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"
41 struct kmem_cache *xfs_inode_cache;
44 * Used in xfs_itruncate_extents(). This is the maximum number of extents
45 * freed from a file in a single transaction.
47 #define XFS_ITRUNC_MAX_EXTENTS 2
49 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
50 STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
54 * helper function to extract extent size hint from inode
61 * No point in aligning allocations if we need to COW to actually
64 if (xfs_is_always_cow_inode(ip))
66 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
68 if (XFS_IS_REALTIME_INODE(ip))
69 return ip->i_mount->m_sb.sb_rextsize;
74 * Helper function to extract CoW extent size hint from inode.
75 * Between the extent size hint and the CoW extent size hint, we
76 * return the greater of the two. If the value is zero (automatic),
77 * use the default size.
80 xfs_get_cowextsz_hint(
86 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
88 b = xfs_get_extsz_hint(ip);
92 return XFS_DEFAULT_COWEXTSZ_HINT;
97 * These two are wrapper routines around the xfs_ilock() routine used to
98 * centralize some grungy code. They are used in places that wish to lock the
99 * inode solely for reading the extents. The reason these places can't just
100 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
101 * bringing in of the extents from disk for a file in b-tree format. If the
102 * inode is in b-tree format, then we need to lock the inode exclusively until
103 * the extents are read in. Locking it exclusively all the time would limit
104 * our parallelism unnecessarily, though. What we do instead is check to see
105 * if the extents have been read in yet, and only lock the inode exclusively
108 * The functions return a value which should be given to the corresponding
109 * xfs_iunlock() call.
112 xfs_ilock_data_map_shared(
113 struct xfs_inode *ip)
115 uint lock_mode = XFS_ILOCK_SHARED;
117 if (xfs_need_iread_extents(&ip->i_df))
118 lock_mode = XFS_ILOCK_EXCL;
119 xfs_ilock(ip, lock_mode);
124 xfs_ilock_attr_map_shared(
125 struct xfs_inode *ip)
127 uint lock_mode = XFS_ILOCK_SHARED;
129 if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
136 * You can't set both SHARED and EXCL for the same lock,
137 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
138 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
139 * to set in lock_flags.
142 xfs_lock_flags_assert(
145 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
146 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
147 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
148 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
149 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
150 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
151 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
152 ASSERT(lock_flags != 0);
156 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
157 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
158 * various combinations of the locks to be obtained.
160 * The 3 locks should always be ordered so that the IO lock is obtained first,
161 * the mmap lock second and the ilock last in order to prevent deadlock.
163 * Basic locking order:
165 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
167 * mmap_lock locking order:
169 * i_rwsem -> page lock -> mmap_lock
170 * mmap_lock -> invalidate_lock -> page_lock
172 * The difference in mmap_lock locking order mean that we cannot hold the
173 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
174 * can fault in pages during copy in/out (for buffered IO) or require the
175 * mmap_lock in get_user_pages() to map the user pages into the kernel address
176 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
177 * fault because page faults already hold the mmap_lock.
179 * Hence to serialise fully against both syscall and mmap based IO, we need to
180 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
181 * both taken in places where we need to invalidate the page cache in a race
182 * free manner (e.g. truncate, hole punch and other extent manipulation
190 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
192 xfs_lock_flags_assert(lock_flags);
194 if (lock_flags & XFS_IOLOCK_EXCL) {
195 down_write_nested(&VFS_I(ip)->i_rwsem,
196 XFS_IOLOCK_DEP(lock_flags));
197 } else if (lock_flags & XFS_IOLOCK_SHARED) {
198 down_read_nested(&VFS_I(ip)->i_rwsem,
199 XFS_IOLOCK_DEP(lock_flags));
202 if (lock_flags & XFS_MMAPLOCK_EXCL) {
203 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
204 XFS_MMAPLOCK_DEP(lock_flags));
205 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
206 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
207 XFS_MMAPLOCK_DEP(lock_flags));
210 if (lock_flags & XFS_ILOCK_EXCL)
211 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 else if (lock_flags & XFS_ILOCK_SHARED)
213 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
217 * This is just like xfs_ilock(), except that the caller
218 * is guaranteed not to sleep. It returns 1 if it gets
219 * the requested locks and 0 otherwise. If the IO lock is
220 * obtained but the inode lock cannot be, then the IO lock
221 * is dropped before returning.
223 * ip -- the inode being locked
224 * lock_flags -- this parameter indicates the inode's locks to be
225 * to be locked. See the comment for xfs_ilock() for a list
233 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
235 xfs_lock_flags_assert(lock_flags);
237 if (lock_flags & XFS_IOLOCK_EXCL) {
238 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
240 } else if (lock_flags & XFS_IOLOCK_SHARED) {
241 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
245 if (lock_flags & XFS_MMAPLOCK_EXCL) {
246 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
247 goto out_undo_iolock;
248 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
249 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
250 goto out_undo_iolock;
253 if (lock_flags & XFS_ILOCK_EXCL) {
254 if (!mrtryupdate(&ip->i_lock))
255 goto out_undo_mmaplock;
256 } else if (lock_flags & XFS_ILOCK_SHARED) {
257 if (!mrtryaccess(&ip->i_lock))
258 goto out_undo_mmaplock;
263 if (lock_flags & XFS_MMAPLOCK_EXCL)
264 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
265 else if (lock_flags & XFS_MMAPLOCK_SHARED)
266 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
268 if (lock_flags & XFS_IOLOCK_EXCL)
269 up_write(&VFS_I(ip)->i_rwsem);
270 else if (lock_flags & XFS_IOLOCK_SHARED)
271 up_read(&VFS_I(ip)->i_rwsem);
277 * xfs_iunlock() is used to drop the inode locks acquired with
278 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
279 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
280 * that we know which locks to drop.
282 * ip -- the inode being unlocked
283 * lock_flags -- this parameter indicates the inode's locks to be
284 * to be unlocked. See the comment for xfs_ilock() for a list
285 * of valid values for this parameter.
293 xfs_lock_flags_assert(lock_flags);
295 if (lock_flags & XFS_IOLOCK_EXCL)
296 up_write(&VFS_I(ip)->i_rwsem);
297 else if (lock_flags & XFS_IOLOCK_SHARED)
298 up_read(&VFS_I(ip)->i_rwsem);
300 if (lock_flags & XFS_MMAPLOCK_EXCL)
301 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
302 else if (lock_flags & XFS_MMAPLOCK_SHARED)
303 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
305 if (lock_flags & XFS_ILOCK_EXCL)
306 mrunlock_excl(&ip->i_lock);
307 else if (lock_flags & XFS_ILOCK_SHARED)
308 mrunlock_shared(&ip->i_lock);
310 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
314 * give up write locks. the i/o lock cannot be held nested
315 * if it is being demoted.
322 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
324 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
326 if (lock_flags & XFS_ILOCK_EXCL)
327 mrdemote(&ip->i_lock);
328 if (lock_flags & XFS_MMAPLOCK_EXCL)
329 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
330 if (lock_flags & XFS_IOLOCK_EXCL)
331 downgrade_write(&VFS_I(ip)->i_rwsem);
333 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
336 #if defined(DEBUG) || defined(XFS_WARN)
338 __xfs_rwsem_islocked(
339 struct rw_semaphore *rwsem,
343 return rwsem_is_locked(rwsem);
346 return lockdep_is_held_type(rwsem, 0);
349 * We are checking that the lock is held at least in shared
350 * mode but don't care that it might be held exclusively
351 * (i.e. shared | excl). Hence we check if the lock is held
352 * in any mode rather than an explicit shared mode.
354 return lockdep_is_held_type(rwsem, -1);
359 struct xfs_inode *ip,
362 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
363 if (!(lock_flags & XFS_ILOCK_SHARED))
364 return !!ip->i_lock.mr_writer;
365 return rwsem_is_locked(&ip->i_lock.mr_lock);
368 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
369 return __xfs_rwsem_islocked(&VFS_I(ip)->i_mapping->invalidate_lock,
370 (lock_flags & XFS_MMAPLOCK_SHARED));
373 if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) {
374 return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
375 (lock_flags & XFS_IOLOCK_SHARED));
384 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
385 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
386 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
387 * errors and warnings.
389 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
391 xfs_lockdep_subclass_ok(
394 return subclass < MAX_LOCKDEP_SUBCLASSES;
397 #define xfs_lockdep_subclass_ok(subclass) (true)
401 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
402 * value. This can be called for any type of inode lock combination, including
403 * parent locking. Care must be taken to ensure we don't overrun the subclass
404 * storage fields in the class mask we build.
413 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
415 ASSERT(xfs_lockdep_subclass_ok(subclass));
417 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
418 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
419 class += subclass << XFS_IOLOCK_SHIFT;
422 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
423 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
424 class += subclass << XFS_MMAPLOCK_SHIFT;
427 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
428 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
429 class += subclass << XFS_ILOCK_SHIFT;
432 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
436 * The following routine will lock n inodes in exclusive mode. We assume the
437 * caller calls us with the inodes in i_ino order.
439 * We need to detect deadlock where an inode that we lock is in the AIL and we
440 * start waiting for another inode that is locked by a thread in a long running
441 * transaction (such as truncate). This can result in deadlock since the long
442 * running trans might need to wait for the inode we just locked in order to
443 * push the tail and free space in the log.
445 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
446 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
447 * lock more than one at a time, lockdep will report false positives saying we
448 * have violated locking orders.
452 struct xfs_inode **ips,
460 struct xfs_log_item *lp;
463 * Currently supports between 2 and 5 inodes with exclusive locking. We
464 * support an arbitrary depth of locking here, but absolute limits on
465 * inodes depend on the type of locking and the limits placed by
466 * lockdep annotations in xfs_lock_inumorder. These are all checked by
469 ASSERT(ips && inodes >= 2 && inodes <= 5);
470 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
472 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
474 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
475 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
476 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
477 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
479 if (lock_mode & XFS_IOLOCK_EXCL) {
480 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
481 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
482 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
487 for (; i < inodes; i++) {
490 if (i && (ips[i] == ips[i - 1])) /* Already locked */
494 * If try_lock is not set yet, make sure all locked inodes are
495 * not in the AIL. If any are, set try_lock to be used later.
498 for (j = (i - 1); j >= 0 && !try_lock; j--) {
499 lp = &ips[j]->i_itemp->ili_item;
500 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
506 * If any of the previous locks we have locked is in the AIL,
507 * we must TRY to get the second and subsequent locks. If
508 * we can't get any, we must release all we have
512 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
516 /* try_lock means we have an inode locked that is in the AIL. */
518 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
522 * Unlock all previous guys and try again. xfs_iunlock will try
523 * to push the tail if the inode is in the AIL.
526 for (j = i - 1; j >= 0; j--) {
528 * Check to see if we've already unlocked this one. Not
529 * the first one going back, and the inode ptr is the
532 if (j != (i - 1) && ips[j] == ips[j + 1])
535 xfs_iunlock(ips[j], lock_mode);
538 if ((attempts % 5) == 0) {
539 delay(1); /* Don't just spin the CPU */
546 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
547 * mmaplock must be double-locked separately since we use i_rwsem and
548 * invalidate_lock for that. We now support taking one lock EXCL and the
553 struct xfs_inode *ip0,
555 struct xfs_inode *ip1,
559 struct xfs_log_item *lp;
561 ASSERT(hweight32(ip0_mode) == 1);
562 ASSERT(hweight32(ip1_mode) == 1);
563 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
565 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
566 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
567 ASSERT(ip0->i_ino != ip1->i_ino);
569 if (ip0->i_ino > ip1->i_ino) {
571 swap(ip0_mode, ip1_mode);
575 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
578 * If the first lock we have locked is in the AIL, we must TRY to get
579 * the second lock. If we can't get it, we must release the first one
582 lp = &ip0->i_itemp->ili_item;
583 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
584 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
585 xfs_iunlock(ip0, ip0_mode);
586 if ((++attempts % 5) == 0)
587 delay(1); /* Don't just spin the CPU */
591 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
597 struct xfs_inode *ip)
601 if (ip->i_diflags & XFS_DIFLAG_ANY) {
602 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
603 flags |= FS_XFLAG_REALTIME;
604 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
605 flags |= FS_XFLAG_PREALLOC;
606 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
607 flags |= FS_XFLAG_IMMUTABLE;
608 if (ip->i_diflags & XFS_DIFLAG_APPEND)
609 flags |= FS_XFLAG_APPEND;
610 if (ip->i_diflags & XFS_DIFLAG_SYNC)
611 flags |= FS_XFLAG_SYNC;
612 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
613 flags |= FS_XFLAG_NOATIME;
614 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
615 flags |= FS_XFLAG_NODUMP;
616 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
617 flags |= FS_XFLAG_RTINHERIT;
618 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
619 flags |= FS_XFLAG_PROJINHERIT;
620 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
621 flags |= FS_XFLAG_NOSYMLINKS;
622 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
623 flags |= FS_XFLAG_EXTSIZE;
624 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
625 flags |= FS_XFLAG_EXTSZINHERIT;
626 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
627 flags |= FS_XFLAG_NODEFRAG;
628 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
629 flags |= FS_XFLAG_FILESTREAM;
632 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
633 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
634 flags |= FS_XFLAG_DAX;
635 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
636 flags |= FS_XFLAG_COWEXTSIZE;
639 if (xfs_inode_has_attr_fork(ip))
640 flags |= FS_XFLAG_HASATTR;
645 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
646 * is allowed, otherwise it has to be an exact match. If a CI match is found,
647 * ci_name->name will point to a the actual name (caller must free) or
648 * will be set to NULL if an exact match is found.
652 struct xfs_inode *dp,
653 const struct xfs_name *name,
654 struct xfs_inode **ipp,
655 struct xfs_name *ci_name)
660 trace_xfs_lookup(dp, name);
662 if (xfs_is_shutdown(dp->i_mount))
665 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
669 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
677 kmem_free(ci_name->name);
683 /* Propagate di_flags from a parent inode to a child inode. */
685 xfs_inode_inherit_flags(
686 struct xfs_inode *ip,
687 const struct xfs_inode *pip)
689 unsigned int di_flags = 0;
690 xfs_failaddr_t failaddr;
691 umode_t mode = VFS_I(ip)->i_mode;
694 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
695 di_flags |= XFS_DIFLAG_RTINHERIT;
696 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
697 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
698 ip->i_extsize = pip->i_extsize;
700 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
701 di_flags |= XFS_DIFLAG_PROJINHERIT;
702 } else if (S_ISREG(mode)) {
703 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
704 xfs_has_realtime(ip->i_mount))
705 di_flags |= XFS_DIFLAG_REALTIME;
706 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
707 di_flags |= XFS_DIFLAG_EXTSIZE;
708 ip->i_extsize = pip->i_extsize;
711 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
713 di_flags |= XFS_DIFLAG_NOATIME;
714 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
716 di_flags |= XFS_DIFLAG_NODUMP;
717 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
719 di_flags |= XFS_DIFLAG_SYNC;
720 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
721 xfs_inherit_nosymlinks)
722 di_flags |= XFS_DIFLAG_NOSYMLINKS;
723 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
724 xfs_inherit_nodefrag)
725 di_flags |= XFS_DIFLAG_NODEFRAG;
726 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
727 di_flags |= XFS_DIFLAG_FILESTREAM;
729 ip->i_diflags |= di_flags;
732 * Inode verifiers on older kernels only check that the extent size
733 * hint is an integer multiple of the rt extent size on realtime files.
734 * They did not check the hint alignment on a directory with both
735 * rtinherit and extszinherit flags set. If the misaligned hint is
736 * propagated from a directory into a new realtime file, new file
737 * allocations will fail due to math errors in the rt allocator and/or
738 * trip the verifiers. Validate the hint settings in the new file so
739 * that we don't let broken hints propagate.
741 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
742 VFS_I(ip)->i_mode, ip->i_diflags);
744 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
745 XFS_DIFLAG_EXTSZINHERIT);
750 /* Propagate di_flags2 from a parent inode to a child inode. */
752 xfs_inode_inherit_flags2(
753 struct xfs_inode *ip,
754 const struct xfs_inode *pip)
756 xfs_failaddr_t failaddr;
758 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
759 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
760 ip->i_cowextsize = pip->i_cowextsize;
762 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
763 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
765 /* Don't let invalid cowextsize hints propagate. */
766 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
767 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
769 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
770 ip->i_cowextsize = 0;
775 * Initialise a newly allocated inode and return the in-core inode to the
776 * caller locked exclusively.
780 struct mnt_idmap *idmap,
781 struct xfs_trans *tp,
782 struct xfs_inode *pip,
789 struct xfs_inode **ipp)
791 struct inode *dir = pip ? VFS_I(pip) : NULL;
792 struct xfs_mount *mp = tp->t_mountp;
793 struct xfs_inode *ip;
796 struct timespec64 tv;
800 * Protect against obviously corrupt allocation btree records. Later
801 * xfs_iget checks will catch re-allocation of other active in-memory
802 * and on-disk inodes. If we don't catch reallocating the parent inode
803 * here we will deadlock in xfs_iget() so we have to do these checks
806 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
807 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
808 return -EFSCORRUPTED;
812 * Get the in-core inode with the lock held exclusively to prevent
813 * others from looking at until we're done.
815 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
821 set_nlink(inode, nlink);
822 inode->i_rdev = rdev;
825 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
826 inode_fsuid_set(inode, idmap);
827 inode->i_gid = dir->i_gid;
828 inode->i_mode = mode;
830 inode_init_owner(idmap, inode, dir, mode);
834 * If the group ID of the new file does not match the effective group
835 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
836 * (and only if the irix_sgid_inherit compatibility variable is set).
838 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
839 !vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode)))
840 inode->i_mode &= ~S_ISGID;
843 ip->i_df.if_nextents = 0;
844 ASSERT(ip->i_nblocks == 0);
846 tv = inode_set_ctime_current(inode);
853 if (xfs_has_v3inodes(mp)) {
854 inode_set_iversion(inode, 1);
855 ip->i_cowextsize = 0;
859 flags = XFS_ILOG_CORE;
860 switch (mode & S_IFMT) {
865 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
866 flags |= XFS_ILOG_DEV;
870 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
871 xfs_inode_inherit_flags(ip, pip);
872 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
873 xfs_inode_inherit_flags2(ip, pip);
876 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
877 ip->i_df.if_bytes = 0;
878 ip->i_df.if_u1.if_root = NULL;
885 * If we need to create attributes immediately after allocating the
886 * inode, initialise an empty attribute fork right now. We use the
887 * default fork offset for attributes here as we don't know exactly what
888 * size or how many attributes we might be adding. We can do this
889 * safely here because we know the data fork is completely empty and
890 * this saves us from needing to run a separate transaction to set the
891 * fork offset in the immediate future.
893 if (init_xattrs && xfs_has_attr(mp)) {
894 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
895 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
899 * Log the new values stuffed into the inode.
901 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
902 xfs_trans_log_inode(tp, ip, flags);
904 /* now that we have an i_mode we can setup the inode structure */
912 * Decrement the link count on an inode & log the change. If this causes the
913 * link count to go to zero, move the inode to AGI unlinked list so that it can
914 * be freed when the last active reference goes away via xfs_inactive().
916 static int /* error */
921 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
923 drop_nlink(VFS_I(ip));
924 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
926 if (VFS_I(ip)->i_nlink)
929 return xfs_iunlink(tp, ip);
933 * Increment the link count on an inode & log the change.
940 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
942 inc_nlink(VFS_I(ip));
943 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
948 struct mnt_idmap *idmap,
950 struct xfs_name *name,
956 int is_dir = S_ISDIR(mode);
957 struct xfs_mount *mp = dp->i_mount;
958 struct xfs_inode *ip = NULL;
959 struct xfs_trans *tp = NULL;
961 bool unlock_dp_on_error = false;
963 struct xfs_dquot *udqp = NULL;
964 struct xfs_dquot *gdqp = NULL;
965 struct xfs_dquot *pdqp = NULL;
966 struct xfs_trans_res *tres;
970 trace_xfs_create(dp, name);
972 if (xfs_is_shutdown(mp))
975 prid = xfs_get_initial_prid(dp);
978 * Make sure that we have allocated dquot(s) on disk.
980 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
981 mapped_fsgid(idmap, &init_user_ns), prid,
982 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
983 &udqp, &gdqp, &pdqp);
988 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
989 tres = &M_RES(mp)->tr_mkdir;
991 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
992 tres = &M_RES(mp)->tr_create;
996 * Initially assume that the file does not exist and
997 * reserve the resources for that case. If that is not
998 * the case we'll drop the one we have and get a more
999 * appropriate transaction later.
1001 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1003 if (error == -ENOSPC) {
1004 /* flush outstanding delalloc blocks and retry */
1005 xfs_flush_inodes(mp);
1006 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1010 goto out_release_dquots;
1012 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1013 unlock_dp_on_error = true;
1016 * A newly created regular or special file just has one directory
1017 * entry pointing to them, but a directory also the "." entry
1018 * pointing to itself.
1020 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1022 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1023 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1025 goto out_trans_cancel;
1028 * Now we join the directory inode to the transaction. We do not do it
1029 * earlier because xfs_dialloc might commit the previous transaction
1030 * (and release all the locks). An error from here on will result in
1031 * the transaction cancel unlocking dp so don't do it explicitly in the
1034 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1035 unlock_dp_on_error = false;
1037 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1038 resblks - XFS_IALLOC_SPACE_RES(mp));
1040 ASSERT(error != -ENOSPC);
1041 goto out_trans_cancel;
1043 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1044 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1047 error = xfs_dir_init(tp, ip, dp);
1049 goto out_trans_cancel;
1051 xfs_bumplink(tp, dp);
1055 * If this is a synchronous mount, make sure that the
1056 * create transaction goes to disk before returning to
1059 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1060 xfs_trans_set_sync(tp);
1063 * Attach the dquot(s) to the inodes and modify them incore.
1064 * These ids of the inode couldn't have changed since the new
1065 * inode has been locked ever since it was created.
1067 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1069 error = xfs_trans_commit(tp);
1071 goto out_release_inode;
1073 xfs_qm_dqrele(udqp);
1074 xfs_qm_dqrele(gdqp);
1075 xfs_qm_dqrele(pdqp);
1081 xfs_trans_cancel(tp);
1084 * Wait until after the current transaction is aborted to finish the
1085 * setup of the inode and release the inode. This prevents recursive
1086 * transactions and deadlocks from xfs_inactive.
1089 xfs_finish_inode_setup(ip);
1093 xfs_qm_dqrele(udqp);
1094 xfs_qm_dqrele(gdqp);
1095 xfs_qm_dqrele(pdqp);
1097 if (unlock_dp_on_error)
1098 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1104 struct mnt_idmap *idmap,
1105 struct xfs_inode *dp,
1107 struct xfs_inode **ipp)
1109 struct xfs_mount *mp = dp->i_mount;
1110 struct xfs_inode *ip = NULL;
1111 struct xfs_trans *tp = NULL;
1114 struct xfs_dquot *udqp = NULL;
1115 struct xfs_dquot *gdqp = NULL;
1116 struct xfs_dquot *pdqp = NULL;
1117 struct xfs_trans_res *tres;
1121 if (xfs_is_shutdown(mp))
1124 prid = xfs_get_initial_prid(dp);
1127 * Make sure that we have allocated dquot(s) on disk.
1129 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
1130 mapped_fsgid(idmap, &init_user_ns), prid,
1131 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1132 &udqp, &gdqp, &pdqp);
1136 resblks = XFS_IALLOC_SPACE_RES(mp);
1137 tres = &M_RES(mp)->tr_create_tmpfile;
1139 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1142 goto out_release_dquots;
1144 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1146 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1147 0, 0, prid, false, &ip);
1149 goto out_trans_cancel;
1151 if (xfs_has_wsync(mp))
1152 xfs_trans_set_sync(tp);
1155 * Attach the dquot(s) to the inodes and modify them incore.
1156 * These ids of the inode couldn't have changed since the new
1157 * inode has been locked ever since it was created.
1159 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1161 error = xfs_iunlink(tp, ip);
1163 goto out_trans_cancel;
1165 error = xfs_trans_commit(tp);
1167 goto out_release_inode;
1169 xfs_qm_dqrele(udqp);
1170 xfs_qm_dqrele(gdqp);
1171 xfs_qm_dqrele(pdqp);
1177 xfs_trans_cancel(tp);
1180 * Wait until after the current transaction is aborted to finish the
1181 * setup of the inode and release the inode. This prevents recursive
1182 * transactions and deadlocks from xfs_inactive.
1185 xfs_finish_inode_setup(ip);
1189 xfs_qm_dqrele(udqp);
1190 xfs_qm_dqrele(gdqp);
1191 xfs_qm_dqrele(pdqp);
1200 struct xfs_name *target_name)
1202 xfs_mount_t *mp = tdp->i_mount;
1204 int error, nospace_error = 0;
1207 trace_xfs_link(tdp, target_name);
1209 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1211 if (xfs_is_shutdown(mp))
1214 error = xfs_qm_dqattach(sip);
1218 error = xfs_qm_dqattach(tdp);
1222 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1223 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1224 &tp, &nospace_error);
1229 * If we are using project inheritance, we only allow hard link
1230 * creation in our tree when the project IDs are the same; else
1231 * the tree quota mechanism could be circumvented.
1233 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1234 tdp->i_projid != sip->i_projid)) {
1240 error = xfs_dir_canenter(tp, tdp, target_name);
1246 * Handle initial link state of O_TMPFILE inode
1248 if (VFS_I(sip)->i_nlink == 0) {
1249 struct xfs_perag *pag;
1251 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1252 error = xfs_iunlink_remove(tp, pag, sip);
1258 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1262 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1263 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1265 xfs_bumplink(tp, sip);
1268 * If this is a synchronous mount, make sure that the
1269 * link transaction goes to disk before returning to
1272 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1273 xfs_trans_set_sync(tp);
1275 return xfs_trans_commit(tp);
1278 xfs_trans_cancel(tp);
1280 if (error == -ENOSPC && nospace_error)
1281 error = nospace_error;
1285 /* Clear the reflink flag and the cowblocks tag if possible. */
1287 xfs_itruncate_clear_reflink_flags(
1288 struct xfs_inode *ip)
1290 struct xfs_ifork *dfork;
1291 struct xfs_ifork *cfork;
1293 if (!xfs_is_reflink_inode(ip))
1295 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1296 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1297 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1298 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1299 if (cfork->if_bytes == 0)
1300 xfs_inode_clear_cowblocks_tag(ip);
1304 * Free up the underlying blocks past new_size. The new size must be smaller
1305 * than the current size. This routine can be used both for the attribute and
1306 * data fork, and does not modify the inode size, which is left to the caller.
1308 * The transaction passed to this routine must have made a permanent log
1309 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1310 * given transaction and start new ones, so make sure everything involved in
1311 * the transaction is tidy before calling here. Some transaction will be
1312 * returned to the caller to be committed. The incoming transaction must
1313 * already include the inode, and both inode locks must be held exclusively.
1314 * The inode must also be "held" within the transaction. On return the inode
1315 * will be "held" within the returned transaction. This routine does NOT
1316 * require any disk space to be reserved for it within the transaction.
1318 * If we get an error, we must return with the inode locked and linked into the
1319 * current transaction. This keeps things simple for the higher level code,
1320 * because it always knows that the inode is locked and held in the transaction
1321 * that returns to it whether errors occur or not. We don't mark the inode
1322 * dirty on error so that transactions can be easily aborted if possible.
1325 xfs_itruncate_extents_flags(
1326 struct xfs_trans **tpp,
1327 struct xfs_inode *ip,
1329 xfs_fsize_t new_size,
1332 struct xfs_mount *mp = ip->i_mount;
1333 struct xfs_trans *tp = *tpp;
1334 xfs_fileoff_t first_unmap_block;
1335 xfs_filblks_t unmap_len;
1338 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1339 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1340 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1341 ASSERT(new_size <= XFS_ISIZE(ip));
1342 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1343 ASSERT(ip->i_itemp != NULL);
1344 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1345 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1347 trace_xfs_itruncate_extents_start(ip, new_size);
1349 flags |= xfs_bmapi_aflag(whichfork);
1352 * Since it is possible for space to become allocated beyond
1353 * the end of the file (in a crash where the space is allocated
1354 * but the inode size is not yet updated), simply remove any
1355 * blocks which show up between the new EOF and the maximum
1356 * possible file size.
1358 * We have to free all the blocks to the bmbt maximum offset, even if
1359 * the page cache can't scale that far.
1361 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1362 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1363 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1367 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1368 while (unmap_len > 0) {
1369 ASSERT(tp->t_highest_agno == NULLAGNUMBER);
1370 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1371 flags, XFS_ITRUNC_MAX_EXTENTS);
1375 /* free the just unmapped extents */
1376 error = xfs_defer_finish(&tp);
1381 if (whichfork == XFS_DATA_FORK) {
1382 /* Remove all pending CoW reservations. */
1383 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1384 first_unmap_block, XFS_MAX_FILEOFF, true);
1388 xfs_itruncate_clear_reflink_flags(ip);
1392 * Always re-log the inode so that our permanent transaction can keep
1393 * on rolling it forward in the log.
1395 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1397 trace_xfs_itruncate_extents_end(ip, new_size);
1408 xfs_mount_t *mp = ip->i_mount;
1411 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1414 /* If this is a read-only mount, don't do this (would generate I/O) */
1415 if (xfs_is_readonly(mp))
1418 if (!xfs_is_shutdown(mp)) {
1422 * If we previously truncated this file and removed old data
1423 * in the process, we want to initiate "early" writeout on
1424 * the last close. This is an attempt to combat the notorious
1425 * NULL files problem which is particularly noticeable from a
1426 * truncate down, buffered (re-)write (delalloc), followed by
1427 * a crash. What we are effectively doing here is
1428 * significantly reducing the time window where we'd otherwise
1429 * be exposed to that problem.
1431 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1433 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1434 if (ip->i_delayed_blks > 0) {
1435 error = filemap_flush(VFS_I(ip)->i_mapping);
1442 if (VFS_I(ip)->i_nlink == 0)
1446 * If we can't get the iolock just skip truncating the blocks past EOF
1447 * because we could deadlock with the mmap_lock otherwise. We'll get
1448 * another chance to drop them once the last reference to the inode is
1449 * dropped, so we'll never leak blocks permanently.
1451 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1454 if (xfs_can_free_eofblocks(ip, false)) {
1456 * Check if the inode is being opened, written and closed
1457 * frequently and we have delayed allocation blocks outstanding
1458 * (e.g. streaming writes from the NFS server), truncating the
1459 * blocks past EOF will cause fragmentation to occur.
1461 * In this case don't do the truncation, but we have to be
1462 * careful how we detect this case. Blocks beyond EOF show up as
1463 * i_delayed_blks even when the inode is clean, so we need to
1464 * truncate them away first before checking for a dirty release.
1465 * Hence on the first dirty close we will still remove the
1466 * speculative allocation, but after that we will leave it in
1469 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1472 error = xfs_free_eofblocks(ip);
1476 /* delalloc blocks after truncation means it really is dirty */
1477 if (ip->i_delayed_blks)
1478 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1482 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1487 * xfs_inactive_truncate
1489 * Called to perform a truncate when an inode becomes unlinked.
1492 xfs_inactive_truncate(
1493 struct xfs_inode *ip)
1495 struct xfs_mount *mp = ip->i_mount;
1496 struct xfs_trans *tp;
1499 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1501 ASSERT(xfs_is_shutdown(mp));
1504 xfs_ilock(ip, XFS_ILOCK_EXCL);
1505 xfs_trans_ijoin(tp, ip, 0);
1508 * Log the inode size first to prevent stale data exposure in the event
1509 * of a system crash before the truncate completes. See the related
1510 * comment in xfs_vn_setattr_size() for details.
1512 ip->i_disk_size = 0;
1513 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1515 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1517 goto error_trans_cancel;
1519 ASSERT(ip->i_df.if_nextents == 0);
1521 error = xfs_trans_commit(tp);
1525 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1529 xfs_trans_cancel(tp);
1531 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1536 * xfs_inactive_ifree()
1538 * Perform the inode free when an inode is unlinked.
1542 struct xfs_inode *ip)
1544 struct xfs_mount *mp = ip->i_mount;
1545 struct xfs_trans *tp;
1549 * We try to use a per-AG reservation for any block needed by the finobt
1550 * tree, but as the finobt feature predates the per-AG reservation
1551 * support a degraded file system might not have enough space for the
1552 * reservation at mount time. In that case try to dip into the reserved
1555 * Send a warning if the reservation does happen to fail, as the inode
1556 * now remains allocated and sits on the unlinked list until the fs is
1559 if (unlikely(mp->m_finobt_nores)) {
1560 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1561 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1564 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1567 if (error == -ENOSPC) {
1568 xfs_warn_ratelimited(mp,
1569 "Failed to remove inode(s) from unlinked list. "
1570 "Please free space, unmount and run xfs_repair.");
1572 ASSERT(xfs_is_shutdown(mp));
1578 * We do not hold the inode locked across the entire rolling transaction
1579 * here. We only need to hold it for the first transaction that
1580 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1581 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1582 * here breaks the relationship between cluster buffer invalidation and
1583 * stale inode invalidation on cluster buffer item journal commit
1584 * completion, and can result in leaving dirty stale inodes hanging
1587 * We have no need for serialising this inode operation against other
1588 * operations - we freed the inode and hence reallocation is required
1589 * and that will serialise on reallocating the space the deferops need
1590 * to free. Hence we can unlock the inode on the first commit of
1591 * the transaction rather than roll it right through the deferops. This
1592 * avoids relogging the XFS_ISTALE inode.
1594 * We check that xfs_ifree() hasn't grown an internal transaction roll
1595 * by asserting that the inode is still locked when it returns.
1597 xfs_ilock(ip, XFS_ILOCK_EXCL);
1598 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1600 error = xfs_ifree(tp, ip);
1601 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1604 * If we fail to free the inode, shut down. The cancel
1605 * might do that, we need to make sure. Otherwise the
1606 * inode might be lost for a long time or forever.
1608 if (!xfs_is_shutdown(mp)) {
1609 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1611 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1613 xfs_trans_cancel(tp);
1618 * Credit the quota account(s). The inode is gone.
1620 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1622 return xfs_trans_commit(tp);
1626 * Returns true if we need to update the on-disk metadata before we can free
1627 * the memory used by this inode. Updates include freeing post-eof
1628 * preallocations; freeing COW staging extents; and marking the inode free in
1629 * the inobt if it is on the unlinked list.
1632 xfs_inode_needs_inactive(
1633 struct xfs_inode *ip)
1635 struct xfs_mount *mp = ip->i_mount;
1636 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1639 * If the inode is already free, then there can be nothing
1642 if (VFS_I(ip)->i_mode == 0)
1645 /* If this is a read-only mount, don't do this (would generate I/O) */
1646 if (xfs_is_readonly(mp))
1649 /* If the log isn't running, push inodes straight to reclaim. */
1650 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1653 /* Metadata inodes require explicit resource cleanup. */
1654 if (xfs_is_metadata_inode(ip))
1657 /* Want to clean out the cow blocks if there are any. */
1658 if (cow_ifp && cow_ifp->if_bytes > 0)
1661 /* Unlinked files must be freed. */
1662 if (VFS_I(ip)->i_nlink == 0)
1666 * This file isn't being freed, so check if there are post-eof blocks
1667 * to free. @force is true because we are evicting an inode from the
1668 * cache. Post-eof blocks must be freed, lest we end up with broken
1669 * free space accounting.
1671 * Note: don't bother with iolock here since lockdep complains about
1672 * acquiring it in reclaim context. We have the only reference to the
1673 * inode at this point anyways.
1675 return xfs_can_free_eofblocks(ip, true);
1681 * This is called when the vnode reference count for the vnode
1682 * goes to zero. If the file has been unlinked, then it must
1683 * now be truncated. Also, we clear all of the read-ahead state
1684 * kept for the inode here since the file is now closed.
1690 struct xfs_mount *mp;
1695 * If the inode is already free, then there can be nothing
1698 if (VFS_I(ip)->i_mode == 0) {
1699 ASSERT(ip->i_df.if_broot_bytes == 0);
1704 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1706 /* If this is a read-only mount, don't do this (would generate I/O) */
1707 if (xfs_is_readonly(mp))
1710 /* Metadata inodes require explicit resource cleanup. */
1711 if (xfs_is_metadata_inode(ip))
1714 /* Try to clean out the cow blocks if there are any. */
1715 if (xfs_inode_has_cow_data(ip))
1716 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1718 if (VFS_I(ip)->i_nlink != 0) {
1720 * force is true because we are evicting an inode from the
1721 * cache. Post-eof blocks must be freed, lest we end up with
1722 * broken free space accounting.
1724 * Note: don't bother with iolock here since lockdep complains
1725 * about acquiring it in reclaim context. We have the only
1726 * reference to the inode at this point anyways.
1728 if (xfs_can_free_eofblocks(ip, true))
1729 error = xfs_free_eofblocks(ip);
1734 if (S_ISREG(VFS_I(ip)->i_mode) &&
1735 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1736 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1739 error = xfs_qm_dqattach(ip);
1743 if (S_ISLNK(VFS_I(ip)->i_mode))
1744 error = xfs_inactive_symlink(ip);
1746 error = xfs_inactive_truncate(ip);
1751 * If there are attributes associated with the file then blow them away
1752 * now. The code calls a routine that recursively deconstructs the
1753 * attribute fork. If also blows away the in-core attribute fork.
1755 if (xfs_inode_has_attr_fork(ip)) {
1756 error = xfs_attr_inactive(ip);
1761 ASSERT(ip->i_forkoff == 0);
1766 error = xfs_inactive_ifree(ip);
1770 * We're done making metadata updates for this inode, so we can release
1771 * the attached dquots.
1773 xfs_qm_dqdetach(ip);
1778 * In-Core Unlinked List Lookups
1779 * =============================
1781 * Every inode is supposed to be reachable from some other piece of metadata
1782 * with the exception of the root directory. Inodes with a connection to a
1783 * file descriptor but not linked from anywhere in the on-disk directory tree
1784 * are collectively known as unlinked inodes, though the filesystem itself
1785 * maintains links to these inodes so that on-disk metadata are consistent.
1787 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1788 * header contains a number of buckets that point to an inode, and each inode
1789 * record has a pointer to the next inode in the hash chain. This
1790 * singly-linked list causes scaling problems in the iunlink remove function
1791 * because we must walk that list to find the inode that points to the inode
1792 * being removed from the unlinked hash bucket list.
1794 * Hence we keep an in-memory double linked list to link each inode on an
1795 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1796 * based lists would require having 64 list heads in the perag, one for each
1797 * list. This is expensive in terms of memory (think millions of AGs) and cache
1798 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1799 * must be referenced at the VFS level to keep them on the list and hence we
1800 * have an existence guarantee for inodes on the unlinked list.
1802 * Given we have an existence guarantee, we can use lockless inode cache lookups
1803 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1804 * for the double linked unlinked list, and we don't need any extra locking to
1805 * keep the list safe as all manipulations are done under the AGI buffer lock.
1806 * Keeping the list up to date does not require memory allocation, just finding
1807 * the XFS inode and updating the next/prev unlinked list aginos.
1811 * Find an inode on the unlinked list. This does not take references to the
1812 * inode as we have existence guarantees by holding the AGI buffer lock and that
1813 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1814 * don't find the inode in cache, then let the caller handle the situation.
1816 static struct xfs_inode *
1818 struct xfs_perag *pag,
1821 struct xfs_inode *ip;
1824 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1827 * Inode not in memory or in RCU freeing limbo should not happen.
1828 * Warn about this and let the caller handle the failure.
1830 if (WARN_ON_ONCE(!ip || !ip->i_ino)) {
1834 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1839 /* Update the prev pointer of the next agino. */
1841 xfs_iunlink_update_backref(
1842 struct xfs_perag *pag,
1843 xfs_agino_t prev_agino,
1844 xfs_agino_t next_agino)
1846 struct xfs_inode *ip;
1848 /* No update necessary if we are at the end of the list. */
1849 if (next_agino == NULLAGINO)
1852 ip = xfs_iunlink_lookup(pag, next_agino);
1854 return -EFSCORRUPTED;
1855 ip->i_prev_unlinked = prev_agino;
1860 * Point the AGI unlinked bucket at an inode and log the results. The caller
1861 * is responsible for validating the old value.
1864 xfs_iunlink_update_bucket(
1865 struct xfs_trans *tp,
1866 struct xfs_perag *pag,
1867 struct xfs_buf *agibp,
1868 unsigned int bucket_index,
1869 xfs_agino_t new_agino)
1871 struct xfs_agi *agi = agibp->b_addr;
1872 xfs_agino_t old_value;
1875 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
1877 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1878 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1879 old_value, new_agino);
1882 * We should never find the head of the list already set to the value
1883 * passed in because either we're adding or removing ourselves from the
1886 if (old_value == new_agino) {
1887 xfs_buf_mark_corrupt(agibp);
1888 return -EFSCORRUPTED;
1891 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
1892 offset = offsetof(struct xfs_agi, agi_unlinked) +
1893 (sizeof(xfs_agino_t) * bucket_index);
1894 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
1899 xfs_iunlink_insert_inode(
1900 struct xfs_trans *tp,
1901 struct xfs_perag *pag,
1902 struct xfs_buf *agibp,
1903 struct xfs_inode *ip)
1905 struct xfs_mount *mp = tp->t_mountp;
1906 struct xfs_agi *agi = agibp->b_addr;
1907 xfs_agino_t next_agino;
1908 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1909 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1913 * Get the index into the agi hash table for the list this inode will
1914 * go on. Make sure the pointer isn't garbage and that this inode
1915 * isn't already on the list.
1917 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1918 if (next_agino == agino ||
1919 !xfs_verify_agino_or_null(pag, next_agino)) {
1920 xfs_buf_mark_corrupt(agibp);
1921 return -EFSCORRUPTED;
1925 * Update the prev pointer in the next inode to point back to this
1928 error = xfs_iunlink_update_backref(pag, agino, next_agino);
1932 if (next_agino != NULLAGINO) {
1934 * There is already another inode in the bucket, so point this
1935 * inode to the current head of the list.
1937 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
1940 ip->i_next_unlinked = next_agino;
1943 /* Point the head of the list to point to this inode. */
1944 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
1948 * This is called when the inode's link count has gone to 0 or we are creating
1949 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
1951 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1952 * list when the inode is freed.
1956 struct xfs_trans *tp,
1957 struct xfs_inode *ip)
1959 struct xfs_mount *mp = tp->t_mountp;
1960 struct xfs_perag *pag;
1961 struct xfs_buf *agibp;
1964 ASSERT(VFS_I(ip)->i_nlink == 0);
1965 ASSERT(VFS_I(ip)->i_mode != 0);
1966 trace_xfs_iunlink(ip);
1968 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1970 /* Get the agi buffer first. It ensures lock ordering on the list. */
1971 error = xfs_read_agi(pag, tp, &agibp);
1975 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
1982 xfs_iunlink_remove_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 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1991 xfs_agino_t head_agino;
1992 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1995 trace_xfs_iunlink_remove(ip);
1998 * Get the index into the agi hash table for the list this inode will
1999 * go on. Make sure the head pointer isn't garbage.
2001 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2002 if (!xfs_verify_agino(pag, head_agino)) {
2003 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2005 return -EFSCORRUPTED;
2009 * Set our inode's next_unlinked pointer to NULL and then return
2010 * the old pointer value so that we can update whatever was previous
2011 * to us in the list to point to whatever was next in the list.
2013 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2018 * Update the prev pointer in the next inode to point back to previous
2019 * inode in the chain.
2021 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2022 ip->i_next_unlinked);
2026 if (head_agino != agino) {
2027 struct xfs_inode *prev_ip;
2029 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2031 return -EFSCORRUPTED;
2033 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2034 ip->i_next_unlinked);
2035 prev_ip->i_next_unlinked = ip->i_next_unlinked;
2037 /* Point the head of the list to the next unlinked inode. */
2038 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2039 ip->i_next_unlinked);
2042 ip->i_next_unlinked = NULLAGINO;
2043 ip->i_prev_unlinked = NULLAGINO;
2048 * Pull the on-disk inode from the AGI unlinked list.
2052 struct xfs_trans *tp,
2053 struct xfs_perag *pag,
2054 struct xfs_inode *ip)
2056 struct xfs_buf *agibp;
2059 trace_xfs_iunlink_remove(ip);
2061 /* Get the agi buffer first. It ensures lock ordering on the list. */
2062 error = xfs_read_agi(pag, tp, &agibp);
2066 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2070 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2071 * mark it stale. We should only find clean inodes in this lookup that aren't
2075 xfs_ifree_mark_inode_stale(
2076 struct xfs_perag *pag,
2077 struct xfs_inode *free_ip,
2080 struct xfs_mount *mp = pag->pag_mount;
2081 struct xfs_inode_log_item *iip;
2082 struct xfs_inode *ip;
2086 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2088 /* Inode not in memory, nothing to do */
2095 * because this is an RCU protected lookup, we could find a recently
2096 * freed or even reallocated inode during the lookup. We need to check
2097 * under the i_flags_lock for a valid inode here. Skip it if it is not
2098 * valid, the wrong inode or stale.
2100 spin_lock(&ip->i_flags_lock);
2101 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2102 goto out_iflags_unlock;
2105 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2106 * other inodes that we did not find in the list attached to the buffer
2107 * and are not already marked stale. If we can't lock it, back off and
2110 if (ip != free_ip) {
2111 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2112 spin_unlock(&ip->i_flags_lock);
2118 ip->i_flags |= XFS_ISTALE;
2121 * If the inode is flushing, it is already attached to the buffer. All
2122 * we needed to do here is mark the inode stale so buffer IO completion
2123 * will remove it from the AIL.
2126 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2127 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2128 ASSERT(iip->ili_last_fields);
2133 * Inodes not attached to the buffer can be released immediately.
2134 * Everything else has to go through xfs_iflush_abort() on journal
2135 * commit as the flock synchronises removal of the inode from the
2136 * cluster buffer against inode reclaim.
2138 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2141 __xfs_iflags_set(ip, XFS_IFLUSHING);
2142 spin_unlock(&ip->i_flags_lock);
2145 /* we have a dirty inode in memory that has not yet been flushed. */
2146 spin_lock(&iip->ili_lock);
2147 iip->ili_last_fields = iip->ili_fields;
2148 iip->ili_fields = 0;
2149 iip->ili_fsync_fields = 0;
2150 spin_unlock(&iip->ili_lock);
2151 ASSERT(iip->ili_last_fields);
2154 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2159 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2161 spin_unlock(&ip->i_flags_lock);
2166 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2167 * inodes that are in memory - they all must be marked stale and attached to
2168 * the cluster buffer.
2172 struct xfs_trans *tp,
2173 struct xfs_perag *pag,
2174 struct xfs_inode *free_ip,
2175 struct xfs_icluster *xic)
2177 struct xfs_mount *mp = free_ip->i_mount;
2178 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2181 xfs_ino_t inum = xic->first_ino;
2187 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2189 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2191 * The allocation bitmap tells us which inodes of the chunk were
2192 * physically allocated. Skip the cluster if an inode falls into
2195 ioffset = inum - xic->first_ino;
2196 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2197 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2201 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2202 XFS_INO_TO_AGBNO(mp, inum));
2205 * We obtain and lock the backing buffer first in the process
2206 * here to ensure dirty inodes attached to the buffer remain in
2207 * the flushing state while we mark them stale.
2209 * If we scan the in-memory inodes first, then buffer IO can
2210 * complete before we get a lock on it, and hence we may fail
2211 * to mark all the active inodes on the buffer stale.
2213 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2214 mp->m_bsize * igeo->blocks_per_cluster,
2220 * This buffer may not have been correctly initialised as we
2221 * didn't read it from disk. That's not important because we are
2222 * only using to mark the buffer as stale in the log, and to
2223 * attach stale cached inodes on it. That means it will never be
2224 * dispatched for IO. If it is, we want to know about it, and we
2225 * want it to fail. We can acheive this by adding a write
2226 * verifier to the buffer.
2228 bp->b_ops = &xfs_inode_buf_ops;
2231 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2232 * too. This requires lookups, and will skip inodes that we've
2233 * already marked XFS_ISTALE.
2235 for (i = 0; i < igeo->inodes_per_cluster; i++)
2236 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2238 xfs_trans_stale_inode_buf(tp, bp);
2239 xfs_trans_binval(tp, bp);
2245 * This is called to return an inode to the inode free list. The inode should
2246 * already be truncated to 0 length and have no pages associated with it. This
2247 * routine also assumes that the inode is already a part of the transaction.
2249 * The on-disk copy of the inode will have been added to the list of unlinked
2250 * inodes in the AGI. We need to remove the inode from that list atomically with
2251 * respect to freeing it here.
2255 struct xfs_trans *tp,
2256 struct xfs_inode *ip)
2258 struct xfs_mount *mp = ip->i_mount;
2259 struct xfs_perag *pag;
2260 struct xfs_icluster xic = { 0 };
2261 struct xfs_inode_log_item *iip = ip->i_itemp;
2264 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2265 ASSERT(VFS_I(ip)->i_nlink == 0);
2266 ASSERT(ip->i_df.if_nextents == 0);
2267 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2268 ASSERT(ip->i_nblocks == 0);
2270 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2273 * Free the inode first so that we guarantee that the AGI lock is going
2274 * to be taken before we remove the inode from the unlinked list. This
2275 * makes the AGI lock -> unlinked list modification order the same as
2276 * used in O_TMPFILE creation.
2278 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2282 error = xfs_iunlink_remove(tp, pag, ip);
2287 * Free any local-format data sitting around before we reset the
2288 * data fork to extents format. Note that the attr fork data has
2289 * already been freed by xfs_attr_inactive.
2291 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2292 kmem_free(ip->i_df.if_u1.if_data);
2293 ip->i_df.if_u1.if_data = NULL;
2294 ip->i_df.if_bytes = 0;
2297 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2299 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2300 ip->i_forkoff = 0; /* mark the attr fork not in use */
2301 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2302 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2303 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2305 /* Don't attempt to replay owner changes for a deleted inode */
2306 spin_lock(&iip->ili_lock);
2307 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2308 spin_unlock(&iip->ili_lock);
2311 * Bump the generation count so no one will be confused
2312 * by reincarnations of this inode.
2314 VFS_I(ip)->i_generation++;
2315 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2318 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2325 * This is called to unpin an inode. The caller must have the inode locked
2326 * in at least shared mode so that the buffer cannot be subsequently pinned
2327 * once someone is waiting for it to be unpinned.
2331 struct xfs_inode *ip)
2333 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2335 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2337 /* Give the log a push to start the unpinning I/O */
2338 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2344 struct xfs_inode *ip)
2346 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2347 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2352 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2353 if (xfs_ipincount(ip))
2355 } while (xfs_ipincount(ip));
2356 finish_wait(wq, &wait.wq_entry);
2361 struct xfs_inode *ip)
2363 if (xfs_ipincount(ip))
2364 __xfs_iunpin_wait(ip);
2368 * Removing an inode from the namespace involves removing the directory entry
2369 * and dropping the link count on the inode. Removing the directory entry can
2370 * result in locking an AGF (directory blocks were freed) and removing a link
2371 * count can result in placing the inode on an unlinked list which results in
2374 * The big problem here is that we have an ordering constraint on AGF and AGI
2375 * locking - inode allocation locks the AGI, then can allocate a new extent for
2376 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2377 * removes the inode from the unlinked list, requiring that we lock the AGI
2378 * first, and then freeing the inode can result in an inode chunk being freed
2379 * and hence freeing disk space requiring that we lock an AGF.
2381 * Hence the ordering that is imposed by other parts of the code is AGI before
2382 * AGF. This means we cannot remove the directory entry before we drop the inode
2383 * reference count and put it on the unlinked list as this results in a lock
2384 * order of AGF then AGI, and this can deadlock against inode allocation and
2385 * freeing. Therefore we must drop the link counts before we remove the
2388 * This is still safe from a transactional point of view - it is not until we
2389 * get to xfs_defer_finish() that we have the possibility of multiple
2390 * transactions in this operation. Hence as long as we remove the directory
2391 * entry and drop the link count in the first transaction of the remove
2392 * operation, there are no transactional constraints on the ordering here.
2397 struct xfs_name *name,
2400 xfs_mount_t *mp = dp->i_mount;
2401 xfs_trans_t *tp = NULL;
2402 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2407 trace_xfs_remove(dp, name);
2409 if (xfs_is_shutdown(mp))
2412 error = xfs_qm_dqattach(dp);
2416 error = xfs_qm_dqattach(ip);
2421 * We try to get the real space reservation first, allowing for
2422 * directory btree deletion(s) implying possible bmap insert(s). If we
2423 * can't get the space reservation then we use 0 instead, and avoid the
2424 * bmap btree insert(s) in the directory code by, if the bmap insert
2425 * tries to happen, instead trimming the LAST block from the directory.
2427 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2428 * the directory code can handle a reservationless update and we don't
2429 * want to prevent a user from trying to free space by deleting things.
2431 resblks = XFS_REMOVE_SPACE_RES(mp);
2432 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2435 ASSERT(error != -ENOSPC);
2440 * If we're removing a directory perform some additional validation.
2443 ASSERT(VFS_I(ip)->i_nlink >= 2);
2444 if (VFS_I(ip)->i_nlink != 2) {
2446 goto out_trans_cancel;
2448 if (!xfs_dir_isempty(ip)) {
2450 goto out_trans_cancel;
2453 /* Drop the link from ip's "..". */
2454 error = xfs_droplink(tp, dp);
2456 goto out_trans_cancel;
2458 /* Drop the "." link from ip to self. */
2459 error = xfs_droplink(tp, ip);
2461 goto out_trans_cancel;
2464 * Point the unlinked child directory's ".." entry to the root
2465 * directory to eliminate back-references to inodes that may
2466 * get freed before the child directory is closed. If the fs
2467 * gets shrunk, this can lead to dirent inode validation errors.
2469 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2470 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2471 tp->t_mountp->m_sb.sb_rootino, 0);
2473 goto out_trans_cancel;
2477 * When removing a non-directory we need to log the parent
2478 * inode here. For a directory this is done implicitly
2479 * by the xfs_droplink call for the ".." entry.
2481 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2483 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2485 /* Drop the link from dp to ip. */
2486 error = xfs_droplink(tp, ip);
2488 goto out_trans_cancel;
2490 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2492 ASSERT(error != -ENOENT);
2493 goto out_trans_cancel;
2497 * If this is a synchronous mount, make sure that the
2498 * remove transaction goes to disk before returning to
2501 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2502 xfs_trans_set_sync(tp);
2504 error = xfs_trans_commit(tp);
2508 if (is_dir && xfs_inode_is_filestream(ip))
2509 xfs_filestream_deassociate(ip);
2514 xfs_trans_cancel(tp);
2520 * Enter all inodes for a rename transaction into a sorted array.
2522 #define __XFS_SORT_INODES 5
2524 xfs_sort_for_rename(
2525 struct xfs_inode *dp1, /* in: old (source) directory inode */
2526 struct xfs_inode *dp2, /* in: new (target) directory inode */
2527 struct xfs_inode *ip1, /* in: inode of old entry */
2528 struct xfs_inode *ip2, /* in: inode of new entry */
2529 struct xfs_inode *wip, /* in: whiteout inode */
2530 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2531 int *num_inodes) /* in/out: inodes in array */
2535 ASSERT(*num_inodes == __XFS_SORT_INODES);
2536 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2539 * i_tab contains a list of pointers to inodes. We initialize
2540 * the table here & we'll sort it. We will then use it to
2541 * order the acquisition of the inode locks.
2543 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2556 * Sort the elements via bubble sort. (Remember, there are at
2557 * most 5 elements to sort, so this is adequate.)
2559 for (i = 0; i < *num_inodes; i++) {
2560 for (j = 1; j < *num_inodes; j++) {
2561 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2562 struct xfs_inode *temp = i_tab[j];
2563 i_tab[j] = i_tab[j-1];
2572 struct xfs_trans *tp)
2575 * If this is a synchronous mount, make sure that the rename transaction
2576 * goes to disk before returning to the user.
2578 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2579 xfs_trans_set_sync(tp);
2581 return xfs_trans_commit(tp);
2585 * xfs_cross_rename()
2587 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2591 struct xfs_trans *tp,
2592 struct xfs_inode *dp1,
2593 struct xfs_name *name1,
2594 struct xfs_inode *ip1,
2595 struct xfs_inode *dp2,
2596 struct xfs_name *name2,
2597 struct xfs_inode *ip2,
2605 /* Swap inode number for dirent in first parent */
2606 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2608 goto out_trans_abort;
2610 /* Swap inode number for dirent in second parent */
2611 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2613 goto out_trans_abort;
2616 * If we're renaming one or more directories across different parents,
2617 * update the respective ".." entries (and link counts) to match the new
2621 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2623 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2624 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2625 dp1->i_ino, spaceres);
2627 goto out_trans_abort;
2629 /* transfer ip2 ".." reference to dp1 */
2630 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2631 error = xfs_droplink(tp, dp2);
2633 goto out_trans_abort;
2634 xfs_bumplink(tp, dp1);
2638 * Although ip1 isn't changed here, userspace needs
2639 * to be warned about the change, so that applications
2640 * relying on it (like backup ones), will properly
2643 ip1_flags |= XFS_ICHGTIME_CHG;
2644 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2647 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2648 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2649 dp2->i_ino, spaceres);
2651 goto out_trans_abort;
2653 /* transfer ip1 ".." reference to dp2 */
2654 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2655 error = xfs_droplink(tp, dp1);
2657 goto out_trans_abort;
2658 xfs_bumplink(tp, dp2);
2662 * Although ip2 isn't changed here, userspace needs
2663 * to be warned about the change, so that applications
2664 * relying on it (like backup ones), will properly
2667 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2668 ip2_flags |= XFS_ICHGTIME_CHG;
2673 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2674 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2677 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2678 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2681 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2682 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2684 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2685 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2686 return xfs_finish_rename(tp);
2689 xfs_trans_cancel(tp);
2694 * xfs_rename_alloc_whiteout()
2696 * Return a referenced, unlinked, unlocked inode that can be used as a
2697 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2698 * crash between allocating the inode and linking it into the rename transaction
2699 * recovery will free the inode and we won't leak it.
2702 xfs_rename_alloc_whiteout(
2703 struct mnt_idmap *idmap,
2704 struct xfs_name *src_name,
2705 struct xfs_inode *dp,
2706 struct xfs_inode **wip)
2708 struct xfs_inode *tmpfile;
2712 error = xfs_create_tmpfile(idmap, dp, S_IFCHR | WHITEOUT_MODE,
2717 name.name = src_name->name;
2718 name.len = src_name->len;
2719 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2721 xfs_finish_inode_setup(tmpfile);
2727 * Prepare the tmpfile inode as if it were created through the VFS.
2728 * Complete the inode setup and flag it as linkable. nlink is already
2729 * zero, so we can skip the drop_nlink.
2731 xfs_setup_iops(tmpfile);
2732 xfs_finish_inode_setup(tmpfile);
2733 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2744 struct mnt_idmap *idmap,
2745 struct xfs_inode *src_dp,
2746 struct xfs_name *src_name,
2747 struct xfs_inode *src_ip,
2748 struct xfs_inode *target_dp,
2749 struct xfs_name *target_name,
2750 struct xfs_inode *target_ip,
2753 struct xfs_mount *mp = src_dp->i_mount;
2754 struct xfs_trans *tp;
2755 struct xfs_inode *wip = NULL; /* whiteout inode */
2756 struct xfs_inode *inodes[__XFS_SORT_INODES];
2758 int num_inodes = __XFS_SORT_INODES;
2759 bool new_parent = (src_dp != target_dp);
2760 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2762 bool retried = false;
2763 int error, nospace_error = 0;
2765 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2767 if ((flags & RENAME_EXCHANGE) && !target_ip)
2771 * If we are doing a whiteout operation, allocate the whiteout inode
2772 * we will be placing at the target and ensure the type is set
2775 if (flags & RENAME_WHITEOUT) {
2776 error = xfs_rename_alloc_whiteout(idmap, src_name,
2781 /* setup target dirent info as whiteout */
2782 src_name->type = XFS_DIR3_FT_CHRDEV;
2785 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2786 inodes, &num_inodes);
2790 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2791 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2792 if (error == -ENOSPC) {
2793 nospace_error = error;
2795 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2799 goto out_release_wip;
2802 * Attach the dquots to the inodes
2804 error = xfs_qm_vop_rename_dqattach(inodes);
2806 goto out_trans_cancel;
2809 * Lock all the participating inodes. Depending upon whether
2810 * the target_name exists in the target directory, and
2811 * whether the target directory is the same as the source
2812 * directory, we can lock from 2 to 5 inodes.
2814 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2817 * Join all the inodes to the transaction. From this point on,
2818 * we can rely on either trans_commit or trans_cancel to unlock
2821 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2823 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2824 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2826 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2828 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2831 * If we are using project inheritance, we only allow renames
2832 * into our tree when the project IDs are the same; else the
2833 * tree quota mechanism would be circumvented.
2835 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2836 target_dp->i_projid != src_ip->i_projid)) {
2838 goto out_trans_cancel;
2841 /* RENAME_EXCHANGE is unique from here on. */
2842 if (flags & RENAME_EXCHANGE)
2843 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2844 target_dp, target_name, target_ip,
2848 * Try to reserve quota to handle an expansion of the target directory.
2849 * We'll allow the rename to continue in reservationless mode if we hit
2850 * a space usage constraint. If we trigger reservationless mode, save
2851 * the errno if there isn't any free space in the target directory.
2853 if (spaceres != 0) {
2854 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2856 if (error == -EDQUOT || error == -ENOSPC) {
2858 xfs_trans_cancel(tp);
2859 xfs_blockgc_free_quota(target_dp, 0);
2864 nospace_error = error;
2869 goto out_trans_cancel;
2873 * Check for expected errors before we dirty the transaction
2874 * so we can return an error without a transaction abort.
2876 if (target_ip == NULL) {
2878 * If there's no space reservation, check the entry will
2879 * fit before actually inserting it.
2882 error = xfs_dir_canenter(tp, target_dp, target_name);
2884 goto out_trans_cancel;
2888 * If target exists and it's a directory, check that whether
2889 * it can be destroyed.
2891 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
2892 (!xfs_dir_isempty(target_ip) ||
2893 (VFS_I(target_ip)->i_nlink > 2))) {
2895 goto out_trans_cancel;
2900 * Lock the AGI buffers we need to handle bumping the nlink of the
2901 * whiteout inode off the unlinked list and to handle dropping the
2902 * nlink of the target inode. Per locking order rules, do this in
2903 * increasing AG order and before directory block allocation tries to
2904 * grab AGFs because we grab AGIs before AGFs.
2906 * The (vfs) caller must ensure that if src is a directory then
2907 * target_ip is either null or an empty directory.
2909 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2910 if (inodes[i] == wip ||
2911 (inodes[i] == target_ip &&
2912 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2913 struct xfs_perag *pag;
2916 pag = xfs_perag_get(mp,
2917 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2918 error = xfs_read_agi(pag, tp, &bp);
2921 goto out_trans_cancel;
2926 * Directory entry creation below may acquire the AGF. Remove
2927 * the whiteout from the unlinked list first to preserve correct
2928 * AGI/AGF locking order. This dirties the transaction so failures
2929 * after this point will abort and log recovery will clean up the
2932 * For whiteouts, we need to bump the link count on the whiteout
2933 * inode. After this point, we have a real link, clear the tmpfile
2934 * state flag from the inode so it doesn't accidentally get misused
2938 struct xfs_perag *pag;
2940 ASSERT(VFS_I(wip)->i_nlink == 0);
2942 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
2943 error = xfs_iunlink_remove(tp, pag, wip);
2946 goto out_trans_cancel;
2948 xfs_bumplink(tp, wip);
2949 VFS_I(wip)->i_state &= ~I_LINKABLE;
2953 * Set up the target.
2955 if (target_ip == NULL) {
2957 * If target does not exist and the rename crosses
2958 * directories, adjust the target directory link count
2959 * to account for the ".." reference from the new entry.
2961 error = xfs_dir_createname(tp, target_dp, target_name,
2962 src_ip->i_ino, spaceres);
2964 goto out_trans_cancel;
2966 xfs_trans_ichgtime(tp, target_dp,
2967 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2969 if (new_parent && src_is_directory) {
2970 xfs_bumplink(tp, target_dp);
2972 } else { /* target_ip != NULL */
2974 * Link the source inode under the target name.
2975 * If the source inode is a directory and we are moving
2976 * it across directories, its ".." entry will be
2977 * inconsistent until we replace that down below.
2979 * In case there is already an entry with the same
2980 * name at the destination directory, remove it first.
2982 error = xfs_dir_replace(tp, target_dp, target_name,
2983 src_ip->i_ino, spaceres);
2985 goto out_trans_cancel;
2987 xfs_trans_ichgtime(tp, target_dp,
2988 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2991 * Decrement the link count on the target since the target
2992 * dir no longer points to it.
2994 error = xfs_droplink(tp, target_ip);
2996 goto out_trans_cancel;
2998 if (src_is_directory) {
3000 * Drop the link from the old "." entry.
3002 error = xfs_droplink(tp, target_ip);
3004 goto out_trans_cancel;
3006 } /* target_ip != NULL */
3009 * Remove the source.
3011 if (new_parent && src_is_directory) {
3013 * Rewrite the ".." entry to point to the new
3016 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3017 target_dp->i_ino, spaceres);
3018 ASSERT(error != -EEXIST);
3020 goto out_trans_cancel;
3024 * We always want to hit the ctime on the source inode.
3026 * This isn't strictly required by the standards since the source
3027 * inode isn't really being changed, but old unix file systems did
3028 * it and some incremental backup programs won't work without it.
3030 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3031 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3034 * Adjust the link count on src_dp. This is necessary when
3035 * renaming a directory, either within one parent when
3036 * the target existed, or across two parent directories.
3038 if (src_is_directory && (new_parent || target_ip != NULL)) {
3041 * Decrement link count on src_directory since the
3042 * entry that's moved no longer points to it.
3044 error = xfs_droplink(tp, src_dp);
3046 goto out_trans_cancel;
3050 * For whiteouts, we only need to update the source dirent with the
3051 * inode number of the whiteout inode rather than removing it
3055 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3058 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3062 goto out_trans_cancel;
3064 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3065 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3067 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3069 error = xfs_finish_rename(tp);
3075 xfs_trans_cancel(tp);
3079 if (error == -ENOSPC && nospace_error)
3080 error = nospace_error;
3086 struct xfs_inode *ip,
3089 struct xfs_inode_log_item *iip = ip->i_itemp;
3090 struct xfs_dinode *dip;
3091 struct xfs_mount *mp = ip->i_mount;
3094 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3095 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3096 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3097 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3098 ASSERT(iip->ili_item.li_buf == bp);
3100 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3103 * We don't flush the inode if any of the following checks fail, but we
3104 * do still update the log item and attach to the backing buffer as if
3105 * the flush happened. This is a formality to facilitate predictable
3106 * error handling as the caller will shutdown and fail the buffer.
3108 error = -EFSCORRUPTED;
3109 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3110 mp, XFS_ERRTAG_IFLUSH_1)) {
3111 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3112 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3113 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3116 if (S_ISREG(VFS_I(ip)->i_mode)) {
3118 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3119 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3120 mp, XFS_ERRTAG_IFLUSH_3)) {
3121 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3122 "%s: Bad regular inode %llu, ptr "PTR_FMT,
3123 __func__, ip->i_ino, ip);
3126 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3128 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3129 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3130 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3131 mp, XFS_ERRTAG_IFLUSH_4)) {
3132 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3133 "%s: Bad directory inode %llu, ptr "PTR_FMT,
3134 __func__, ip->i_ino, ip);
3138 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3139 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3140 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3141 "%s: detected corrupt incore inode %llu, "
3142 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3143 __func__, ip->i_ino,
3144 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3148 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3149 mp, XFS_ERRTAG_IFLUSH_6)) {
3150 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3151 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3152 __func__, ip->i_ino, ip->i_forkoff, ip);
3157 * Inode item log recovery for v2 inodes are dependent on the flushiter
3158 * count for correct sequencing. We bump the flush iteration count so
3159 * we can detect flushes which postdate a log record during recovery.
3160 * This is redundant as we now log every change and hence this can't
3161 * happen but we need to still do it to ensure backwards compatibility
3162 * with old kernels that predate logging all inode changes.
3164 if (!xfs_has_v3inodes(mp))
3168 * If there are inline format data / attr forks attached to this inode,
3169 * make sure they are not corrupt.
3171 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3172 xfs_ifork_verify_local_data(ip))
3174 if (xfs_inode_has_attr_fork(ip) &&
3175 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3176 xfs_ifork_verify_local_attr(ip))
3180 * Copy the dirty parts of the inode into the on-disk inode. We always
3181 * copy out the core of the inode, because if the inode is dirty at all
3184 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3186 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3187 if (!xfs_has_v3inodes(mp)) {
3188 if (ip->i_flushiter == DI_MAX_FLUSH)
3189 ip->i_flushiter = 0;
3192 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3193 if (xfs_inode_has_attr_fork(ip))
3194 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3197 * We've recorded everything logged in the inode, so we'd like to clear
3198 * the ili_fields bits so we don't log and flush things unnecessarily.
3199 * However, we can't stop logging all this information until the data
3200 * we've copied into the disk buffer is written to disk. If we did we
3201 * might overwrite the copy of the inode in the log with all the data
3202 * after re-logging only part of it, and in the face of a crash we
3203 * wouldn't have all the data we need to recover.
3205 * What we do is move the bits to the ili_last_fields field. When
3206 * logging the inode, these bits are moved back to the ili_fields field.
3207 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3208 * we know that the information those bits represent is permanently on
3209 * disk. As long as the flush completes before the inode is logged
3210 * again, then both ili_fields and ili_last_fields will be cleared.
3214 spin_lock(&iip->ili_lock);
3215 iip->ili_last_fields = iip->ili_fields;
3216 iip->ili_fields = 0;
3217 iip->ili_fsync_fields = 0;
3218 spin_unlock(&iip->ili_lock);
3221 * Store the current LSN of the inode so that we can tell whether the
3222 * item has moved in the AIL from xfs_buf_inode_iodone().
3224 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3225 &iip->ili_item.li_lsn);
3227 /* generate the checksum. */
3228 xfs_dinode_calc_crc(mp, dip);
3233 * Non-blocking flush of dirty inode metadata into the backing buffer.
3235 * The caller must have a reference to the inode and hold the cluster buffer
3236 * locked. The function will walk across all the inodes on the cluster buffer it
3237 * can find and lock without blocking, and flush them to the cluster buffer.
3239 * On successful flushing of at least one inode, the caller must write out the
3240 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3241 * the caller needs to release the buffer. On failure, the filesystem will be
3242 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3249 struct xfs_mount *mp = bp->b_mount;
3250 struct xfs_log_item *lip, *n;
3251 struct xfs_inode *ip;
3252 struct xfs_inode_log_item *iip;
3257 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3258 * will remove itself from the list.
3260 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3261 iip = (struct xfs_inode_log_item *)lip;
3262 ip = iip->ili_inode;
3265 * Quick and dirty check to avoid locks if possible.
3267 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3269 if (xfs_ipincount(ip))
3273 * The inode is still attached to the buffer, which means it is
3274 * dirty but reclaim might try to grab it. Check carefully for
3275 * that, and grab the ilock while still holding the i_flags_lock
3276 * to guarantee reclaim will not be able to reclaim this inode
3277 * once we drop the i_flags_lock.
3279 spin_lock(&ip->i_flags_lock);
3280 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3281 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3282 spin_unlock(&ip->i_flags_lock);
3287 * ILOCK will pin the inode against reclaim and prevent
3288 * concurrent transactions modifying the inode while we are
3289 * flushing the inode. If we get the lock, set the flushing
3290 * state before we drop the i_flags_lock.
3292 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3293 spin_unlock(&ip->i_flags_lock);
3296 __xfs_iflags_set(ip, XFS_IFLUSHING);
3297 spin_unlock(&ip->i_flags_lock);
3300 * Abort flushing this inode if we are shut down because the
3301 * inode may not currently be in the AIL. This can occur when
3302 * log I/O failure unpins the inode without inserting into the
3303 * AIL, leaving a dirty/unpinned inode attached to the buffer
3304 * that otherwise looks like it should be flushed.
3306 if (xlog_is_shutdown(mp->m_log)) {
3307 xfs_iunpin_wait(ip);
3308 xfs_iflush_abort(ip);
3309 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3314 /* don't block waiting on a log force to unpin dirty inodes */
3315 if (xfs_ipincount(ip)) {
3316 xfs_iflags_clear(ip, XFS_IFLUSHING);
3317 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3321 if (!xfs_inode_clean(ip))
3322 error = xfs_iflush(ip, bp);
3324 xfs_iflags_clear(ip, XFS_IFLUSHING);
3325 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3333 * Shutdown first so we kill the log before we release this
3334 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3335 * of the log, failing it before the _log_ is shut down can
3336 * result in the log tail being moved forward in the journal
3337 * on disk because log writes can still be taking place. Hence
3338 * unpinning the tail will allow the ICREATE intent to be
3339 * removed from the log an recovery will fail with uninitialised
3340 * inode cluster buffers.
3342 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3343 bp->b_flags |= XBF_ASYNC;
3344 xfs_buf_ioend_fail(bp);
3351 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3352 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3357 /* Release an inode. */
3360 struct xfs_inode *ip)
3362 trace_xfs_irele(ip, _RET_IP_);
3367 * Ensure all commited transactions touching the inode are written to the log.
3370 xfs_log_force_inode(
3371 struct xfs_inode *ip)
3375 xfs_ilock(ip, XFS_ILOCK_SHARED);
3376 if (xfs_ipincount(ip))
3377 seq = ip->i_itemp->ili_commit_seq;
3378 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3382 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3386 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3387 * abide vfs locking order (lowest pointer value goes first) and breaking the
3388 * layout leases before proceeding. The loop is needed because we cannot call
3389 * the blocking break_layout() with the iolocks held, and therefore have to
3390 * back out both locks.
3393 xfs_iolock_two_inodes_and_break_layout(
3403 /* Wait to break both inodes' layouts before we start locking. */
3404 error = break_layout(src, true);
3408 error = break_layout(dest, true);
3413 /* Lock one inode and make sure nobody got in and leased it. */
3415 error = break_layout(src, false);
3418 if (error == -EWOULDBLOCK)
3426 /* Lock the other inode and make sure nobody got in and leased it. */
3427 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3428 error = break_layout(dest, false);
3432 if (error == -EWOULDBLOCK)
3441 xfs_mmaplock_two_inodes_and_break_dax_layout(
3442 struct xfs_inode *ip1,
3443 struct xfs_inode *ip2)
3449 if (ip1->i_ino > ip2->i_ino)
3454 /* Lock the first inode */
3455 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3456 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3457 if (error || retry) {
3458 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3459 if (error == 0 && retry)
3467 /* Nested lock the second inode */
3468 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3470 * We cannot use xfs_break_dax_layouts() directly here because it may
3471 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3472 * for this nested lock case.
3474 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3475 if (page && page_ref_count(page) != 1) {
3476 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3477 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3485 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3490 struct xfs_inode *ip1,
3491 struct xfs_inode *ip2)
3495 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3499 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3500 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3502 inode_unlock(VFS_I(ip2));
3504 inode_unlock(VFS_I(ip1));
3508 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3509 VFS_I(ip2)->i_mapping);
3514 /* Unlock both inodes to allow IO and mmap activity. */
3516 xfs_iunlock2_io_mmap(
3517 struct xfs_inode *ip1,
3518 struct xfs_inode *ip2)
3520 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3521 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3523 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3525 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3526 VFS_I(ip2)->i_mapping);
3528 inode_unlock(VFS_I(ip2));
3530 inode_unlock(VFS_I(ip1));