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_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
64 return ip->i_d.di_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_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
84 a = ip->i_d.di_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 struct xfs_icdinode *dic = &ip->i_d;
659 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
663 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
664 * is allowed, otherwise it has to be an exact match. If a CI match is found,
665 * ci_name->name will point to a the actual name (caller must free) or
666 * will be set to NULL if an exact match is found.
671 struct xfs_name *name,
673 struct xfs_name *ci_name)
678 trace_xfs_lookup(dp, name);
680 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
683 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
687 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
695 kmem_free(ci_name->name);
701 /* Propagate di_flags from a parent inode to a child inode. */
703 xfs_inode_inherit_flags(
704 struct xfs_inode *ip,
705 const struct xfs_inode *pip)
707 unsigned int di_flags = 0;
708 umode_t mode = VFS_I(ip)->i_mode;
711 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
712 di_flags |= XFS_DIFLAG_RTINHERIT;
713 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
714 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
715 ip->i_d.di_extsize = pip->i_d.di_extsize;
717 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
718 di_flags |= XFS_DIFLAG_PROJINHERIT;
719 } else if (S_ISREG(mode)) {
720 if ((pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) &&
721 xfs_sb_version_hasrealtime(&ip->i_mount->m_sb))
722 di_flags |= XFS_DIFLAG_REALTIME;
723 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
724 di_flags |= XFS_DIFLAG_EXTSIZE;
725 ip->i_d.di_extsize = pip->i_d.di_extsize;
728 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
730 di_flags |= XFS_DIFLAG_NOATIME;
731 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
733 di_flags |= XFS_DIFLAG_NODUMP;
734 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
736 di_flags |= XFS_DIFLAG_SYNC;
737 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
738 xfs_inherit_nosymlinks)
739 di_flags |= XFS_DIFLAG_NOSYMLINKS;
740 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
741 xfs_inherit_nodefrag)
742 di_flags |= XFS_DIFLAG_NODEFRAG;
743 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
744 di_flags |= XFS_DIFLAG_FILESTREAM;
746 ip->i_d.di_flags |= di_flags;
749 /* Propagate di_flags2 from a parent inode to a child inode. */
751 xfs_inode_inherit_flags2(
752 struct xfs_inode *ip,
753 const struct xfs_inode *pip)
755 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
756 ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
757 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
759 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
760 ip->i_d.di_flags2 |= XFS_DIFLAG2_DAX;
764 * Initialise a newly allocated inode and return the in-core inode to the
765 * caller locked exclusively.
769 struct user_namespace *mnt_userns,
770 struct xfs_trans *tp,
771 struct xfs_inode *pip,
777 struct xfs_inode **ipp)
779 struct inode *dir = pip ? VFS_I(pip) : NULL;
780 struct xfs_mount *mp = tp->t_mountp;
781 struct xfs_inode *ip;
784 struct timespec64 tv;
788 * Protect against obviously corrupt allocation btree records. Later
789 * xfs_iget checks will catch re-allocation of other active in-memory
790 * and on-disk inodes. If we don't catch reallocating the parent inode
791 * here we will deadlock in xfs_iget() so we have to do these checks
794 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
795 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
796 return -EFSCORRUPTED;
800 * Get the in-core inode with the lock held exclusively to prevent
801 * others from looking at until we're done.
803 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
809 set_nlink(inode, nlink);
810 inode->i_rdev = rdev;
811 ip->i_d.di_projid = prid;
813 if (dir && !(dir->i_mode & S_ISGID) &&
814 (mp->m_flags & XFS_MOUNT_GRPID)) {
815 inode->i_uid = fsuid_into_mnt(mnt_userns);
816 inode->i_gid = dir->i_gid;
817 inode->i_mode = mode;
819 inode_init_owner(mnt_userns, inode, dir, mode);
823 * If the group ID of the new file does not match the effective group
824 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
825 * (and only if the irix_sgid_inherit compatibility variable is set).
827 if (irix_sgid_inherit &&
828 (inode->i_mode & S_ISGID) &&
829 !in_group_p(i_gid_into_mnt(mnt_userns, inode)))
830 inode->i_mode &= ~S_ISGID;
833 ip->i_df.if_nextents = 0;
834 ASSERT(ip->i_d.di_nblocks == 0);
836 tv = current_time(inode);
841 ip->i_d.di_extsize = 0;
842 ip->i_d.di_dmevmask = 0;
843 ip->i_d.di_dmstate = 0;
844 ip->i_d.di_flags = 0;
846 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
847 inode_set_iversion(inode, 1);
848 ip->i_d.di_flags2 = mp->m_ino_geo.new_diflags2;
849 ip->i_d.di_cowextsize = 0;
850 ip->i_d.di_crtime = tv;
853 flags = XFS_ILOG_CORE;
854 switch (mode & S_IFMT) {
859 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
860 ip->i_df.if_flags = 0;
861 flags |= XFS_ILOG_DEV;
865 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY))
866 xfs_inode_inherit_flags(ip, pip);
867 if (pip && (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY))
868 xfs_inode_inherit_flags2(ip, pip);
871 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
872 ip->i_df.if_flags = XFS_IFEXTENTS;
873 ip->i_df.if_bytes = 0;
874 ip->i_df.if_u1.if_root = NULL;
881 * Log the new values stuffed into the inode.
883 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
884 xfs_trans_log_inode(tp, ip, flags);
886 /* now that we have an i_mode we can setup the inode structure */
894 * Allocates a new inode from disk and return a pointer to the incore copy. This
895 * routine will internally commit the current transaction and allocate a new one
896 * if we needed to allocate more on-disk free inodes to perform the requested
899 * If we are allocating quota inodes, we do not have a parent inode to attach to
900 * or associate with (i.e. dp == NULL) because they are not linked into the
901 * directory structure - they are attached directly to the superblock - and so
906 struct user_namespace *mnt_userns,
907 struct xfs_trans **tpp,
908 struct xfs_inode *dp,
913 struct xfs_inode **ipp)
915 struct xfs_buf *agibp;
916 xfs_ino_t parent_ino = dp ? dp->i_ino : 0;
920 ASSERT((*tpp)->t_flags & XFS_TRANS_PERM_LOG_RES);
923 * Call the space management code to pick the on-disk inode to be
926 error = xfs_dialloc_select_ag(tpp, parent_ino, mode, &agibp);
933 /* Allocate an inode from the selected AG */
934 error = xfs_dialloc_ag(*tpp, agibp, parent_ino, &ino);
937 ASSERT(ino != NULLFSINO);
939 return xfs_init_new_inode(mnt_userns, *tpp, dp, ino, mode, nlink, rdev,
944 * Decrement the link count on an inode & log the change. If this causes the
945 * link count to go to zero, move the inode to AGI unlinked list so that it can
946 * be freed when the last active reference goes away via xfs_inactive().
948 static int /* error */
953 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
955 drop_nlink(VFS_I(ip));
956 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
958 if (VFS_I(ip)->i_nlink)
961 return xfs_iunlink(tp, ip);
965 * Increment the link count on an inode & log the change.
972 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
974 inc_nlink(VFS_I(ip));
975 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
980 struct user_namespace *mnt_userns,
982 struct xfs_name *name,
987 int is_dir = S_ISDIR(mode);
988 struct xfs_mount *mp = dp->i_mount;
989 struct xfs_inode *ip = NULL;
990 struct xfs_trans *tp = NULL;
992 bool unlock_dp_on_error = false;
994 struct xfs_dquot *udqp = NULL;
995 struct xfs_dquot *gdqp = NULL;
996 struct xfs_dquot *pdqp = NULL;
997 struct xfs_trans_res *tres;
1000 trace_xfs_create(dp, name);
1002 if (XFS_FORCED_SHUTDOWN(mp))
1005 prid = xfs_get_initial_prid(dp);
1008 * Make sure that we have allocated dquot(s) on disk.
1010 error = xfs_qm_vop_dqalloc(dp, fsuid_into_mnt(mnt_userns),
1011 fsgid_into_mnt(mnt_userns), prid,
1012 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1013 &udqp, &gdqp, &pdqp);
1018 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1019 tres = &M_RES(mp)->tr_mkdir;
1021 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1022 tres = &M_RES(mp)->tr_create;
1026 * Initially assume that the file does not exist and
1027 * reserve the resources for that case. If that is not
1028 * the case we'll drop the one we have and get a more
1029 * appropriate transaction later.
1031 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1033 if (error == -ENOSPC) {
1034 /* flush outstanding delalloc blocks and retry */
1035 xfs_flush_inodes(mp);
1036 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1040 goto out_release_dquots;
1042 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1043 unlock_dp_on_error = true;
1045 error = xfs_iext_count_may_overflow(dp, XFS_DATA_FORK,
1046 XFS_IEXT_DIR_MANIP_CNT(mp));
1048 goto out_trans_cancel;
1051 * A newly created regular or special file just has one directory
1052 * entry pointing to them, but a directory also the "." entry
1053 * pointing to itself.
1055 error = xfs_dir_ialloc(mnt_userns, &tp, dp, mode, is_dir ? 2 : 1, rdev,
1058 goto out_trans_cancel;
1061 * Now we join the directory inode to the transaction. We do not do it
1062 * earlier because xfs_dir_ialloc might commit the previous transaction
1063 * (and release all the locks). An error from here on will result in
1064 * the transaction cancel unlocking dp so don't do it explicitly in the
1067 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1068 unlock_dp_on_error = false;
1070 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1071 resblks - XFS_IALLOC_SPACE_RES(mp));
1073 ASSERT(error != -ENOSPC);
1074 goto out_trans_cancel;
1076 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1077 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1080 error = xfs_dir_init(tp, ip, dp);
1082 goto out_trans_cancel;
1084 xfs_bumplink(tp, dp);
1088 * If this is a synchronous mount, make sure that the
1089 * create transaction goes to disk before returning to
1092 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1093 xfs_trans_set_sync(tp);
1096 * Attach the dquot(s) to the inodes and modify them incore.
1097 * These ids of the inode couldn't have changed since the new
1098 * inode has been locked ever since it was created.
1100 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1102 error = xfs_trans_commit(tp);
1104 goto out_release_inode;
1106 xfs_qm_dqrele(udqp);
1107 xfs_qm_dqrele(gdqp);
1108 xfs_qm_dqrele(pdqp);
1114 xfs_trans_cancel(tp);
1117 * Wait until after the current transaction is aborted to finish the
1118 * setup of the inode and release the inode. This prevents recursive
1119 * transactions and deadlocks from xfs_inactive.
1122 xfs_finish_inode_setup(ip);
1126 xfs_qm_dqrele(udqp);
1127 xfs_qm_dqrele(gdqp);
1128 xfs_qm_dqrele(pdqp);
1130 if (unlock_dp_on_error)
1131 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1137 struct user_namespace *mnt_userns,
1138 struct xfs_inode *dp,
1140 struct xfs_inode **ipp)
1142 struct xfs_mount *mp = dp->i_mount;
1143 struct xfs_inode *ip = NULL;
1144 struct xfs_trans *tp = NULL;
1147 struct xfs_dquot *udqp = NULL;
1148 struct xfs_dquot *gdqp = NULL;
1149 struct xfs_dquot *pdqp = NULL;
1150 struct xfs_trans_res *tres;
1153 if (XFS_FORCED_SHUTDOWN(mp))
1156 prid = xfs_get_initial_prid(dp);
1159 * Make sure that we have allocated dquot(s) on disk.
1161 error = xfs_qm_vop_dqalloc(dp, fsuid_into_mnt(mnt_userns),
1162 fsgid_into_mnt(mnt_userns), prid,
1163 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1164 &udqp, &gdqp, &pdqp);
1168 resblks = XFS_IALLOC_SPACE_RES(mp);
1169 tres = &M_RES(mp)->tr_create_tmpfile;
1171 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1174 goto out_release_dquots;
1176 error = xfs_dir_ialloc(mnt_userns, &tp, dp, mode, 0, 0, prid, &ip);
1178 goto out_trans_cancel;
1180 if (mp->m_flags & XFS_MOUNT_WSYNC)
1181 xfs_trans_set_sync(tp);
1184 * Attach the dquot(s) to the inodes and modify them incore.
1185 * These ids of the inode couldn't have changed since the new
1186 * inode has been locked ever since it was created.
1188 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1190 error = xfs_iunlink(tp, ip);
1192 goto out_trans_cancel;
1194 error = xfs_trans_commit(tp);
1196 goto out_release_inode;
1198 xfs_qm_dqrele(udqp);
1199 xfs_qm_dqrele(gdqp);
1200 xfs_qm_dqrele(pdqp);
1206 xfs_trans_cancel(tp);
1209 * Wait until after the current transaction is aborted to finish the
1210 * setup of the inode and release the inode. This prevents recursive
1211 * transactions and deadlocks from xfs_inactive.
1214 xfs_finish_inode_setup(ip);
1218 xfs_qm_dqrele(udqp);
1219 xfs_qm_dqrele(gdqp);
1220 xfs_qm_dqrele(pdqp);
1229 struct xfs_name *target_name)
1231 xfs_mount_t *mp = tdp->i_mount;
1236 trace_xfs_link(tdp, target_name);
1238 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1240 if (XFS_FORCED_SHUTDOWN(mp))
1243 error = xfs_qm_dqattach(sip);
1247 error = xfs_qm_dqattach(tdp);
1251 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1252 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1253 if (error == -ENOSPC) {
1255 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1260 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1262 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1263 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1265 error = xfs_iext_count_may_overflow(tdp, XFS_DATA_FORK,
1266 XFS_IEXT_DIR_MANIP_CNT(mp));
1271 * If we are using project inheritance, we only allow hard link
1272 * creation in our tree when the project IDs are the same; else
1273 * the tree quota mechanism could be circumvented.
1275 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1276 tdp->i_d.di_projid != sip->i_d.di_projid)) {
1282 error = xfs_dir_canenter(tp, tdp, target_name);
1288 * Handle initial link state of O_TMPFILE inode
1290 if (VFS_I(sip)->i_nlink == 0) {
1291 error = xfs_iunlink_remove(tp, sip);
1296 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1300 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1301 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1303 xfs_bumplink(tp, sip);
1306 * If this is a synchronous mount, make sure that the
1307 * link transaction goes to disk before returning to
1310 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1311 xfs_trans_set_sync(tp);
1313 return xfs_trans_commit(tp);
1316 xfs_trans_cancel(tp);
1321 /* Clear the reflink flag and the cowblocks tag if possible. */
1323 xfs_itruncate_clear_reflink_flags(
1324 struct xfs_inode *ip)
1326 struct xfs_ifork *dfork;
1327 struct xfs_ifork *cfork;
1329 if (!xfs_is_reflink_inode(ip))
1331 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1332 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1333 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1334 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1335 if (cfork->if_bytes == 0)
1336 xfs_inode_clear_cowblocks_tag(ip);
1340 * Free up the underlying blocks past new_size. The new size must be smaller
1341 * than the current size. This routine can be used both for the attribute and
1342 * data fork, and does not modify the inode size, which is left to the caller.
1344 * The transaction passed to this routine must have made a permanent log
1345 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1346 * given transaction and start new ones, so make sure everything involved in
1347 * the transaction is tidy before calling here. Some transaction will be
1348 * returned to the caller to be committed. The incoming transaction must
1349 * already include the inode, and both inode locks must be held exclusively.
1350 * The inode must also be "held" within the transaction. On return the inode
1351 * will be "held" within the returned transaction. This routine does NOT
1352 * require any disk space to be reserved for it within the transaction.
1354 * If we get an error, we must return with the inode locked and linked into the
1355 * current transaction. This keeps things simple for the higher level code,
1356 * because it always knows that the inode is locked and held in the transaction
1357 * that returns to it whether errors occur or not. We don't mark the inode
1358 * dirty on error so that transactions can be easily aborted if possible.
1361 xfs_itruncate_extents_flags(
1362 struct xfs_trans **tpp,
1363 struct xfs_inode *ip,
1365 xfs_fsize_t new_size,
1368 struct xfs_mount *mp = ip->i_mount;
1369 struct xfs_trans *tp = *tpp;
1370 xfs_fileoff_t first_unmap_block;
1371 xfs_filblks_t unmap_len;
1374 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1375 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1376 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1377 ASSERT(new_size <= XFS_ISIZE(ip));
1378 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1379 ASSERT(ip->i_itemp != NULL);
1380 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1381 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1383 trace_xfs_itruncate_extents_start(ip, new_size);
1385 flags |= xfs_bmapi_aflag(whichfork);
1388 * Since it is possible for space to become allocated beyond
1389 * the end of the file (in a crash where the space is allocated
1390 * but the inode size is not yet updated), simply remove any
1391 * blocks which show up between the new EOF and the maximum
1392 * possible file size.
1394 * We have to free all the blocks to the bmbt maximum offset, even if
1395 * the page cache can't scale that far.
1397 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1398 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1399 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1403 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1404 while (unmap_len > 0) {
1405 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1406 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1407 flags, XFS_ITRUNC_MAX_EXTENTS);
1411 /* free the just unmapped extents */
1412 error = xfs_defer_finish(&tp);
1417 if (whichfork == XFS_DATA_FORK) {
1418 /* Remove all pending CoW reservations. */
1419 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1420 first_unmap_block, XFS_MAX_FILEOFF, true);
1424 xfs_itruncate_clear_reflink_flags(ip);
1428 * Always re-log the inode so that our permanent transaction can keep
1429 * on rolling it forward in the log.
1431 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1433 trace_xfs_itruncate_extents_end(ip, new_size);
1444 xfs_mount_t *mp = ip->i_mount;
1447 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1450 /* If this is a read-only mount, don't do this (would generate I/O) */
1451 if (mp->m_flags & XFS_MOUNT_RDONLY)
1454 if (!XFS_FORCED_SHUTDOWN(mp)) {
1458 * If we previously truncated this file and removed old data
1459 * in the process, we want to initiate "early" writeout on
1460 * the last close. This is an attempt to combat the notorious
1461 * NULL files problem which is particularly noticeable from a
1462 * truncate down, buffered (re-)write (delalloc), followed by
1463 * a crash. What we are effectively doing here is
1464 * significantly reducing the time window where we'd otherwise
1465 * be exposed to that problem.
1467 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1469 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1470 if (ip->i_delayed_blks > 0) {
1471 error = filemap_flush(VFS_I(ip)->i_mapping);
1478 if (VFS_I(ip)->i_nlink == 0)
1481 if (xfs_can_free_eofblocks(ip, false)) {
1484 * Check if the inode is being opened, written and closed
1485 * frequently and we have delayed allocation blocks outstanding
1486 * (e.g. streaming writes from the NFS server), truncating the
1487 * blocks past EOF will cause fragmentation to occur.
1489 * In this case don't do the truncation, but we have to be
1490 * careful how we detect this case. Blocks beyond EOF show up as
1491 * i_delayed_blks even when the inode is clean, so we need to
1492 * truncate them away first before checking for a dirty release.
1493 * Hence on the first dirty close we will still remove the
1494 * speculative allocation, but after that we will leave it in
1497 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1500 * If we can't get the iolock just skip truncating the blocks
1501 * past EOF because we could deadlock with the mmap_lock
1502 * otherwise. We'll get another chance to drop them once the
1503 * last reference to the inode is dropped, so we'll never leak
1504 * blocks permanently.
1506 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1507 error = xfs_free_eofblocks(ip);
1508 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1513 /* delalloc blocks after truncation means it really is dirty */
1514 if (ip->i_delayed_blks)
1515 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1521 * xfs_inactive_truncate
1523 * Called to perform a truncate when an inode becomes unlinked.
1526 xfs_inactive_truncate(
1527 struct xfs_inode *ip)
1529 struct xfs_mount *mp = ip->i_mount;
1530 struct xfs_trans *tp;
1533 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1535 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1538 xfs_ilock(ip, XFS_ILOCK_EXCL);
1539 xfs_trans_ijoin(tp, ip, 0);
1542 * Log the inode size first to prevent stale data exposure in the event
1543 * of a system crash before the truncate completes. See the related
1544 * comment in xfs_vn_setattr_size() for details.
1546 ip->i_d.di_size = 0;
1547 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1549 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1551 goto error_trans_cancel;
1553 ASSERT(ip->i_df.if_nextents == 0);
1555 error = xfs_trans_commit(tp);
1559 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1563 xfs_trans_cancel(tp);
1565 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1570 * xfs_inactive_ifree()
1572 * Perform the inode free when an inode is unlinked.
1576 struct xfs_inode *ip)
1578 struct xfs_mount *mp = ip->i_mount;
1579 struct xfs_trans *tp;
1583 * We try to use a per-AG reservation for any block needed by the finobt
1584 * tree, but as the finobt feature predates the per-AG reservation
1585 * support a degraded file system might not have enough space for the
1586 * reservation at mount time. In that case try to dip into the reserved
1589 * Send a warning if the reservation does happen to fail, as the inode
1590 * now remains allocated and sits on the unlinked list until the fs is
1593 if (unlikely(mp->m_finobt_nores)) {
1594 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1595 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1598 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1601 if (error == -ENOSPC) {
1602 xfs_warn_ratelimited(mp,
1603 "Failed to remove inode(s) from unlinked list. "
1604 "Please free space, unmount and run xfs_repair.");
1606 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1612 * We do not hold the inode locked across the entire rolling transaction
1613 * here. We only need to hold it for the first transaction that
1614 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1615 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1616 * here breaks the relationship between cluster buffer invalidation and
1617 * stale inode invalidation on cluster buffer item journal commit
1618 * completion, and can result in leaving dirty stale inodes hanging
1621 * We have no need for serialising this inode operation against other
1622 * operations - we freed the inode and hence reallocation is required
1623 * and that will serialise on reallocating the space the deferops need
1624 * to free. Hence we can unlock the inode on the first commit of
1625 * the transaction rather than roll it right through the deferops. This
1626 * avoids relogging the XFS_ISTALE inode.
1628 * We check that xfs_ifree() hasn't grown an internal transaction roll
1629 * by asserting that the inode is still locked when it returns.
1631 xfs_ilock(ip, XFS_ILOCK_EXCL);
1632 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1634 error = xfs_ifree(tp, ip);
1635 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1638 * If we fail to free the inode, shut down. The cancel
1639 * might do that, we need to make sure. Otherwise the
1640 * inode might be lost for a long time or forever.
1642 if (!XFS_FORCED_SHUTDOWN(mp)) {
1643 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1645 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1647 xfs_trans_cancel(tp);
1652 * Credit the quota account(s). The inode is gone.
1654 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1657 * Just ignore errors at this point. There is nothing we can do except
1658 * to try to keep going. Make sure it's not a silent error.
1660 error = xfs_trans_commit(tp);
1662 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1671 * This is called when the vnode reference count for the vnode
1672 * goes to zero. If the file has been unlinked, then it must
1673 * now be truncated. Also, we clear all of the read-ahead state
1674 * kept for the inode here since the file is now closed.
1680 struct xfs_mount *mp;
1685 * If the inode is already free, then there can be nothing
1688 if (VFS_I(ip)->i_mode == 0) {
1689 ASSERT(ip->i_df.if_broot_bytes == 0);
1694 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1696 /* If this is a read-only mount, don't do this (would generate I/O) */
1697 if (mp->m_flags & XFS_MOUNT_RDONLY)
1700 /* Try to clean out the cow blocks if there are any. */
1701 if (xfs_inode_has_cow_data(ip))
1702 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1704 if (VFS_I(ip)->i_nlink != 0) {
1706 * force is true because we are evicting an inode from the
1707 * cache. Post-eof blocks must be freed, lest we end up with
1708 * broken free space accounting.
1710 * Note: don't bother with iolock here since lockdep complains
1711 * about acquiring it in reclaim context. We have the only
1712 * reference to the inode at this point anyways.
1714 if (xfs_can_free_eofblocks(ip, true))
1715 xfs_free_eofblocks(ip);
1720 if (S_ISREG(VFS_I(ip)->i_mode) &&
1721 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1722 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1725 error = xfs_qm_dqattach(ip);
1729 if (S_ISLNK(VFS_I(ip)->i_mode))
1730 error = xfs_inactive_symlink(ip);
1732 error = xfs_inactive_truncate(ip);
1737 * If there are attributes associated with the file then blow them away
1738 * now. The code calls a routine that recursively deconstructs the
1739 * attribute fork. If also blows away the in-core attribute fork.
1741 if (XFS_IFORK_Q(ip)) {
1742 error = xfs_attr_inactive(ip);
1748 ASSERT(ip->i_d.di_forkoff == 0);
1753 error = xfs_inactive_ifree(ip);
1758 * Release the dquots held by inode, if any.
1760 xfs_qm_dqdetach(ip);
1764 * In-Core Unlinked List Lookups
1765 * =============================
1767 * Every inode is supposed to be reachable from some other piece of metadata
1768 * with the exception of the root directory. Inodes with a connection to a
1769 * file descriptor but not linked from anywhere in the on-disk directory tree
1770 * are collectively known as unlinked inodes, though the filesystem itself
1771 * maintains links to these inodes so that on-disk metadata are consistent.
1773 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1774 * header contains a number of buckets that point to an inode, and each inode
1775 * record has a pointer to the next inode in the hash chain. This
1776 * singly-linked list causes scaling problems in the iunlink remove function
1777 * because we must walk that list to find the inode that points to the inode
1778 * being removed from the unlinked hash bucket list.
1780 * What if we modelled the unlinked list as a collection of records capturing
1781 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1782 * have a fast way to look up unlinked list predecessors, which avoids the
1783 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1786 * Because this is a backref cache, we ignore operational failures since the
1787 * iunlink code can fall back to the slow bucket walk. The only errors that
1788 * should bubble out are for obviously incorrect situations.
1790 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1791 * access or have otherwise provided for concurrency control.
1794 /* Capture a "X.next_unlinked = Y" relationship. */
1795 struct xfs_iunlink {
1796 struct rhash_head iu_rhash_head;
1797 xfs_agino_t iu_agino; /* X */
1798 xfs_agino_t iu_next_unlinked; /* Y */
1801 /* Unlinked list predecessor lookup hashtable construction */
1803 xfs_iunlink_obj_cmpfn(
1804 struct rhashtable_compare_arg *arg,
1807 const xfs_agino_t *key = arg->key;
1808 const struct xfs_iunlink *iu = obj;
1810 if (iu->iu_next_unlinked != *key)
1815 static const struct rhashtable_params xfs_iunlink_hash_params = {
1816 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1817 .key_len = sizeof(xfs_agino_t),
1818 .key_offset = offsetof(struct xfs_iunlink,
1820 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1821 .automatic_shrinking = true,
1822 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1826 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1827 * relation is found.
1830 xfs_iunlink_lookup_backref(
1831 struct xfs_perag *pag,
1834 struct xfs_iunlink *iu;
1836 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1837 xfs_iunlink_hash_params);
1838 return iu ? iu->iu_agino : NULLAGINO;
1842 * Take ownership of an iunlink cache entry and insert it into the hash table.
1843 * If successful, the entry will be owned by the cache; if not, it is freed.
1844 * Either way, the caller does not own @iu after this call.
1847 xfs_iunlink_insert_backref(
1848 struct xfs_perag *pag,
1849 struct xfs_iunlink *iu)
1853 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1854 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1856 * Fail loudly if there already was an entry because that's a sign of
1857 * corruption of in-memory data. Also fail loudly if we see an error
1858 * code we didn't anticipate from the rhashtable code. Currently we
1859 * only anticipate ENOMEM.
1862 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1866 * Absorb any runtime errors that aren't a result of corruption because
1867 * this is a cache and we can always fall back to bucket list scanning.
1869 if (error != 0 && error != -EEXIST)
1874 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1876 xfs_iunlink_add_backref(
1877 struct xfs_perag *pag,
1878 xfs_agino_t prev_agino,
1879 xfs_agino_t this_agino)
1881 struct xfs_iunlink *iu;
1883 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
1886 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
1887 iu->iu_agino = prev_agino;
1888 iu->iu_next_unlinked = this_agino;
1890 return xfs_iunlink_insert_backref(pag, iu);
1894 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
1895 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
1896 * wasn't any such entry then we don't bother.
1899 xfs_iunlink_change_backref(
1900 struct xfs_perag *pag,
1902 xfs_agino_t next_unlinked)
1904 struct xfs_iunlink *iu;
1907 /* Look up the old entry; if there wasn't one then exit. */
1908 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1909 xfs_iunlink_hash_params);
1914 * Remove the entry. This shouldn't ever return an error, but if we
1915 * couldn't remove the old entry we don't want to add it again to the
1916 * hash table, and if the entry disappeared on us then someone's
1917 * violated the locking rules and we need to fail loudly. Either way
1918 * we cannot remove the inode because internal state is or would have
1921 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
1922 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1926 /* If there is no new next entry just free our item and return. */
1927 if (next_unlinked == NULLAGINO) {
1932 /* Update the entry and re-add it to the hash table. */
1933 iu->iu_next_unlinked = next_unlinked;
1934 return xfs_iunlink_insert_backref(pag, iu);
1937 /* Set up the in-core predecessor structures. */
1940 struct xfs_perag *pag)
1942 return rhashtable_init(&pag->pagi_unlinked_hash,
1943 &xfs_iunlink_hash_params);
1946 /* Free the in-core predecessor structures. */
1948 xfs_iunlink_free_item(
1952 struct xfs_iunlink *iu = ptr;
1953 bool *freed_anything = arg;
1955 *freed_anything = true;
1960 xfs_iunlink_destroy(
1961 struct xfs_perag *pag)
1963 bool freed_anything = false;
1965 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
1966 xfs_iunlink_free_item, &freed_anything);
1968 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
1972 * Point the AGI unlinked bucket at an inode and log the results. The caller
1973 * is responsible for validating the old value.
1976 xfs_iunlink_update_bucket(
1977 struct xfs_trans *tp,
1978 xfs_agnumber_t agno,
1979 struct xfs_buf *agibp,
1980 unsigned int bucket_index,
1981 xfs_agino_t new_agino)
1983 struct xfs_agi *agi = agibp->b_addr;
1984 xfs_agino_t old_value;
1987 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
1989 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1990 trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
1991 old_value, new_agino);
1994 * We should never find the head of the list already set to the value
1995 * passed in because either we're adding or removing ourselves from the
1998 if (old_value == new_agino) {
1999 xfs_buf_mark_corrupt(agibp);
2000 return -EFSCORRUPTED;
2003 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2004 offset = offsetof(struct xfs_agi, agi_unlinked) +
2005 (sizeof(xfs_agino_t) * bucket_index);
2006 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2010 /* Set an on-disk inode's next_unlinked pointer. */
2012 xfs_iunlink_update_dinode(
2013 struct xfs_trans *tp,
2014 xfs_agnumber_t agno,
2016 struct xfs_buf *ibp,
2017 struct xfs_dinode *dip,
2018 struct xfs_imap *imap,
2019 xfs_agino_t next_agino)
2021 struct xfs_mount *mp = tp->t_mountp;
2024 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2026 trace_xfs_iunlink_update_dinode(mp, agno, agino,
2027 be32_to_cpu(dip->di_next_unlinked), next_agino);
2029 dip->di_next_unlinked = cpu_to_be32(next_agino);
2030 offset = imap->im_boffset +
2031 offsetof(struct xfs_dinode, di_next_unlinked);
2033 /* need to recalc the inode CRC if appropriate */
2034 xfs_dinode_calc_crc(mp, dip);
2035 xfs_trans_inode_buf(tp, ibp);
2036 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2039 /* Set an in-core inode's unlinked pointer and return the old value. */
2041 xfs_iunlink_update_inode(
2042 struct xfs_trans *tp,
2043 struct xfs_inode *ip,
2044 xfs_agnumber_t agno,
2045 xfs_agino_t next_agino,
2046 xfs_agino_t *old_next_agino)
2048 struct xfs_mount *mp = tp->t_mountp;
2049 struct xfs_dinode *dip;
2050 struct xfs_buf *ibp;
2051 xfs_agino_t old_value;
2054 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2056 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 0);
2060 /* Make sure the old pointer isn't garbage. */
2061 old_value = be32_to_cpu(dip->di_next_unlinked);
2062 if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2063 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2064 sizeof(*dip), __this_address);
2065 error = -EFSCORRUPTED;
2070 * Since we're updating a linked list, we should never find that the
2071 * current pointer is the same as the new value, unless we're
2072 * terminating the list.
2074 *old_next_agino = old_value;
2075 if (old_value == next_agino) {
2076 if (next_agino != NULLAGINO) {
2077 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2078 dip, sizeof(*dip), __this_address);
2079 error = -EFSCORRUPTED;
2084 /* Ok, update the new pointer. */
2085 xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2086 ibp, dip, &ip->i_imap, next_agino);
2089 xfs_trans_brelse(tp, ibp);
2094 * This is called when the inode's link count has gone to 0 or we are creating
2095 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2097 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2098 * list when the inode is freed.
2102 struct xfs_trans *tp,
2103 struct xfs_inode *ip)
2105 struct xfs_mount *mp = tp->t_mountp;
2106 struct xfs_agi *agi;
2107 struct xfs_buf *agibp;
2108 xfs_agino_t next_agino;
2109 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2110 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2111 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2114 ASSERT(VFS_I(ip)->i_nlink == 0);
2115 ASSERT(VFS_I(ip)->i_mode != 0);
2116 trace_xfs_iunlink(ip);
2118 /* Get the agi buffer first. It ensures lock ordering on the list. */
2119 error = xfs_read_agi(mp, tp, agno, &agibp);
2122 agi = agibp->b_addr;
2125 * Get the index into the agi hash table for the list this inode will
2126 * go on. Make sure the pointer isn't garbage and that this inode
2127 * isn't already on the list.
2129 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2130 if (next_agino == agino ||
2131 !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2132 xfs_buf_mark_corrupt(agibp);
2133 return -EFSCORRUPTED;
2136 if (next_agino != NULLAGINO) {
2137 xfs_agino_t old_agino;
2140 * There is already another inode in the bucket, so point this
2141 * inode to the current head of the list.
2143 error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2147 ASSERT(old_agino == NULLAGINO);
2150 * agino has been unlinked, add a backref from the next inode
2153 error = xfs_iunlink_add_backref(agibp->b_pag, agino, next_agino);
2158 /* Point the head of the list to point to this inode. */
2159 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2162 /* Return the imap, dinode pointer, and buffer for an inode. */
2164 xfs_iunlink_map_ino(
2165 struct xfs_trans *tp,
2166 xfs_agnumber_t agno,
2168 struct xfs_imap *imap,
2169 struct xfs_dinode **dipp,
2170 struct xfs_buf **bpp)
2172 struct xfs_mount *mp = tp->t_mountp;
2176 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2178 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2183 error = xfs_imap_to_bp(mp, tp, imap, dipp, bpp, 0);
2185 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2194 * Walk the unlinked chain from @head_agino until we find the inode that
2195 * points to @target_agino. Return the inode number, map, dinode pointer,
2196 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2198 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2199 * @agino, @imap, @dipp, and @bpp are all output parameters.
2201 * Do not call this function if @target_agino is the head of the list.
2204 xfs_iunlink_map_prev(
2205 struct xfs_trans *tp,
2206 xfs_agnumber_t agno,
2207 xfs_agino_t head_agino,
2208 xfs_agino_t target_agino,
2210 struct xfs_imap *imap,
2211 struct xfs_dinode **dipp,
2212 struct xfs_buf **bpp,
2213 struct xfs_perag *pag)
2215 struct xfs_mount *mp = tp->t_mountp;
2216 xfs_agino_t next_agino;
2219 ASSERT(head_agino != target_agino);
2222 /* See if our backref cache can find it faster. */
2223 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2224 if (*agino != NULLAGINO) {
2225 error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2229 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2233 * If we get here the cache contents were corrupt, so drop the
2234 * buffer and fall back to walking the bucket list.
2236 xfs_trans_brelse(tp, *bpp);
2241 trace_xfs_iunlink_map_prev_fallback(mp, agno);
2243 /* Otherwise, walk the entire bucket until we find it. */
2244 next_agino = head_agino;
2245 while (next_agino != target_agino) {
2246 xfs_agino_t unlinked_agino;
2249 xfs_trans_brelse(tp, *bpp);
2251 *agino = next_agino;
2252 error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2257 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2259 * Make sure this pointer is valid and isn't an obvious
2262 if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2263 next_agino == unlinked_agino) {
2264 XFS_CORRUPTION_ERROR(__func__,
2265 XFS_ERRLEVEL_LOW, mp,
2266 *dipp, sizeof(**dipp));
2267 error = -EFSCORRUPTED;
2270 next_agino = unlinked_agino;
2277 * Pull the on-disk inode from the AGI unlinked list.
2281 struct xfs_trans *tp,
2282 struct xfs_inode *ip)
2284 struct xfs_mount *mp = tp->t_mountp;
2285 struct xfs_agi *agi;
2286 struct xfs_buf *agibp;
2287 struct xfs_buf *last_ibp;
2288 struct xfs_dinode *last_dip = NULL;
2289 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2290 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2291 xfs_agino_t next_agino;
2292 xfs_agino_t head_agino;
2293 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2296 trace_xfs_iunlink_remove(ip);
2298 /* Get the agi buffer first. It ensures lock ordering on the list. */
2299 error = xfs_read_agi(mp, tp, agno, &agibp);
2302 agi = agibp->b_addr;
2305 * Get the index into the agi hash table for the list this inode will
2306 * go on. Make sure the head pointer isn't garbage.
2308 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2309 if (!xfs_verify_agino(mp, agno, head_agino)) {
2310 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2312 return -EFSCORRUPTED;
2316 * Set our inode's next_unlinked pointer to NULL and then return
2317 * the old pointer value so that we can update whatever was previous
2318 * to us in the list to point to whatever was next in the list.
2320 error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2325 * If there was a backref pointing from the next inode back to this
2326 * one, remove it because we've removed this inode from the list.
2328 * Later, if this inode was in the middle of the list we'll update
2329 * this inode's backref to point from the next inode.
2331 if (next_agino != NULLAGINO) {
2332 error = xfs_iunlink_change_backref(agibp->b_pag, next_agino,
2338 if (head_agino != agino) {
2339 struct xfs_imap imap;
2340 xfs_agino_t prev_agino;
2342 /* We need to search the list for the inode being freed. */
2343 error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2344 &prev_agino, &imap, &last_dip, &last_ibp,
2349 /* Point the previous inode on the list to the next inode. */
2350 xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2351 last_dip, &imap, next_agino);
2354 * Now we deal with the backref for this inode. If this inode
2355 * pointed at a real inode, change the backref that pointed to
2356 * us to point to our old next. If this inode was the end of
2357 * the list, delete the backref that pointed to us. Note that
2358 * change_backref takes care of deleting the backref if
2359 * next_agino is NULLAGINO.
2361 return xfs_iunlink_change_backref(agibp->b_pag, agino,
2365 /* Point the head of the list to the next unlinked inode. */
2366 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2371 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2372 * mark it stale. We should only find clean inodes in this lookup that aren't
2376 xfs_ifree_mark_inode_stale(
2378 struct xfs_inode *free_ip,
2381 struct xfs_mount *mp = bp->b_mount;
2382 struct xfs_perag *pag = bp->b_pag;
2383 struct xfs_inode_log_item *iip;
2384 struct xfs_inode *ip;
2388 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2390 /* Inode not in memory, nothing to do */
2397 * because this is an RCU protected lookup, we could find a recently
2398 * freed or even reallocated inode during the lookup. We need to check
2399 * under the i_flags_lock for a valid inode here. Skip it if it is not
2400 * valid, the wrong inode or stale.
2402 spin_lock(&ip->i_flags_lock);
2403 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2404 goto out_iflags_unlock;
2407 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2408 * other inodes that we did not find in the list attached to the buffer
2409 * and are not already marked stale. If we can't lock it, back off and
2412 if (ip != free_ip) {
2413 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2414 spin_unlock(&ip->i_flags_lock);
2420 ip->i_flags |= XFS_ISTALE;
2423 * If the inode is flushing, it is already attached to the buffer. All
2424 * we needed to do here is mark the inode stale so buffer IO completion
2425 * will remove it from the AIL.
2428 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2429 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2430 ASSERT(iip->ili_last_fields);
2435 * Inodes not attached to the buffer can be released immediately.
2436 * Everything else has to go through xfs_iflush_abort() on journal
2437 * commit as the flock synchronises removal of the inode from the
2438 * cluster buffer against inode reclaim.
2440 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2443 __xfs_iflags_set(ip, XFS_IFLUSHING);
2444 spin_unlock(&ip->i_flags_lock);
2447 /* we have a dirty inode in memory that has not yet been flushed. */
2448 spin_lock(&iip->ili_lock);
2449 iip->ili_last_fields = iip->ili_fields;
2450 iip->ili_fields = 0;
2451 iip->ili_fsync_fields = 0;
2452 spin_unlock(&iip->ili_lock);
2453 ASSERT(iip->ili_last_fields);
2456 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2461 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2463 spin_unlock(&ip->i_flags_lock);
2468 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2469 * inodes that are in memory - they all must be marked stale and attached to
2470 * the cluster buffer.
2474 struct xfs_inode *free_ip,
2475 struct xfs_trans *tp,
2476 struct xfs_icluster *xic)
2478 struct xfs_mount *mp = free_ip->i_mount;
2479 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2482 xfs_ino_t inum = xic->first_ino;
2488 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2490 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2492 * The allocation bitmap tells us which inodes of the chunk were
2493 * physically allocated. Skip the cluster if an inode falls into
2496 ioffset = inum - xic->first_ino;
2497 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2498 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2502 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2503 XFS_INO_TO_AGBNO(mp, inum));
2506 * We obtain and lock the backing buffer first in the process
2507 * here to ensure dirty inodes attached to the buffer remain in
2508 * the flushing state while we mark them stale.
2510 * If we scan the in-memory inodes first, then buffer IO can
2511 * complete before we get a lock on it, and hence we may fail
2512 * to mark all the active inodes on the buffer stale.
2514 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2515 mp->m_bsize * igeo->blocks_per_cluster,
2521 * This buffer may not have been correctly initialised as we
2522 * didn't read it from disk. That's not important because we are
2523 * only using to mark the buffer as stale in the log, and to
2524 * attach stale cached inodes on it. That means it will never be
2525 * dispatched for IO. If it is, we want to know about it, and we
2526 * want it to fail. We can acheive this by adding a write
2527 * verifier to the buffer.
2529 bp->b_ops = &xfs_inode_buf_ops;
2532 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2533 * too. This requires lookups, and will skip inodes that we've
2534 * already marked XFS_ISTALE.
2536 for (i = 0; i < igeo->inodes_per_cluster; i++)
2537 xfs_ifree_mark_inode_stale(bp, free_ip, inum + i);
2539 xfs_trans_stale_inode_buf(tp, bp);
2540 xfs_trans_binval(tp, bp);
2546 * This is called to return an inode to the inode free list.
2547 * The inode should already be truncated to 0 length and have
2548 * no pages associated with it. This routine also assumes that
2549 * the inode is already a part of the transaction.
2551 * The on-disk copy of the inode will have been added to the list
2552 * of unlinked inodes in the AGI. We need to remove the inode from
2553 * that list atomically with respect to freeing it here.
2557 struct xfs_trans *tp,
2558 struct xfs_inode *ip)
2561 struct xfs_icluster xic = { 0 };
2562 struct xfs_inode_log_item *iip = ip->i_itemp;
2564 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2565 ASSERT(VFS_I(ip)->i_nlink == 0);
2566 ASSERT(ip->i_df.if_nextents == 0);
2567 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2568 ASSERT(ip->i_d.di_nblocks == 0);
2571 * Pull the on-disk inode from the AGI unlinked list.
2573 error = xfs_iunlink_remove(tp, ip);
2577 error = xfs_difree(tp, ip->i_ino, &xic);
2582 * Free any local-format data sitting around before we reset the
2583 * data fork to extents format. Note that the attr fork data has
2584 * already been freed by xfs_attr_inactive.
2586 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2587 kmem_free(ip->i_df.if_u1.if_data);
2588 ip->i_df.if_u1.if_data = NULL;
2589 ip->i_df.if_bytes = 0;
2592 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2593 ip->i_d.di_flags = 0;
2594 ip->i_d.di_flags2 = ip->i_mount->m_ino_geo.new_diflags2;
2595 ip->i_d.di_dmevmask = 0;
2596 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2597 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2599 /* Don't attempt to replay owner changes for a deleted inode */
2600 spin_lock(&iip->ili_lock);
2601 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2602 spin_unlock(&iip->ili_lock);
2605 * Bump the generation count so no one will be confused
2606 * by reincarnations of this inode.
2608 VFS_I(ip)->i_generation++;
2609 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2612 error = xfs_ifree_cluster(ip, tp, &xic);
2618 * This is called to unpin an inode. The caller must have the inode locked
2619 * in at least shared mode so that the buffer cannot be subsequently pinned
2620 * once someone is waiting for it to be unpinned.
2624 struct xfs_inode *ip)
2626 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2628 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2630 /* Give the log a push to start the unpinning I/O */
2631 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2637 struct xfs_inode *ip)
2639 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2640 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2645 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2646 if (xfs_ipincount(ip))
2648 } while (xfs_ipincount(ip));
2649 finish_wait(wq, &wait.wq_entry);
2654 struct xfs_inode *ip)
2656 if (xfs_ipincount(ip))
2657 __xfs_iunpin_wait(ip);
2661 * Removing an inode from the namespace involves removing the directory entry
2662 * and dropping the link count on the inode. Removing the directory entry can
2663 * result in locking an AGF (directory blocks were freed) and removing a link
2664 * count can result in placing the inode on an unlinked list which results in
2667 * The big problem here is that we have an ordering constraint on AGF and AGI
2668 * locking - inode allocation locks the AGI, then can allocate a new extent for
2669 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2670 * removes the inode from the unlinked list, requiring that we lock the AGI
2671 * first, and then freeing the inode can result in an inode chunk being freed
2672 * and hence freeing disk space requiring that we lock an AGF.
2674 * Hence the ordering that is imposed by other parts of the code is AGI before
2675 * AGF. This means we cannot remove the directory entry before we drop the inode
2676 * reference count and put it on the unlinked list as this results in a lock
2677 * order of AGF then AGI, and this can deadlock against inode allocation and
2678 * freeing. Therefore we must drop the link counts before we remove the
2681 * This is still safe from a transactional point of view - it is not until we
2682 * get to xfs_defer_finish() that we have the possibility of multiple
2683 * transactions in this operation. Hence as long as we remove the directory
2684 * entry and drop the link count in the first transaction of the remove
2685 * operation, there are no transactional constraints on the ordering here.
2690 struct xfs_name *name,
2693 xfs_mount_t *mp = dp->i_mount;
2694 xfs_trans_t *tp = NULL;
2695 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2699 trace_xfs_remove(dp, name);
2701 if (XFS_FORCED_SHUTDOWN(mp))
2704 error = xfs_qm_dqattach(dp);
2708 error = xfs_qm_dqattach(ip);
2713 * We try to get the real space reservation first,
2714 * allowing for directory btree deletion(s) implying
2715 * possible bmap insert(s). If we can't get the space
2716 * reservation then we use 0 instead, and avoid the bmap
2717 * btree insert(s) in the directory code by, if the bmap
2718 * insert tries to happen, instead trimming the LAST
2719 * block from the directory.
2721 resblks = XFS_REMOVE_SPACE_RES(mp);
2722 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2723 if (error == -ENOSPC) {
2725 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2729 ASSERT(error != -ENOSPC);
2733 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2735 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2736 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2739 * If we're removing a directory perform some additional validation.
2742 ASSERT(VFS_I(ip)->i_nlink >= 2);
2743 if (VFS_I(ip)->i_nlink != 2) {
2745 goto out_trans_cancel;
2747 if (!xfs_dir_isempty(ip)) {
2749 goto out_trans_cancel;
2752 /* Drop the link from ip's "..". */
2753 error = xfs_droplink(tp, dp);
2755 goto out_trans_cancel;
2757 /* Drop the "." link from ip to self. */
2758 error = xfs_droplink(tp, ip);
2760 goto out_trans_cancel;
2763 * When removing a non-directory we need to log the parent
2764 * inode here. For a directory this is done implicitly
2765 * by the xfs_droplink call for the ".." entry.
2767 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2769 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2771 /* Drop the link from dp to ip. */
2772 error = xfs_droplink(tp, ip);
2774 goto out_trans_cancel;
2776 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2778 ASSERT(error != -ENOENT);
2779 goto out_trans_cancel;
2783 * If this is a synchronous mount, make sure that the
2784 * remove transaction goes to disk before returning to
2787 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2788 xfs_trans_set_sync(tp);
2790 error = xfs_trans_commit(tp);
2794 if (is_dir && xfs_inode_is_filestream(ip))
2795 xfs_filestream_deassociate(ip);
2800 xfs_trans_cancel(tp);
2806 * Enter all inodes for a rename transaction into a sorted array.
2808 #define __XFS_SORT_INODES 5
2810 xfs_sort_for_rename(
2811 struct xfs_inode *dp1, /* in: old (source) directory inode */
2812 struct xfs_inode *dp2, /* in: new (target) directory inode */
2813 struct xfs_inode *ip1, /* in: inode of old entry */
2814 struct xfs_inode *ip2, /* in: inode of new entry */
2815 struct xfs_inode *wip, /* in: whiteout inode */
2816 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2817 int *num_inodes) /* in/out: inodes in array */
2821 ASSERT(*num_inodes == __XFS_SORT_INODES);
2822 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2825 * i_tab contains a list of pointers to inodes. We initialize
2826 * the table here & we'll sort it. We will then use it to
2827 * order the acquisition of the inode locks.
2829 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2842 * Sort the elements via bubble sort. (Remember, there are at
2843 * most 5 elements to sort, so this is adequate.)
2845 for (i = 0; i < *num_inodes; i++) {
2846 for (j = 1; j < *num_inodes; j++) {
2847 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2848 struct xfs_inode *temp = i_tab[j];
2849 i_tab[j] = i_tab[j-1];
2858 struct xfs_trans *tp)
2861 * If this is a synchronous mount, make sure that the rename transaction
2862 * goes to disk before returning to the user.
2864 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2865 xfs_trans_set_sync(tp);
2867 return xfs_trans_commit(tp);
2871 * xfs_cross_rename()
2873 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2877 struct xfs_trans *tp,
2878 struct xfs_inode *dp1,
2879 struct xfs_name *name1,
2880 struct xfs_inode *ip1,
2881 struct xfs_inode *dp2,
2882 struct xfs_name *name2,
2883 struct xfs_inode *ip2,
2891 /* Swap inode number for dirent in first parent */
2892 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2894 goto out_trans_abort;
2896 /* Swap inode number for dirent in second parent */
2897 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2899 goto out_trans_abort;
2902 * If we're renaming one or more directories across different parents,
2903 * update the respective ".." entries (and link counts) to match the new
2907 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2909 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2910 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2911 dp1->i_ino, spaceres);
2913 goto out_trans_abort;
2915 /* transfer ip2 ".." reference to dp1 */
2916 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2917 error = xfs_droplink(tp, dp2);
2919 goto out_trans_abort;
2920 xfs_bumplink(tp, dp1);
2924 * Although ip1 isn't changed here, userspace needs
2925 * to be warned about the change, so that applications
2926 * relying on it (like backup ones), will properly
2929 ip1_flags |= XFS_ICHGTIME_CHG;
2930 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2933 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2934 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2935 dp2->i_ino, spaceres);
2937 goto out_trans_abort;
2939 /* transfer ip1 ".." reference to dp2 */
2940 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2941 error = xfs_droplink(tp, dp1);
2943 goto out_trans_abort;
2944 xfs_bumplink(tp, dp2);
2948 * Although ip2 isn't changed here, userspace needs
2949 * to be warned about the change, so that applications
2950 * relying on it (like backup ones), will properly
2953 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2954 ip2_flags |= XFS_ICHGTIME_CHG;
2959 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2960 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2963 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2964 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2967 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2968 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2970 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2971 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2972 return xfs_finish_rename(tp);
2975 xfs_trans_cancel(tp);
2980 * xfs_rename_alloc_whiteout()
2982 * Return a referenced, unlinked, unlocked inode that can be used as a
2983 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2984 * crash between allocating the inode and linking it into the rename transaction
2985 * recovery will free the inode and we won't leak it.
2988 xfs_rename_alloc_whiteout(
2989 struct user_namespace *mnt_userns,
2990 struct xfs_inode *dp,
2991 struct xfs_inode **wip)
2993 struct xfs_inode *tmpfile;
2996 error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
3002 * Prepare the tmpfile inode as if it were created through the VFS.
3003 * Complete the inode setup and flag it as linkable. nlink is already
3004 * zero, so we can skip the drop_nlink.
3006 xfs_setup_iops(tmpfile);
3007 xfs_finish_inode_setup(tmpfile);
3008 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3019 struct user_namespace *mnt_userns,
3020 struct xfs_inode *src_dp,
3021 struct xfs_name *src_name,
3022 struct xfs_inode *src_ip,
3023 struct xfs_inode *target_dp,
3024 struct xfs_name *target_name,
3025 struct xfs_inode *target_ip,
3028 struct xfs_mount *mp = src_dp->i_mount;
3029 struct xfs_trans *tp;
3030 struct xfs_inode *wip = NULL; /* whiteout inode */
3031 struct xfs_inode *inodes[__XFS_SORT_INODES];
3033 int num_inodes = __XFS_SORT_INODES;
3034 bool new_parent = (src_dp != target_dp);
3035 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3039 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3041 if ((flags & RENAME_EXCHANGE) && !target_ip)
3045 * If we are doing a whiteout operation, allocate the whiteout inode
3046 * we will be placing at the target and ensure the type is set
3049 if (flags & RENAME_WHITEOUT) {
3050 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3051 error = xfs_rename_alloc_whiteout(mnt_userns, target_dp, &wip);
3055 /* setup target dirent info as whiteout */
3056 src_name->type = XFS_DIR3_FT_CHRDEV;
3059 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3060 inodes, &num_inodes);
3062 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3063 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3064 if (error == -ENOSPC) {
3066 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3070 goto out_release_wip;
3073 * Attach the dquots to the inodes
3075 error = xfs_qm_vop_rename_dqattach(inodes);
3077 goto out_trans_cancel;
3080 * Lock all the participating inodes. Depending upon whether
3081 * the target_name exists in the target directory, and
3082 * whether the target directory is the same as the source
3083 * directory, we can lock from 2 to 4 inodes.
3085 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3088 * Join all the inodes to the transaction. From this point on,
3089 * we can rely on either trans_commit or trans_cancel to unlock
3092 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3094 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3095 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3097 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3099 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3102 * If we are using project inheritance, we only allow renames
3103 * into our tree when the project IDs are the same; else the
3104 * tree quota mechanism would be circumvented.
3106 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3107 target_dp->i_d.di_projid != src_ip->i_d.di_projid)) {
3109 goto out_trans_cancel;
3112 /* RENAME_EXCHANGE is unique from here on. */
3113 if (flags & RENAME_EXCHANGE)
3114 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3115 target_dp, target_name, target_ip,
3119 * Check for expected errors before we dirty the transaction
3120 * so we can return an error without a transaction abort.
3122 * Extent count overflow check:
3124 * From the perspective of src_dp, a rename operation is essentially a
3125 * directory entry remove operation. Hence the only place where we check
3126 * for extent count overflow for src_dp is in
3127 * xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns
3128 * -ENOSPC when it detects a possible extent count overflow and in
3129 * response, the higher layers of directory handling code do the
3131 * 1. Data/Free blocks: XFS lets these blocks linger until a
3132 * future remove operation removes them.
3133 * 2. Dabtree blocks: XFS swaps the blocks with the last block in the
3134 * Leaf space and unmaps the last block.
3136 * For target_dp, there are two cases depending on whether the
3137 * destination directory entry exists or not.
3139 * When destination directory entry does not exist (i.e. target_ip ==
3140 * NULL), extent count overflow check is performed only when transaction
3141 * has a non-zero sized space reservation associated with it. With a
3142 * zero-sized space reservation, XFS allows a rename operation to
3143 * continue only when the directory has sufficient free space in its
3144 * data/leaf/free space blocks to hold the new entry.
3146 * When destination directory entry exists (i.e. target_ip != NULL), all
3147 * we need to do is change the inode number associated with the already
3148 * existing entry. Hence there is no need to perform an extent count
3151 if (target_ip == NULL) {
3153 * If there's no space reservation, check the entry will
3154 * fit before actually inserting it.
3157 error = xfs_dir_canenter(tp, target_dp, target_name);
3159 goto out_trans_cancel;
3161 error = xfs_iext_count_may_overflow(target_dp,
3163 XFS_IEXT_DIR_MANIP_CNT(mp));
3165 goto out_trans_cancel;
3169 * If target exists and it's a directory, check that whether
3170 * it can be destroyed.
3172 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3173 (!xfs_dir_isempty(target_ip) ||
3174 (VFS_I(target_ip)->i_nlink > 2))) {
3176 goto out_trans_cancel;
3181 * Lock the AGI buffers we need to handle bumping the nlink of the
3182 * whiteout inode off the unlinked list and to handle dropping the
3183 * nlink of the target inode. Per locking order rules, do this in
3184 * increasing AG order and before directory block allocation tries to
3185 * grab AGFs because we grab AGIs before AGFs.
3187 * The (vfs) caller must ensure that if src is a directory then
3188 * target_ip is either null or an empty directory.
3190 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3191 if (inodes[i] == wip ||
3192 (inodes[i] == target_ip &&
3193 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3195 xfs_agnumber_t agno;
3197 agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
3198 error = xfs_read_agi(mp, tp, agno, &bp);
3200 goto out_trans_cancel;
3205 * Directory entry creation below may acquire the AGF. Remove
3206 * the whiteout from the unlinked list first to preserve correct
3207 * AGI/AGF locking order. This dirties the transaction so failures
3208 * after this point will abort and log recovery will clean up the
3211 * For whiteouts, we need to bump the link count on the whiteout
3212 * inode. After this point, we have a real link, clear the tmpfile
3213 * state flag from the inode so it doesn't accidentally get misused
3217 ASSERT(VFS_I(wip)->i_nlink == 0);
3218 error = xfs_iunlink_remove(tp, wip);
3220 goto out_trans_cancel;
3222 xfs_bumplink(tp, wip);
3223 VFS_I(wip)->i_state &= ~I_LINKABLE;
3227 * Set up the target.
3229 if (target_ip == NULL) {
3231 * If target does not exist and the rename crosses
3232 * directories, adjust the target directory link count
3233 * to account for the ".." reference from the new entry.
3235 error = xfs_dir_createname(tp, target_dp, target_name,
3236 src_ip->i_ino, spaceres);
3238 goto out_trans_cancel;
3240 xfs_trans_ichgtime(tp, target_dp,
3241 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3243 if (new_parent && src_is_directory) {
3244 xfs_bumplink(tp, target_dp);
3246 } else { /* target_ip != NULL */
3248 * Link the source inode under the target name.
3249 * If the source inode is a directory and we are moving
3250 * it across directories, its ".." entry will be
3251 * inconsistent until we replace that down below.
3253 * In case there is already an entry with the same
3254 * name at the destination directory, remove it first.
3256 error = xfs_dir_replace(tp, target_dp, target_name,
3257 src_ip->i_ino, spaceres);
3259 goto out_trans_cancel;
3261 xfs_trans_ichgtime(tp, target_dp,
3262 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3265 * Decrement the link count on the target since the target
3266 * dir no longer points to it.
3268 error = xfs_droplink(tp, target_ip);
3270 goto out_trans_cancel;
3272 if (src_is_directory) {
3274 * Drop the link from the old "." entry.
3276 error = xfs_droplink(tp, target_ip);
3278 goto out_trans_cancel;
3280 } /* target_ip != NULL */
3283 * Remove the source.
3285 if (new_parent && src_is_directory) {
3287 * Rewrite the ".." entry to point to the new
3290 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3291 target_dp->i_ino, spaceres);
3292 ASSERT(error != -EEXIST);
3294 goto out_trans_cancel;
3298 * We always want to hit the ctime on the source inode.
3300 * This isn't strictly required by the standards since the source
3301 * inode isn't really being changed, but old unix file systems did
3302 * it and some incremental backup programs won't work without it.
3304 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3305 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3308 * Adjust the link count on src_dp. This is necessary when
3309 * renaming a directory, either within one parent when
3310 * the target existed, or across two parent directories.
3312 if (src_is_directory && (new_parent || target_ip != NULL)) {
3315 * Decrement link count on src_directory since the
3316 * entry that's moved no longer points to it.
3318 error = xfs_droplink(tp, src_dp);
3320 goto out_trans_cancel;
3324 * For whiteouts, we only need to update the source dirent with the
3325 * inode number of the whiteout inode rather than removing it
3329 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3333 * NOTE: We don't need to check for extent count overflow here
3334 * because the dir remove name code will leave the dir block in
3335 * place if the extent count would overflow.
3337 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3342 goto out_trans_cancel;
3344 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3345 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3347 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3349 error = xfs_finish_rename(tp);
3355 xfs_trans_cancel(tp);
3364 struct xfs_inode *ip,
3367 struct xfs_inode_log_item *iip = ip->i_itemp;
3368 struct xfs_dinode *dip;
3369 struct xfs_mount *mp = ip->i_mount;
3372 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3373 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3374 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3375 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3376 ASSERT(iip->ili_item.li_buf == bp);
3378 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3381 * We don't flush the inode if any of the following checks fail, but we
3382 * do still update the log item and attach to the backing buffer as if
3383 * the flush happened. This is a formality to facilitate predictable
3384 * error handling as the caller will shutdown and fail the buffer.
3386 error = -EFSCORRUPTED;
3387 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3388 mp, XFS_ERRTAG_IFLUSH_1)) {
3389 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3390 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3391 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3394 if (S_ISREG(VFS_I(ip)->i_mode)) {
3396 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3397 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3398 mp, XFS_ERRTAG_IFLUSH_3)) {
3399 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3400 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3401 __func__, ip->i_ino, ip);
3404 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3406 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3407 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3408 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3409 mp, XFS_ERRTAG_IFLUSH_4)) {
3410 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3411 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3412 __func__, ip->i_ino, ip);
3416 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3417 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3418 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3419 "%s: detected corrupt incore inode %Lu, "
3420 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3421 __func__, ip->i_ino,
3422 ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3423 ip->i_d.di_nblocks, ip);
3426 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3427 mp, XFS_ERRTAG_IFLUSH_6)) {
3428 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3429 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3430 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3435 * Inode item log recovery for v2 inodes are dependent on the
3436 * di_flushiter count for correct sequencing. We bump the flush
3437 * iteration count so we can detect flushes which postdate a log record
3438 * during recovery. This is redundant as we now log every change and
3439 * hence this can't happen but we need to still do it to ensure
3440 * backwards compatibility with old kernels that predate logging all
3443 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3444 ip->i_d.di_flushiter++;
3447 * If there are inline format data / attr forks attached to this inode,
3448 * make sure they are not corrupt.
3450 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3451 xfs_ifork_verify_local_data(ip))
3453 if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3454 xfs_ifork_verify_local_attr(ip))
3458 * Copy the dirty parts of the inode into the on-disk inode. We always
3459 * copy out the core of the inode, because if the inode is dirty at all
3462 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3464 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3465 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3466 ip->i_d.di_flushiter = 0;
3468 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3469 if (XFS_IFORK_Q(ip))
3470 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3473 * We've recorded everything logged in the inode, so we'd like to clear
3474 * the ili_fields bits so we don't log and flush things unnecessarily.
3475 * However, we can't stop logging all this information until the data
3476 * we've copied into the disk buffer is written to disk. If we did we
3477 * might overwrite the copy of the inode in the log with all the data
3478 * after re-logging only part of it, and in the face of a crash we
3479 * wouldn't have all the data we need to recover.
3481 * What we do is move the bits to the ili_last_fields field. When
3482 * logging the inode, these bits are moved back to the ili_fields field.
3483 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3484 * we know that the information those bits represent is permanently on
3485 * disk. As long as the flush completes before the inode is logged
3486 * again, then both ili_fields and ili_last_fields will be cleared.
3490 spin_lock(&iip->ili_lock);
3491 iip->ili_last_fields = iip->ili_fields;
3492 iip->ili_fields = 0;
3493 iip->ili_fsync_fields = 0;
3494 spin_unlock(&iip->ili_lock);
3497 * Store the current LSN of the inode so that we can tell whether the
3498 * item has moved in the AIL from xfs_buf_inode_iodone().
3500 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3501 &iip->ili_item.li_lsn);
3503 /* generate the checksum. */
3504 xfs_dinode_calc_crc(mp, dip);
3509 * Non-blocking flush of dirty inode metadata into the backing buffer.
3511 * The caller must have a reference to the inode and hold the cluster buffer
3512 * locked. The function will walk across all the inodes on the cluster buffer it
3513 * can find and lock without blocking, and flush them to the cluster buffer.
3515 * On successful flushing of at least one inode, the caller must write out the
3516 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3517 * the caller needs to release the buffer. On failure, the filesystem will be
3518 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3525 struct xfs_mount *mp = bp->b_mount;
3526 struct xfs_log_item *lip, *n;
3527 struct xfs_inode *ip;
3528 struct xfs_inode_log_item *iip;
3533 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3534 * can remove itself from the list.
3536 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3537 iip = (struct xfs_inode_log_item *)lip;
3538 ip = iip->ili_inode;
3541 * Quick and dirty check to avoid locks if possible.
3543 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3545 if (xfs_ipincount(ip))
3549 * The inode is still attached to the buffer, which means it is
3550 * dirty but reclaim might try to grab it. Check carefully for
3551 * that, and grab the ilock while still holding the i_flags_lock
3552 * to guarantee reclaim will not be able to reclaim this inode
3553 * once we drop the i_flags_lock.
3555 spin_lock(&ip->i_flags_lock);
3556 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3557 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3558 spin_unlock(&ip->i_flags_lock);
3563 * ILOCK will pin the inode against reclaim and prevent
3564 * concurrent transactions modifying the inode while we are
3565 * flushing the inode. If we get the lock, set the flushing
3566 * state before we drop the i_flags_lock.
3568 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3569 spin_unlock(&ip->i_flags_lock);
3572 __xfs_iflags_set(ip, XFS_IFLUSHING);
3573 spin_unlock(&ip->i_flags_lock);
3576 * Abort flushing this inode if we are shut down because the
3577 * inode may not currently be in the AIL. This can occur when
3578 * log I/O failure unpins the inode without inserting into the
3579 * AIL, leaving a dirty/unpinned inode attached to the buffer
3580 * that otherwise looks like it should be flushed.
3582 if (XFS_FORCED_SHUTDOWN(mp)) {
3583 xfs_iunpin_wait(ip);
3584 xfs_iflush_abort(ip);
3585 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3590 /* don't block waiting on a log force to unpin dirty inodes */
3591 if (xfs_ipincount(ip)) {
3592 xfs_iflags_clear(ip, XFS_IFLUSHING);
3593 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3597 if (!xfs_inode_clean(ip))
3598 error = xfs_iflush(ip, bp);
3600 xfs_iflags_clear(ip, XFS_IFLUSHING);
3601 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3608 bp->b_flags |= XBF_ASYNC;
3609 xfs_buf_ioend_fail(bp);
3610 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3617 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3618 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3623 /* Release an inode. */
3626 struct xfs_inode *ip)
3628 trace_xfs_irele(ip, _RET_IP_);
3633 * Ensure all commited transactions touching the inode are written to the log.
3636 xfs_log_force_inode(
3637 struct xfs_inode *ip)
3641 xfs_ilock(ip, XFS_ILOCK_SHARED);
3642 if (xfs_ipincount(ip))
3643 lsn = ip->i_itemp->ili_last_lsn;
3644 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3648 return xfs_log_force_lsn(ip->i_mount, lsn, XFS_LOG_SYNC, NULL);
3652 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3653 * abide vfs locking order (lowest pointer value goes first) and breaking the
3654 * layout leases before proceeding. The loop is needed because we cannot call
3655 * the blocking break_layout() with the iolocks held, and therefore have to
3656 * back out both locks.
3659 xfs_iolock_two_inodes_and_break_layout(
3669 /* Wait to break both inodes' layouts before we start locking. */
3670 error = break_layout(src, true);
3674 error = break_layout(dest, true);
3679 /* Lock one inode and make sure nobody got in and leased it. */
3681 error = break_layout(src, false);
3684 if (error == -EWOULDBLOCK)
3692 /* Lock the other inode and make sure nobody got in and leased it. */
3693 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3694 error = break_layout(dest, false);
3698 if (error == -EWOULDBLOCK)
3707 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3712 struct xfs_inode *ip1,
3713 struct xfs_inode *ip2)
3717 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3721 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3723 xfs_lock_two_inodes(ip1, XFS_MMAPLOCK_EXCL,
3724 ip2, XFS_MMAPLOCK_EXCL);
3728 /* Unlock both inodes to allow IO and mmap activity. */
3730 xfs_iunlock2_io_mmap(
3731 struct xfs_inode *ip1,
3732 struct xfs_inode *ip2)
3734 bool same_inode = (ip1 == ip2);
3736 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3738 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3739 inode_unlock(VFS_I(ip2));
3741 inode_unlock(VFS_I(ip1));