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_);
331 #if defined(DEBUG) || defined(XFS_WARN)
333 __xfs_rwsem_islocked(
334 struct rw_semaphore *rwsem,
338 return rwsem_is_locked(rwsem);
341 return lockdep_is_held_type(rwsem, 0);
344 * We are checking that the lock is held at least in shared
345 * mode but don't care that it might be held exclusively
346 * (i.e. shared | excl). Hence we check if the lock is held
347 * in any mode rather than an explicit shared mode.
349 return lockdep_is_held_type(rwsem, -1);
354 struct xfs_inode *ip,
357 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
358 if (!(lock_flags & XFS_ILOCK_SHARED))
359 return !!ip->i_lock.mr_writer;
360 return rwsem_is_locked(&ip->i_lock.mr_lock);
363 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
364 return __xfs_rwsem_islocked(&VFS_I(ip)->i_mapping->invalidate_lock,
365 (lock_flags & XFS_MMAPLOCK_SHARED));
368 if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) {
369 return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
370 (lock_flags & XFS_IOLOCK_SHARED));
379 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
380 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
381 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
382 * errors and warnings.
384 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
386 xfs_lockdep_subclass_ok(
389 return subclass < MAX_LOCKDEP_SUBCLASSES;
392 #define xfs_lockdep_subclass_ok(subclass) (true)
396 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
397 * value. This can be called for any type of inode lock combination, including
398 * parent locking. Care must be taken to ensure we don't overrun the subclass
399 * storage fields in the class mask we build.
408 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
410 ASSERT(xfs_lockdep_subclass_ok(subclass));
412 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
413 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
414 class += subclass << XFS_IOLOCK_SHIFT;
417 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
418 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
419 class += subclass << XFS_MMAPLOCK_SHIFT;
422 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
423 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
424 class += subclass << XFS_ILOCK_SHIFT;
427 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
431 * The following routine will lock n inodes in exclusive mode. We assume the
432 * caller calls us with the inodes in i_ino order.
434 * We need to detect deadlock where an inode that we lock is in the AIL and we
435 * start waiting for another inode that is locked by a thread in a long running
436 * transaction (such as truncate). This can result in deadlock since the long
437 * running trans might need to wait for the inode we just locked in order to
438 * push the tail and free space in the log.
440 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
441 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
442 * lock more than one at a time, lockdep will report false positives saying we
443 * have violated locking orders.
447 struct xfs_inode **ips,
455 struct xfs_log_item *lp;
458 * Currently supports between 2 and 5 inodes with exclusive locking. We
459 * support an arbitrary depth of locking here, but absolute limits on
460 * inodes depend on the type of locking and the limits placed by
461 * lockdep annotations in xfs_lock_inumorder. These are all checked by
464 ASSERT(ips && inodes >= 2 && inodes <= 5);
465 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
467 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
469 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
470 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
471 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
472 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
474 if (lock_mode & XFS_IOLOCK_EXCL) {
475 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
476 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
477 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
482 for (; i < inodes; i++) {
485 if (i && (ips[i] == ips[i - 1])) /* Already locked */
489 * If try_lock is not set yet, make sure all locked inodes are
490 * not in the AIL. If any are, set try_lock to be used later.
493 for (j = (i - 1); j >= 0 && !try_lock; j--) {
494 lp = &ips[j]->i_itemp->ili_item;
495 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
501 * If any of the previous locks we have locked is in the AIL,
502 * we must TRY to get the second and subsequent locks. If
503 * we can't get any, we must release all we have
507 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
511 /* try_lock means we have an inode locked that is in the AIL. */
513 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
517 * Unlock all previous guys and try again. xfs_iunlock will try
518 * to push the tail if the inode is in the AIL.
521 for (j = i - 1; j >= 0; j--) {
523 * Check to see if we've already unlocked this one. Not
524 * the first one going back, and the inode ptr is the
527 if (j != (i - 1) && ips[j] == ips[j + 1])
530 xfs_iunlock(ips[j], lock_mode);
533 if ((attempts % 5) == 0) {
534 delay(1); /* Don't just spin the CPU */
541 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
542 * mmaplock must be double-locked separately since we use i_rwsem and
543 * invalidate_lock for that. We now support taking one lock EXCL and the
548 struct xfs_inode *ip0,
550 struct xfs_inode *ip1,
554 struct xfs_log_item *lp;
556 ASSERT(hweight32(ip0_mode) == 1);
557 ASSERT(hweight32(ip1_mode) == 1);
558 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
559 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
560 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
561 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
562 ASSERT(ip0->i_ino != ip1->i_ino);
564 if (ip0->i_ino > ip1->i_ino) {
566 swap(ip0_mode, ip1_mode);
570 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
573 * If the first lock we have locked is in the AIL, we must TRY to get
574 * the second lock. If we can't get it, we must release the first one
577 lp = &ip0->i_itemp->ili_item;
578 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
579 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
580 xfs_iunlock(ip0, ip0_mode);
581 if ((++attempts % 5) == 0)
582 delay(1); /* Don't just spin the CPU */
586 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
592 struct xfs_inode *ip)
596 if (ip->i_diflags & XFS_DIFLAG_ANY) {
597 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
598 flags |= FS_XFLAG_REALTIME;
599 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
600 flags |= FS_XFLAG_PREALLOC;
601 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
602 flags |= FS_XFLAG_IMMUTABLE;
603 if (ip->i_diflags & XFS_DIFLAG_APPEND)
604 flags |= FS_XFLAG_APPEND;
605 if (ip->i_diflags & XFS_DIFLAG_SYNC)
606 flags |= FS_XFLAG_SYNC;
607 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
608 flags |= FS_XFLAG_NOATIME;
609 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
610 flags |= FS_XFLAG_NODUMP;
611 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
612 flags |= FS_XFLAG_RTINHERIT;
613 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
614 flags |= FS_XFLAG_PROJINHERIT;
615 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
616 flags |= FS_XFLAG_NOSYMLINKS;
617 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
618 flags |= FS_XFLAG_EXTSIZE;
619 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
620 flags |= FS_XFLAG_EXTSZINHERIT;
621 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
622 flags |= FS_XFLAG_NODEFRAG;
623 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
624 flags |= FS_XFLAG_FILESTREAM;
627 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
628 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
629 flags |= FS_XFLAG_DAX;
630 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
631 flags |= FS_XFLAG_COWEXTSIZE;
634 if (xfs_inode_has_attr_fork(ip))
635 flags |= FS_XFLAG_HASATTR;
640 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
641 * is allowed, otherwise it has to be an exact match. If a CI match is found,
642 * ci_name->name will point to a the actual name (caller must free) or
643 * will be set to NULL if an exact match is found.
647 struct xfs_inode *dp,
648 const struct xfs_name *name,
649 struct xfs_inode **ipp,
650 struct xfs_name *ci_name)
655 trace_xfs_lookup(dp, name);
657 if (xfs_is_shutdown(dp->i_mount))
659 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
662 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
666 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
674 kmem_free(ci_name->name);
680 /* Propagate di_flags from a parent inode to a child inode. */
682 xfs_inode_inherit_flags(
683 struct xfs_inode *ip,
684 const struct xfs_inode *pip)
686 unsigned int di_flags = 0;
687 xfs_failaddr_t failaddr;
688 umode_t mode = VFS_I(ip)->i_mode;
691 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
692 di_flags |= XFS_DIFLAG_RTINHERIT;
693 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
694 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
695 ip->i_extsize = pip->i_extsize;
697 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
698 di_flags |= XFS_DIFLAG_PROJINHERIT;
699 } else if (S_ISREG(mode)) {
700 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
701 xfs_has_realtime(ip->i_mount))
702 di_flags |= XFS_DIFLAG_REALTIME;
703 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
704 di_flags |= XFS_DIFLAG_EXTSIZE;
705 ip->i_extsize = pip->i_extsize;
708 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
710 di_flags |= XFS_DIFLAG_NOATIME;
711 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
713 di_flags |= XFS_DIFLAG_NODUMP;
714 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
716 di_flags |= XFS_DIFLAG_SYNC;
717 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
718 xfs_inherit_nosymlinks)
719 di_flags |= XFS_DIFLAG_NOSYMLINKS;
720 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
721 xfs_inherit_nodefrag)
722 di_flags |= XFS_DIFLAG_NODEFRAG;
723 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
724 di_flags |= XFS_DIFLAG_FILESTREAM;
726 ip->i_diflags |= di_flags;
729 * Inode verifiers on older kernels only check that the extent size
730 * hint is an integer multiple of the rt extent size on realtime files.
731 * They did not check the hint alignment on a directory with both
732 * rtinherit and extszinherit flags set. If the misaligned hint is
733 * propagated from a directory into a new realtime file, new file
734 * allocations will fail due to math errors in the rt allocator and/or
735 * trip the verifiers. Validate the hint settings in the new file so
736 * that we don't let broken hints propagate.
738 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
739 VFS_I(ip)->i_mode, ip->i_diflags);
741 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
742 XFS_DIFLAG_EXTSZINHERIT);
747 /* Propagate di_flags2 from a parent inode to a child inode. */
749 xfs_inode_inherit_flags2(
750 struct xfs_inode *ip,
751 const struct xfs_inode *pip)
753 xfs_failaddr_t failaddr;
755 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
756 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
757 ip->i_cowextsize = pip->i_cowextsize;
759 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
760 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
762 /* Don't let invalid cowextsize hints propagate. */
763 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
764 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
766 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
767 ip->i_cowextsize = 0;
772 * Initialise a newly allocated inode and return the in-core inode to the
773 * caller locked exclusively.
777 struct mnt_idmap *idmap,
778 struct xfs_trans *tp,
779 struct xfs_inode *pip,
786 struct xfs_inode **ipp)
788 struct inode *dir = pip ? VFS_I(pip) : NULL;
789 struct xfs_mount *mp = tp->t_mountp;
790 struct xfs_inode *ip;
793 struct timespec64 tv;
797 * Protect against obviously corrupt allocation btree records. Later
798 * xfs_iget checks will catch re-allocation of other active in-memory
799 * and on-disk inodes. If we don't catch reallocating the parent inode
800 * here we will deadlock in xfs_iget() so we have to do these checks
803 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
804 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
805 return -EFSCORRUPTED;
809 * Get the in-core inode with the lock held exclusively to prevent
810 * others from looking at until we're done.
812 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
818 set_nlink(inode, nlink);
819 inode->i_rdev = rdev;
822 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
823 inode_fsuid_set(inode, idmap);
824 inode->i_gid = dir->i_gid;
825 inode->i_mode = mode;
827 inode_init_owner(idmap, inode, dir, mode);
831 * If the group ID of the new file does not match the effective group
832 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
833 * (and only if the irix_sgid_inherit compatibility variable is set).
835 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
836 !vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode)))
837 inode->i_mode &= ~S_ISGID;
840 ip->i_df.if_nextents = 0;
841 ASSERT(ip->i_nblocks == 0);
843 tv = inode_set_ctime_current(inode);
844 inode_set_mtime_to_ts(inode, tv);
845 inode_set_atime_to_ts(inode, tv);
850 if (xfs_has_v3inodes(mp)) {
851 inode_set_iversion(inode, 1);
852 ip->i_cowextsize = 0;
856 flags = XFS_ILOG_CORE;
857 switch (mode & S_IFMT) {
862 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
863 flags |= XFS_ILOG_DEV;
867 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
868 xfs_inode_inherit_flags(ip, pip);
869 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
870 xfs_inode_inherit_flags2(ip, pip);
873 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
874 ip->i_df.if_bytes = 0;
875 ip->i_df.if_data = NULL;
882 * If we need to create attributes immediately after allocating the
883 * inode, initialise an empty attribute fork right now. We use the
884 * default fork offset for attributes here as we don't know exactly what
885 * size or how many attributes we might be adding. We can do this
886 * safely here because we know the data fork is completely empty and
887 * this saves us from needing to run a separate transaction to set the
888 * fork offset in the immediate future.
890 if (init_xattrs && xfs_has_attr(mp)) {
891 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
892 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
896 * Log the new values stuffed into the inode.
898 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
899 xfs_trans_log_inode(tp, ip, flags);
901 /* now that we have an i_mode we can setup the inode structure */
909 * Decrement the link count on an inode & log the change. If this causes the
910 * link count to go to zero, move the inode to AGI unlinked list so that it can
911 * be freed when the last active reference goes away via xfs_inactive().
913 static int /* error */
918 if (VFS_I(ip)->i_nlink == 0) {
919 xfs_alert(ip->i_mount,
920 "%s: Attempt to drop inode (%llu) with nlink zero.",
921 __func__, ip->i_ino);
922 return -EFSCORRUPTED;
925 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
927 drop_nlink(VFS_I(ip));
928 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
930 if (VFS_I(ip)->i_nlink)
933 return xfs_iunlink(tp, ip);
937 * Increment the link count on an inode & log the change.
944 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
946 inc_nlink(VFS_I(ip));
947 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
952 struct mnt_idmap *idmap,
954 struct xfs_name *name,
960 int is_dir = S_ISDIR(mode);
961 struct xfs_mount *mp = dp->i_mount;
962 struct xfs_inode *ip = NULL;
963 struct xfs_trans *tp = NULL;
965 bool unlock_dp_on_error = false;
967 struct xfs_dquot *udqp = NULL;
968 struct xfs_dquot *gdqp = NULL;
969 struct xfs_dquot *pdqp = NULL;
970 struct xfs_trans_res *tres;
974 trace_xfs_create(dp, name);
976 if (xfs_is_shutdown(mp))
978 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
981 prid = xfs_get_initial_prid(dp);
984 * Make sure that we have allocated dquot(s) on disk.
986 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
987 mapped_fsgid(idmap, &init_user_ns), prid,
988 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
989 &udqp, &gdqp, &pdqp);
994 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
995 tres = &M_RES(mp)->tr_mkdir;
997 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
998 tres = &M_RES(mp)->tr_create;
1002 * Initially assume that the file does not exist and
1003 * reserve the resources for that case. If that is not
1004 * the case we'll drop the one we have and get a more
1005 * appropriate transaction later.
1007 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1009 if (error == -ENOSPC) {
1010 /* flush outstanding delalloc blocks and retry */
1011 xfs_flush_inodes(mp);
1012 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1016 goto out_release_dquots;
1018 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1019 unlock_dp_on_error = true;
1022 * A newly created regular or special file just has one directory
1023 * entry pointing to them, but a directory also the "." entry
1024 * pointing to itself.
1026 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1028 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1029 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1031 goto out_trans_cancel;
1034 * Now we join the directory inode to the transaction. We do not do it
1035 * earlier because xfs_dialloc might commit the previous transaction
1036 * (and release all the locks). An error from here on will result in
1037 * the transaction cancel unlocking dp so don't do it explicitly in the
1040 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1041 unlock_dp_on_error = false;
1043 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1044 resblks - XFS_IALLOC_SPACE_RES(mp));
1046 ASSERT(error != -ENOSPC);
1047 goto out_trans_cancel;
1049 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1050 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1053 error = xfs_dir_init(tp, ip, dp);
1055 goto out_trans_cancel;
1057 xfs_bumplink(tp, dp);
1061 * If this is a synchronous mount, make sure that the
1062 * create transaction goes to disk before returning to
1065 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1066 xfs_trans_set_sync(tp);
1069 * Attach the dquot(s) to the inodes and modify them incore.
1070 * These ids of the inode couldn't have changed since the new
1071 * inode has been locked ever since it was created.
1073 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1075 error = xfs_trans_commit(tp);
1077 goto out_release_inode;
1079 xfs_qm_dqrele(udqp);
1080 xfs_qm_dqrele(gdqp);
1081 xfs_qm_dqrele(pdqp);
1087 xfs_trans_cancel(tp);
1090 * Wait until after the current transaction is aborted to finish the
1091 * setup of the inode and release the inode. This prevents recursive
1092 * transactions and deadlocks from xfs_inactive.
1095 xfs_finish_inode_setup(ip);
1099 xfs_qm_dqrele(udqp);
1100 xfs_qm_dqrele(gdqp);
1101 xfs_qm_dqrele(pdqp);
1103 if (unlock_dp_on_error)
1104 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1110 struct mnt_idmap *idmap,
1111 struct xfs_inode *dp,
1113 struct xfs_inode **ipp)
1115 struct xfs_mount *mp = dp->i_mount;
1116 struct xfs_inode *ip = NULL;
1117 struct xfs_trans *tp = NULL;
1120 struct xfs_dquot *udqp = NULL;
1121 struct xfs_dquot *gdqp = NULL;
1122 struct xfs_dquot *pdqp = NULL;
1123 struct xfs_trans_res *tres;
1127 if (xfs_is_shutdown(mp))
1130 prid = xfs_get_initial_prid(dp);
1133 * Make sure that we have allocated dquot(s) on disk.
1135 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
1136 mapped_fsgid(idmap, &init_user_ns), prid,
1137 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1138 &udqp, &gdqp, &pdqp);
1142 resblks = XFS_IALLOC_SPACE_RES(mp);
1143 tres = &M_RES(mp)->tr_create_tmpfile;
1145 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1148 goto out_release_dquots;
1150 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1152 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1153 0, 0, prid, false, &ip);
1155 goto out_trans_cancel;
1157 if (xfs_has_wsync(mp))
1158 xfs_trans_set_sync(tp);
1161 * Attach the dquot(s) to the inodes and modify them incore.
1162 * These ids of the inode couldn't have changed since the new
1163 * inode has been locked ever since it was created.
1165 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1167 error = xfs_iunlink(tp, ip);
1169 goto out_trans_cancel;
1171 error = xfs_trans_commit(tp);
1173 goto out_release_inode;
1175 xfs_qm_dqrele(udqp);
1176 xfs_qm_dqrele(gdqp);
1177 xfs_qm_dqrele(pdqp);
1183 xfs_trans_cancel(tp);
1186 * Wait until after the current transaction is aborted to finish the
1187 * setup of the inode and release the inode. This prevents recursive
1188 * transactions and deadlocks from xfs_inactive.
1191 xfs_finish_inode_setup(ip);
1195 xfs_qm_dqrele(udqp);
1196 xfs_qm_dqrele(gdqp);
1197 xfs_qm_dqrele(pdqp);
1206 struct xfs_name *target_name)
1208 xfs_mount_t *mp = tdp->i_mount;
1210 int error, nospace_error = 0;
1213 trace_xfs_link(tdp, target_name);
1215 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1217 if (xfs_is_shutdown(mp))
1219 if (xfs_ifork_zapped(tdp, XFS_DATA_FORK))
1222 error = xfs_qm_dqattach(sip);
1226 error = xfs_qm_dqattach(tdp);
1230 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1231 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1232 &tp, &nospace_error);
1237 * If we are using project inheritance, we only allow hard link
1238 * creation in our tree when the project IDs are the same; else
1239 * the tree quota mechanism could be circumvented.
1241 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1242 tdp->i_projid != sip->i_projid)) {
1248 error = xfs_dir_canenter(tp, tdp, target_name);
1254 * Handle initial link state of O_TMPFILE inode
1256 if (VFS_I(sip)->i_nlink == 0) {
1257 struct xfs_perag *pag;
1259 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1260 error = xfs_iunlink_remove(tp, pag, sip);
1266 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1270 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1271 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1273 xfs_bumplink(tp, sip);
1276 * If this is a synchronous mount, make sure that the
1277 * link transaction goes to disk before returning to
1280 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1281 xfs_trans_set_sync(tp);
1283 return xfs_trans_commit(tp);
1286 xfs_trans_cancel(tp);
1288 if (error == -ENOSPC && nospace_error)
1289 error = nospace_error;
1293 /* Clear the reflink flag and the cowblocks tag if possible. */
1295 xfs_itruncate_clear_reflink_flags(
1296 struct xfs_inode *ip)
1298 struct xfs_ifork *dfork;
1299 struct xfs_ifork *cfork;
1301 if (!xfs_is_reflink_inode(ip))
1303 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1304 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1305 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1306 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1307 if (cfork->if_bytes == 0)
1308 xfs_inode_clear_cowblocks_tag(ip);
1312 * Free up the underlying blocks past new_size. The new size must be smaller
1313 * than the current size. This routine can be used both for the attribute and
1314 * data fork, and does not modify the inode size, which is left to the caller.
1316 * The transaction passed to this routine must have made a permanent log
1317 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1318 * given transaction and start new ones, so make sure everything involved in
1319 * the transaction is tidy before calling here. Some transaction will be
1320 * returned to the caller to be committed. The incoming transaction must
1321 * already include the inode, and both inode locks must be held exclusively.
1322 * The inode must also be "held" within the transaction. On return the inode
1323 * will be "held" within the returned transaction. This routine does NOT
1324 * require any disk space to be reserved for it within the transaction.
1326 * If we get an error, we must return with the inode locked and linked into the
1327 * current transaction. This keeps things simple for the higher level code,
1328 * because it always knows that the inode is locked and held in the transaction
1329 * that returns to it whether errors occur or not. We don't mark the inode
1330 * dirty on error so that transactions can be easily aborted if possible.
1333 xfs_itruncate_extents_flags(
1334 struct xfs_trans **tpp,
1335 struct xfs_inode *ip,
1337 xfs_fsize_t new_size,
1340 struct xfs_mount *mp = ip->i_mount;
1341 struct xfs_trans *tp = *tpp;
1342 xfs_fileoff_t first_unmap_block;
1345 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1346 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1347 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1348 ASSERT(new_size <= XFS_ISIZE(ip));
1349 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1350 ASSERT(ip->i_itemp != NULL);
1351 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1352 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1354 trace_xfs_itruncate_extents_start(ip, new_size);
1356 flags |= xfs_bmapi_aflag(whichfork);
1359 * Since it is possible for space to become allocated beyond
1360 * the end of the file (in a crash where the space is allocated
1361 * but the inode size is not yet updated), simply remove any
1362 * blocks which show up between the new EOF and the maximum
1363 * possible file size.
1365 * We have to free all the blocks to the bmbt maximum offset, even if
1366 * the page cache can't scale that far.
1368 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1369 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1370 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1374 error = xfs_bunmapi_range(&tp, ip, flags, first_unmap_block,
1379 if (whichfork == XFS_DATA_FORK) {
1380 /* Remove all pending CoW reservations. */
1381 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1382 first_unmap_block, XFS_MAX_FILEOFF, true);
1386 xfs_itruncate_clear_reflink_flags(ip);
1390 * Always re-log the inode so that our permanent transaction can keep
1391 * on rolling it forward in the log.
1393 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1395 trace_xfs_itruncate_extents_end(ip, new_size);
1406 xfs_mount_t *mp = ip->i_mount;
1409 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1412 /* If this is a read-only mount, don't do this (would generate I/O) */
1413 if (xfs_is_readonly(mp))
1416 if (!xfs_is_shutdown(mp)) {
1420 * If we previously truncated this file and removed old data
1421 * in the process, we want to initiate "early" writeout on
1422 * the last close. This is an attempt to combat the notorious
1423 * NULL files problem which is particularly noticeable from a
1424 * truncate down, buffered (re-)write (delalloc), followed by
1425 * a crash. What we are effectively doing here is
1426 * significantly reducing the time window where we'd otherwise
1427 * be exposed to that problem.
1429 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1431 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1432 if (ip->i_delayed_blks > 0) {
1433 error = filemap_flush(VFS_I(ip)->i_mapping);
1440 if (VFS_I(ip)->i_nlink == 0)
1444 * If we can't get the iolock just skip truncating the blocks past EOF
1445 * because we could deadlock with the mmap_lock otherwise. We'll get
1446 * another chance to drop them once the last reference to the inode is
1447 * dropped, so we'll never leak blocks permanently.
1449 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1452 if (xfs_can_free_eofblocks(ip, false)) {
1454 * Check if the inode is being opened, written and closed
1455 * frequently and we have delayed allocation blocks outstanding
1456 * (e.g. streaming writes from the NFS server), truncating the
1457 * blocks past EOF will cause fragmentation to occur.
1459 * In this case don't do the truncation, but we have to be
1460 * careful how we detect this case. Blocks beyond EOF show up as
1461 * i_delayed_blks even when the inode is clean, so we need to
1462 * truncate them away first before checking for a dirty release.
1463 * Hence on the first dirty close we will still remove the
1464 * speculative allocation, but after that we will leave it in
1467 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1470 error = xfs_free_eofblocks(ip);
1474 /* delalloc blocks after truncation means it really is dirty */
1475 if (ip->i_delayed_blks)
1476 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1480 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1485 * xfs_inactive_truncate
1487 * Called to perform a truncate when an inode becomes unlinked.
1490 xfs_inactive_truncate(
1491 struct xfs_inode *ip)
1493 struct xfs_mount *mp = ip->i_mount;
1494 struct xfs_trans *tp;
1497 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1499 ASSERT(xfs_is_shutdown(mp));
1502 xfs_ilock(ip, XFS_ILOCK_EXCL);
1503 xfs_trans_ijoin(tp, ip, 0);
1506 * Log the inode size first to prevent stale data exposure in the event
1507 * of a system crash before the truncate completes. See the related
1508 * comment in xfs_vn_setattr_size() for details.
1510 ip->i_disk_size = 0;
1511 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1513 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1515 goto error_trans_cancel;
1517 ASSERT(ip->i_df.if_nextents == 0);
1519 error = xfs_trans_commit(tp);
1523 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527 xfs_trans_cancel(tp);
1529 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1534 * xfs_inactive_ifree()
1536 * Perform the inode free when an inode is unlinked.
1540 struct xfs_inode *ip)
1542 struct xfs_mount *mp = ip->i_mount;
1543 struct xfs_trans *tp;
1547 * We try to use a per-AG reservation for any block needed by the finobt
1548 * tree, but as the finobt feature predates the per-AG reservation
1549 * support a degraded file system might not have enough space for the
1550 * reservation at mount time. In that case try to dip into the reserved
1553 * Send a warning if the reservation does happen to fail, as the inode
1554 * now remains allocated and sits on the unlinked list until the fs is
1557 if (unlikely(mp->m_finobt_nores)) {
1558 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1559 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1562 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1565 if (error == -ENOSPC) {
1566 xfs_warn_ratelimited(mp,
1567 "Failed to remove inode(s) from unlinked list. "
1568 "Please free space, unmount and run xfs_repair.");
1570 ASSERT(xfs_is_shutdown(mp));
1576 * We do not hold the inode locked across the entire rolling transaction
1577 * here. We only need to hold it for the first transaction that
1578 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1579 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1580 * here breaks the relationship between cluster buffer invalidation and
1581 * stale inode invalidation on cluster buffer item journal commit
1582 * completion, and can result in leaving dirty stale inodes hanging
1585 * We have no need for serialising this inode operation against other
1586 * operations - we freed the inode and hence reallocation is required
1587 * and that will serialise on reallocating the space the deferops need
1588 * to free. Hence we can unlock the inode on the first commit of
1589 * the transaction rather than roll it right through the deferops. This
1590 * avoids relogging the XFS_ISTALE inode.
1592 * We check that xfs_ifree() hasn't grown an internal transaction roll
1593 * by asserting that the inode is still locked when it returns.
1595 xfs_ilock(ip, XFS_ILOCK_EXCL);
1596 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1598 error = xfs_ifree(tp, ip);
1599 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1602 * If we fail to free the inode, shut down. The cancel
1603 * might do that, we need to make sure. Otherwise the
1604 * inode might be lost for a long time or forever.
1606 if (!xfs_is_shutdown(mp)) {
1607 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1609 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1611 xfs_trans_cancel(tp);
1616 * Credit the quota account(s). The inode is gone.
1618 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1620 return xfs_trans_commit(tp);
1624 * Returns true if we need to update the on-disk metadata before we can free
1625 * the memory used by this inode. Updates include freeing post-eof
1626 * preallocations; freeing COW staging extents; and marking the inode free in
1627 * the inobt if it is on the unlinked list.
1630 xfs_inode_needs_inactive(
1631 struct xfs_inode *ip)
1633 struct xfs_mount *mp = ip->i_mount;
1634 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1637 * If the inode is already free, then there can be nothing
1640 if (VFS_I(ip)->i_mode == 0)
1644 * If this is a read-only mount, don't do this (would generate I/O)
1645 * unless we're in log recovery and cleaning the iunlinked list.
1647 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1650 /* If the log isn't running, push inodes straight to reclaim. */
1651 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1654 /* Metadata inodes require explicit resource cleanup. */
1655 if (xfs_is_metadata_inode(ip))
1658 /* Want to clean out the cow blocks if there are any. */
1659 if (cow_ifp && cow_ifp->if_bytes > 0)
1662 /* Unlinked files must be freed. */
1663 if (VFS_I(ip)->i_nlink == 0)
1667 * This file isn't being freed, so check if there are post-eof blocks
1668 * to free. @force is true because we are evicting an inode from the
1669 * cache. Post-eof blocks must be freed, lest we end up with broken
1670 * free space accounting.
1672 * Note: don't bother with iolock here since lockdep complains about
1673 * acquiring it in reclaim context. We have the only reference to the
1674 * inode at this point anyways.
1676 return xfs_can_free_eofblocks(ip, true);
1682 * This is called when the vnode reference count for the vnode
1683 * goes to zero. If the file has been unlinked, then it must
1684 * now be truncated. Also, we clear all of the read-ahead state
1685 * kept for the inode here since the file is now closed.
1691 struct xfs_mount *mp;
1696 * If the inode is already free, then there can be nothing
1699 if (VFS_I(ip)->i_mode == 0) {
1700 ASSERT(ip->i_df.if_broot_bytes == 0);
1705 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1708 * If this is a read-only mount, don't do this (would generate I/O)
1709 * unless we're in log recovery and cleaning the iunlinked list.
1711 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1714 /* Metadata inodes require explicit resource cleanup. */
1715 if (xfs_is_metadata_inode(ip))
1718 /* Try to clean out the cow blocks if there are any. */
1719 if (xfs_inode_has_cow_data(ip))
1720 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1722 if (VFS_I(ip)->i_nlink != 0) {
1724 * force is true because we are evicting an inode from the
1725 * cache. Post-eof blocks must be freed, lest we end up with
1726 * broken free space accounting.
1728 * Note: don't bother with iolock here since lockdep complains
1729 * about acquiring it in reclaim context. We have the only
1730 * reference to the inode at this point anyways.
1732 if (xfs_can_free_eofblocks(ip, true))
1733 error = xfs_free_eofblocks(ip);
1738 if (S_ISREG(VFS_I(ip)->i_mode) &&
1739 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1740 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1743 if (xfs_iflags_test(ip, XFS_IQUOTAUNCHECKED)) {
1745 * If this inode is being inactivated during a quotacheck and
1746 * has not yet been scanned by quotacheck, we /must/ remove
1747 * the dquots from the inode before inactivation changes the
1748 * block and inode counts. Most probably this is a result of
1749 * reloading the incore iunlinked list to purge unrecovered
1752 xfs_qm_dqdetach(ip);
1754 error = xfs_qm_dqattach(ip);
1759 if (S_ISLNK(VFS_I(ip)->i_mode))
1760 error = xfs_inactive_symlink(ip);
1762 error = xfs_inactive_truncate(ip);
1767 * If there are attributes associated with the file then blow them away
1768 * now. The code calls a routine that recursively deconstructs the
1769 * attribute fork. If also blows away the in-core attribute fork.
1771 if (xfs_inode_has_attr_fork(ip)) {
1772 error = xfs_attr_inactive(ip);
1777 ASSERT(ip->i_forkoff == 0);
1782 error = xfs_inactive_ifree(ip);
1786 * We're done making metadata updates for this inode, so we can release
1787 * the attached dquots.
1789 xfs_qm_dqdetach(ip);
1794 * In-Core Unlinked List Lookups
1795 * =============================
1797 * Every inode is supposed to be reachable from some other piece of metadata
1798 * with the exception of the root directory. Inodes with a connection to a
1799 * file descriptor but not linked from anywhere in the on-disk directory tree
1800 * are collectively known as unlinked inodes, though the filesystem itself
1801 * maintains links to these inodes so that on-disk metadata are consistent.
1803 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1804 * header contains a number of buckets that point to an inode, and each inode
1805 * record has a pointer to the next inode in the hash chain. This
1806 * singly-linked list causes scaling problems in the iunlink remove function
1807 * because we must walk that list to find the inode that points to the inode
1808 * being removed from the unlinked hash bucket list.
1810 * Hence we keep an in-memory double linked list to link each inode on an
1811 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1812 * based lists would require having 64 list heads in the perag, one for each
1813 * list. This is expensive in terms of memory (think millions of AGs) and cache
1814 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1815 * must be referenced at the VFS level to keep them on the list and hence we
1816 * have an existence guarantee for inodes on the unlinked list.
1818 * Given we have an existence guarantee, we can use lockless inode cache lookups
1819 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1820 * for the double linked unlinked list, and we don't need any extra locking to
1821 * keep the list safe as all manipulations are done under the AGI buffer lock.
1822 * Keeping the list up to date does not require memory allocation, just finding
1823 * the XFS inode and updating the next/prev unlinked list aginos.
1827 * Find an inode on the unlinked list. This does not take references to the
1828 * inode as we have existence guarantees by holding the AGI buffer lock and that
1829 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1830 * don't find the inode in cache, then let the caller handle the situation.
1832 static struct xfs_inode *
1834 struct xfs_perag *pag,
1837 struct xfs_inode *ip;
1840 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1842 /* Caller can handle inode not being in memory. */
1848 * Inode in RCU freeing limbo should not happen. Warn about this and
1849 * let the caller handle the failure.
1851 if (WARN_ON_ONCE(!ip->i_ino)) {
1855 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1861 * Update the prev pointer of the next agino. Returns -ENOLINK if the inode
1865 xfs_iunlink_update_backref(
1866 struct xfs_perag *pag,
1867 xfs_agino_t prev_agino,
1868 xfs_agino_t next_agino)
1870 struct xfs_inode *ip;
1872 /* No update necessary if we are at the end of the list. */
1873 if (next_agino == NULLAGINO)
1876 ip = xfs_iunlink_lookup(pag, next_agino);
1880 ip->i_prev_unlinked = prev_agino;
1885 * Point the AGI unlinked bucket at an inode and log the results. The caller
1886 * is responsible for validating the old value.
1889 xfs_iunlink_update_bucket(
1890 struct xfs_trans *tp,
1891 struct xfs_perag *pag,
1892 struct xfs_buf *agibp,
1893 unsigned int bucket_index,
1894 xfs_agino_t new_agino)
1896 struct xfs_agi *agi = agibp->b_addr;
1897 xfs_agino_t old_value;
1900 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
1902 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1903 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1904 old_value, new_agino);
1907 * We should never find the head of the list already set to the value
1908 * passed in because either we're adding or removing ourselves from the
1911 if (old_value == new_agino) {
1912 xfs_buf_mark_corrupt(agibp);
1913 return -EFSCORRUPTED;
1916 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
1917 offset = offsetof(struct xfs_agi, agi_unlinked) +
1918 (sizeof(xfs_agino_t) * bucket_index);
1919 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
1924 * Load the inode @next_agino into the cache and set its prev_unlinked pointer
1925 * to @prev_agino. Caller must hold the AGI to synchronize with other changes
1926 * to the unlinked list.
1929 xfs_iunlink_reload_next(
1930 struct xfs_trans *tp,
1931 struct xfs_buf *agibp,
1932 xfs_agino_t prev_agino,
1933 xfs_agino_t next_agino)
1935 struct xfs_perag *pag = agibp->b_pag;
1936 struct xfs_mount *mp = pag->pag_mount;
1937 struct xfs_inode *next_ip = NULL;
1941 ASSERT(next_agino != NULLAGINO);
1945 next_ip = radix_tree_lookup(&pag->pag_ici_root, next_agino);
1946 ASSERT(next_ip == NULL);
1950 xfs_info_ratelimited(mp,
1951 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating recovery.",
1952 next_agino, pag->pag_agno);
1955 * Use an untrusted lookup just to be cautious in case the AGI has been
1956 * corrupted and now points at a free inode. That shouldn't happen,
1957 * but we'd rather shut down now since we're already running in a weird
1960 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, next_agino);
1961 error = xfs_iget(mp, tp, ino, XFS_IGET_UNTRUSTED, 0, &next_ip);
1965 /* If this is not an unlinked inode, something is very wrong. */
1966 if (VFS_I(next_ip)->i_nlink != 0) {
1967 error = -EFSCORRUPTED;
1971 next_ip->i_prev_unlinked = prev_agino;
1972 trace_xfs_iunlink_reload_next(next_ip);
1974 ASSERT(!(VFS_I(next_ip)->i_state & I_DONTCACHE));
1975 if (xfs_is_quotacheck_running(mp) && next_ip)
1976 xfs_iflags_set(next_ip, XFS_IQUOTAUNCHECKED);
1982 xfs_iunlink_insert_inode(
1983 struct xfs_trans *tp,
1984 struct xfs_perag *pag,
1985 struct xfs_buf *agibp,
1986 struct xfs_inode *ip)
1988 struct xfs_mount *mp = tp->t_mountp;
1989 struct xfs_agi *agi = agibp->b_addr;
1990 xfs_agino_t next_agino;
1991 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1992 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1996 * Get the index into the agi hash table for the list this inode will
1997 * go on. Make sure the pointer isn't garbage and that this inode
1998 * isn't already on the list.
2000 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2001 if (next_agino == agino ||
2002 !xfs_verify_agino_or_null(pag, next_agino)) {
2003 xfs_buf_mark_corrupt(agibp);
2004 return -EFSCORRUPTED;
2008 * Update the prev pointer in the next inode to point back to this
2011 error = xfs_iunlink_update_backref(pag, agino, next_agino);
2012 if (error == -ENOLINK)
2013 error = xfs_iunlink_reload_next(tp, agibp, agino, next_agino);
2017 if (next_agino != NULLAGINO) {
2019 * There is already another inode in the bucket, so point this
2020 * inode to the current head of the list.
2022 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
2025 ip->i_next_unlinked = next_agino;
2028 /* Point the head of the list to point to this inode. */
2029 ip->i_prev_unlinked = NULLAGINO;
2030 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
2034 * This is called when the inode's link count has gone to 0 or we are creating
2035 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2037 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2038 * list when the inode is freed.
2042 struct xfs_trans *tp,
2043 struct xfs_inode *ip)
2045 struct xfs_mount *mp = tp->t_mountp;
2046 struct xfs_perag *pag;
2047 struct xfs_buf *agibp;
2050 ASSERT(VFS_I(ip)->i_nlink == 0);
2051 ASSERT(VFS_I(ip)->i_mode != 0);
2052 trace_xfs_iunlink(ip);
2054 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2056 /* Get the agi buffer first. It ensures lock ordering on the list. */
2057 error = xfs_read_agi(pag, tp, &agibp);
2061 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
2068 xfs_iunlink_remove_inode(
2069 struct xfs_trans *tp,
2070 struct xfs_perag *pag,
2071 struct xfs_buf *agibp,
2072 struct xfs_inode *ip)
2074 struct xfs_mount *mp = tp->t_mountp;
2075 struct xfs_agi *agi = agibp->b_addr;
2076 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2077 xfs_agino_t head_agino;
2078 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2081 trace_xfs_iunlink_remove(ip);
2084 * Get the index into the agi hash table for the list this inode will
2085 * go on. Make sure the head pointer isn't garbage.
2087 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2088 if (!xfs_verify_agino(pag, head_agino)) {
2089 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2091 return -EFSCORRUPTED;
2095 * Set our inode's next_unlinked pointer to NULL and then return
2096 * the old pointer value so that we can update whatever was previous
2097 * to us in the list to point to whatever was next in the list.
2099 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2104 * Update the prev pointer in the next inode to point back to previous
2105 * inode in the chain.
2107 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2108 ip->i_next_unlinked);
2109 if (error == -ENOLINK)
2110 error = xfs_iunlink_reload_next(tp, agibp, ip->i_prev_unlinked,
2111 ip->i_next_unlinked);
2115 if (head_agino != agino) {
2116 struct xfs_inode *prev_ip;
2118 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2120 return -EFSCORRUPTED;
2122 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2123 ip->i_next_unlinked);
2124 prev_ip->i_next_unlinked = ip->i_next_unlinked;
2126 /* Point the head of the list to the next unlinked inode. */
2127 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2128 ip->i_next_unlinked);
2131 ip->i_next_unlinked = NULLAGINO;
2132 ip->i_prev_unlinked = 0;
2137 * Pull the on-disk inode from the AGI unlinked list.
2141 struct xfs_trans *tp,
2142 struct xfs_perag *pag,
2143 struct xfs_inode *ip)
2145 struct xfs_buf *agibp;
2148 trace_xfs_iunlink_remove(ip);
2150 /* Get the agi buffer first. It ensures lock ordering on the list. */
2151 error = xfs_read_agi(pag, tp, &agibp);
2155 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2159 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2160 * mark it stale. We should only find clean inodes in this lookup that aren't
2164 xfs_ifree_mark_inode_stale(
2165 struct xfs_perag *pag,
2166 struct xfs_inode *free_ip,
2169 struct xfs_mount *mp = pag->pag_mount;
2170 struct xfs_inode_log_item *iip;
2171 struct xfs_inode *ip;
2175 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2177 /* Inode not in memory, nothing to do */
2184 * because this is an RCU protected lookup, we could find a recently
2185 * freed or even reallocated inode during the lookup. We need to check
2186 * under the i_flags_lock for a valid inode here. Skip it if it is not
2187 * valid, the wrong inode or stale.
2189 spin_lock(&ip->i_flags_lock);
2190 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2191 goto out_iflags_unlock;
2194 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2195 * other inodes that we did not find in the list attached to the buffer
2196 * and are not already marked stale. If we can't lock it, back off and
2199 if (ip != free_ip) {
2200 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2201 spin_unlock(&ip->i_flags_lock);
2207 ip->i_flags |= XFS_ISTALE;
2210 * If the inode is flushing, it is already attached to the buffer. All
2211 * we needed to do here is mark the inode stale so buffer IO completion
2212 * will remove it from the AIL.
2215 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2216 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2217 ASSERT(iip->ili_last_fields);
2222 * Inodes not attached to the buffer can be released immediately.
2223 * Everything else has to go through xfs_iflush_abort() on journal
2224 * commit as the flock synchronises removal of the inode from the
2225 * cluster buffer against inode reclaim.
2227 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2230 __xfs_iflags_set(ip, XFS_IFLUSHING);
2231 spin_unlock(&ip->i_flags_lock);
2234 /* we have a dirty inode in memory that has not yet been flushed. */
2235 spin_lock(&iip->ili_lock);
2236 iip->ili_last_fields = iip->ili_fields;
2237 iip->ili_fields = 0;
2238 iip->ili_fsync_fields = 0;
2239 spin_unlock(&iip->ili_lock);
2240 ASSERT(iip->ili_last_fields);
2243 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2248 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2250 spin_unlock(&ip->i_flags_lock);
2255 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2256 * inodes that are in memory - they all must be marked stale and attached to
2257 * the cluster buffer.
2261 struct xfs_trans *tp,
2262 struct xfs_perag *pag,
2263 struct xfs_inode *free_ip,
2264 struct xfs_icluster *xic)
2266 struct xfs_mount *mp = free_ip->i_mount;
2267 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2270 xfs_ino_t inum = xic->first_ino;
2276 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2278 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2280 * The allocation bitmap tells us which inodes of the chunk were
2281 * physically allocated. Skip the cluster if an inode falls into
2284 ioffset = inum - xic->first_ino;
2285 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2286 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2290 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2291 XFS_INO_TO_AGBNO(mp, inum));
2294 * We obtain and lock the backing buffer first in the process
2295 * here to ensure dirty inodes attached to the buffer remain in
2296 * the flushing state while we mark them stale.
2298 * If we scan the in-memory inodes first, then buffer IO can
2299 * complete before we get a lock on it, and hence we may fail
2300 * to mark all the active inodes on the buffer stale.
2302 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2303 mp->m_bsize * igeo->blocks_per_cluster,
2309 * This buffer may not have been correctly initialised as we
2310 * didn't read it from disk. That's not important because we are
2311 * only using to mark the buffer as stale in the log, and to
2312 * attach stale cached inodes on it. That means it will never be
2313 * dispatched for IO. If it is, we want to know about it, and we
2314 * want it to fail. We can acheive this by adding a write
2315 * verifier to the buffer.
2317 bp->b_ops = &xfs_inode_buf_ops;
2320 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2321 * too. This requires lookups, and will skip inodes that we've
2322 * already marked XFS_ISTALE.
2324 for (i = 0; i < igeo->inodes_per_cluster; i++)
2325 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2327 xfs_trans_stale_inode_buf(tp, bp);
2328 xfs_trans_binval(tp, bp);
2334 * This is called to return an inode to the inode free list. The inode should
2335 * already be truncated to 0 length and have no pages associated with it. This
2336 * routine also assumes that the inode is already a part of the transaction.
2338 * The on-disk copy of the inode will have been added to the list of unlinked
2339 * inodes in the AGI. We need to remove the inode from that list atomically with
2340 * respect to freeing it here.
2344 struct xfs_trans *tp,
2345 struct xfs_inode *ip)
2347 struct xfs_mount *mp = ip->i_mount;
2348 struct xfs_perag *pag;
2349 struct xfs_icluster xic = { 0 };
2350 struct xfs_inode_log_item *iip = ip->i_itemp;
2353 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2354 ASSERT(VFS_I(ip)->i_nlink == 0);
2355 ASSERT(ip->i_df.if_nextents == 0);
2356 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2357 ASSERT(ip->i_nblocks == 0);
2359 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2362 * Free the inode first so that we guarantee that the AGI lock is going
2363 * to be taken before we remove the inode from the unlinked list. This
2364 * makes the AGI lock -> unlinked list modification order the same as
2365 * used in O_TMPFILE creation.
2367 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2371 error = xfs_iunlink_remove(tp, pag, ip);
2376 * Free any local-format data sitting around before we reset the
2377 * data fork to extents format. Note that the attr fork data has
2378 * already been freed by xfs_attr_inactive.
2380 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2381 kmem_free(ip->i_df.if_data);
2382 ip->i_df.if_data = NULL;
2383 ip->i_df.if_bytes = 0;
2386 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2388 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2389 ip->i_forkoff = 0; /* mark the attr fork not in use */
2390 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2391 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2392 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2394 /* Don't attempt to replay owner changes for a deleted inode */
2395 spin_lock(&iip->ili_lock);
2396 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2397 spin_unlock(&iip->ili_lock);
2400 * Bump the generation count so no one will be confused
2401 * by reincarnations of this inode.
2403 VFS_I(ip)->i_generation++;
2404 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2407 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2414 * This is called to unpin an inode. The caller must have the inode locked
2415 * in at least shared mode so that the buffer cannot be subsequently pinned
2416 * once someone is waiting for it to be unpinned.
2420 struct xfs_inode *ip)
2422 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2424 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2426 /* Give the log a push to start the unpinning I/O */
2427 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2433 struct xfs_inode *ip)
2435 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2436 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2441 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2442 if (xfs_ipincount(ip))
2444 } while (xfs_ipincount(ip));
2445 finish_wait(wq, &wait.wq_entry);
2450 struct xfs_inode *ip)
2452 if (xfs_ipincount(ip))
2453 __xfs_iunpin_wait(ip);
2457 * Removing an inode from the namespace involves removing the directory entry
2458 * and dropping the link count on the inode. Removing the directory entry can
2459 * result in locking an AGF (directory blocks were freed) and removing a link
2460 * count can result in placing the inode on an unlinked list which results in
2463 * The big problem here is that we have an ordering constraint on AGF and AGI
2464 * locking - inode allocation locks the AGI, then can allocate a new extent for
2465 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2466 * removes the inode from the unlinked list, requiring that we lock the AGI
2467 * first, and then freeing the inode can result in an inode chunk being freed
2468 * and hence freeing disk space requiring that we lock an AGF.
2470 * Hence the ordering that is imposed by other parts of the code is AGI before
2471 * AGF. This means we cannot remove the directory entry before we drop the inode
2472 * reference count and put it on the unlinked list as this results in a lock
2473 * order of AGF then AGI, and this can deadlock against inode allocation and
2474 * freeing. Therefore we must drop the link counts before we remove the
2477 * This is still safe from a transactional point of view - it is not until we
2478 * get to xfs_defer_finish() that we have the possibility of multiple
2479 * transactions in this operation. Hence as long as we remove the directory
2480 * entry and drop the link count in the first transaction of the remove
2481 * operation, there are no transactional constraints on the ordering here.
2486 struct xfs_name *name,
2489 xfs_mount_t *mp = dp->i_mount;
2490 xfs_trans_t *tp = NULL;
2491 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2496 trace_xfs_remove(dp, name);
2498 if (xfs_is_shutdown(mp))
2500 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
2503 error = xfs_qm_dqattach(dp);
2507 error = xfs_qm_dqattach(ip);
2512 * We try to get the real space reservation first, allowing for
2513 * directory btree deletion(s) implying possible bmap insert(s). If we
2514 * can't get the space reservation then we use 0 instead, and avoid the
2515 * bmap btree insert(s) in the directory code by, if the bmap insert
2516 * tries to happen, instead trimming the LAST block from the directory.
2518 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2519 * the directory code can handle a reservationless update and we don't
2520 * want to prevent a user from trying to free space by deleting things.
2522 resblks = XFS_REMOVE_SPACE_RES(mp);
2523 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2526 ASSERT(error != -ENOSPC);
2531 * If we're removing a directory perform some additional validation.
2534 ASSERT(VFS_I(ip)->i_nlink >= 2);
2535 if (VFS_I(ip)->i_nlink != 2) {
2537 goto out_trans_cancel;
2539 if (!xfs_dir_isempty(ip)) {
2541 goto out_trans_cancel;
2544 /* Drop the link from ip's "..". */
2545 error = xfs_droplink(tp, dp);
2547 goto out_trans_cancel;
2549 /* Drop the "." link from ip to self. */
2550 error = xfs_droplink(tp, ip);
2552 goto out_trans_cancel;
2555 * Point the unlinked child directory's ".." entry to the root
2556 * directory to eliminate back-references to inodes that may
2557 * get freed before the child directory is closed. If the fs
2558 * gets shrunk, this can lead to dirent inode validation errors.
2560 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2561 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2562 tp->t_mountp->m_sb.sb_rootino, 0);
2564 goto out_trans_cancel;
2568 * When removing a non-directory we need to log the parent
2569 * inode here. For a directory this is done implicitly
2570 * by the xfs_droplink call for the ".." entry.
2572 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2574 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2576 /* Drop the link from dp to ip. */
2577 error = xfs_droplink(tp, ip);
2579 goto out_trans_cancel;
2581 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2583 ASSERT(error != -ENOENT);
2584 goto out_trans_cancel;
2588 * If this is a synchronous mount, make sure that the
2589 * remove transaction goes to disk before returning to
2592 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2593 xfs_trans_set_sync(tp);
2595 error = xfs_trans_commit(tp);
2599 if (is_dir && xfs_inode_is_filestream(ip))
2600 xfs_filestream_deassociate(ip);
2605 xfs_trans_cancel(tp);
2611 * Enter all inodes for a rename transaction into a sorted array.
2613 #define __XFS_SORT_INODES 5
2615 xfs_sort_for_rename(
2616 struct xfs_inode *dp1, /* in: old (source) directory inode */
2617 struct xfs_inode *dp2, /* in: new (target) directory inode */
2618 struct xfs_inode *ip1, /* in: inode of old entry */
2619 struct xfs_inode *ip2, /* in: inode of new entry */
2620 struct xfs_inode *wip, /* in: whiteout inode */
2621 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2622 int *num_inodes) /* in/out: inodes in array */
2626 ASSERT(*num_inodes == __XFS_SORT_INODES);
2627 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2630 * i_tab contains a list of pointers to inodes. We initialize
2631 * the table here & we'll sort it. We will then use it to
2632 * order the acquisition of the inode locks.
2634 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2647 * Sort the elements via bubble sort. (Remember, there are at
2648 * most 5 elements to sort, so this is adequate.)
2650 for (i = 0; i < *num_inodes; i++) {
2651 for (j = 1; j < *num_inodes; j++) {
2652 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2653 struct xfs_inode *temp = i_tab[j];
2654 i_tab[j] = i_tab[j-1];
2663 struct xfs_trans *tp)
2666 * If this is a synchronous mount, make sure that the rename transaction
2667 * goes to disk before returning to the user.
2669 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2670 xfs_trans_set_sync(tp);
2672 return xfs_trans_commit(tp);
2676 * xfs_cross_rename()
2678 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2682 struct xfs_trans *tp,
2683 struct xfs_inode *dp1,
2684 struct xfs_name *name1,
2685 struct xfs_inode *ip1,
2686 struct xfs_inode *dp2,
2687 struct xfs_name *name2,
2688 struct xfs_inode *ip2,
2696 /* Swap inode number for dirent in first parent */
2697 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2699 goto out_trans_abort;
2701 /* Swap inode number for dirent in second parent */
2702 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2704 goto out_trans_abort;
2707 * If we're renaming one or more directories across different parents,
2708 * update the respective ".." entries (and link counts) to match the new
2712 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2714 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2715 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2716 dp1->i_ino, spaceres);
2718 goto out_trans_abort;
2720 /* transfer ip2 ".." reference to dp1 */
2721 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2722 error = xfs_droplink(tp, dp2);
2724 goto out_trans_abort;
2725 xfs_bumplink(tp, dp1);
2729 * Although ip1 isn't changed here, userspace needs
2730 * to be warned about the change, so that applications
2731 * relying on it (like backup ones), will properly
2734 ip1_flags |= XFS_ICHGTIME_CHG;
2735 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2738 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2739 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2740 dp2->i_ino, spaceres);
2742 goto out_trans_abort;
2744 /* transfer ip1 ".." reference to dp2 */
2745 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2746 error = xfs_droplink(tp, dp1);
2748 goto out_trans_abort;
2749 xfs_bumplink(tp, dp2);
2753 * Although ip2 isn't changed here, userspace needs
2754 * to be warned about the change, so that applications
2755 * relying on it (like backup ones), will properly
2758 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2759 ip2_flags |= XFS_ICHGTIME_CHG;
2764 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2765 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2768 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2769 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2772 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2773 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2775 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2776 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2777 return xfs_finish_rename(tp);
2780 xfs_trans_cancel(tp);
2785 * xfs_rename_alloc_whiteout()
2787 * Return a referenced, unlinked, unlocked inode that can be used as a
2788 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2789 * crash between allocating the inode and linking it into the rename transaction
2790 * recovery will free the inode and we won't leak it.
2793 xfs_rename_alloc_whiteout(
2794 struct mnt_idmap *idmap,
2795 struct xfs_name *src_name,
2796 struct xfs_inode *dp,
2797 struct xfs_inode **wip)
2799 struct xfs_inode *tmpfile;
2803 error = xfs_create_tmpfile(idmap, dp, S_IFCHR | WHITEOUT_MODE,
2808 name.name = src_name->name;
2809 name.len = src_name->len;
2810 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2812 xfs_finish_inode_setup(tmpfile);
2818 * Prepare the tmpfile inode as if it were created through the VFS.
2819 * Complete the inode setup and flag it as linkable. nlink is already
2820 * zero, so we can skip the drop_nlink.
2822 xfs_setup_iops(tmpfile);
2823 xfs_finish_inode_setup(tmpfile);
2824 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2835 struct mnt_idmap *idmap,
2836 struct xfs_inode *src_dp,
2837 struct xfs_name *src_name,
2838 struct xfs_inode *src_ip,
2839 struct xfs_inode *target_dp,
2840 struct xfs_name *target_name,
2841 struct xfs_inode *target_ip,
2844 struct xfs_mount *mp = src_dp->i_mount;
2845 struct xfs_trans *tp;
2846 struct xfs_inode *wip = NULL; /* whiteout inode */
2847 struct xfs_inode *inodes[__XFS_SORT_INODES];
2849 int num_inodes = __XFS_SORT_INODES;
2850 bool new_parent = (src_dp != target_dp);
2851 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2853 bool retried = false;
2854 int error, nospace_error = 0;
2856 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2858 if ((flags & RENAME_EXCHANGE) && !target_ip)
2862 * If we are doing a whiteout operation, allocate the whiteout inode
2863 * we will be placing at the target and ensure the type is set
2866 if (flags & RENAME_WHITEOUT) {
2867 error = xfs_rename_alloc_whiteout(idmap, src_name,
2872 /* setup target dirent info as whiteout */
2873 src_name->type = XFS_DIR3_FT_CHRDEV;
2876 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2877 inodes, &num_inodes);
2881 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2882 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2883 if (error == -ENOSPC) {
2884 nospace_error = error;
2886 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2890 goto out_release_wip;
2893 * Attach the dquots to the inodes
2895 error = xfs_qm_vop_rename_dqattach(inodes);
2897 goto out_trans_cancel;
2900 * Lock all the participating inodes. Depending upon whether
2901 * the target_name exists in the target directory, and
2902 * whether the target directory is the same as the source
2903 * directory, we can lock from 2 to 5 inodes.
2905 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2908 * Join all the inodes to the transaction. From this point on,
2909 * we can rely on either trans_commit or trans_cancel to unlock
2912 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2914 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2915 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2917 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2919 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2922 * If we are using project inheritance, we only allow renames
2923 * into our tree when the project IDs are the same; else the
2924 * tree quota mechanism would be circumvented.
2926 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2927 target_dp->i_projid != src_ip->i_projid)) {
2929 goto out_trans_cancel;
2932 /* RENAME_EXCHANGE is unique from here on. */
2933 if (flags & RENAME_EXCHANGE)
2934 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2935 target_dp, target_name, target_ip,
2939 * Try to reserve quota to handle an expansion of the target directory.
2940 * We'll allow the rename to continue in reservationless mode if we hit
2941 * a space usage constraint. If we trigger reservationless mode, save
2942 * the errno if there isn't any free space in the target directory.
2944 if (spaceres != 0) {
2945 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2947 if (error == -EDQUOT || error == -ENOSPC) {
2949 xfs_trans_cancel(tp);
2950 xfs_blockgc_free_quota(target_dp, 0);
2955 nospace_error = error;
2960 goto out_trans_cancel;
2964 * Check for expected errors before we dirty the transaction
2965 * so we can return an error without a transaction abort.
2967 if (target_ip == NULL) {
2969 * If there's no space reservation, check the entry will
2970 * fit before actually inserting it.
2973 error = xfs_dir_canenter(tp, target_dp, target_name);
2975 goto out_trans_cancel;
2979 * If target exists and it's a directory, check that whether
2980 * it can be destroyed.
2982 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
2983 (!xfs_dir_isempty(target_ip) ||
2984 (VFS_I(target_ip)->i_nlink > 2))) {
2986 goto out_trans_cancel;
2991 * Lock the AGI buffers we need to handle bumping the nlink of the
2992 * whiteout inode off the unlinked list and to handle dropping the
2993 * nlink of the target inode. Per locking order rules, do this in
2994 * increasing AG order and before directory block allocation tries to
2995 * grab AGFs because we grab AGIs before AGFs.
2997 * The (vfs) caller must ensure that if src is a directory then
2998 * target_ip is either null or an empty directory.
3000 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3001 if (inodes[i] == wip ||
3002 (inodes[i] == target_ip &&
3003 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3004 struct xfs_perag *pag;
3007 pag = xfs_perag_get(mp,
3008 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
3009 error = xfs_read_agi(pag, tp, &bp);
3012 goto out_trans_cancel;
3017 * Directory entry creation below may acquire the AGF. Remove
3018 * the whiteout from the unlinked list first to preserve correct
3019 * AGI/AGF locking order. This dirties the transaction so failures
3020 * after this point will abort and log recovery will clean up the
3023 * For whiteouts, we need to bump the link count on the whiteout
3024 * inode. After this point, we have a real link, clear the tmpfile
3025 * state flag from the inode so it doesn't accidentally get misused
3029 struct xfs_perag *pag;
3031 ASSERT(VFS_I(wip)->i_nlink == 0);
3033 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
3034 error = xfs_iunlink_remove(tp, pag, wip);
3037 goto out_trans_cancel;
3039 xfs_bumplink(tp, wip);
3040 VFS_I(wip)->i_state &= ~I_LINKABLE;
3044 * Set up the target.
3046 if (target_ip == NULL) {
3048 * If target does not exist and the rename crosses
3049 * directories, adjust the target directory link count
3050 * to account for the ".." reference from the new entry.
3052 error = xfs_dir_createname(tp, target_dp, target_name,
3053 src_ip->i_ino, spaceres);
3055 goto out_trans_cancel;
3057 xfs_trans_ichgtime(tp, target_dp,
3058 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3060 if (new_parent && src_is_directory) {
3061 xfs_bumplink(tp, target_dp);
3063 } else { /* target_ip != NULL */
3065 * Link the source inode under the target name.
3066 * If the source inode is a directory and we are moving
3067 * it across directories, its ".." entry will be
3068 * inconsistent until we replace that down below.
3070 * In case there is already an entry with the same
3071 * name at the destination directory, remove it first.
3073 error = xfs_dir_replace(tp, target_dp, target_name,
3074 src_ip->i_ino, spaceres);
3076 goto out_trans_cancel;
3078 xfs_trans_ichgtime(tp, target_dp,
3079 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3082 * Decrement the link count on the target since the target
3083 * dir no longer points to it.
3085 error = xfs_droplink(tp, target_ip);
3087 goto out_trans_cancel;
3089 if (src_is_directory) {
3091 * Drop the link from the old "." entry.
3093 error = xfs_droplink(tp, target_ip);
3095 goto out_trans_cancel;
3097 } /* target_ip != NULL */
3100 * Remove the source.
3102 if (new_parent && src_is_directory) {
3104 * Rewrite the ".." entry to point to the new
3107 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3108 target_dp->i_ino, spaceres);
3109 ASSERT(error != -EEXIST);
3111 goto out_trans_cancel;
3115 * We always want to hit the ctime on the source inode.
3117 * This isn't strictly required by the standards since the source
3118 * inode isn't really being changed, but old unix file systems did
3119 * it and some incremental backup programs won't work without it.
3121 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3122 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3125 * Adjust the link count on src_dp. This is necessary when
3126 * renaming a directory, either within one parent when
3127 * the target existed, or across two parent directories.
3129 if (src_is_directory && (new_parent || target_ip != NULL)) {
3132 * Decrement link count on src_directory since the
3133 * entry that's moved no longer points to it.
3135 error = xfs_droplink(tp, src_dp);
3137 goto out_trans_cancel;
3141 * For whiteouts, we only need to update the source dirent with the
3142 * inode number of the whiteout inode rather than removing it
3146 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3149 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3153 goto out_trans_cancel;
3155 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3156 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3158 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3160 error = xfs_finish_rename(tp);
3166 xfs_trans_cancel(tp);
3170 if (error == -ENOSPC && nospace_error)
3171 error = nospace_error;
3177 struct xfs_inode *ip,
3180 struct xfs_inode_log_item *iip = ip->i_itemp;
3181 struct xfs_dinode *dip;
3182 struct xfs_mount *mp = ip->i_mount;
3185 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3186 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3187 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3188 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3189 ASSERT(iip->ili_item.li_buf == bp);
3191 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3194 * We don't flush the inode if any of the following checks fail, but we
3195 * do still update the log item and attach to the backing buffer as if
3196 * the flush happened. This is a formality to facilitate predictable
3197 * error handling as the caller will shutdown and fail the buffer.
3199 error = -EFSCORRUPTED;
3200 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3201 mp, XFS_ERRTAG_IFLUSH_1)) {
3202 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3203 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3204 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3207 if (S_ISREG(VFS_I(ip)->i_mode)) {
3209 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3210 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3211 mp, XFS_ERRTAG_IFLUSH_3)) {
3212 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3213 "%s: Bad regular inode %llu, ptr "PTR_FMT,
3214 __func__, ip->i_ino, ip);
3217 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3219 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3220 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3221 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3222 mp, XFS_ERRTAG_IFLUSH_4)) {
3223 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3224 "%s: Bad directory inode %llu, ptr "PTR_FMT,
3225 __func__, ip->i_ino, ip);
3229 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3230 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3231 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3232 "%s: detected corrupt incore inode %llu, "
3233 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3234 __func__, ip->i_ino,
3235 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3239 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3240 mp, XFS_ERRTAG_IFLUSH_6)) {
3241 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3242 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3243 __func__, ip->i_ino, ip->i_forkoff, ip);
3248 * Inode item log recovery for v2 inodes are dependent on the flushiter
3249 * count for correct sequencing. We bump the flush iteration count so
3250 * we can detect flushes which postdate a log record during recovery.
3251 * This is redundant as we now log every change and hence this can't
3252 * happen but we need to still do it to ensure backwards compatibility
3253 * with old kernels that predate logging all inode changes.
3255 if (!xfs_has_v3inodes(mp))
3259 * If there are inline format data / attr forks attached to this inode,
3260 * make sure they are not corrupt.
3262 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3263 xfs_ifork_verify_local_data(ip))
3265 if (xfs_inode_has_attr_fork(ip) &&
3266 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3267 xfs_ifork_verify_local_attr(ip))
3271 * Copy the dirty parts of the inode into the on-disk inode. We always
3272 * copy out the core of the inode, because if the inode is dirty at all
3275 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3277 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3278 if (!xfs_has_v3inodes(mp)) {
3279 if (ip->i_flushiter == DI_MAX_FLUSH)
3280 ip->i_flushiter = 0;
3283 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3284 if (xfs_inode_has_attr_fork(ip))
3285 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3288 * We've recorded everything logged in the inode, so we'd like to clear
3289 * the ili_fields bits so we don't log and flush things unnecessarily.
3290 * However, we can't stop logging all this information until the data
3291 * we've copied into the disk buffer is written to disk. If we did we
3292 * might overwrite the copy of the inode in the log with all the data
3293 * after re-logging only part of it, and in the face of a crash we
3294 * wouldn't have all the data we need to recover.
3296 * What we do is move the bits to the ili_last_fields field. When
3297 * logging the inode, these bits are moved back to the ili_fields field.
3298 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3299 * we know that the information those bits represent is permanently on
3300 * disk. As long as the flush completes before the inode is logged
3301 * again, then both ili_fields and ili_last_fields will be cleared.
3305 spin_lock(&iip->ili_lock);
3306 iip->ili_last_fields = iip->ili_fields;
3307 iip->ili_fields = 0;
3308 iip->ili_fsync_fields = 0;
3309 spin_unlock(&iip->ili_lock);
3312 * Store the current LSN of the inode so that we can tell whether the
3313 * item has moved in the AIL from xfs_buf_inode_iodone().
3315 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3316 &iip->ili_item.li_lsn);
3318 /* generate the checksum. */
3319 xfs_dinode_calc_crc(mp, dip);
3324 * Non-blocking flush of dirty inode metadata into the backing buffer.
3326 * The caller must have a reference to the inode and hold the cluster buffer
3327 * locked. The function will walk across all the inodes on the cluster buffer it
3328 * can find and lock without blocking, and flush them to the cluster buffer.
3330 * On successful flushing of at least one inode, the caller must write out the
3331 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3332 * the caller needs to release the buffer. On failure, the filesystem will be
3333 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3340 struct xfs_mount *mp = bp->b_mount;
3341 struct xfs_log_item *lip, *n;
3342 struct xfs_inode *ip;
3343 struct xfs_inode_log_item *iip;
3348 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3349 * will remove itself from the list.
3351 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3352 iip = (struct xfs_inode_log_item *)lip;
3353 ip = iip->ili_inode;
3356 * Quick and dirty check to avoid locks if possible.
3358 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3360 if (xfs_ipincount(ip))
3364 * The inode is still attached to the buffer, which means it is
3365 * dirty but reclaim might try to grab it. Check carefully for
3366 * that, and grab the ilock while still holding the i_flags_lock
3367 * to guarantee reclaim will not be able to reclaim this inode
3368 * once we drop the i_flags_lock.
3370 spin_lock(&ip->i_flags_lock);
3371 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3372 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3373 spin_unlock(&ip->i_flags_lock);
3378 * ILOCK will pin the inode against reclaim and prevent
3379 * concurrent transactions modifying the inode while we are
3380 * flushing the inode. If we get the lock, set the flushing
3381 * state before we drop the i_flags_lock.
3383 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3384 spin_unlock(&ip->i_flags_lock);
3387 __xfs_iflags_set(ip, XFS_IFLUSHING);
3388 spin_unlock(&ip->i_flags_lock);
3391 * Abort flushing this inode if we are shut down because the
3392 * inode may not currently be in the AIL. This can occur when
3393 * log I/O failure unpins the inode without inserting into the
3394 * AIL, leaving a dirty/unpinned inode attached to the buffer
3395 * that otherwise looks like it should be flushed.
3397 if (xlog_is_shutdown(mp->m_log)) {
3398 xfs_iunpin_wait(ip);
3399 xfs_iflush_abort(ip);
3400 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3405 /* don't block waiting on a log force to unpin dirty inodes */
3406 if (xfs_ipincount(ip)) {
3407 xfs_iflags_clear(ip, XFS_IFLUSHING);
3408 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3412 if (!xfs_inode_clean(ip))
3413 error = xfs_iflush(ip, bp);
3415 xfs_iflags_clear(ip, XFS_IFLUSHING);
3416 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3424 * Shutdown first so we kill the log before we release this
3425 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3426 * of the log, failing it before the _log_ is shut down can
3427 * result in the log tail being moved forward in the journal
3428 * on disk because log writes can still be taking place. Hence
3429 * unpinning the tail will allow the ICREATE intent to be
3430 * removed from the log an recovery will fail with uninitialised
3431 * inode cluster buffers.
3433 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3434 bp->b_flags |= XBF_ASYNC;
3435 xfs_buf_ioend_fail(bp);
3442 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3443 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3448 /* Release an inode. */
3451 struct xfs_inode *ip)
3453 trace_xfs_irele(ip, _RET_IP_);
3458 * Ensure all commited transactions touching the inode are written to the log.
3461 xfs_log_force_inode(
3462 struct xfs_inode *ip)
3466 xfs_ilock(ip, XFS_ILOCK_SHARED);
3467 if (xfs_ipincount(ip))
3468 seq = ip->i_itemp->ili_commit_seq;
3469 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3473 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3477 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3478 * abide vfs locking order (lowest pointer value goes first) and breaking the
3479 * layout leases before proceeding. The loop is needed because we cannot call
3480 * the blocking break_layout() with the iolocks held, and therefore have to
3481 * back out both locks.
3484 xfs_iolock_two_inodes_and_break_layout(
3494 /* Wait to break both inodes' layouts before we start locking. */
3495 error = break_layout(src, true);
3499 error = break_layout(dest, true);
3504 /* Lock one inode and make sure nobody got in and leased it. */
3506 error = break_layout(src, false);
3509 if (error == -EWOULDBLOCK)
3517 /* Lock the other inode and make sure nobody got in and leased it. */
3518 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3519 error = break_layout(dest, false);
3523 if (error == -EWOULDBLOCK)
3532 xfs_mmaplock_two_inodes_and_break_dax_layout(
3533 struct xfs_inode *ip1,
3534 struct xfs_inode *ip2)
3540 if (ip1->i_ino > ip2->i_ino)
3545 /* Lock the first inode */
3546 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3547 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3548 if (error || retry) {
3549 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3550 if (error == 0 && retry)
3558 /* Nested lock the second inode */
3559 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3561 * We cannot use xfs_break_dax_layouts() directly here because it may
3562 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3563 * for this nested lock case.
3565 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3566 if (page && page_ref_count(page) != 1) {
3567 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3568 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3576 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3581 struct xfs_inode *ip1,
3582 struct xfs_inode *ip2)
3586 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3590 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3591 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3593 inode_unlock(VFS_I(ip2));
3595 inode_unlock(VFS_I(ip1));
3599 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3600 VFS_I(ip2)->i_mapping);
3605 /* Unlock both inodes to allow IO and mmap activity. */
3607 xfs_iunlock2_io_mmap(
3608 struct xfs_inode *ip1,
3609 struct xfs_inode *ip2)
3611 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3612 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3614 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3616 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3617 VFS_I(ip2)->i_mapping);
3619 inode_unlock(VFS_I(ip2));
3621 inode_unlock(VFS_I(ip1));
3624 /* Drop the MMAPLOCK and the IOLOCK after a remap completes. */
3626 xfs_iunlock2_remapping(
3627 struct xfs_inode *ip1,
3628 struct xfs_inode *ip2)
3630 xfs_iflags_clear(ip1, XFS_IREMAPPING);
3633 xfs_iunlock(ip1, XFS_MMAPLOCK_SHARED);
3634 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3637 inode_unlock_shared(VFS_I(ip1));
3638 inode_unlock(VFS_I(ip2));
3642 * Reload the incore inode list for this inode. Caller should ensure that
3643 * the link count cannot change, either by taking ILOCK_SHARED or otherwise
3644 * preventing other threads from executing.
3647 xfs_inode_reload_unlinked_bucket(
3648 struct xfs_trans *tp,
3649 struct xfs_inode *ip)
3651 struct xfs_mount *mp = tp->t_mountp;
3652 struct xfs_buf *agibp;
3653 struct xfs_agi *agi;
3654 struct xfs_perag *pag;
3655 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
3656 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
3657 xfs_agino_t prev_agino, next_agino;
3658 unsigned int bucket;
3659 bool foundit = false;
3662 /* Grab the first inode in the list */
3663 pag = xfs_perag_get(mp, agno);
3664 error = xfs_ialloc_read_agi(pag, tp, &agibp);
3670 * We've taken ILOCK_SHARED and the AGI buffer lock to stabilize the
3671 * incore unlinked list pointers for this inode. Check once more to
3672 * see if we raced with anyone else to reload the unlinked list.
3674 if (!xfs_inode_unlinked_incomplete(ip)) {
3679 bucket = agino % XFS_AGI_UNLINKED_BUCKETS;
3680 agi = agibp->b_addr;
3682 trace_xfs_inode_reload_unlinked_bucket(ip);
3684 xfs_info_ratelimited(mp,
3685 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating list recovery.",
3688 prev_agino = NULLAGINO;
3689 next_agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3690 while (next_agino != NULLAGINO) {
3691 struct xfs_inode *next_ip = NULL;
3693 /* Found this caller's inode, set its backlink. */
3694 if (next_agino == agino) {
3696 next_ip->i_prev_unlinked = prev_agino;
3701 /* Try in-memory lookup first. */
3702 next_ip = xfs_iunlink_lookup(pag, next_agino);
3706 /* Inode not in memory, try reloading it. */
3707 error = xfs_iunlink_reload_next(tp, agibp, prev_agino,
3712 /* Grab the reloaded inode. */
3713 next_ip = xfs_iunlink_lookup(pag, next_agino);
3715 /* No incore inode at all? We reloaded it... */
3716 ASSERT(next_ip != NULL);
3717 error = -EFSCORRUPTED;
3722 prev_agino = next_agino;
3723 next_agino = next_ip->i_next_unlinked;
3727 xfs_trans_brelse(tp, agibp);
3728 /* Should have found this inode somewhere in the iunlinked bucket. */
3729 if (!error && !foundit)
3730 error = -EFSCORRUPTED;
3734 /* Decide if this inode is missing its unlinked list and reload it. */
3736 xfs_inode_reload_unlinked(
3737 struct xfs_inode *ip)
3739 struct xfs_trans *tp;
3742 error = xfs_trans_alloc_empty(ip->i_mount, &tp);
3746 xfs_ilock(ip, XFS_ILOCK_SHARED);
3747 if (xfs_inode_unlinked_incomplete(ip))
3748 error = xfs_inode_reload_unlinked_bucket(tp, ip);
3749 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3750 xfs_trans_cancel(tp);
3755 /* Has this inode fork been zapped by repair? */
3758 const struct xfs_inode *ip,
3761 unsigned int datamask = 0;
3763 switch (whichfork) {
3765 switch (ip->i_vnode.i_mode & S_IFMT) {
3767 datamask = XFS_SICK_INO_DIR_ZAPPED;
3770 datamask = XFS_SICK_INO_SYMLINK_ZAPPED;
3773 return ip->i_sick & (XFS_SICK_INO_BMBTD_ZAPPED | datamask);
3775 return ip->i_sick & XFS_SICK_INO_BMBTA_ZAPPED;