodes attached to the buffer remain in
2270 * the flushing state while we mark them stale.
2272 * If we scan the in-memory inodes first, then buffer IO can
2273 * complete before we get a lock on it, and hence we may fail
2274 * to mark all the active inodes on the buffer stale.
2276 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2277 mp->m_bsize * igeo->blocks_per_cluster,
2283 * This buffer may not have been correctly initialised as we
2284 * didn't read it from disk. That's not important because we are
2285 * only using to mark the buffer as stale in the log, and to
2286 * attach stale cached inodes on it. That means it will never be
2287 * dispatched for IO. If it is, we want to know about it, and we
2288 * want it to fail. We can acheive this by adding a write
2289 * verifier to the buffer.
2291 bp->b_ops = &xfs_inode_buf_ops;
2294 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2295 * too. This requires lookups, and will skip inodes that we've
2296 * already marked XFS_ISTALE.
2298 for (i = 0; i < igeo->inodes_per_cluster; i++)
2299 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2301 xfs_trans_stale_inode_buf(tp, bp);
2302 xfs_trans_binval(tp, bp);
2308 * This is called to return an inode to the inode free list. The inode should
2309 * already be truncated to 0 length and have no pages associated with it. This
2310 * routine also assumes that the inode is already a part of the transaction.
2312 * The on-disk copy of the inode will have been added to the list of unlinked
2313 * inodes in the AGI. We need to remove the inode from that list atomically with
2314 * respect to freeing it here.
2318 struct xfs_trans *tp,
2319 struct xfs_inode *ip)
2321 struct xfs_mount *mp = ip->i_mount;
2322 struct xfs_perag *pag;
2323 struct xfs_icluster xic = { 0 };
2324 struct xfs_inode_log_item *iip = ip->i_itemp;
2327 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
2328 ASSERT(VFS_I(ip)->i_nlink == 0);
2329 ASSERT(ip->i_df.if_nextents == 0);
2330 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2331 ASSERT(ip->i_nblocks == 0);
2333 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2336 * Free the inode first so that we guarantee that the AGI lock is going
2337 * to be taken before we remove the inode from the unlinked list. This
2338 * makes the AGI lock -> unlinked list modification order the same as
2339 * used in O_TMPFILE creation.
2341 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2345 error = xfs_iunlink_remove(tp, pag, ip);
2350 * Free any local-format data sitting around before we reset the
2351 * data fork to extents format. Note that the attr fork data has
2352 * already been freed by xfs_attr_inactive.
2354 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2355 kfree(ip->i_df.if_data);
2356 ip->i_df.if_data = NULL;
2357 ip->i_df.if_bytes = 0;
2360 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2362 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2363 ip->i_forkoff = 0; /* mark the attr fork not in use */
2364 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2365 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2366 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2368 /* Don't attempt to replay owner changes for a deleted inode */
2369 spin_lock(&iip->ili_lock);
2370 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2371 spin_unlock(&iip->ili_lock);
2374 * Bump the generation count so no one will be confused
2375 * by reincarnations of this inode.
2377 VFS_I(ip)->i_generation++;
2378 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2381 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2388 * This is called to unpin an inode. The caller must have the inode locked
2389 * in at least shared mode so that the buffer cannot be subsequently pinned
2390 * once someone is waiting for it to be unpinned.
2394 struct xfs_inode *ip)
2396 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
2398 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2400 /* Give the log a push to start the unpinning I/O */
2401 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2407 struct xfs_inode *ip)
2409 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2410 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2415 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2416 if (xfs_ipincount(ip))
2418 } while (xfs_ipincount(ip));
2419 finish_wait(wq, &wait.wq_entry);
2424 struct xfs_inode *ip)
2426 if (xfs_ipincount(ip))
2427 __xfs_iunpin_wait(ip);
2431 * Removing an inode from the namespace involves removing the directory entry
2432 * and dropping the link count on the inode. Removing the directory entry can
2433 * result in locking an AGF (directory blocks were freed) and removing a link
2434 * count can result in placing the inode on an unlinked list which results in
2437 * The big problem here is that we have an ordering constraint on AGF and AGI
2438 * locking - inode allocation locks the AGI, then can allocate a new extent for
2439 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2440 * removes the inode from the unlinked list, requiring that we lock the AGI
2441 * first, and then freeing the inode can result in an inode chunk being freed
2442 * and hence freeing disk space requiring that we lock an AGF.
2444 * Hence the ordering that is imposed by other parts of the code is AGI before
2445 * AGF. This means we cannot remove the directory entry before we drop the inode
2446 * reference count and put it on the unlinked list as this results in a lock
2447 * order of AGF then AGI, and this can deadlock against inode allocation and
2448 * freeing. Therefore we must drop the link counts before we remove the
2451 * This is still safe from a transactional point of view - it is not until we
2452 * get to xfs_defer_finish() that we have the possibility of multiple
2453 * transactions in this operation. Hence as long as we remove the directory
2454 * entry and drop the link count in the first transaction of the remove
2455 * operation, there are no transactional constraints on the ordering here.
2460 struct xfs_name *name,
2463 xfs_mount_t *mp = dp->i_mount;
2464 xfs_trans_t *tp = NULL;
2465 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2470 trace_xfs_remove(dp, name);
2472 if (xfs_is_shutdown(mp))
2474 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
2477 error = xfs_qm_dqattach(dp);
2481 error = xfs_qm_dqattach(ip);
2486 * We try to get the real space reservation first, allowing for
2487 * directory btree deletion(s) implying possible bmap insert(s). If we
2488 * can't get the space reservation then we use 0 instead, and avoid the
2489 * bmap btree insert(s) in the directory code by, if the bmap insert
2490 * tries to happen, instead trimming the LAST block from the directory.
2492 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2493 * the directory code can handle a reservationless update and we don't
2494 * want to prevent a user from trying to free space by deleting things.
2496 resblks = XFS_REMOVE_SPACE_RES(mp);
2497 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2500 ASSERT(error != -ENOSPC);
2505 * If we're removing a directory perform some additional validation.
2508 ASSERT(VFS_I(ip)->i_nlink >= 2);
2509 if (VFS_I(ip)->i_nlink != 2) {
2511 goto out_trans_cancel;
2513 if (!xfs_dir_isempty(ip)) {
2515 goto out_trans_cancel;
2518 /* Drop the link from ip's "..". */
2519 error = xfs_droplink(tp, dp);
2521 goto out_trans_cancel;
2523 /* Drop the "." link from ip to self. */
2524 error = xfs_droplink(tp, ip);
2526 goto out_trans_cancel;
2529 * Point the unlinked child directory's ".." entry to the root
2530 * directory to eliminate back-references to inodes that may
2531 * get freed before the child directory is closed. If the fs
2532 * gets shrunk, this can lead to dirent inode validation errors.
2534 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2535 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2536 tp->t_mountp->m_sb.sb_rootino, 0);
2538 goto out_trans_cancel;
2542 * When removing a non-directory we need to log the parent
2543 * inode here. For a directory this is done implicitly
2544 * by the xfs_droplink call for the ".." entry.
2546 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2548 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2550 /* Drop the link from dp to ip. */
2551 error = xfs_droplink(tp, ip);
2553 goto out_trans_cancel;
2555 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2557 ASSERT(error != -ENOENT);
2558 goto out_trans_cancel;
2562 * If this is a synchronous mount, make sure that the
2563 * remove transaction goes to disk before returning to
2566 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2567 xfs_trans_set_sync(tp);
2569 error = xfs_trans_commit(tp);
2573 if (is_dir && xfs_inode_is_filestream(ip))
2574 xfs_filestream_deassociate(ip);
2579 xfs_trans_cancel(tp);
2585 * Enter all inodes for a rename transaction into a sorted array.
2587 #define __XFS_SORT_INODES 5
2589 xfs_sort_for_rename(
2590 struct xfs_inode *dp1, /* in: old (source) directory inode */
2591 struct xfs_inode *dp2, /* in: new (target) directory inode */
2592 struct xfs_inode *ip1, /* in: inode of old entry */
2593 struct xfs_inode *ip2, /* in: inode of new entry */
2594 struct xfs_inode *wip, /* in: whiteout inode */
2595 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2596 int *num_inodes) /* in/out: inodes in array */
2600 ASSERT(*num_inodes == __XFS_SORT_INODES);
2601 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2604 * i_tab contains a list of pointers to inodes. We initialize
2605 * the table here & we'll sort it. We will then use it to
2606 * order the acquisition of the inode locks.
2608 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2621 * Sort the elements via bubble sort. (Remember, there are at
2622 * most 5 elements to sort, so this is adequate.)
2624 for (i = 0; i < *num_inodes; i++) {
2625 for (j = 1; j < *num_inodes; j++) {
2626 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2627 struct xfs_inode *temp = i_tab[j];
2628 i_tab[j] = i_tab[j-1];
2637 struct xfs_trans *tp)
2640 * If this is a synchronous mount, make sure that the rename transaction
2641 * goes to disk before returning to the user.
2643 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2644 xfs_trans_set_sync(tp);
2646 return xfs_trans_commit(tp);
2650 * xfs_cross_rename()
2652 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2656 struct xfs_trans *tp,
2657 struct xfs_inode *dp1,
2658 struct xfs_name *name1,
2659 struct xfs_inode *ip1,
2660 struct xfs_inode *dp2,
2661 struct xfs_name *name2,
2662 struct xfs_inode *ip2,
2670 /* Swap inode number for dirent in first parent */
2671 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2673 goto out_trans_abort;
2675 /* Swap inode number for dirent in second parent */
2676 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2678 goto out_trans_abort;
2681 * If we're renaming one or more directories across different parents,
2682 * update the respective ".." entries (and link counts) to match the new
2686 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2688 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2689 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2690 dp1->i_ino, spaceres);
2692 goto out_trans_abort;
2694 /* transfer ip2 ".." reference to dp1 */
2695 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2696 error = xfs_droplink(tp, dp2);
2698 goto out_trans_abort;
2699 xfs_bumplink(tp, dp1);
2703 * Although ip1 isn't changed here, userspace needs
2704 * to be warned about the change, so that applications
2705 * relying on it (like backup ones), will properly
2708 ip1_flags |= XFS_ICHGTIME_CHG;
2709 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2712 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2713 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2714 dp2->i_ino, spaceres);
2716 goto out_trans_abort;
2718 /* transfer ip1 ".." reference to dp2 */
2719 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2720 error = xfs_droplink(tp, dp1);
2722 goto out_trans_abort;
2723 xfs_bumplink(tp, dp2);
2727 * Although ip2 isn't changed here, userspace needs
2728 * to be warned about the change, so that applications
2729 * relying on it (like backup ones), will properly
2732 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2733 ip2_flags |= XFS_ICHGTIME_CHG;
2738 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2739 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2742 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2743 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2746 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2747 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2749 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2750 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2751 return xfs_finish_rename(tp);
2754 xfs_trans_cancel(tp);
2759 * xfs_rename_alloc_whiteout()
2761 * Return a referenced, unlinked, unlocked inode that can be used as a
2762 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2763 * crash between allocating the inode and linking it into the rename transaction
2764 * recovery will free the inode and we won't leak it.
2767 xfs_rename_alloc_whiteout(
2768 struct mnt_idmap *idmap,
2769 struct xfs_name *src_name,
2770 struct xfs_inode *dp,
2771 struct xfs_inode **wip)
2773 struct xfs_inode *tmpfile;
2777 error = xfs_create_tmpfile(idmap, dp, S_IFCHR | WHITEOUT_MODE,
2782 name.name = src_name->name;
2783 name.len = src_name->len;
2784 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2786 xfs_finish_inode_setup(tmpfile);
2792 * Prepare the tmpfile inode as if it were created through the VFS.
2793 * Complete the inode setup and flag it as linkable. nlink is already
2794 * zero, so we can skip the drop_nlink.
2796 xfs_setup_iops(tmpfile);
2797 xfs_finish_inode_setup(tmpfile);
2798 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2809 struct mnt_idmap *idmap,
2810 struct xfs_inode *src_dp,
2811 struct xfs_name *src_name,
2812 struct xfs_inode *src_ip,
2813 struct xfs_inode *target_dp,
2814 struct xfs_name *target_name,
2815 struct xfs_inode *target_ip,
2818 struct xfs_mount *mp = src_dp->i_mount;
2819 struct xfs_trans *tp;
2820 struct xfs_inode *wip = NULL; /* whiteout inode */
2821 struct xfs_inode *inodes[__XFS_SORT_INODES];
2823 int num_inodes = __XFS_SORT_INODES;
2824 bool new_parent = (src_dp != target_dp);
2825 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2827 bool retried = false;
2828 int error, nospace_error = 0;
2830 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2832 if ((flags & RENAME_EXCHANGE) && !target_ip)
2836 * If we are doing a whiteout operation, allocate the whiteout inode
2837 * we will be placing at the target and ensure the type is set
2840 if (flags & RENAME_WHITEOUT) {
2841 error = xfs_rename_alloc_whiteout(idmap, src_name,
2846 /* setup target dirent info as whiteout */
2847 src_name->type = XFS_DIR3_FT_CHRDEV;
2850 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2851 inodes, &num_inodes);
2855 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2856 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2857 if (error == -ENOSPC) {
2858 nospace_error = error;
2860 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2864 goto out_release_wip;
2867 * Attach the dquots to the inodes
2869 error = xfs_qm_vop_rename_dqattach(inodes);
2871 goto out_trans_cancel;
2874 * Lock all the participating inodes. Depending upon whether
2875 * the target_name exists in the target directory, and
2876 * whether the target directory is the same as the source
2877 * directory, we can lock from 2 to 5 inodes.
2879 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2882 * Join all the inodes to the transaction. From this point on,
2883 * we can rely on either trans_commit or trans_cancel to unlock
2886 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2888 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2889 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2891 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2893 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2896 * If we are using project inheritance, we only allow renames
2897 * into our tree when the project IDs are the same; else the
2898 * tree quota mechanism would be circumvented.
2900 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2901 target_dp->i_projid != src_ip->i_projid)) {
2903 goto out_trans_cancel;
2906 /* RENAME_EXCHANGE is unique from here on. */
2907 if (flags & RENAME_EXCHANGE)
2908 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2909 target_dp, target_name, target_ip,
2913 * Try to reserve quota to handle an expansion of the target directory.
2914 * We'll allow the rename to continue in reservationless mode if we hit
2915 * a space usage constraint. If we trigger reservationless mode, save
2916 * the errno if there isn't any free space in the target directory.
2918 if (spaceres != 0) {
2919 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2921 if (error == -EDQUOT || error == -ENOSPC) {
2923 xfs_trans_cancel(tp);
2924 xfs_blockgc_free_quota(target_dp, 0);
2929 nospace_error = error;
2934 goto out_trans_cancel;
2938 * Check for expected errors before we dirty the transaction
2939 * so we can return an error without a transaction abort.
2941 if (target_ip == NULL) {
2943 * If there's no space reservation, check the entry will
2944 * fit before actually inserting it.
2947 error = xfs_dir_canenter(tp, target_dp, target_name);
2949 goto out_trans_cancel;
2953 * If target exists and it's a directory, check that whether
2954 * it can be destroyed.
2956 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
2957 (!xfs_dir_isempty(target_ip) ||
2958 (VFS_I(target_ip)->i_nlink > 2))) {
2960 goto out_trans_cancel;
2965 * Lock the AGI buffers we need to handle bumping the nlink of the
2966 * whiteout inode off the unlinked list and to handle dropping the
2967 * nlink of the target inode. Per locking order rules, do this in
2968 * increasing AG order and before directory block allocation tries to
2969 * grab AGFs because we grab AGIs before AGFs.
2971 * The (vfs) caller must ensure that if src is a directory then
2972 * target_ip is either null or an empty directory.
2974 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2975 if (inodes[i] == wip ||
2976 (inodes[i] == target_ip &&
2977 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2978 struct xfs_perag *pag;
2981 pag = xfs_perag_get(mp,
2982 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2983 error = xfs_read_agi(pag, tp, &bp);
2986 goto out_trans_cancel;
2991 * Directory entry creation below may acquire the AGF. Remove
2992 * the whiteout from the unlinked list first to preserve correct
2993 * AGI/AGF locking order. This dirties the transaction so failures
2994 * after this point will abort and log recovery will clean up the
2997 * For whiteouts, we need to bump the link count on the whiteout
2998 * inode. After this point, we have a real link, clear the tmpfile
2999 * state flag from the inode so it doesn't accidentally get misused
3003 struct xfs_perag *pag;
3005 ASSERT(VFS_I(wip)->i_nlink == 0);
3007 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
3008 error = xfs_iunlink_remove(tp, pag, wip);
3011 goto out_trans_cancel;
3013 xfs_bumplink(tp, wip);
3014 VFS_I(wip)->i_state &= ~I_LINKABLE;
3018 * Set up the target.
3020 if (target_ip == NULL) {
3022 * If target does not exist and the rename crosses
3023 * directories, adjust the target directory link count
3024 * to account for the ".." reference from the new entry.
3026 error = xfs_dir_createname(tp, target_dp, target_name,
3027 src_ip->i_ino, spaceres);
3029 goto out_trans_cancel;
3031 xfs_trans_ichgtime(tp, target_dp,
3032 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3034 if (new_parent && src_is_directory) {
3035 xfs_bumplink(tp, target_dp);
3037 } else { /* target_ip != NULL */
3039 * Link the source inode under the target name.
3040 * If the source inode is a directory and we are moving
3041 * it across directories, its ".." entry will be
3042 * inconsistent until we replace that down below.
3044 * In case there is already an entry with the same
3045 * name at the destination directory, remove it first.
3047 error = xfs_dir_replace(tp, target_dp, target_name,
3048 src_ip->i_ino, spaceres);
3050 goto out_trans_cancel;
3052 xfs_trans_ichgtime(tp, target_dp,
3053 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3056 * Decrement the link count on the target since the target
3057 * dir no longer points to it.
3059 error = xfs_droplink(tp, target_ip);
3061 goto out_trans_cancel;
3063 if (src_is_directory) {
3065 * Drop the link from the old "." entry.
3067 error = xfs_droplink(tp, target_ip);
3069 goto out_trans_cancel;
3071 } /* target_ip != NULL */
3074 * Remove the source.
3076 if (new_parent && src_is_directory) {
3078 * Rewrite the ".." entry to point to the new
3081 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3082 target_dp->i_ino, spaceres);
3083 ASSERT(error != -EEXIST);
3085 goto out_trans_cancel;
3089 * We always want to hit the ctime on the source inode.
3091 * This isn't strictly required by the standards since the source
3092 * inode isn't really being changed, but old unix file systems did
3093 * it and some incremental backup programs won't work without it.
3095 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3096 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3099 * Adjust the link count on src_dp. This is necessary when
3100 * renaming a directory, either within one parent when
3101 * the target existed, or across two parent directories.
3103 if (src_is_directory && (new_parent || target_ip != NULL)) {
3106 * Decrement link count on src_directory since the
3107 * entry that's moved no longer points to it.
3109 error = xfs_droplink(tp, src_dp);
3111 goto out_trans_cancel;
3115 * For whiteouts, we only need to update the source dirent with the
3116 * inode number of the whiteout inode rather than removing it
3120 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3123 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3127 goto out_trans_cancel;
3129 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3130 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3132 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3134 error = xfs_finish_rename(tp);
3140 xfs_trans_cancel(tp);
3144 if (error == -ENOSPC && nospace_error)
3145 error = nospace_error;
3151 struct xfs_inode *ip,
3154 struct xfs_inode_log_item *iip = ip->i_itemp;
3155 struct xfs_dinode *dip;
3156 struct xfs_mount *mp = ip->i_mount;
3159 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
3160 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3161 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3162 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3163 ASSERT(iip->ili_item.li_buf == bp);
3165 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3168 * We don't flush the inode if any of the following checks fail, but we
3169 * do still update the log item and attach to the backing buffer as if
3170 * the flush happened. This is a formality to facilitate predictable
3171 * error handling as the caller will shutdown and fail the buffer.
3173 error = -EFSCORRUPTED;
3174 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3175 mp, XFS_ERRTAG_IFLUSH_1)) {
3176 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3177 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3178 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3181 if (S_ISREG(VFS_I(ip)->i_mode)) {
3183 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3184 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3185 mp, XFS_ERRTAG_IFLUSH_3)) {
3186 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3187 "%s: Bad regular inode %llu, ptr "PTR_FMT,
3188 __func__, ip->i_ino, ip);
3191 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3193 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3194 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3195 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3196 mp, XFS_ERRTAG_IFLUSH_4)) {
3197 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3198 "%s: Bad directory inode %llu, ptr "PTR_FMT,
3199 __func__, ip->i_ino, ip);
3203 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3204 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3205 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3206 "%s: detected corrupt incore inode %llu, "
3207 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3208 __func__, ip->i_ino,
3209 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3213 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3214 mp, XFS_ERRTAG_IFLUSH_6)) {
3215 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3216 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3217 __func__, ip->i_ino, ip->i_forkoff, ip);
3222 * Inode item log recovery for v2 inodes are dependent on the flushiter
3223 * count for correct sequencing. We bump the flush iteration count so
3224 * we can detect flushes which postdate a log record during recovery.
3225 * This is redundant as we now log every change and hence this can't
3226 * happen but we need to still do it to ensure backwards compatibility
3227 * with old kernels that predate logging all inode changes.
3229 if (!xfs_has_v3inodes(mp))
3233 * If there are inline format data / attr forks attached to this inode,
3234 * make sure they are not corrupt.
3236 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3237 xfs_ifork_verify_local_data(ip))
3239 if (xfs_inode_has_attr_fork(ip) &&
3240 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3241 xfs_ifork_verify_local_attr(ip))
3245 * Copy the dirty parts of the inode into the on-disk inode. We always
3246 * copy out the core of the inode, because if the inode is dirty at all
3249 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3251 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3252 if (!xfs_has_v3inodes(mp)) {
3253 if (ip->i_flushiter == DI_MAX_FLUSH)
3254 ip->i_flushiter = 0;
3257 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3258 if (xfs_inode_has_attr_fork(ip))
3259 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3262 * We've recorded everything logged in the inode, so we'd like to clear
3263 * the ili_fields bits so we don't log and flush things unnecessarily.
3264 * However, we can't stop logging all this information until the data
3265 * we've copied into the disk buffer is written to disk. If we did we
3266 * might overwrite the copy of the inode in the log with all the data
3267 * after re-logging only part of it, and in the face of a crash we
3268 * wouldn't have all the data we need to recover.
3270 * What we do is move the bits to the ili_last_fields field. When
3271 * logging the inode, these bits are moved back to the ili_fields field.
3272 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3273 * we know that the information those bits represent is permanently on
3274 * disk. As long as the flush completes before the inode is logged
3275 * again, then both ili_fields and ili_last_fields will be cleared.
3279 spin_lock(&iip->ili_lock);
3280 iip->ili_last_fields = iip->ili_fields;
3281 iip->ili_fields = 0;
3282 iip->ili_fsync_fields = 0;
3283 spin_unlock(&iip->ili_lock);
3286 * Store the current LSN of the inode so that we can tell whether the
3287 * item has moved in the AIL from xfs_buf_inode_iodone().
3289 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3290 &iip->ili_item.li_lsn);
3292 /* generate the checksum. */
3293 xfs_dinode_calc_crc(mp, dip);
3298 * Non-blocking flush of dirty inode metadata into the backing buffer.
3300 * The caller must have a reference to the inode and hold the cluster buffer
3301 * locked. The function will walk across all the inodes on the cluster buffer it
3302 * can find and lock without blocking, and flush them to the cluster buffer.
3304 * On successful flushing of at least one inode, the caller must write out the
3305 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3306 * the caller needs to release the buffer. On failure, the filesystem will be
3307 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3314 struct xfs_mount *mp = bp->b_mount;
3315 struct xfs_log_item *lip, *n;
3316 struct xfs_inode *ip;
3317 struct xfs_inode_log_item *iip;
3322 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3323 * will remove itself from the list.
3325 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3326 iip = (struct xfs_inode_log_item *)lip;
3327 ip = iip->ili_inode;
3330 * Quick and dirty check to avoid locks if possible.
3332 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3334 if (xfs_ipincount(ip))
3338 * The inode is still attached to the buffer, which means it is
3339 * dirty but reclaim might try to grab it. Check carefully for
3340 * that, and grab the ilock while still holding the i_flags_lock
3341 * to guarantee reclaim will not be able to reclaim this inode
3342 * once we drop the i_flags_lock.
3344 spin_lock(&ip->i_flags_lock);
3345 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3346 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3347 spin_unlock(&ip->i_flags_lock);
3352 * ILOCK will pin the inode against reclaim and prevent
3353 * concurrent transactions modifying the inode while we are
3354 * flushing the inode. If we get the lock, set the flushing
3355 * state before we drop the i_flags_lock.
3357 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3358 spin_unlock(&ip->i_flags_lock);
3361 __xfs_iflags_set(ip, XFS_IFLUSHING);
3362 spin_unlock(&ip->i_flags_lock);
3365 * Abort flushing this inode if we are shut down because the
3366 * inode may not currently be in the AIL. This can occur when
3367 * log I/O failure unpins the inode without inserting into the
3368 * AIL, leaving a dirty/unpinned inode attached to the buffer
3369 * that otherwise looks like it should be flushed.
3371 if (xlog_is_shutdown(mp->m_log)) {
3372 xfs_iunpin_wait(ip);
3373 xfs_iflush_abort(ip);
3374 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3379 /* don't block waiting on a log force to unpin dirty inodes */
3380 if (xfs_ipincount(ip)) {
3381 xfs_iflags_clear(ip, XFS_IFLUSHING);
3382 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3386 if (!xfs_inode_clean(ip))
3387 error = xfs_iflush(ip, bp);
3389 xfs_iflags_clear(ip, XFS_IFLUSHING);
3390 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3398 * Shutdown first so we kill the log before we release this
3399 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3400 * of the log, failing it before the _log_ is shut down can
3401 * result in the log tail being moved forward in the journal
3402 * on disk because log writes can still be taking place. Hence
3403 * unpinning the tail will allow the ICREATE intent to be
3404 * removed from the log an recovery will fail with uninitialised
3405 * inode cluster buffers.
3407 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3408 bp->b_flags |= XBF_ASYNC;
3409 xfs_buf_ioend_fail(bp);
3416 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3417 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3422 /* Release an inode. */
3425 struct xfs_inode *ip)
3427 trace_xfs_irele(ip, _RET_IP_);
3432 * Ensure all commited transactions touching the inode are written to the log.
3435 xfs_log_force_inode(
3436 struct xfs_inode *ip)
3440 xfs_ilock(ip, XFS_ILOCK_SHARED);
3441 if (xfs_ipincount(ip))
3442 seq = ip->i_itemp->ili_commit_seq;
3443 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3447 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3451 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3452 * abide vfs locking order (lowest pointer value goes first) and breaking the
3453 * layout leases before proceeding. The loop is needed because we cannot call
3454 * the blocking break_layout() with the iolocks held, and therefore have to
3455 * back out both locks.
3458 xfs_iolock_two_inodes_and_break_layout(
3468 /* Wait to break both inodes' layouts before we start locking. */
3469 error = break_layout(src, true);
3473 error = break_layout(dest, true);
3478 /* Lock one inode and make sure nobody got in and leased it. */
3480 error = break_layout(src, false);
3483 if (error == -EWOULDBLOCK)
3491 /* Lock the other inode and make sure nobody got in and leased it. */
3492 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3493 error = break_layout(dest, false);
3497 if (error == -EWOULDBLOCK)
3506 xfs_mmaplock_two_inodes_and_break_dax_layout(
3507 struct xfs_inode *ip1,
3508 struct xfs_inode *ip2)
3514 if (ip1->i_ino > ip2->i_ino)
3519 /* Lock the first inode */
3520 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3521 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3522 if (error || retry) {
3523 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3524 if (error == 0 && retry)
3532 /* Nested lock the second inode */
3533 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3535 * We cannot use xfs_break_dax_layouts() directly here because it may
3536 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3537 * for this nested lock case.
3539 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3540 if (page && page_ref_count(page) != 1) {
3541 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3542 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3550 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3555 struct xfs_inode *ip1,
3556 struct xfs_inode *ip2)
3560 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3564 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3565 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3567 inode_unlock(VFS_I(ip2));
3569 inode_unlock(VFS_I(ip1));
3573 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3574 VFS_I(ip2)->i_mapping);
3579 /* Unlock both inodes to allow IO and mmap activity. */
3581 xfs_iunlock2_io_mmap(
3582 struct xfs_inode *ip1,
3583 struct xfs_inode *ip2)
3585 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3586 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3588 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3590 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3591 VFS_I(ip2)->i_mapping);
3593 inode_unlock(VFS_I(ip2));
3595 inode_unlock(VFS_I(ip1));
3598 /* Drop the MMAPLOCK and the IOLOCK after a remap completes. */
3600 xfs_iunlock2_remapping(
3601 struct xfs_inode *ip1,
3602 struct xfs_inode *ip2)
3604 xfs_iflags_clear(ip1, XFS_IREMAPPING);
3607 xfs_iunlock(ip1, XFS_MMAPLOCK_SHARED);
3608 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3611 inode_unlock_shared(VFS_I(ip1));
3612 inode_unlock(VFS_I(ip2));
3616 * Reload the incore inode list for this inode. Caller should ensure that
3617 * the link count cannot change, either by taking ILOCK_SHARED or otherwise
3618 * preventing other threads from executing.
3621 xfs_inode_reload_unlinked_bucket(
3622 struct xfs_trans *tp,
3623 struct xfs_inode *ip)
3625 struct xfs_mount *mp = tp->t_mountp;
3626 struct xfs_buf *agibp;
3627 struct xfs_agi *agi;
3628 struct xfs_perag *pag;
3629 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
3630 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
3631 xfs_agino_t prev_agino, next_agino;
3632 unsigned int bucket;
3633 bool foundit = false;
3636 /* Grab the first inode in the list */
3637 pag = xfs_perag_get(mp, agno);
3638 error = xfs_ialloc_read_agi(pag, tp, &agibp);
3644 * We've taken ILOCK_SHARED and the AGI buffer lock to stabilize the
3645 * incore unlinked list pointers for this inode. Check once more to
3646 * see if we raced with anyone else to reload the unlinked list.
3648 if (!xfs_inode_unlinked_incomplete(ip)) {
3653 bucket = agino % XFS_AGI_UNLINKED_BUCKETS;
3654 agi = agibp->b_addr;
3656 trace_xfs_inode_reload_unlinked_bucket(ip);
3658 xfs_info_ratelimited(mp,
3659 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating list recovery.",
3662 prev_agino = NULLAGINO;
3663 next_agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3664 while (next_agino != NULLAGINO) {
3665 struct xfs_inode *next_ip = NULL;
3667 /* Found this caller's inode, set its backlink. */
3668 if (next_agino == agino) {
3670 next_ip->i_prev_unlinked = prev_agino;
3675 /* Try in-memory lookup first. */
3676 next_ip = xfs_iunlink_lookup(pag, next_agino);
3680 /* Inode not in memory, try reloading it. */
3681 error = xfs_iunlink_reload_next(tp, agibp, prev_agino,
3686 /* Grab the reloaded inode. */
3687 next_ip = xfs_iunlink_lookup(pag, next_agino);
3689 /* No incore inode at all? We reloaded it... */
3690 ASSERT(next_ip != NULL);
3691 error = -EFSCORRUPTED;
3696 prev_agino = next_agino;
3697 next_agino = next_ip->i_next_unlinked;
3701 xfs_trans_brelse(tp, agibp);
3702 /* Should have found this inode somewhere in the iunlinked bucket. */
3703 if (!error && !foundit)
3704 error = -EFSCORRUPTED;
3708 /* Decide if this inode is missing its unlinked list and reload it. */
3710 xfs_inode_reload_unlinked(
3711 struct xfs_inode *ip)
3713 struct xfs_trans *tp;
3716 error = xfs_trans_alloc_empty(ip->i_mount, &tp);
3720 xfs_ilock(ip, XFS_ILOCK_SHARED);
3721 if (xfs_inode_unlinked_incomplete(ip))
3722 error = xfs_inode_reload_unlinked_bucket(tp, ip);
3723 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3724 xfs_trans_cancel(tp);
3729 /* Has this inode fork been zapped by repair? */
3732 const struct xfs_inode *ip,
3735 unsigned int datamask = 0;
3737 switch (whichfork) {
3739 switch (ip->i_vnode.i_mode & S_IFMT) {
3741 datamask = XFS_SICK_INO_DIR_ZAPPED;
3744 datamask = XFS_SICK_INO_SYMLINK_ZAPPED;
3747 return ip->i_sick & (XFS_SICK_INO_BMBTD_ZAPPED | datamask);
3749 return ip->i_sick & XFS_SICK_INO_BMBTA_ZAPPED;