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"
15 #include "xfs_mount.h"
16 #include "xfs_defer.h"
17 #include "xfs_inode.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_inode_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 kmem_zone_t *xfs_inode_zone;
42 * Used in xfs_itruncate_extents(). This is the maximum number of extents
43 * freed from a file in a single transaction.
45 #define XFS_ITRUNC_MAX_EXTENTS 2
47 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
48 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
51 * helper function to extract extent size hint from inode
58 * No point in aligning allocations if we need to COW to actually
61 if (xfs_is_always_cow_inode(ip))
63 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
65 if (XFS_IS_REALTIME_INODE(ip))
66 return ip->i_mount->m_sb.sb_rextsize;
71 * Helper function to extract CoW extent size hint from inode.
72 * Between the extent size hint and the CoW extent size hint, we
73 * return the greater of the two. If the value is zero (automatic),
74 * use the default size.
77 xfs_get_cowextsz_hint(
83 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
85 b = xfs_get_extsz_hint(ip);
89 return XFS_DEFAULT_COWEXTSZ_HINT;
94 * These two are wrapper routines around the xfs_ilock() routine used to
95 * centralize some grungy code. They are used in places that wish to lock the
96 * inode solely for reading the extents. The reason these places can't just
97 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
98 * bringing in of the extents from disk for a file in b-tree format. If the
99 * inode is in b-tree format, then we need to lock the inode exclusively until
100 * the extents are read in. Locking it exclusively all the time would limit
101 * our parallelism unnecessarily, though. What we do instead is check to see
102 * if the extents have been read in yet, and only lock the inode exclusively
105 * The functions return a value which should be given to the corresponding
106 * xfs_iunlock() call.
109 xfs_ilock_data_map_shared(
110 struct xfs_inode *ip)
112 uint lock_mode = XFS_ILOCK_SHARED;
114 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE &&
115 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
116 lock_mode = XFS_ILOCK_EXCL;
117 xfs_ilock(ip, lock_mode);
122 xfs_ilock_attr_map_shared(
123 struct xfs_inode *ip)
125 uint lock_mode = XFS_ILOCK_SHARED;
128 ip->i_afp->if_format == XFS_DINODE_FMT_BTREE &&
129 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
136 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
137 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
138 * various combinations of the locks to be obtained.
140 * The 3 locks should always be ordered so that the IO lock is obtained first,
141 * the mmap lock second and the ilock last in order to prevent deadlock.
143 * Basic locking order:
145 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
147 * mmap_lock locking order:
149 * i_rwsem -> page lock -> mmap_lock
150 * mmap_lock -> i_mmap_lock -> page_lock
152 * The difference in mmap_lock locking order mean that we cannot hold the
153 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
154 * fault in pages during copy in/out (for buffered IO) or require the mmap_lock
155 * in get_user_pages() to map the user pages into the kernel address space for
156 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
157 * page faults already hold the mmap_lock.
159 * Hence to serialise fully against both syscall and mmap based IO, we need to
160 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
161 * taken in places where we need to invalidate the page cache in a race
162 * free manner (e.g. truncate, hole punch and other extent manipulation
170 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
173 * You can't set both SHARED and EXCL for the same lock,
174 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
175 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
177 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
178 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
179 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
180 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
181 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
182 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
183 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
185 if (lock_flags & XFS_IOLOCK_EXCL) {
186 down_write_nested(&VFS_I(ip)->i_rwsem,
187 XFS_IOLOCK_DEP(lock_flags));
188 } else if (lock_flags & XFS_IOLOCK_SHARED) {
189 down_read_nested(&VFS_I(ip)->i_rwsem,
190 XFS_IOLOCK_DEP(lock_flags));
193 if (lock_flags & XFS_MMAPLOCK_EXCL)
194 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
195 else if (lock_flags & XFS_MMAPLOCK_SHARED)
196 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
198 if (lock_flags & XFS_ILOCK_EXCL)
199 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
200 else if (lock_flags & XFS_ILOCK_SHARED)
201 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
205 * This is just like xfs_ilock(), except that the caller
206 * is guaranteed not to sleep. It returns 1 if it gets
207 * the requested locks and 0 otherwise. If the IO lock is
208 * obtained but the inode lock cannot be, then the IO lock
209 * is dropped before returning.
211 * ip -- the inode being locked
212 * lock_flags -- this parameter indicates the inode's locks to be
213 * to be locked. See the comment for xfs_ilock() for a list
221 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
224 * You can't set both SHARED and EXCL for the same lock,
225 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
226 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
228 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
229 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
230 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
231 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
232 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
233 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
234 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
236 if (lock_flags & XFS_IOLOCK_EXCL) {
237 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
239 } else if (lock_flags & XFS_IOLOCK_SHARED) {
240 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
244 if (lock_flags & XFS_MMAPLOCK_EXCL) {
245 if (!mrtryupdate(&ip->i_mmaplock))
246 goto out_undo_iolock;
247 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
248 if (!mrtryaccess(&ip->i_mmaplock))
249 goto out_undo_iolock;
252 if (lock_flags & XFS_ILOCK_EXCL) {
253 if (!mrtryupdate(&ip->i_lock))
254 goto out_undo_mmaplock;
255 } else if (lock_flags & XFS_ILOCK_SHARED) {
256 if (!mrtryaccess(&ip->i_lock))
257 goto out_undo_mmaplock;
262 if (lock_flags & XFS_MMAPLOCK_EXCL)
263 mrunlock_excl(&ip->i_mmaplock);
264 else if (lock_flags & XFS_MMAPLOCK_SHARED)
265 mrunlock_shared(&ip->i_mmaplock);
267 if (lock_flags & XFS_IOLOCK_EXCL)
268 up_write(&VFS_I(ip)->i_rwsem);
269 else if (lock_flags & XFS_IOLOCK_SHARED)
270 up_read(&VFS_I(ip)->i_rwsem);
276 * xfs_iunlock() is used to drop the inode locks acquired with
277 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
278 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
279 * that we know which locks to drop.
281 * ip -- the inode being unlocked
282 * lock_flags -- this parameter indicates the inode's locks to be
283 * to be unlocked. See the comment for xfs_ilock() for a list
284 * of valid values for this parameter.
293 * You can't set both SHARED and EXCL for the same lock,
294 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
295 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
297 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
298 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
299 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
300 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
301 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
302 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
303 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
304 ASSERT(lock_flags != 0);
306 if (lock_flags & XFS_IOLOCK_EXCL)
307 up_write(&VFS_I(ip)->i_rwsem);
308 else if (lock_flags & XFS_IOLOCK_SHARED)
309 up_read(&VFS_I(ip)->i_rwsem);
311 if (lock_flags & XFS_MMAPLOCK_EXCL)
312 mrunlock_excl(&ip->i_mmaplock);
313 else if (lock_flags & XFS_MMAPLOCK_SHARED)
314 mrunlock_shared(&ip->i_mmaplock);
316 if (lock_flags & XFS_ILOCK_EXCL)
317 mrunlock_excl(&ip->i_lock);
318 else if (lock_flags & XFS_ILOCK_SHARED)
319 mrunlock_shared(&ip->i_lock);
321 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
325 * give up write locks. the i/o lock cannot be held nested
326 * if it is being demoted.
333 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
335 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
337 if (lock_flags & XFS_ILOCK_EXCL)
338 mrdemote(&ip->i_lock);
339 if (lock_flags & XFS_MMAPLOCK_EXCL)
340 mrdemote(&ip->i_mmaplock);
341 if (lock_flags & XFS_IOLOCK_EXCL)
342 downgrade_write(&VFS_I(ip)->i_rwsem);
344 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
347 #if defined(DEBUG) || defined(XFS_WARN)
353 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
354 if (!(lock_flags & XFS_ILOCK_SHARED))
355 return !!ip->i_lock.mr_writer;
356 return rwsem_is_locked(&ip->i_lock.mr_lock);
359 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
360 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
361 return !!ip->i_mmaplock.mr_writer;
362 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
365 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
366 if (!(lock_flags & XFS_IOLOCK_SHARED))
367 return !debug_locks ||
368 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
369 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
378 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
379 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
380 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
381 * errors and warnings.
383 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
385 xfs_lockdep_subclass_ok(
388 return subclass < MAX_LOCKDEP_SUBCLASSES;
391 #define xfs_lockdep_subclass_ok(subclass) (true)
395 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
396 * value. This can be called for any type of inode lock combination, including
397 * parent locking. Care must be taken to ensure we don't overrun the subclass
398 * storage fields in the class mask we build.
401 xfs_lock_inumorder(int lock_mode, int subclass)
405 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
407 ASSERT(xfs_lockdep_subclass_ok(subclass));
409 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
410 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
411 class += subclass << XFS_IOLOCK_SHIFT;
414 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
415 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
416 class += subclass << XFS_MMAPLOCK_SHIFT;
419 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
420 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
421 class += subclass << XFS_ILOCK_SHIFT;
424 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
428 * The following routine will lock n inodes in exclusive mode. We assume the
429 * caller calls us with the inodes in i_ino order.
431 * We need to detect deadlock where an inode that we lock is in the AIL and we
432 * start waiting for another inode that is locked by a thread in a long running
433 * transaction (such as truncate). This can result in deadlock since the long
434 * running trans might need to wait for the inode we just locked in order to
435 * push the tail and free space in the log.
437 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
438 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
439 * lock more than one at a time, lockdep will report false positives saying we
440 * have violated locking orders.
444 struct xfs_inode **ips,
448 int attempts = 0, i, j, try_lock;
449 struct xfs_log_item *lp;
452 * Currently supports between 2 and 5 inodes with exclusive locking. We
453 * support an arbitrary depth of locking here, but absolute limits on
454 * inodes depend on the type of locking and the limits placed by
455 * lockdep annotations in xfs_lock_inumorder. These are all checked by
458 ASSERT(ips && inodes >= 2 && inodes <= 5);
459 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
461 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
463 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
464 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
466 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
468 if (lock_mode & XFS_IOLOCK_EXCL) {
469 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
470 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
471 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
476 for (; i < inodes; i++) {
479 if (i && (ips[i] == ips[i - 1])) /* Already locked */
483 * If try_lock is not set yet, make sure all locked inodes are
484 * not in the AIL. If any are, set try_lock to be used later.
487 for (j = (i - 1); j >= 0 && !try_lock; j--) {
488 lp = &ips[j]->i_itemp->ili_item;
489 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
495 * If any of the previous locks we have locked is in the AIL,
496 * we must TRY to get the second and subsequent locks. If
497 * we can't get any, we must release all we have
501 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
505 /* try_lock means we have an inode locked that is in the AIL. */
507 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
511 * Unlock all previous guys and try again. xfs_iunlock will try
512 * to push the tail if the inode is in the AIL.
515 for (j = i - 1; j >= 0; j--) {
517 * Check to see if we've already unlocked this one. Not
518 * the first one going back, and the inode ptr is the
521 if (j != (i - 1) && ips[j] == ips[j + 1])
524 xfs_iunlock(ips[j], lock_mode);
527 if ((attempts % 5) == 0) {
528 delay(1); /* Don't just spin the CPU */
537 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
538 * the mmaplock or the ilock, but not more than one type at a time. If we lock
539 * more than one at a time, lockdep will report false positives saying we have
540 * violated locking orders. The iolock must be double-locked separately since
541 * we use i_rwsem for that. We now support taking one lock EXCL and the other
546 struct xfs_inode *ip0,
548 struct xfs_inode *ip1,
551 struct xfs_inode *temp;
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 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
563 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
565 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
566 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
567 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
569 ASSERT(ip0->i_ino != ip1->i_ino);
571 if (ip0->i_ino > ip1->i_ino) {
575 mode_temp = ip0_mode;
577 ip1_mode = mode_temp;
581 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
584 * If the first lock we have locked is in the AIL, we must TRY to get
585 * the second lock. If we can't get it, we must release the first one
588 lp = &ip0->i_itemp->ili_item;
589 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
590 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
591 xfs_iunlock(ip0, ip0_mode);
592 if ((++attempts % 5) == 0)
593 delay(1); /* Don't just spin the CPU */
597 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
609 if (di_flags & XFS_DIFLAG_ANY) {
610 if (di_flags & XFS_DIFLAG_REALTIME)
611 flags |= FS_XFLAG_REALTIME;
612 if (di_flags & XFS_DIFLAG_PREALLOC)
613 flags |= FS_XFLAG_PREALLOC;
614 if (di_flags & XFS_DIFLAG_IMMUTABLE)
615 flags |= FS_XFLAG_IMMUTABLE;
616 if (di_flags & XFS_DIFLAG_APPEND)
617 flags |= FS_XFLAG_APPEND;
618 if (di_flags & XFS_DIFLAG_SYNC)
619 flags |= FS_XFLAG_SYNC;
620 if (di_flags & XFS_DIFLAG_NOATIME)
621 flags |= FS_XFLAG_NOATIME;
622 if (di_flags & XFS_DIFLAG_NODUMP)
623 flags |= FS_XFLAG_NODUMP;
624 if (di_flags & XFS_DIFLAG_RTINHERIT)
625 flags |= FS_XFLAG_RTINHERIT;
626 if (di_flags & XFS_DIFLAG_PROJINHERIT)
627 flags |= FS_XFLAG_PROJINHERIT;
628 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
629 flags |= FS_XFLAG_NOSYMLINKS;
630 if (di_flags & XFS_DIFLAG_EXTSIZE)
631 flags |= FS_XFLAG_EXTSIZE;
632 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
633 flags |= FS_XFLAG_EXTSZINHERIT;
634 if (di_flags & XFS_DIFLAG_NODEFRAG)
635 flags |= FS_XFLAG_NODEFRAG;
636 if (di_flags & XFS_DIFLAG_FILESTREAM)
637 flags |= FS_XFLAG_FILESTREAM;
640 if (di_flags2 & XFS_DIFLAG2_ANY) {
641 if (di_flags2 & XFS_DIFLAG2_DAX)
642 flags |= FS_XFLAG_DAX;
643 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
644 flags |= FS_XFLAG_COWEXTSIZE;
648 flags |= FS_XFLAG_HASATTR;
655 struct xfs_inode *ip)
657 return _xfs_dic2xflags(ip->i_diflags, ip->i_diflags2, XFS_IFORK_Q(ip));
661 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
662 * is allowed, otherwise it has to be an exact match. If a CI match is found,
663 * ci_name->name will point to a the actual name (caller must free) or
664 * will be set to NULL if an exact match is found.
669 struct xfs_name *name,
671 struct xfs_name *ci_name)
676 trace_xfs_lookup(dp, name);
678 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
681 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
685 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
693 kmem_free(ci_name->name);
699 /* Propagate di_flags from a parent inode to a child inode. */
701 xfs_inode_inherit_flags(
702 struct xfs_inode *ip,
703 const struct xfs_inode *pip)
705 unsigned int di_flags = 0;
706 umode_t mode = VFS_I(ip)->i_mode;
709 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
710 di_flags |= XFS_DIFLAG_RTINHERIT;
711 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
712 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
713 ip->i_extsize = pip->i_extsize;
715 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
716 di_flags |= XFS_DIFLAG_PROJINHERIT;
717 } else if (S_ISREG(mode)) {
718 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
719 xfs_sb_version_hasrealtime(&ip->i_mount->m_sb))
720 di_flags |= XFS_DIFLAG_REALTIME;
721 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
722 di_flags |= XFS_DIFLAG_EXTSIZE;
723 ip->i_extsize = pip->i_extsize;
726 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
728 di_flags |= XFS_DIFLAG_NOATIME;
729 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
731 di_flags |= XFS_DIFLAG_NODUMP;
732 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
734 di_flags |= XFS_DIFLAG_SYNC;
735 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
736 xfs_inherit_nosymlinks)
737 di_flags |= XFS_DIFLAG_NOSYMLINKS;
738 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
739 xfs_inherit_nodefrag)
740 di_flags |= XFS_DIFLAG_NODEFRAG;
741 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
742 di_flags |= XFS_DIFLAG_FILESTREAM;
744 ip->i_diflags |= di_flags;
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 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
754 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
755 ip->i_cowextsize = pip->i_cowextsize;
757 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
758 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
762 * Initialise a newly allocated inode and return the in-core inode to the
763 * caller locked exclusively.
767 struct user_namespace *mnt_userns,
768 struct xfs_trans *tp,
769 struct xfs_inode *pip,
776 struct xfs_inode **ipp)
778 struct inode *dir = pip ? VFS_I(pip) : NULL;
779 struct xfs_mount *mp = tp->t_mountp;
780 struct xfs_inode *ip;
783 struct timespec64 tv;
787 * Protect against obviously corrupt allocation btree records. Later
788 * xfs_iget checks will catch re-allocation of other active in-memory
789 * and on-disk inodes. If we don't catch reallocating the parent inode
790 * here we will deadlock in xfs_iget() so we have to do these checks
793 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
794 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
795 return -EFSCORRUPTED;
799 * Get the in-core inode with the lock held exclusively to prevent
800 * others from looking at until we're done.
802 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
808 set_nlink(inode, nlink);
809 inode->i_rdev = rdev;
812 if (dir && !(dir->i_mode & S_ISGID) &&
813 (mp->m_flags & XFS_MOUNT_GRPID)) {
814 inode->i_uid = fsuid_into_mnt(mnt_userns);
815 inode->i_gid = dir->i_gid;
816 inode->i_mode = mode;
818 inode_init_owner(mnt_userns, inode, dir, mode);
822 * If the group ID of the new file does not match the effective group
823 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
824 * (and only if the irix_sgid_inherit compatibility variable is set).
826 if (irix_sgid_inherit &&
827 (inode->i_mode & S_ISGID) &&
828 !in_group_p(i_gid_into_mnt(mnt_userns, inode)))
829 inode->i_mode &= ~S_ISGID;
832 ip->i_df.if_nextents = 0;
833 ASSERT(ip->i_nblocks == 0);
835 tv = current_time(inode);
843 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
844 inode_set_iversion(inode, 1);
845 ip->i_cowextsize = 0;
849 flags = XFS_ILOG_CORE;
850 switch (mode & S_IFMT) {
855 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
856 ip->i_df.if_flags = 0;
857 flags |= XFS_ILOG_DEV;
861 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
862 xfs_inode_inherit_flags(ip, pip);
863 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
864 xfs_inode_inherit_flags2(ip, pip);
867 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
868 ip->i_df.if_flags = XFS_IFEXTENTS;
869 ip->i_df.if_bytes = 0;
870 ip->i_df.if_u1.if_root = NULL;
877 * If we need to create attributes immediately after allocating the
878 * inode, initialise an empty attribute fork right now. We use the
879 * default fork offset for attributes here as we don't know exactly what
880 * size or how many attributes we might be adding. We can do this
881 * safely here because we know the data fork is completely empty and
882 * this saves us from needing to run a separate transaction to set the
883 * fork offset in the immediate future.
886 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
887 ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0);
891 * Log the new values stuffed into the inode.
893 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
894 xfs_trans_log_inode(tp, ip, flags);
896 /* now that we have an i_mode we can setup the inode structure */
904 * Allocates a new inode from disk and return a pointer to the incore copy. This
905 * routine will internally commit the current transaction and allocate a new one
906 * if we needed to allocate more on-disk free inodes to perform the requested
909 * If we are allocating quota inodes, we do not have a parent inode to attach to
910 * or associate with (i.e. dp == NULL) because they are not linked into the
911 * directory structure - they are attached directly to the superblock - and so
916 struct user_namespace *mnt_userns,
917 struct xfs_trans **tpp,
918 struct xfs_inode *dp,
924 struct xfs_inode **ipp)
926 struct xfs_buf *agibp;
927 xfs_ino_t parent_ino = dp ? dp->i_ino : 0;
931 ASSERT((*tpp)->t_flags & XFS_TRANS_PERM_LOG_RES);
934 * Call the space management code to pick the on-disk inode to be
937 error = xfs_dialloc_select_ag(tpp, parent_ino, mode, &agibp);
944 /* Allocate an inode from the selected AG */
945 error = xfs_dialloc_ag(*tpp, agibp, parent_ino, &ino);
948 ASSERT(ino != NULLFSINO);
950 return xfs_init_new_inode(mnt_userns, *tpp, dp, ino, mode, nlink, rdev,
951 prid, init_xattrs, ipp);
955 * Decrement the link count on an inode & log the change. If this causes the
956 * link count to go to zero, move the inode to AGI unlinked list so that it can
957 * be freed when the last active reference goes away via xfs_inactive().
959 static int /* error */
964 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
966 drop_nlink(VFS_I(ip));
967 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
969 if (VFS_I(ip)->i_nlink)
972 return xfs_iunlink(tp, ip);
976 * Increment the link count on an inode & log the change.
983 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
985 inc_nlink(VFS_I(ip));
986 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
991 struct user_namespace *mnt_userns,
993 struct xfs_name *name,
999 int is_dir = S_ISDIR(mode);
1000 struct xfs_mount *mp = dp->i_mount;
1001 struct xfs_inode *ip = NULL;
1002 struct xfs_trans *tp = NULL;
1004 bool unlock_dp_on_error = false;
1006 struct xfs_dquot *udqp = NULL;
1007 struct xfs_dquot *gdqp = NULL;
1008 struct xfs_dquot *pdqp = NULL;
1009 struct xfs_trans_res *tres;
1012 trace_xfs_create(dp, name);
1014 if (XFS_FORCED_SHUTDOWN(mp))
1017 prid = xfs_get_initial_prid(dp);
1020 * Make sure that we have allocated dquot(s) on disk.
1022 error = xfs_qm_vop_dqalloc(dp, fsuid_into_mnt(mnt_userns),
1023 fsgid_into_mnt(mnt_userns), prid,
1024 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1025 &udqp, &gdqp, &pdqp);
1030 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1031 tres = &M_RES(mp)->tr_mkdir;
1033 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1034 tres = &M_RES(mp)->tr_create;
1038 * Initially assume that the file does not exist and
1039 * reserve the resources for that case. If that is not
1040 * the case we'll drop the one we have and get a more
1041 * appropriate transaction later.
1043 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1045 if (error == -ENOSPC) {
1046 /* flush outstanding delalloc blocks and retry */
1047 xfs_flush_inodes(mp);
1048 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1052 goto out_release_dquots;
1054 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1055 unlock_dp_on_error = true;
1057 error = xfs_iext_count_may_overflow(dp, XFS_DATA_FORK,
1058 XFS_IEXT_DIR_MANIP_CNT(mp));
1060 goto out_trans_cancel;
1063 * A newly created regular or special file just has one directory
1064 * entry pointing to them, but a directory also the "." entry
1065 * pointing to itself.
1067 error = xfs_dir_ialloc(mnt_userns, &tp, dp, mode, is_dir ? 2 : 1, rdev,
1068 prid, init_xattrs, &ip);
1070 goto out_trans_cancel;
1073 * Now we join the directory inode to the transaction. We do not do it
1074 * earlier because xfs_dir_ialloc might commit the previous transaction
1075 * (and release all the locks). An error from here on will result in
1076 * the transaction cancel unlocking dp so don't do it explicitly in the
1079 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1080 unlock_dp_on_error = false;
1082 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1083 resblks - XFS_IALLOC_SPACE_RES(mp));
1085 ASSERT(error != -ENOSPC);
1086 goto out_trans_cancel;
1088 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1089 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1092 error = xfs_dir_init(tp, ip, dp);
1094 goto out_trans_cancel;
1096 xfs_bumplink(tp, dp);
1100 * If this is a synchronous mount, make sure that the
1101 * create transaction goes to disk before returning to
1104 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1105 xfs_trans_set_sync(tp);
1108 * Attach the dquot(s) to the inodes and modify them incore.
1109 * These ids of the inode couldn't have changed since the new
1110 * inode has been locked ever since it was created.
1112 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1114 error = xfs_trans_commit(tp);
1116 goto out_release_inode;
1118 xfs_qm_dqrele(udqp);
1119 xfs_qm_dqrele(gdqp);
1120 xfs_qm_dqrele(pdqp);
1126 xfs_trans_cancel(tp);
1129 * Wait until after the current transaction is aborted to finish the
1130 * setup of the inode and release the inode. This prevents recursive
1131 * transactions and deadlocks from xfs_inactive.
1134 xfs_finish_inode_setup(ip);
1138 xfs_qm_dqrele(udqp);
1139 xfs_qm_dqrele(gdqp);
1140 xfs_qm_dqrele(pdqp);
1142 if (unlock_dp_on_error)
1143 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1149 struct user_namespace *mnt_userns,
1150 struct xfs_inode *dp,
1152 struct xfs_inode **ipp)
1154 struct xfs_mount *mp = dp->i_mount;
1155 struct xfs_inode *ip = NULL;
1156 struct xfs_trans *tp = NULL;
1159 struct xfs_dquot *udqp = NULL;
1160 struct xfs_dquot *gdqp = NULL;
1161 struct xfs_dquot *pdqp = NULL;
1162 struct xfs_trans_res *tres;
1165 if (XFS_FORCED_SHUTDOWN(mp))
1168 prid = xfs_get_initial_prid(dp);
1171 * Make sure that we have allocated dquot(s) on disk.
1173 error = xfs_qm_vop_dqalloc(dp, fsuid_into_mnt(mnt_userns),
1174 fsgid_into_mnt(mnt_userns), prid,
1175 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1176 &udqp, &gdqp, &pdqp);
1180 resblks = XFS_IALLOC_SPACE_RES(mp);
1181 tres = &M_RES(mp)->tr_create_tmpfile;
1183 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1186 goto out_release_dquots;
1188 error = xfs_dir_ialloc(mnt_userns, &tp, dp, mode, 0, 0, prid,
1191 goto out_trans_cancel;
1193 if (mp->m_flags & XFS_MOUNT_WSYNC)
1194 xfs_trans_set_sync(tp);
1197 * Attach the dquot(s) to the inodes and modify them incore.
1198 * These ids of the inode couldn't have changed since the new
1199 * inode has been locked ever since it was created.
1201 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1203 error = xfs_iunlink(tp, ip);
1205 goto out_trans_cancel;
1207 error = xfs_trans_commit(tp);
1209 goto out_release_inode;
1211 xfs_qm_dqrele(udqp);
1212 xfs_qm_dqrele(gdqp);
1213 xfs_qm_dqrele(pdqp);
1219 xfs_trans_cancel(tp);
1222 * Wait until after the current transaction is aborted to finish the
1223 * setup of the inode and release the inode. This prevents recursive
1224 * transactions and deadlocks from xfs_inactive.
1227 xfs_finish_inode_setup(ip);
1231 xfs_qm_dqrele(udqp);
1232 xfs_qm_dqrele(gdqp);
1233 xfs_qm_dqrele(pdqp);
1242 struct xfs_name *target_name)
1244 xfs_mount_t *mp = tdp->i_mount;
1249 trace_xfs_link(tdp, target_name);
1251 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1253 if (XFS_FORCED_SHUTDOWN(mp))
1256 error = xfs_qm_dqattach(sip);
1260 error = xfs_qm_dqattach(tdp);
1264 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1265 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1266 if (error == -ENOSPC) {
1268 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1273 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1275 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1276 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1278 error = xfs_iext_count_may_overflow(tdp, XFS_DATA_FORK,
1279 XFS_IEXT_DIR_MANIP_CNT(mp));
1284 * If we are using project inheritance, we only allow hard link
1285 * creation in our tree when the project IDs are the same; else
1286 * the tree quota mechanism could be circumvented.
1288 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1289 tdp->i_projid != sip->i_projid)) {
1295 error = xfs_dir_canenter(tp, tdp, target_name);
1301 * Handle initial link state of O_TMPFILE inode
1303 if (VFS_I(sip)->i_nlink == 0) {
1304 error = xfs_iunlink_remove(tp, sip);
1309 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1313 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1314 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1316 xfs_bumplink(tp, sip);
1319 * If this is a synchronous mount, make sure that the
1320 * link transaction goes to disk before returning to
1323 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1324 xfs_trans_set_sync(tp);
1326 return xfs_trans_commit(tp);
1329 xfs_trans_cancel(tp);
1334 /* Clear the reflink flag and the cowblocks tag if possible. */
1336 xfs_itruncate_clear_reflink_flags(
1337 struct xfs_inode *ip)
1339 struct xfs_ifork *dfork;
1340 struct xfs_ifork *cfork;
1342 if (!xfs_is_reflink_inode(ip))
1344 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1345 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1346 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1347 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1348 if (cfork->if_bytes == 0)
1349 xfs_inode_clear_cowblocks_tag(ip);
1353 * Free up the underlying blocks past new_size. The new size must be smaller
1354 * than the current size. This routine can be used both for the attribute and
1355 * data fork, and does not modify the inode size, which is left to the caller.
1357 * The transaction passed to this routine must have made a permanent log
1358 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1359 * given transaction and start new ones, so make sure everything involved in
1360 * the transaction is tidy before calling here. Some transaction will be
1361 * returned to the caller to be committed. The incoming transaction must
1362 * already include the inode, and both inode locks must be held exclusively.
1363 * The inode must also be "held" within the transaction. On return the inode
1364 * will be "held" within the returned transaction. This routine does NOT
1365 * require any disk space to be reserved for it within the transaction.
1367 * If we get an error, we must return with the inode locked and linked into the
1368 * current transaction. This keeps things simple for the higher level code,
1369 * because it always knows that the inode is locked and held in the transaction
1370 * that returns to it whether errors occur or not. We don't mark the inode
1371 * dirty on error so that transactions can be easily aborted if possible.
1374 xfs_itruncate_extents_flags(
1375 struct xfs_trans **tpp,
1376 struct xfs_inode *ip,
1378 xfs_fsize_t new_size,
1381 struct xfs_mount *mp = ip->i_mount;
1382 struct xfs_trans *tp = *tpp;
1383 xfs_fileoff_t first_unmap_block;
1384 xfs_filblks_t unmap_len;
1387 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1388 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1389 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1390 ASSERT(new_size <= XFS_ISIZE(ip));
1391 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1392 ASSERT(ip->i_itemp != NULL);
1393 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1394 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1396 trace_xfs_itruncate_extents_start(ip, new_size);
1398 flags |= xfs_bmapi_aflag(whichfork);
1401 * Since it is possible for space to become allocated beyond
1402 * the end of the file (in a crash where the space is allocated
1403 * but the inode size is not yet updated), simply remove any
1404 * blocks which show up between the new EOF and the maximum
1405 * possible file size.
1407 * We have to free all the blocks to the bmbt maximum offset, even if
1408 * the page cache can't scale that far.
1410 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1411 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1412 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1416 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1417 while (unmap_len > 0) {
1418 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1419 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1420 flags, XFS_ITRUNC_MAX_EXTENTS);
1424 /* free the just unmapped extents */
1425 error = xfs_defer_finish(&tp);
1430 if (whichfork == XFS_DATA_FORK) {
1431 /* Remove all pending CoW reservations. */
1432 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1433 first_unmap_block, XFS_MAX_FILEOFF, true);
1437 xfs_itruncate_clear_reflink_flags(ip);
1441 * Always re-log the inode so that our permanent transaction can keep
1442 * on rolling it forward in the log.
1444 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1446 trace_xfs_itruncate_extents_end(ip, new_size);
1457 xfs_mount_t *mp = ip->i_mount;
1460 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1463 /* If this is a read-only mount, don't do this (would generate I/O) */
1464 if (mp->m_flags & XFS_MOUNT_RDONLY)
1467 if (!XFS_FORCED_SHUTDOWN(mp)) {
1471 * If we previously truncated this file and removed old data
1472 * in the process, we want to initiate "early" writeout on
1473 * the last close. This is an attempt to combat the notorious
1474 * NULL files problem which is particularly noticeable from a
1475 * truncate down, buffered (re-)write (delalloc), followed by
1476 * a crash. What we are effectively doing here is
1477 * significantly reducing the time window where we'd otherwise
1478 * be exposed to that problem.
1480 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1482 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1483 if (ip->i_delayed_blks > 0) {
1484 error = filemap_flush(VFS_I(ip)->i_mapping);
1491 if (VFS_I(ip)->i_nlink == 0)
1494 if (xfs_can_free_eofblocks(ip, false)) {
1497 * Check if the inode is being opened, written and closed
1498 * frequently and we have delayed allocation blocks outstanding
1499 * (e.g. streaming writes from the NFS server), truncating the
1500 * blocks past EOF will cause fragmentation to occur.
1502 * In this case don't do the truncation, but we have to be
1503 * careful how we detect this case. Blocks beyond EOF show up as
1504 * i_delayed_blks even when the inode is clean, so we need to
1505 * truncate them away first before checking for a dirty release.
1506 * Hence on the first dirty close we will still remove the
1507 * speculative allocation, but after that we will leave it in
1510 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1513 * If we can't get the iolock just skip truncating the blocks
1514 * past EOF because we could deadlock with the mmap_lock
1515 * otherwise. We'll get another chance to drop them once the
1516 * last reference to the inode is dropped, so we'll never leak
1517 * blocks permanently.
1519 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1520 error = xfs_free_eofblocks(ip);
1521 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1526 /* delalloc blocks after truncation means it really is dirty */
1527 if (ip->i_delayed_blks)
1528 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1534 * xfs_inactive_truncate
1536 * Called to perform a truncate when an inode becomes unlinked.
1539 xfs_inactive_truncate(
1540 struct xfs_inode *ip)
1542 struct xfs_mount *mp = ip->i_mount;
1543 struct xfs_trans *tp;
1546 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1548 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1551 xfs_ilock(ip, XFS_ILOCK_EXCL);
1552 xfs_trans_ijoin(tp, ip, 0);
1555 * Log the inode size first to prevent stale data exposure in the event
1556 * of a system crash before the truncate completes. See the related
1557 * comment in xfs_vn_setattr_size() for details.
1559 ip->i_disk_size = 0;
1560 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1562 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1564 goto error_trans_cancel;
1566 ASSERT(ip->i_df.if_nextents == 0);
1568 error = xfs_trans_commit(tp);
1572 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1576 xfs_trans_cancel(tp);
1578 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1583 * xfs_inactive_ifree()
1585 * Perform the inode free when an inode is unlinked.
1589 struct xfs_inode *ip)
1591 struct xfs_mount *mp = ip->i_mount;
1592 struct xfs_trans *tp;
1596 * We try to use a per-AG reservation for any block needed by the finobt
1597 * tree, but as the finobt feature predates the per-AG reservation
1598 * support a degraded file system might not have enough space for the
1599 * reservation at mount time. In that case try to dip into the reserved
1602 * Send a warning if the reservation does happen to fail, as the inode
1603 * now remains allocated and sits on the unlinked list until the fs is
1606 if (unlikely(mp->m_finobt_nores)) {
1607 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1608 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1611 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1614 if (error == -ENOSPC) {
1615 xfs_warn_ratelimited(mp,
1616 "Failed to remove inode(s) from unlinked list. "
1617 "Please free space, unmount and run xfs_repair.");
1619 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1625 * We do not hold the inode locked across the entire rolling transaction
1626 * here. We only need to hold it for the first transaction that
1627 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1628 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1629 * here breaks the relationship between cluster buffer invalidation and
1630 * stale inode invalidation on cluster buffer item journal commit
1631 * completion, and can result in leaving dirty stale inodes hanging
1634 * We have no need for serialising this inode operation against other
1635 * operations - we freed the inode and hence reallocation is required
1636 * and that will serialise on reallocating the space the deferops need
1637 * to free. Hence we can unlock the inode on the first commit of
1638 * the transaction rather than roll it right through the deferops. This
1639 * avoids relogging the XFS_ISTALE inode.
1641 * We check that xfs_ifree() hasn't grown an internal transaction roll
1642 * by asserting that the inode is still locked when it returns.
1644 xfs_ilock(ip, XFS_ILOCK_EXCL);
1645 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1647 error = xfs_ifree(tp, ip);
1648 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1651 * If we fail to free the inode, shut down. The cancel
1652 * might do that, we need to make sure. Otherwise the
1653 * inode might be lost for a long time or forever.
1655 if (!XFS_FORCED_SHUTDOWN(mp)) {
1656 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1658 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1660 xfs_trans_cancel(tp);
1665 * Credit the quota account(s). The inode is gone.
1667 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1670 * Just ignore errors at this point. There is nothing we can do except
1671 * to try to keep going. Make sure it's not a silent error.
1673 error = xfs_trans_commit(tp);
1675 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1684 * This is called when the vnode reference count for the vnode
1685 * goes to zero. If the file has been unlinked, then it must
1686 * now be truncated. Also, we clear all of the read-ahead state
1687 * kept for the inode here since the file is now closed.
1693 struct xfs_mount *mp;
1698 * If the inode is already free, then there can be nothing
1701 if (VFS_I(ip)->i_mode == 0) {
1702 ASSERT(ip->i_df.if_broot_bytes == 0);
1707 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1709 /* If this is a read-only mount, don't do this (would generate I/O) */
1710 if (mp->m_flags & XFS_MOUNT_RDONLY)
1713 /* Metadata inodes require explicit resource cleanup. */
1714 if (xfs_is_metadata_inode(ip))
1717 /* Try to clean out the cow blocks if there are any. */
1718 if (xfs_inode_has_cow_data(ip))
1719 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1721 if (VFS_I(ip)->i_nlink != 0) {
1723 * force is true because we are evicting an inode from the
1724 * cache. Post-eof blocks must be freed, lest we end up with
1725 * broken free space accounting.
1727 * Note: don't bother with iolock here since lockdep complains
1728 * about acquiring it in reclaim context. We have the only
1729 * reference to the inode at this point anyways.
1731 if (xfs_can_free_eofblocks(ip, true))
1732 xfs_free_eofblocks(ip);
1737 if (S_ISREG(VFS_I(ip)->i_mode) &&
1738 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1739 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1742 error = xfs_qm_dqattach(ip);
1746 if (S_ISLNK(VFS_I(ip)->i_mode))
1747 error = xfs_inactive_symlink(ip);
1749 error = xfs_inactive_truncate(ip);
1754 * If there are attributes associated with the file then blow them away
1755 * now. The code calls a routine that recursively deconstructs the
1756 * attribute fork. If also blows away the in-core attribute fork.
1758 if (XFS_IFORK_Q(ip)) {
1759 error = xfs_attr_inactive(ip);
1765 ASSERT(ip->i_forkoff == 0);
1770 error = xfs_inactive_ifree(ip);
1775 * Release the dquots held by inode, if any.
1777 xfs_qm_dqdetach(ip);
1781 * In-Core Unlinked List Lookups
1782 * =============================
1784 * Every inode is supposed to be reachable from some other piece of metadata
1785 * with the exception of the root directory. Inodes with a connection to a
1786 * file descriptor but not linked from anywhere in the on-disk directory tree
1787 * are collectively known as unlinked inodes, though the filesystem itself
1788 * maintains links to these inodes so that on-disk metadata are consistent.
1790 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1791 * header contains a number of buckets that point to an inode, and each inode
1792 * record has a pointer to the next inode in the hash chain. This
1793 * singly-linked list causes scaling problems in the iunlink remove function
1794 * because we must walk that list to find the inode that points to the inode
1795 * being removed from the unlinked hash bucket list.
1797 * What if we modelled the unlinked list as a collection of records capturing
1798 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1799 * have a fast way to look up unlinked list predecessors, which avoids the
1800 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1803 * Because this is a backref cache, we ignore operational failures since the
1804 * iunlink code can fall back to the slow bucket walk. The only errors that
1805 * should bubble out are for obviously incorrect situations.
1807 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1808 * access or have otherwise provided for concurrency control.
1811 /* Capture a "X.next_unlinked = Y" relationship. */
1812 struct xfs_iunlink {
1813 struct rhash_head iu_rhash_head;
1814 xfs_agino_t iu_agino; /* X */
1815 xfs_agino_t iu_next_unlinked; /* Y */
1818 /* Unlinked list predecessor lookup hashtable construction */
1820 xfs_iunlink_obj_cmpfn(
1821 struct rhashtable_compare_arg *arg,
1824 const xfs_agino_t *key = arg->key;
1825 const struct xfs_iunlink *iu = obj;
1827 if (iu->iu_next_unlinked != *key)
1832 static const struct rhashtable_params xfs_iunlink_hash_params = {
1833 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1834 .key_len = sizeof(xfs_agino_t),
1835 .key_offset = offsetof(struct xfs_iunlink,
1837 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1838 .automatic_shrinking = true,
1839 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1843 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1844 * relation is found.
1847 xfs_iunlink_lookup_backref(
1848 struct xfs_perag *pag,
1851 struct xfs_iunlink *iu;
1853 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1854 xfs_iunlink_hash_params);
1855 return iu ? iu->iu_agino : NULLAGINO;
1859 * Take ownership of an iunlink cache entry and insert it into the hash table.
1860 * If successful, the entry will be owned by the cache; if not, it is freed.
1861 * Either way, the caller does not own @iu after this call.
1864 xfs_iunlink_insert_backref(
1865 struct xfs_perag *pag,
1866 struct xfs_iunlink *iu)
1870 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1871 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1873 * Fail loudly if there already was an entry because that's a sign of
1874 * corruption of in-memory data. Also fail loudly if we see an error
1875 * code we didn't anticipate from the rhashtable code. Currently we
1876 * only anticipate ENOMEM.
1879 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1883 * Absorb any runtime errors that aren't a result of corruption because
1884 * this is a cache and we can always fall back to bucket list scanning.
1886 if (error != 0 && error != -EEXIST)
1891 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1893 xfs_iunlink_add_backref(
1894 struct xfs_perag *pag,
1895 xfs_agino_t prev_agino,
1896 xfs_agino_t this_agino)
1898 struct xfs_iunlink *iu;
1900 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
1903 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
1904 iu->iu_agino = prev_agino;
1905 iu->iu_next_unlinked = this_agino;
1907 return xfs_iunlink_insert_backref(pag, iu);
1911 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
1912 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
1913 * wasn't any such entry then we don't bother.
1916 xfs_iunlink_change_backref(
1917 struct xfs_perag *pag,
1919 xfs_agino_t next_unlinked)
1921 struct xfs_iunlink *iu;
1924 /* Look up the old entry; if there wasn't one then exit. */
1925 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1926 xfs_iunlink_hash_params);
1931 * Remove the entry. This shouldn't ever return an error, but if we
1932 * couldn't remove the old entry we don't want to add it again to the
1933 * hash table, and if the entry disappeared on us then someone's
1934 * violated the locking rules and we need to fail loudly. Either way
1935 * we cannot remove the inode because internal state is or would have
1938 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
1939 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1943 /* If there is no new next entry just free our item and return. */
1944 if (next_unlinked == NULLAGINO) {
1949 /* Update the entry and re-add it to the hash table. */
1950 iu->iu_next_unlinked = next_unlinked;
1951 return xfs_iunlink_insert_backref(pag, iu);
1954 /* Set up the in-core predecessor structures. */
1957 struct xfs_perag *pag)
1959 return rhashtable_init(&pag->pagi_unlinked_hash,
1960 &xfs_iunlink_hash_params);
1963 /* Free the in-core predecessor structures. */
1965 xfs_iunlink_free_item(
1969 struct xfs_iunlink *iu = ptr;
1970 bool *freed_anything = arg;
1972 *freed_anything = true;
1977 xfs_iunlink_destroy(
1978 struct xfs_perag *pag)
1980 bool freed_anything = false;
1982 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
1983 xfs_iunlink_free_item, &freed_anything);
1985 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
1989 * Point the AGI unlinked bucket at an inode and log the results. The caller
1990 * is responsible for validating the old value.
1993 xfs_iunlink_update_bucket(
1994 struct xfs_trans *tp,
1995 xfs_agnumber_t agno,
1996 struct xfs_buf *agibp,
1997 unsigned int bucket_index,
1998 xfs_agino_t new_agino)
2000 struct xfs_agi *agi = agibp->b_addr;
2001 xfs_agino_t old_value;
2004 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
2006 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2007 trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
2008 old_value, new_agino);
2011 * We should never find the head of the list already set to the value
2012 * passed in because either we're adding or removing ourselves from the
2015 if (old_value == new_agino) {
2016 xfs_buf_mark_corrupt(agibp);
2017 return -EFSCORRUPTED;
2020 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2021 offset = offsetof(struct xfs_agi, agi_unlinked) +
2022 (sizeof(xfs_agino_t) * bucket_index);
2023 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2027 /* Set an on-disk inode's next_unlinked pointer. */
2029 xfs_iunlink_update_dinode(
2030 struct xfs_trans *tp,
2031 xfs_agnumber_t agno,
2033 struct xfs_buf *ibp,
2034 struct xfs_dinode *dip,
2035 struct xfs_imap *imap,
2036 xfs_agino_t next_agino)
2038 struct xfs_mount *mp = tp->t_mountp;
2041 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2043 trace_xfs_iunlink_update_dinode(mp, agno, agino,
2044 be32_to_cpu(dip->di_next_unlinked), next_agino);
2046 dip->di_next_unlinked = cpu_to_be32(next_agino);
2047 offset = imap->im_boffset +
2048 offsetof(struct xfs_dinode, di_next_unlinked);
2050 /* need to recalc the inode CRC if appropriate */
2051 xfs_dinode_calc_crc(mp, dip);
2052 xfs_trans_inode_buf(tp, ibp);
2053 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2056 /* Set an in-core inode's unlinked pointer and return the old value. */
2058 xfs_iunlink_update_inode(
2059 struct xfs_trans *tp,
2060 struct xfs_inode *ip,
2061 xfs_agnumber_t agno,
2062 xfs_agino_t next_agino,
2063 xfs_agino_t *old_next_agino)
2065 struct xfs_mount *mp = tp->t_mountp;
2066 struct xfs_dinode *dip;
2067 struct xfs_buf *ibp;
2068 xfs_agino_t old_value;
2071 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2073 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &ibp);
2076 dip = xfs_buf_offset(ibp, ip->i_imap.im_boffset);
2078 /* Make sure the old pointer isn't garbage. */
2079 old_value = be32_to_cpu(dip->di_next_unlinked);
2080 if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2081 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2082 sizeof(*dip), __this_address);
2083 error = -EFSCORRUPTED;
2088 * Since we're updating a linked list, we should never find that the
2089 * current pointer is the same as the new value, unless we're
2090 * terminating the list.
2092 *old_next_agino = old_value;
2093 if (old_value == next_agino) {
2094 if (next_agino != NULLAGINO) {
2095 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2096 dip, sizeof(*dip), __this_address);
2097 error = -EFSCORRUPTED;
2102 /* Ok, update the new pointer. */
2103 xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2104 ibp, dip, &ip->i_imap, next_agino);
2107 xfs_trans_brelse(tp, ibp);
2112 * This is called when the inode's link count has gone to 0 or we are creating
2113 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2115 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2116 * list when the inode is freed.
2120 struct xfs_trans *tp,
2121 struct xfs_inode *ip)
2123 struct xfs_mount *mp = tp->t_mountp;
2124 struct xfs_agi *agi;
2125 struct xfs_buf *agibp;
2126 xfs_agino_t next_agino;
2127 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2128 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2129 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2132 ASSERT(VFS_I(ip)->i_nlink == 0);
2133 ASSERT(VFS_I(ip)->i_mode != 0);
2134 trace_xfs_iunlink(ip);
2136 /* Get the agi buffer first. It ensures lock ordering on the list. */
2137 error = xfs_read_agi(mp, tp, agno, &agibp);
2140 agi = agibp->b_addr;
2143 * Get the index into the agi hash table for the list this inode will
2144 * go on. Make sure the pointer isn't garbage and that this inode
2145 * isn't already on the list.
2147 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2148 if (next_agino == agino ||
2149 !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2150 xfs_buf_mark_corrupt(agibp);
2151 return -EFSCORRUPTED;
2154 if (next_agino != NULLAGINO) {
2155 xfs_agino_t old_agino;
2158 * There is already another inode in the bucket, so point this
2159 * inode to the current head of the list.
2161 error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2165 ASSERT(old_agino == NULLAGINO);
2168 * agino has been unlinked, add a backref from the next inode
2171 error = xfs_iunlink_add_backref(agibp->b_pag, agino, next_agino);
2176 /* Point the head of the list to point to this inode. */
2177 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2180 /* Return the imap, dinode pointer, and buffer for an inode. */
2182 xfs_iunlink_map_ino(
2183 struct xfs_trans *tp,
2184 xfs_agnumber_t agno,
2186 struct xfs_imap *imap,
2187 struct xfs_dinode **dipp,
2188 struct xfs_buf **bpp)
2190 struct xfs_mount *mp = tp->t_mountp;
2194 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2196 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2201 error = xfs_imap_to_bp(mp, tp, imap, bpp);
2203 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2208 *dipp = xfs_buf_offset(*bpp, imap->im_boffset);
2213 * Walk the unlinked chain from @head_agino until we find the inode that
2214 * points to @target_agino. Return the inode number, map, dinode pointer,
2215 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2217 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2218 * @agino, @imap, @dipp, and @bpp are all output parameters.
2220 * Do not call this function if @target_agino is the head of the list.
2223 xfs_iunlink_map_prev(
2224 struct xfs_trans *tp,
2225 xfs_agnumber_t agno,
2226 xfs_agino_t head_agino,
2227 xfs_agino_t target_agino,
2229 struct xfs_imap *imap,
2230 struct xfs_dinode **dipp,
2231 struct xfs_buf **bpp,
2232 struct xfs_perag *pag)
2234 struct xfs_mount *mp = tp->t_mountp;
2235 xfs_agino_t next_agino;
2238 ASSERT(head_agino != target_agino);
2241 /* See if our backref cache can find it faster. */
2242 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2243 if (*agino != NULLAGINO) {
2244 error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2248 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2252 * If we get here the cache contents were corrupt, so drop the
2253 * buffer and fall back to walking the bucket list.
2255 xfs_trans_brelse(tp, *bpp);
2260 trace_xfs_iunlink_map_prev_fallback(mp, agno);
2262 /* Otherwise, walk the entire bucket until we find it. */
2263 next_agino = head_agino;
2264 while (next_agino != target_agino) {
2265 xfs_agino_t unlinked_agino;
2268 xfs_trans_brelse(tp, *bpp);
2270 *agino = next_agino;
2271 error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2276 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2278 * Make sure this pointer is valid and isn't an obvious
2281 if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2282 next_agino == unlinked_agino) {
2283 XFS_CORRUPTION_ERROR(__func__,
2284 XFS_ERRLEVEL_LOW, mp,
2285 *dipp, sizeof(**dipp));
2286 error = -EFSCORRUPTED;
2289 next_agino = unlinked_agino;
2296 * Pull the on-disk inode from the AGI unlinked list.
2300 struct xfs_trans *tp,
2301 struct xfs_inode *ip)
2303 struct xfs_mount *mp = tp->t_mountp;
2304 struct xfs_agi *agi;
2305 struct xfs_buf *agibp;
2306 struct xfs_buf *last_ibp;
2307 struct xfs_dinode *last_dip = NULL;
2308 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2309 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2310 xfs_agino_t next_agino;
2311 xfs_agino_t head_agino;
2312 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2315 trace_xfs_iunlink_remove(ip);
2317 /* Get the agi buffer first. It ensures lock ordering on the list. */
2318 error = xfs_read_agi(mp, tp, agno, &agibp);
2321 agi = agibp->b_addr;
2324 * Get the index into the agi hash table for the list this inode will
2325 * go on. Make sure the head pointer isn't garbage.
2327 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2328 if (!xfs_verify_agino(mp, agno, head_agino)) {
2329 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2331 return -EFSCORRUPTED;
2335 * Set our inode's next_unlinked pointer to NULL and then return
2336 * the old pointer value so that we can update whatever was previous
2337 * to us in the list to point to whatever was next in the list.
2339 error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2344 * If there was a backref pointing from the next inode back to this
2345 * one, remove it because we've removed this inode from the list.
2347 * Later, if this inode was in the middle of the list we'll update
2348 * this inode's backref to point from the next inode.
2350 if (next_agino != NULLAGINO) {
2351 error = xfs_iunlink_change_backref(agibp->b_pag, next_agino,
2357 if (head_agino != agino) {
2358 struct xfs_imap imap;
2359 xfs_agino_t prev_agino;
2361 /* We need to search the list for the inode being freed. */
2362 error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2363 &prev_agino, &imap, &last_dip, &last_ibp,
2368 /* Point the previous inode on the list to the next inode. */
2369 xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2370 last_dip, &imap, next_agino);
2373 * Now we deal with the backref for this inode. If this inode
2374 * pointed at a real inode, change the backref that pointed to
2375 * us to point to our old next. If this inode was the end of
2376 * the list, delete the backref that pointed to us. Note that
2377 * change_backref takes care of deleting the backref if
2378 * next_agino is NULLAGINO.
2380 return xfs_iunlink_change_backref(agibp->b_pag, agino,
2384 /* Point the head of the list to the next unlinked inode. */
2385 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2390 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2391 * mark it stale. We should only find clean inodes in this lookup that aren't
2395 xfs_ifree_mark_inode_stale(
2397 struct xfs_inode *free_ip,
2400 struct xfs_mount *mp = bp->b_mount;
2401 struct xfs_perag *pag = bp->b_pag;
2402 struct xfs_inode_log_item *iip;
2403 struct xfs_inode *ip;
2407 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2409 /* Inode not in memory, nothing to do */
2416 * because this is an RCU protected lookup, we could find a recently
2417 * freed or even reallocated inode during the lookup. We need to check
2418 * under the i_flags_lock for a valid inode here. Skip it if it is not
2419 * valid, the wrong inode or stale.
2421 spin_lock(&ip->i_flags_lock);
2422 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2423 goto out_iflags_unlock;
2426 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2427 * other inodes that we did not find in the list attached to the buffer
2428 * and are not already marked stale. If we can't lock it, back off and
2431 if (ip != free_ip) {
2432 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2433 spin_unlock(&ip->i_flags_lock);
2439 ip->i_flags |= XFS_ISTALE;
2442 * If the inode is flushing, it is already attached to the buffer. All
2443 * we needed to do here is mark the inode stale so buffer IO completion
2444 * will remove it from the AIL.
2447 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2448 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2449 ASSERT(iip->ili_last_fields);
2454 * Inodes not attached to the buffer can be released immediately.
2455 * Everything else has to go through xfs_iflush_abort() on journal
2456 * commit as the flock synchronises removal of the inode from the
2457 * cluster buffer against inode reclaim.
2459 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2462 __xfs_iflags_set(ip, XFS_IFLUSHING);
2463 spin_unlock(&ip->i_flags_lock);
2466 /* we have a dirty inode in memory that has not yet been flushed. */
2467 spin_lock(&iip->ili_lock);
2468 iip->ili_last_fields = iip->ili_fields;
2469 iip->ili_fields = 0;
2470 iip->ili_fsync_fields = 0;
2471 spin_unlock(&iip->ili_lock);
2472 ASSERT(iip->ili_last_fields);
2475 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2480 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2482 spin_unlock(&ip->i_flags_lock);
2487 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2488 * inodes that are in memory - they all must be marked stale and attached to
2489 * the cluster buffer.
2493 struct xfs_inode *free_ip,
2494 struct xfs_trans *tp,
2495 struct xfs_icluster *xic)
2497 struct xfs_mount *mp = free_ip->i_mount;
2498 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2501 xfs_ino_t inum = xic->first_ino;
2507 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2509 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2511 * The allocation bitmap tells us which inodes of the chunk were
2512 * physically allocated. Skip the cluster if an inode falls into
2515 ioffset = inum - xic->first_ino;
2516 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2517 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2521 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2522 XFS_INO_TO_AGBNO(mp, inum));
2525 * We obtain and lock the backing buffer first in the process
2526 * here to ensure dirty inodes attached to the buffer remain in
2527 * the flushing state while we mark them stale.
2529 * If we scan the in-memory inodes first, then buffer IO can
2530 * complete before we get a lock on it, and hence we may fail
2531 * to mark all the active inodes on the buffer stale.
2533 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2534 mp->m_bsize * igeo->blocks_per_cluster,
2540 * This buffer may not have been correctly initialised as we
2541 * didn't read it from disk. That's not important because we are
2542 * only using to mark the buffer as stale in the log, and to
2543 * attach stale cached inodes on it. That means it will never be
2544 * dispatched for IO. If it is, we want to know about it, and we
2545 * want it to fail. We can acheive this by adding a write
2546 * verifier to the buffer.
2548 bp->b_ops = &xfs_inode_buf_ops;
2551 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2552 * too. This requires lookups, and will skip inodes that we've
2553 * already marked XFS_ISTALE.
2555 for (i = 0; i < igeo->inodes_per_cluster; i++)
2556 xfs_ifree_mark_inode_stale(bp, free_ip, inum + i);
2558 xfs_trans_stale_inode_buf(tp, bp);
2559 xfs_trans_binval(tp, bp);
2565 * This is called to return an inode to the inode free list.
2566 * The inode should already be truncated to 0 length and have
2567 * no pages associated with it. This routine also assumes that
2568 * the inode is already a part of the transaction.
2570 * The on-disk copy of the inode will have been added to the list
2571 * of unlinked inodes in the AGI. We need to remove the inode from
2572 * that list atomically with respect to freeing it here.
2576 struct xfs_trans *tp,
2577 struct xfs_inode *ip)
2580 struct xfs_icluster xic = { 0 };
2581 struct xfs_inode_log_item *iip = ip->i_itemp;
2583 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2584 ASSERT(VFS_I(ip)->i_nlink == 0);
2585 ASSERT(ip->i_df.if_nextents == 0);
2586 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2587 ASSERT(ip->i_nblocks == 0);
2590 * Pull the on-disk inode from the AGI unlinked list.
2592 error = xfs_iunlink_remove(tp, ip);
2596 error = xfs_difree(tp, ip->i_ino, &xic);
2601 * Free any local-format data sitting around before we reset the
2602 * data fork to extents format. Note that the attr fork data has
2603 * already been freed by xfs_attr_inactive.
2605 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2606 kmem_free(ip->i_df.if_u1.if_data);
2607 ip->i_df.if_u1.if_data = NULL;
2608 ip->i_df.if_bytes = 0;
2611 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2613 ip->i_diflags2 = ip->i_mount->m_ino_geo.new_diflags2;
2614 ip->i_forkoff = 0; /* mark the attr fork not in use */
2615 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2616 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2617 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2619 /* Don't attempt to replay owner changes for a deleted inode */
2620 spin_lock(&iip->ili_lock);
2621 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2622 spin_unlock(&iip->ili_lock);
2625 * Bump the generation count so no one will be confused
2626 * by reincarnations of this inode.
2628 VFS_I(ip)->i_generation++;
2629 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2632 error = xfs_ifree_cluster(ip, tp, &xic);
2638 * This is called to unpin an inode. The caller must have the inode locked
2639 * in at least shared mode so that the buffer cannot be subsequently pinned
2640 * once someone is waiting for it to be unpinned.
2644 struct xfs_inode *ip)
2646 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2648 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2650 /* Give the log a push to start the unpinning I/O */
2651 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2657 struct xfs_inode *ip)
2659 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2660 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2665 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2666 if (xfs_ipincount(ip))
2668 } while (xfs_ipincount(ip));
2669 finish_wait(wq, &wait.wq_entry);
2674 struct xfs_inode *ip)
2676 if (xfs_ipincount(ip))
2677 __xfs_iunpin_wait(ip);
2681 * Removing an inode from the namespace involves removing the directory entry
2682 * and dropping the link count on the inode. Removing the directory entry can
2683 * result in locking an AGF (directory blocks were freed) and removing a link
2684 * count can result in placing the inode on an unlinked list which results in
2687 * The big problem here is that we have an ordering constraint on AGF and AGI
2688 * locking - inode allocation locks the AGI, then can allocate a new extent for
2689 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2690 * removes the inode from the unlinked list, requiring that we lock the AGI
2691 * first, and then freeing the inode can result in an inode chunk being freed
2692 * and hence freeing disk space requiring that we lock an AGF.
2694 * Hence the ordering that is imposed by other parts of the code is AGI before
2695 * AGF. This means we cannot remove the directory entry before we drop the inode
2696 * reference count and put it on the unlinked list as this results in a lock
2697 * order of AGF then AGI, and this can deadlock against inode allocation and
2698 * freeing. Therefore we must drop the link counts before we remove the
2701 * This is still safe from a transactional point of view - it is not until we
2702 * get to xfs_defer_finish() that we have the possibility of multiple
2703 * transactions in this operation. Hence as long as we remove the directory
2704 * entry and drop the link count in the first transaction of the remove
2705 * operation, there are no transactional constraints on the ordering here.
2710 struct xfs_name *name,
2713 xfs_mount_t *mp = dp->i_mount;
2714 xfs_trans_t *tp = NULL;
2715 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2719 trace_xfs_remove(dp, name);
2721 if (XFS_FORCED_SHUTDOWN(mp))
2724 error = xfs_qm_dqattach(dp);
2728 error = xfs_qm_dqattach(ip);
2733 * We try to get the real space reservation first,
2734 * allowing for directory btree deletion(s) implying
2735 * possible bmap insert(s). If we can't get the space
2736 * reservation then we use 0 instead, and avoid the bmap
2737 * btree insert(s) in the directory code by, if the bmap
2738 * insert tries to happen, instead trimming the LAST
2739 * block from the directory.
2741 resblks = XFS_REMOVE_SPACE_RES(mp);
2742 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2743 if (error == -ENOSPC) {
2745 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2749 ASSERT(error != -ENOSPC);
2753 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2755 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2756 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2759 * If we're removing a directory perform some additional validation.
2762 ASSERT(VFS_I(ip)->i_nlink >= 2);
2763 if (VFS_I(ip)->i_nlink != 2) {
2765 goto out_trans_cancel;
2767 if (!xfs_dir_isempty(ip)) {
2769 goto out_trans_cancel;
2772 /* Drop the link from ip's "..". */
2773 error = xfs_droplink(tp, dp);
2775 goto out_trans_cancel;
2777 /* Drop the "." link from ip to self. */
2778 error = xfs_droplink(tp, ip);
2780 goto out_trans_cancel;
2783 * When removing a non-directory we need to log the parent
2784 * inode here. For a directory this is done implicitly
2785 * by the xfs_droplink call for the ".." entry.
2787 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2789 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2791 /* Drop the link from dp to ip. */
2792 error = xfs_droplink(tp, ip);
2794 goto out_trans_cancel;
2796 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2798 ASSERT(error != -ENOENT);
2799 goto out_trans_cancel;
2803 * If this is a synchronous mount, make sure that the
2804 * remove transaction goes to disk before returning to
2807 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2808 xfs_trans_set_sync(tp);
2810 error = xfs_trans_commit(tp);
2814 if (is_dir && xfs_inode_is_filestream(ip))
2815 xfs_filestream_deassociate(ip);
2820 xfs_trans_cancel(tp);
2826 * Enter all inodes for a rename transaction into a sorted array.
2828 #define __XFS_SORT_INODES 5
2830 xfs_sort_for_rename(
2831 struct xfs_inode *dp1, /* in: old (source) directory inode */
2832 struct xfs_inode *dp2, /* in: new (target) directory inode */
2833 struct xfs_inode *ip1, /* in: inode of old entry */
2834 struct xfs_inode *ip2, /* in: inode of new entry */
2835 struct xfs_inode *wip, /* in: whiteout inode */
2836 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2837 int *num_inodes) /* in/out: inodes in array */
2841 ASSERT(*num_inodes == __XFS_SORT_INODES);
2842 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2845 * i_tab contains a list of pointers to inodes. We initialize
2846 * the table here & we'll sort it. We will then use it to
2847 * order the acquisition of the inode locks.
2849 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2862 * Sort the elements via bubble sort. (Remember, there are at
2863 * most 5 elements to sort, so this is adequate.)
2865 for (i = 0; i < *num_inodes; i++) {
2866 for (j = 1; j < *num_inodes; j++) {
2867 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2868 struct xfs_inode *temp = i_tab[j];
2869 i_tab[j] = i_tab[j-1];
2878 struct xfs_trans *tp)
2881 * If this is a synchronous mount, make sure that the rename transaction
2882 * goes to disk before returning to the user.
2884 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2885 xfs_trans_set_sync(tp);
2887 return xfs_trans_commit(tp);
2891 * xfs_cross_rename()
2893 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2897 struct xfs_trans *tp,
2898 struct xfs_inode *dp1,
2899 struct xfs_name *name1,
2900 struct xfs_inode *ip1,
2901 struct xfs_inode *dp2,
2902 struct xfs_name *name2,
2903 struct xfs_inode *ip2,
2911 /* Swap inode number for dirent in first parent */
2912 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2914 goto out_trans_abort;
2916 /* Swap inode number for dirent in second parent */
2917 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2919 goto out_trans_abort;
2922 * If we're renaming one or more directories across different parents,
2923 * update the respective ".." entries (and link counts) to match the new
2927 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2929 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2930 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2931 dp1->i_ino, spaceres);
2933 goto out_trans_abort;
2935 /* transfer ip2 ".." reference to dp1 */
2936 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2937 error = xfs_droplink(tp, dp2);
2939 goto out_trans_abort;
2940 xfs_bumplink(tp, dp1);
2944 * Although ip1 isn't changed here, userspace needs
2945 * to be warned about the change, so that applications
2946 * relying on it (like backup ones), will properly
2949 ip1_flags |= XFS_ICHGTIME_CHG;
2950 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2953 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2954 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2955 dp2->i_ino, spaceres);
2957 goto out_trans_abort;
2959 /* transfer ip1 ".." reference to dp2 */
2960 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2961 error = xfs_droplink(tp, dp1);
2963 goto out_trans_abort;
2964 xfs_bumplink(tp, dp2);
2968 * Although ip2 isn't changed here, userspace needs
2969 * to be warned about the change, so that applications
2970 * relying on it (like backup ones), will properly
2973 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2974 ip2_flags |= XFS_ICHGTIME_CHG;
2979 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2980 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2983 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2984 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2987 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2988 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2990 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2991 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2992 return xfs_finish_rename(tp);
2995 xfs_trans_cancel(tp);
3000 * xfs_rename_alloc_whiteout()
3002 * Return a referenced, unlinked, unlocked inode that can be used as a
3003 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3004 * crash between allocating the inode and linking it into the rename transaction
3005 * recovery will free the inode and we won't leak it.
3008 xfs_rename_alloc_whiteout(
3009 struct user_namespace *mnt_userns,
3010 struct xfs_inode *dp,
3011 struct xfs_inode **wip)
3013 struct xfs_inode *tmpfile;
3016 error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
3022 * Prepare the tmpfile inode as if it were created through the VFS.
3023 * Complete the inode setup and flag it as linkable. nlink is already
3024 * zero, so we can skip the drop_nlink.
3026 xfs_setup_iops(tmpfile);
3027 xfs_finish_inode_setup(tmpfile);
3028 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3039 struct user_namespace *mnt_userns,
3040 struct xfs_inode *src_dp,
3041 struct xfs_name *src_name,
3042 struct xfs_inode *src_ip,
3043 struct xfs_inode *target_dp,
3044 struct xfs_name *target_name,
3045 struct xfs_inode *target_ip,
3048 struct xfs_mount *mp = src_dp->i_mount;
3049 struct xfs_trans *tp;
3050 struct xfs_inode *wip = NULL; /* whiteout inode */
3051 struct xfs_inode *inodes[__XFS_SORT_INODES];
3053 int num_inodes = __XFS_SORT_INODES;
3054 bool new_parent = (src_dp != target_dp);
3055 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3059 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3061 if ((flags & RENAME_EXCHANGE) && !target_ip)
3065 * If we are doing a whiteout operation, allocate the whiteout inode
3066 * we will be placing at the target and ensure the type is set
3069 if (flags & RENAME_WHITEOUT) {
3070 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3071 error = xfs_rename_alloc_whiteout(mnt_userns, target_dp, &wip);
3075 /* setup target dirent info as whiteout */
3076 src_name->type = XFS_DIR3_FT_CHRDEV;
3079 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3080 inodes, &num_inodes);
3082 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3083 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3084 if (error == -ENOSPC) {
3086 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3090 goto out_release_wip;
3093 * Attach the dquots to the inodes
3095 error = xfs_qm_vop_rename_dqattach(inodes);
3097 goto out_trans_cancel;
3100 * Lock all the participating inodes. Depending upon whether
3101 * the target_name exists in the target directory, and
3102 * whether the target directory is the same as the source
3103 * directory, we can lock from 2 to 4 inodes.
3105 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3108 * Join all the inodes to the transaction. From this point on,
3109 * we can rely on either trans_commit or trans_cancel to unlock
3112 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3114 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3115 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3117 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3119 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3122 * If we are using project inheritance, we only allow renames
3123 * into our tree when the project IDs are the same; else the
3124 * tree quota mechanism would be circumvented.
3126 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
3127 target_dp->i_projid != src_ip->i_projid)) {
3129 goto out_trans_cancel;
3132 /* RENAME_EXCHANGE is unique from here on. */
3133 if (flags & RENAME_EXCHANGE)
3134 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3135 target_dp, target_name, target_ip,
3139 * Check for expected errors before we dirty the transaction
3140 * so we can return an error without a transaction abort.
3142 * Extent count overflow check:
3144 * From the perspective of src_dp, a rename operation is essentially a
3145 * directory entry remove operation. Hence the only place where we check
3146 * for extent count overflow for src_dp is in
3147 * xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns
3148 * -ENOSPC when it detects a possible extent count overflow and in
3149 * response, the higher layers of directory handling code do the
3151 * 1. Data/Free blocks: XFS lets these blocks linger until a
3152 * future remove operation removes them.
3153 * 2. Dabtree blocks: XFS swaps the blocks with the last block in the
3154 * Leaf space and unmaps the last block.
3156 * For target_dp, there are two cases depending on whether the
3157 * destination directory entry exists or not.
3159 * When destination directory entry does not exist (i.e. target_ip ==
3160 * NULL), extent count overflow check is performed only when transaction
3161 * has a non-zero sized space reservation associated with it. With a
3162 * zero-sized space reservation, XFS allows a rename operation to
3163 * continue only when the directory has sufficient free space in its
3164 * data/leaf/free space blocks to hold the new entry.
3166 * When destination directory entry exists (i.e. target_ip != NULL), all
3167 * we need to do is change the inode number associated with the already
3168 * existing entry. Hence there is no need to perform an extent count
3171 if (target_ip == NULL) {
3173 * If there's no space reservation, check the entry will
3174 * fit before actually inserting it.
3177 error = xfs_dir_canenter(tp, target_dp, target_name);
3179 goto out_trans_cancel;
3181 error = xfs_iext_count_may_overflow(target_dp,
3183 XFS_IEXT_DIR_MANIP_CNT(mp));
3185 goto out_trans_cancel;
3189 * If target exists and it's a directory, check that whether
3190 * it can be destroyed.
3192 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3193 (!xfs_dir_isempty(target_ip) ||
3194 (VFS_I(target_ip)->i_nlink > 2))) {
3196 goto out_trans_cancel;
3201 * Lock the AGI buffers we need to handle bumping the nlink of the
3202 * whiteout inode off the unlinked list and to handle dropping the
3203 * nlink of the target inode. Per locking order rules, do this in
3204 * increasing AG order and before directory block allocation tries to
3205 * grab AGFs because we grab AGIs before AGFs.
3207 * The (vfs) caller must ensure that if src is a directory then
3208 * target_ip is either null or an empty directory.
3210 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3211 if (inodes[i] == wip ||
3212 (inodes[i] == target_ip &&
3213 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3215 xfs_agnumber_t agno;
3217 agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
3218 error = xfs_read_agi(mp, tp, agno, &bp);
3220 goto out_trans_cancel;
3225 * Directory entry creation below may acquire the AGF. Remove
3226 * the whiteout from the unlinked list first to preserve correct
3227 * AGI/AGF locking order. This dirties the transaction so failures
3228 * after this point will abort and log recovery will clean up the
3231 * For whiteouts, we need to bump the link count on the whiteout
3232 * inode. After this point, we have a real link, clear the tmpfile
3233 * state flag from the inode so it doesn't accidentally get misused
3237 ASSERT(VFS_I(wip)->i_nlink == 0);
3238 error = xfs_iunlink_remove(tp, wip);
3240 goto out_trans_cancel;
3242 xfs_bumplink(tp, wip);
3243 VFS_I(wip)->i_state &= ~I_LINKABLE;
3247 * Set up the target.
3249 if (target_ip == NULL) {
3251 * If target does not exist and the rename crosses
3252 * directories, adjust the target directory link count
3253 * to account for the ".." reference from the new entry.
3255 error = xfs_dir_createname(tp, target_dp, target_name,
3256 src_ip->i_ino, spaceres);
3258 goto out_trans_cancel;
3260 xfs_trans_ichgtime(tp, target_dp,
3261 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3263 if (new_parent && src_is_directory) {
3264 xfs_bumplink(tp, target_dp);
3266 } else { /* target_ip != NULL */
3268 * Link the source inode under the target name.
3269 * If the source inode is a directory and we are moving
3270 * it across directories, its ".." entry will be
3271 * inconsistent until we replace that down below.
3273 * In case there is already an entry with the same
3274 * name at the destination directory, remove it first.
3276 error = xfs_dir_replace(tp, target_dp, target_name,
3277 src_ip->i_ino, spaceres);
3279 goto out_trans_cancel;
3281 xfs_trans_ichgtime(tp, target_dp,
3282 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3285 * Decrement the link count on the target since the target
3286 * dir no longer points to it.
3288 error = xfs_droplink(tp, target_ip);
3290 goto out_trans_cancel;
3292 if (src_is_directory) {
3294 * Drop the link from the old "." entry.
3296 error = xfs_droplink(tp, target_ip);
3298 goto out_trans_cancel;
3300 } /* target_ip != NULL */
3303 * Remove the source.
3305 if (new_parent && src_is_directory) {
3307 * Rewrite the ".." entry to point to the new
3310 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3311 target_dp->i_ino, spaceres);
3312 ASSERT(error != -EEXIST);
3314 goto out_trans_cancel;
3318 * We always want to hit the ctime on the source inode.
3320 * This isn't strictly required by the standards since the source
3321 * inode isn't really being changed, but old unix file systems did
3322 * it and some incremental backup programs won't work without it.
3324 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3325 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3328 * Adjust the link count on src_dp. This is necessary when
3329 * renaming a directory, either within one parent when
3330 * the target existed, or across two parent directories.
3332 if (src_is_directory && (new_parent || target_ip != NULL)) {
3335 * Decrement link count on src_directory since the
3336 * entry that's moved no longer points to it.
3338 error = xfs_droplink(tp, src_dp);
3340 goto out_trans_cancel;
3344 * For whiteouts, we only need to update the source dirent with the
3345 * inode number of the whiteout inode rather than removing it
3349 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3353 * NOTE: We don't need to check for extent count overflow here
3354 * because the dir remove name code will leave the dir block in
3355 * place if the extent count would overflow.
3357 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3362 goto out_trans_cancel;
3364 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3365 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3367 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3369 error = xfs_finish_rename(tp);
3375 xfs_trans_cancel(tp);
3384 struct xfs_inode *ip,
3387 struct xfs_inode_log_item *iip = ip->i_itemp;
3388 struct xfs_dinode *dip;
3389 struct xfs_mount *mp = ip->i_mount;
3392 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3393 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3394 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3395 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3396 ASSERT(iip->ili_item.li_buf == bp);
3398 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3401 * We don't flush the inode if any of the following checks fail, but we
3402 * do still update the log item and attach to the backing buffer as if
3403 * the flush happened. This is a formality to facilitate predictable
3404 * error handling as the caller will shutdown and fail the buffer.
3406 error = -EFSCORRUPTED;
3407 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3408 mp, XFS_ERRTAG_IFLUSH_1)) {
3409 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3410 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3411 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3414 if (S_ISREG(VFS_I(ip)->i_mode)) {
3416 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3417 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3418 mp, XFS_ERRTAG_IFLUSH_3)) {
3419 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3420 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3421 __func__, ip->i_ino, ip);
3424 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3426 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3427 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3428 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3429 mp, XFS_ERRTAG_IFLUSH_4)) {
3430 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3431 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3432 __func__, ip->i_ino, ip);
3436 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3437 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3438 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3439 "%s: detected corrupt incore inode %Lu, "
3440 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3441 __func__, ip->i_ino,
3442 ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3446 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3447 mp, XFS_ERRTAG_IFLUSH_6)) {
3448 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3449 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3450 __func__, ip->i_ino, ip->i_forkoff, ip);
3455 * Inode item log recovery for v2 inodes are dependent on the flushiter
3456 * count for correct sequencing. We bump the flush iteration count so
3457 * we can detect flushes which postdate a log record during recovery.
3458 * This is redundant as we now log every change and hence this can't
3459 * happen but we need to still do it to ensure backwards compatibility
3460 * with old kernels that predate logging all inode changes.
3462 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3466 * If there are inline format data / attr forks attached to this inode,
3467 * make sure they are not corrupt.
3469 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3470 xfs_ifork_verify_local_data(ip))
3472 if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3473 xfs_ifork_verify_local_attr(ip))
3477 * Copy the dirty parts of the inode into the on-disk inode. We always
3478 * copy out the core of the inode, because if the inode is dirty at all
3481 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3483 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3484 if (!xfs_sb_version_has_v3inode(&mp->m_sb)) {
3485 if (ip->i_flushiter == DI_MAX_FLUSH)
3486 ip->i_flushiter = 0;
3489 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3490 if (XFS_IFORK_Q(ip))
3491 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3494 * We've recorded everything logged in the inode, so we'd like to clear
3495 * the ili_fields bits so we don't log and flush things unnecessarily.
3496 * However, we can't stop logging all this information until the data
3497 * we've copied into the disk buffer is written to disk. If we did we
3498 * might overwrite the copy of the inode in the log with all the data
3499 * after re-logging only part of it, and in the face of a crash we
3500 * wouldn't have all the data we need to recover.
3502 * What we do is move the bits to the ili_last_fields field. When
3503 * logging the inode, these bits are moved back to the ili_fields field.
3504 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3505 * we know that the information those bits represent is permanently on
3506 * disk. As long as the flush completes before the inode is logged
3507 * again, then both ili_fields and ili_last_fields will be cleared.
3511 spin_lock(&iip->ili_lock);
3512 iip->ili_last_fields = iip->ili_fields;
3513 iip->ili_fields = 0;
3514 iip->ili_fsync_fields = 0;
3515 spin_unlock(&iip->ili_lock);
3518 * Store the current LSN of the inode so that we can tell whether the
3519 * item has moved in the AIL from xfs_buf_inode_iodone().
3521 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3522 &iip->ili_item.li_lsn);
3524 /* generate the checksum. */
3525 xfs_dinode_calc_crc(mp, dip);
3530 * Non-blocking flush of dirty inode metadata into the backing buffer.
3532 * The caller must have a reference to the inode and hold the cluster buffer
3533 * locked. The function will walk across all the inodes on the cluster buffer it
3534 * can find and lock without blocking, and flush them to the cluster buffer.
3536 * On successful flushing of at least one inode, the caller must write out the
3537 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3538 * the caller needs to release the buffer. On failure, the filesystem will be
3539 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3546 struct xfs_mount *mp = bp->b_mount;
3547 struct xfs_log_item *lip, *n;
3548 struct xfs_inode *ip;
3549 struct xfs_inode_log_item *iip;
3554 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3555 * can remove itself from the list.
3557 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3558 iip = (struct xfs_inode_log_item *)lip;
3559 ip = iip->ili_inode;
3562 * Quick and dirty check to avoid locks if possible.
3564 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3566 if (xfs_ipincount(ip))
3570 * The inode is still attached to the buffer, which means it is
3571 * dirty but reclaim might try to grab it. Check carefully for
3572 * that, and grab the ilock while still holding the i_flags_lock
3573 * to guarantee reclaim will not be able to reclaim this inode
3574 * once we drop the i_flags_lock.
3576 spin_lock(&ip->i_flags_lock);
3577 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3578 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3579 spin_unlock(&ip->i_flags_lock);
3584 * ILOCK will pin the inode against reclaim and prevent
3585 * concurrent transactions modifying the inode while we are
3586 * flushing the inode. If we get the lock, set the flushing
3587 * state before we drop the i_flags_lock.
3589 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3590 spin_unlock(&ip->i_flags_lock);
3593 __xfs_iflags_set(ip, XFS_IFLUSHING);
3594 spin_unlock(&ip->i_flags_lock);
3597 * Abort flushing this inode if we are shut down because the
3598 * inode may not currently be in the AIL. This can occur when
3599 * log I/O failure unpins the inode without inserting into the
3600 * AIL, leaving a dirty/unpinned inode attached to the buffer
3601 * that otherwise looks like it should be flushed.
3603 if (XFS_FORCED_SHUTDOWN(mp)) {
3604 xfs_iunpin_wait(ip);
3605 xfs_iflush_abort(ip);
3606 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3611 /* don't block waiting on a log force to unpin dirty inodes */
3612 if (xfs_ipincount(ip)) {
3613 xfs_iflags_clear(ip, XFS_IFLUSHING);
3614 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3618 if (!xfs_inode_clean(ip))
3619 error = xfs_iflush(ip, bp);
3621 xfs_iflags_clear(ip, XFS_IFLUSHING);
3622 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3629 bp->b_flags |= XBF_ASYNC;
3630 xfs_buf_ioend_fail(bp);
3631 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3638 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3639 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3644 /* Release an inode. */
3647 struct xfs_inode *ip)
3649 trace_xfs_irele(ip, _RET_IP_);
3654 * Ensure all commited transactions touching the inode are written to the log.
3657 xfs_log_force_inode(
3658 struct xfs_inode *ip)
3662 xfs_ilock(ip, XFS_ILOCK_SHARED);
3663 if (xfs_ipincount(ip))
3664 lsn = ip->i_itemp->ili_last_lsn;
3665 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3669 return xfs_log_force_lsn(ip->i_mount, lsn, XFS_LOG_SYNC, NULL);
3673 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3674 * abide vfs locking order (lowest pointer value goes first) and breaking the
3675 * layout leases before proceeding. The loop is needed because we cannot call
3676 * the blocking break_layout() with the iolocks held, and therefore have to
3677 * back out both locks.
3680 xfs_iolock_two_inodes_and_break_layout(
3690 /* Wait to break both inodes' layouts before we start locking. */
3691 error = break_layout(src, true);
3695 error = break_layout(dest, true);
3700 /* Lock one inode and make sure nobody got in and leased it. */
3702 error = break_layout(src, false);
3705 if (error == -EWOULDBLOCK)
3713 /* Lock the other inode and make sure nobody got in and leased it. */
3714 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3715 error = break_layout(dest, false);
3719 if (error == -EWOULDBLOCK)
3728 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3733 struct xfs_inode *ip1,
3734 struct xfs_inode *ip2)
3738 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3742 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3744 xfs_lock_two_inodes(ip1, XFS_MMAPLOCK_EXCL,
3745 ip2, XFS_MMAPLOCK_EXCL);
3749 /* Unlock both inodes to allow IO and mmap activity. */
3751 xfs_iunlock2_io_mmap(
3752 struct xfs_inode *ip1,
3753 struct xfs_inode *ip2)
3755 bool same_inode = (ip1 == ip2);
3757 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3759 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3760 inode_unlock(VFS_I(ip2));
3762 inode_unlock(VFS_I(ip1));