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 * Allocate an inode on disk and return a copy of its in-core version.
765 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
766 * appropriately within the inode. The uid and gid for the inode are
767 * set according to the contents of the given cred structure.
769 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
770 * has a free inode available, call xfs_iget() to obtain the in-core
771 * version of the allocated inode. Finally, fill in the inode and
772 * log its initial contents. In this case, ialloc_context would be
775 * If xfs_dialloc() does not have an available inode, it will replenish
776 * its supply by doing an allocation. Since we can only do one
777 * allocation within a transaction without deadlocks, we must commit
778 * the current transaction before returning the inode itself.
779 * In this case, therefore, we will set ialloc_context and return.
780 * The caller should then commit the current transaction, start a new
781 * transaction, and call xfs_ialloc() again to actually get the inode.
783 * To ensure that some other process does not grab the inode that
784 * was allocated during the first call to xfs_ialloc(), this routine
785 * also returns the [locked] bp pointing to the head of the freelist
786 * as ialloc_context. The caller should hold this buffer across
787 * the commit and pass it back into this routine on the second call.
789 * If we are allocating quota inodes, we do not have a parent inode
790 * to attach to or associate with (i.e. pip == NULL) because they
791 * are not linked into the directory structure - they are attached
792 * directly to the superblock - and so have no parent.
802 xfs_buf_t **ialloc_context,
805 struct inode *dir = pip ? VFS_I(pip) : NULL;
806 struct xfs_mount *mp = tp->t_mountp;
811 struct timespec64 tv;
815 * Call the space management code to pick
816 * the on-disk inode to be allocated.
818 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
819 ialloc_context, &ino);
822 if (*ialloc_context || ino == NULLFSINO) {
826 ASSERT(*ialloc_context == NULL);
829 * Protect against obviously corrupt allocation btree records. Later
830 * xfs_iget checks will catch re-allocation of other active in-memory
831 * and on-disk inodes. If we don't catch reallocating the parent inode
832 * here we will deadlock in xfs_iget() so we have to do these checks
835 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
836 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
837 return -EFSCORRUPTED;
841 * Get the in-core inode with the lock held exclusively.
842 * This is because we're setting fields here we need
843 * to prevent others from looking at until we're done.
845 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
846 XFS_ILOCK_EXCL, &ip);
851 set_nlink(inode, nlink);
852 inode->i_rdev = rdev;
853 ip->i_d.di_projid = prid;
855 if (dir && !(dir->i_mode & S_ISGID) &&
856 (mp->m_flags & XFS_MOUNT_GRPID)) {
857 inode->i_uid = current_fsuid();
858 inode->i_gid = dir->i_gid;
859 inode->i_mode = mode;
861 inode_init_owner(inode, dir, mode);
865 * If the group ID of the new file does not match the effective group
866 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
867 * (and only if the irix_sgid_inherit compatibility variable is set).
869 if (irix_sgid_inherit &&
870 (inode->i_mode & S_ISGID) && !in_group_p(inode->i_gid))
871 inode->i_mode &= ~S_ISGID;
874 ip->i_df.if_nextents = 0;
875 ASSERT(ip->i_d.di_nblocks == 0);
877 tv = current_time(inode);
882 ip->i_d.di_extsize = 0;
883 ip->i_d.di_dmevmask = 0;
884 ip->i_d.di_dmstate = 0;
885 ip->i_d.di_flags = 0;
887 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
888 inode_set_iversion(inode, 1);
889 ip->i_d.di_flags2 = mp->m_ino_geo.new_diflags2;
890 ip->i_d.di_cowextsize = 0;
891 ip->i_d.di_crtime = tv;
894 flags = XFS_ILOG_CORE;
895 switch (mode & S_IFMT) {
900 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
901 ip->i_df.if_flags = 0;
902 flags |= XFS_ILOG_DEV;
906 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY))
907 xfs_inode_inherit_flags(ip, pip);
908 if (pip && (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY))
909 xfs_inode_inherit_flags2(ip, pip);
912 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
913 ip->i_df.if_flags = XFS_IFEXTENTS;
914 ip->i_df.if_bytes = 0;
915 ip->i_df.if_u1.if_root = NULL;
922 * Log the new values stuffed into the inode.
924 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
925 xfs_trans_log_inode(tp, ip, flags);
927 /* now that we have an i_mode we can setup the inode structure */
935 * Allocates a new inode from disk and return a pointer to the
936 * incore copy. This routine will internally commit the current
937 * transaction and allocate a new one if the Space Manager needed
938 * to do an allocation to replenish the inode free-list.
940 * This routine is designed to be called from xfs_create and
946 xfs_trans_t **tpp, /* input: current transaction;
947 output: may be a new transaction. */
948 xfs_inode_t *dp, /* directory within whose allocate
953 prid_t prid, /* project id */
954 xfs_inode_t **ipp) /* pointer to inode; it will be
959 xfs_buf_t *ialloc_context = NULL;
965 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
968 * xfs_ialloc will return a pointer to an incore inode if
969 * the Space Manager has an available inode on the free
970 * list. Otherwise, it will do an allocation and replenish
971 * the freelist. Since we can only do one allocation per
972 * transaction without deadlocks, we will need to commit the
973 * current transaction and start a new one. We will then
974 * need to call xfs_ialloc again to get the inode.
976 * If xfs_ialloc did an allocation to replenish the freelist,
977 * it returns the bp containing the head of the freelist as
978 * ialloc_context. We will hold a lock on it across the
979 * transaction commit so that no other process can steal
980 * the inode(s) that we've just allocated.
982 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
986 * Return an error if we were unable to allocate a new inode.
987 * This should only happen if we run out of space on disk or
988 * encounter a disk error.
994 if (!ialloc_context && !ip) {
1000 * If the AGI buffer is non-NULL, then we were unable to get an
1001 * inode in one operation. We need to commit the current
1002 * transaction and call xfs_ialloc() again. It is guaranteed
1003 * to succeed the second time.
1005 if (ialloc_context) {
1007 * Normally, xfs_trans_commit releases all the locks.
1008 * We call bhold to hang on to the ialloc_context across
1009 * the commit. Holding this buffer prevents any other
1010 * processes from doing any allocations in this
1013 xfs_trans_bhold(tp, ialloc_context);
1016 * We want the quota changes to be associated with the next
1017 * transaction, NOT this one. So, detach the dqinfo from this
1018 * and attach it to the next transaction.
1023 dqinfo = (void *)tp->t_dqinfo;
1024 tp->t_dqinfo = NULL;
1025 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1026 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1029 code = xfs_trans_roll(&tp);
1032 * Re-attach the quota info that we detached from prev trx.
1035 tp->t_dqinfo = dqinfo;
1036 tp->t_flags |= tflags;
1040 xfs_buf_relse(ialloc_context);
1045 xfs_trans_bjoin(tp, ialloc_context);
1048 * Call ialloc again. Since we've locked out all
1049 * other allocations in this allocation group,
1050 * this call should always succeed.
1052 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1053 &ialloc_context, &ip);
1056 * If we get an error at this point, return to the caller
1057 * so that the current transaction can be aborted.
1064 ASSERT(!ialloc_context && ip);
1075 * Decrement the link count on an inode & log the change. If this causes the
1076 * link count to go to zero, move the inode to AGI unlinked list so that it can
1077 * be freed when the last active reference goes away via xfs_inactive().
1079 static int /* error */
1084 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1086 drop_nlink(VFS_I(ip));
1087 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1089 if (VFS_I(ip)->i_nlink)
1092 return xfs_iunlink(tp, ip);
1096 * Increment the link count on an inode & log the change.
1103 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1105 inc_nlink(VFS_I(ip));
1106 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1112 struct xfs_name *name,
1117 int is_dir = S_ISDIR(mode);
1118 struct xfs_mount *mp = dp->i_mount;
1119 struct xfs_inode *ip = NULL;
1120 struct xfs_trans *tp = NULL;
1122 bool unlock_dp_on_error = false;
1124 struct xfs_dquot *udqp = NULL;
1125 struct xfs_dquot *gdqp = NULL;
1126 struct xfs_dquot *pdqp = NULL;
1127 struct xfs_trans_res *tres;
1130 trace_xfs_create(dp, name);
1132 if (XFS_FORCED_SHUTDOWN(mp))
1135 prid = xfs_get_initial_prid(dp);
1138 * Make sure that we have allocated dquot(s) on disk.
1140 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1141 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1142 &udqp, &gdqp, &pdqp);
1147 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1148 tres = &M_RES(mp)->tr_mkdir;
1150 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1151 tres = &M_RES(mp)->tr_create;
1155 * Initially assume that the file does not exist and
1156 * reserve the resources for that case. If that is not
1157 * the case we'll drop the one we have and get a more
1158 * appropriate transaction later.
1160 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1161 if (error == -ENOSPC) {
1162 /* flush outstanding delalloc blocks and retry */
1163 xfs_flush_inodes(mp);
1164 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1167 goto out_release_inode;
1169 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1170 unlock_dp_on_error = true;
1173 * Reserve disk quota and the inode.
1175 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1176 pdqp, resblks, 1, 0);
1178 goto out_trans_cancel;
1181 * A newly created regular or special file just has one directory
1182 * entry pointing to them, but a directory also the "." entry
1183 * pointing to itself.
1185 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
1187 goto out_trans_cancel;
1190 * Now we join the directory inode to the transaction. We do not do it
1191 * earlier because xfs_dir_ialloc might commit the previous transaction
1192 * (and release all the locks). An error from here on will result in
1193 * the transaction cancel unlocking dp so don't do it explicitly in the
1196 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1197 unlock_dp_on_error = false;
1199 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1200 resblks - XFS_IALLOC_SPACE_RES(mp));
1202 ASSERT(error != -ENOSPC);
1203 goto out_trans_cancel;
1205 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1206 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1209 error = xfs_dir_init(tp, ip, dp);
1211 goto out_trans_cancel;
1213 xfs_bumplink(tp, dp);
1217 * If this is a synchronous mount, make sure that the
1218 * create transaction goes to disk before returning to
1221 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1222 xfs_trans_set_sync(tp);
1225 * Attach the dquot(s) to the inodes and modify them incore.
1226 * These ids of the inode couldn't have changed since the new
1227 * inode has been locked ever since it was created.
1229 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1231 error = xfs_trans_commit(tp);
1233 goto out_release_inode;
1235 xfs_qm_dqrele(udqp);
1236 xfs_qm_dqrele(gdqp);
1237 xfs_qm_dqrele(pdqp);
1243 xfs_trans_cancel(tp);
1246 * Wait until after the current transaction is aborted to finish the
1247 * setup of the inode and release the inode. This prevents recursive
1248 * transactions and deadlocks from xfs_inactive.
1251 xfs_finish_inode_setup(ip);
1255 xfs_qm_dqrele(udqp);
1256 xfs_qm_dqrele(gdqp);
1257 xfs_qm_dqrele(pdqp);
1259 if (unlock_dp_on_error)
1260 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1266 struct xfs_inode *dp,
1268 struct xfs_inode **ipp)
1270 struct xfs_mount *mp = dp->i_mount;
1271 struct xfs_inode *ip = NULL;
1272 struct xfs_trans *tp = NULL;
1275 struct xfs_dquot *udqp = NULL;
1276 struct xfs_dquot *gdqp = NULL;
1277 struct xfs_dquot *pdqp = NULL;
1278 struct xfs_trans_res *tres;
1281 if (XFS_FORCED_SHUTDOWN(mp))
1284 prid = xfs_get_initial_prid(dp);
1287 * Make sure that we have allocated dquot(s) on disk.
1289 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1290 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1291 &udqp, &gdqp, &pdqp);
1295 resblks = XFS_IALLOC_SPACE_RES(mp);
1296 tres = &M_RES(mp)->tr_create_tmpfile;
1298 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1300 goto out_release_inode;
1302 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1303 pdqp, resblks, 1, 0);
1305 goto out_trans_cancel;
1307 error = xfs_dir_ialloc(&tp, dp, mode, 0, 0, prid, &ip);
1309 goto out_trans_cancel;
1311 if (mp->m_flags & XFS_MOUNT_WSYNC)
1312 xfs_trans_set_sync(tp);
1315 * Attach the dquot(s) to the inodes and modify them incore.
1316 * These ids of the inode couldn't have changed since the new
1317 * inode has been locked ever since it was created.
1319 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1321 error = xfs_iunlink(tp, ip);
1323 goto out_trans_cancel;
1325 error = xfs_trans_commit(tp);
1327 goto out_release_inode;
1329 xfs_qm_dqrele(udqp);
1330 xfs_qm_dqrele(gdqp);
1331 xfs_qm_dqrele(pdqp);
1337 xfs_trans_cancel(tp);
1340 * Wait until after the current transaction is aborted to finish the
1341 * setup of the inode and release the inode. This prevents recursive
1342 * transactions and deadlocks from xfs_inactive.
1345 xfs_finish_inode_setup(ip);
1349 xfs_qm_dqrele(udqp);
1350 xfs_qm_dqrele(gdqp);
1351 xfs_qm_dqrele(pdqp);
1360 struct xfs_name *target_name)
1362 xfs_mount_t *mp = tdp->i_mount;
1367 trace_xfs_link(tdp, target_name);
1369 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1371 if (XFS_FORCED_SHUTDOWN(mp))
1374 error = xfs_qm_dqattach(sip);
1378 error = xfs_qm_dqattach(tdp);
1382 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1383 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1384 if (error == -ENOSPC) {
1386 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1391 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1393 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1394 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1397 * If we are using project inheritance, we only allow hard link
1398 * creation in our tree when the project IDs are the same; else
1399 * the tree quota mechanism could be circumvented.
1401 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1402 tdp->i_d.di_projid != sip->i_d.di_projid)) {
1408 error = xfs_dir_canenter(tp, tdp, target_name);
1414 * Handle initial link state of O_TMPFILE inode
1416 if (VFS_I(sip)->i_nlink == 0) {
1417 error = xfs_iunlink_remove(tp, sip);
1422 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1426 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1427 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1429 xfs_bumplink(tp, sip);
1432 * If this is a synchronous mount, make sure that the
1433 * link transaction goes to disk before returning to
1436 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1437 xfs_trans_set_sync(tp);
1439 return xfs_trans_commit(tp);
1442 xfs_trans_cancel(tp);
1447 /* Clear the reflink flag and the cowblocks tag if possible. */
1449 xfs_itruncate_clear_reflink_flags(
1450 struct xfs_inode *ip)
1452 struct xfs_ifork *dfork;
1453 struct xfs_ifork *cfork;
1455 if (!xfs_is_reflink_inode(ip))
1457 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1458 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1459 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1460 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1461 if (cfork->if_bytes == 0)
1462 xfs_inode_clear_cowblocks_tag(ip);
1466 * Free up the underlying blocks past new_size. The new size must be smaller
1467 * than the current size. This routine can be used both for the attribute and
1468 * data fork, and does not modify the inode size, which is left to the caller.
1470 * The transaction passed to this routine must have made a permanent log
1471 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1472 * given transaction and start new ones, so make sure everything involved in
1473 * the transaction is tidy before calling here. Some transaction will be
1474 * returned to the caller to be committed. The incoming transaction must
1475 * already include the inode, and both inode locks must be held exclusively.
1476 * The inode must also be "held" within the transaction. On return the inode
1477 * will be "held" within the returned transaction. This routine does NOT
1478 * require any disk space to be reserved for it within the transaction.
1480 * If we get an error, we must return with the inode locked and linked into the
1481 * current transaction. This keeps things simple for the higher level code,
1482 * because it always knows that the inode is locked and held in the transaction
1483 * that returns to it whether errors occur or not. We don't mark the inode
1484 * dirty on error so that transactions can be easily aborted if possible.
1487 xfs_itruncate_extents_flags(
1488 struct xfs_trans **tpp,
1489 struct xfs_inode *ip,
1491 xfs_fsize_t new_size,
1494 struct xfs_mount *mp = ip->i_mount;
1495 struct xfs_trans *tp = *tpp;
1496 xfs_fileoff_t first_unmap_block;
1497 xfs_filblks_t unmap_len;
1500 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1501 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1502 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1503 ASSERT(new_size <= XFS_ISIZE(ip));
1504 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1505 ASSERT(ip->i_itemp != NULL);
1506 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1507 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1509 trace_xfs_itruncate_extents_start(ip, new_size);
1511 flags |= xfs_bmapi_aflag(whichfork);
1514 * Since it is possible for space to become allocated beyond
1515 * the end of the file (in a crash where the space is allocated
1516 * but the inode size is not yet updated), simply remove any
1517 * blocks which show up between the new EOF and the maximum
1518 * possible file size.
1520 * We have to free all the blocks to the bmbt maximum offset, even if
1521 * the page cache can't scale that far.
1523 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1524 if (first_unmap_block >= XFS_MAX_FILEOFF) {
1525 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1529 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1530 while (unmap_len > 0) {
1531 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1532 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1533 flags, XFS_ITRUNC_MAX_EXTENTS);
1537 /* free the just unmapped extents */
1538 error = xfs_defer_finish(&tp);
1543 if (whichfork == XFS_DATA_FORK) {
1544 /* Remove all pending CoW reservations. */
1545 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1546 first_unmap_block, XFS_MAX_FILEOFF, true);
1550 xfs_itruncate_clear_reflink_flags(ip);
1554 * Always re-log the inode so that our permanent transaction can keep
1555 * on rolling it forward in the log.
1557 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1559 trace_xfs_itruncate_extents_end(ip, new_size);
1570 xfs_mount_t *mp = ip->i_mount;
1573 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1576 /* If this is a read-only mount, don't do this (would generate I/O) */
1577 if (mp->m_flags & XFS_MOUNT_RDONLY)
1580 if (!XFS_FORCED_SHUTDOWN(mp)) {
1584 * If we previously truncated this file and removed old data
1585 * in the process, we want to initiate "early" writeout on
1586 * the last close. This is an attempt to combat the notorious
1587 * NULL files problem which is particularly noticeable from a
1588 * truncate down, buffered (re-)write (delalloc), followed by
1589 * a crash. What we are effectively doing here is
1590 * significantly reducing the time window where we'd otherwise
1591 * be exposed to that problem.
1593 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1595 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1596 if (ip->i_delayed_blks > 0) {
1597 error = filemap_flush(VFS_I(ip)->i_mapping);
1604 if (VFS_I(ip)->i_nlink == 0)
1607 if (xfs_can_free_eofblocks(ip, false)) {
1610 * Check if the inode is being opened, written and closed
1611 * frequently and we have delayed allocation blocks outstanding
1612 * (e.g. streaming writes from the NFS server), truncating the
1613 * blocks past EOF will cause fragmentation to occur.
1615 * In this case don't do the truncation, but we have to be
1616 * careful how we detect this case. Blocks beyond EOF show up as
1617 * i_delayed_blks even when the inode is clean, so we need to
1618 * truncate them away first before checking for a dirty release.
1619 * Hence on the first dirty close we will still remove the
1620 * speculative allocation, but after that we will leave it in
1623 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1626 * If we can't get the iolock just skip truncating the blocks
1627 * past EOF because we could deadlock with the mmap_lock
1628 * otherwise. We'll get another chance to drop them once the
1629 * last reference to the inode is dropped, so we'll never leak
1630 * blocks permanently.
1632 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1633 error = xfs_free_eofblocks(ip);
1634 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1639 /* delalloc blocks after truncation means it really is dirty */
1640 if (ip->i_delayed_blks)
1641 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1647 * xfs_inactive_truncate
1649 * Called to perform a truncate when an inode becomes unlinked.
1652 xfs_inactive_truncate(
1653 struct xfs_inode *ip)
1655 struct xfs_mount *mp = ip->i_mount;
1656 struct xfs_trans *tp;
1659 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1661 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1664 xfs_ilock(ip, XFS_ILOCK_EXCL);
1665 xfs_trans_ijoin(tp, ip, 0);
1668 * Log the inode size first to prevent stale data exposure in the event
1669 * of a system crash before the truncate completes. See the related
1670 * comment in xfs_vn_setattr_size() for details.
1672 ip->i_d.di_size = 0;
1673 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1675 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1677 goto error_trans_cancel;
1679 ASSERT(ip->i_df.if_nextents == 0);
1681 error = xfs_trans_commit(tp);
1685 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1689 xfs_trans_cancel(tp);
1691 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1696 * xfs_inactive_ifree()
1698 * Perform the inode free when an inode is unlinked.
1702 struct xfs_inode *ip)
1704 struct xfs_mount *mp = ip->i_mount;
1705 struct xfs_trans *tp;
1709 * We try to use a per-AG reservation for any block needed by the finobt
1710 * tree, but as the finobt feature predates the per-AG reservation
1711 * support a degraded file system might not have enough space for the
1712 * reservation at mount time. In that case try to dip into the reserved
1715 * Send a warning if the reservation does happen to fail, as the inode
1716 * now remains allocated and sits on the unlinked list until the fs is
1719 if (unlikely(mp->m_finobt_nores)) {
1720 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1721 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1724 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1727 if (error == -ENOSPC) {
1728 xfs_warn_ratelimited(mp,
1729 "Failed to remove inode(s) from unlinked list. "
1730 "Please free space, unmount and run xfs_repair.");
1732 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1738 * We do not hold the inode locked across the entire rolling transaction
1739 * here. We only need to hold it for the first transaction that
1740 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1741 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1742 * here breaks the relationship between cluster buffer invalidation and
1743 * stale inode invalidation on cluster buffer item journal commit
1744 * completion, and can result in leaving dirty stale inodes hanging
1747 * We have no need for serialising this inode operation against other
1748 * operations - we freed the inode and hence reallocation is required
1749 * and that will serialise on reallocating the space the deferops need
1750 * to free. Hence we can unlock the inode on the first commit of
1751 * the transaction rather than roll it right through the deferops. This
1752 * avoids relogging the XFS_ISTALE inode.
1754 * We check that xfs_ifree() hasn't grown an internal transaction roll
1755 * by asserting that the inode is still locked when it returns.
1757 xfs_ilock(ip, XFS_ILOCK_EXCL);
1758 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1760 error = xfs_ifree(tp, ip);
1761 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1764 * If we fail to free the inode, shut down. The cancel
1765 * might do that, we need to make sure. Otherwise the
1766 * inode might be lost for a long time or forever.
1768 if (!XFS_FORCED_SHUTDOWN(mp)) {
1769 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1771 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1773 xfs_trans_cancel(tp);
1778 * Credit the quota account(s). The inode is gone.
1780 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1783 * Just ignore errors at this point. There is nothing we can do except
1784 * to try to keep going. Make sure it's not a silent error.
1786 error = xfs_trans_commit(tp);
1788 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1797 * This is called when the vnode reference count for the vnode
1798 * goes to zero. If the file has been unlinked, then it must
1799 * now be truncated. Also, we clear all of the read-ahead state
1800 * kept for the inode here since the file is now closed.
1806 struct xfs_mount *mp;
1811 * If the inode is already free, then there can be nothing
1814 if (VFS_I(ip)->i_mode == 0) {
1815 ASSERT(ip->i_df.if_broot_bytes == 0);
1820 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1822 /* If this is a read-only mount, don't do this (would generate I/O) */
1823 if (mp->m_flags & XFS_MOUNT_RDONLY)
1826 /* Try to clean out the cow blocks if there are any. */
1827 if (xfs_inode_has_cow_data(ip))
1828 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1830 if (VFS_I(ip)->i_nlink != 0) {
1832 * force is true because we are evicting an inode from the
1833 * cache. Post-eof blocks must be freed, lest we end up with
1834 * broken free space accounting.
1836 * Note: don't bother with iolock here since lockdep complains
1837 * about acquiring it in reclaim context. We have the only
1838 * reference to the inode at this point anyways.
1840 if (xfs_can_free_eofblocks(ip, true))
1841 xfs_free_eofblocks(ip);
1846 if (S_ISREG(VFS_I(ip)->i_mode) &&
1847 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1848 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1851 error = xfs_qm_dqattach(ip);
1855 if (S_ISLNK(VFS_I(ip)->i_mode))
1856 error = xfs_inactive_symlink(ip);
1858 error = xfs_inactive_truncate(ip);
1863 * If there are attributes associated with the file then blow them away
1864 * now. The code calls a routine that recursively deconstructs the
1865 * attribute fork. If also blows away the in-core attribute fork.
1867 if (XFS_IFORK_Q(ip)) {
1868 error = xfs_attr_inactive(ip);
1874 ASSERT(ip->i_d.di_forkoff == 0);
1879 error = xfs_inactive_ifree(ip);
1884 * Release the dquots held by inode, if any.
1886 xfs_qm_dqdetach(ip);
1890 * In-Core Unlinked List Lookups
1891 * =============================
1893 * Every inode is supposed to be reachable from some other piece of metadata
1894 * with the exception of the root directory. Inodes with a connection to a
1895 * file descriptor but not linked from anywhere in the on-disk directory tree
1896 * are collectively known as unlinked inodes, though the filesystem itself
1897 * maintains links to these inodes so that on-disk metadata are consistent.
1899 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1900 * header contains a number of buckets that point to an inode, and each inode
1901 * record has a pointer to the next inode in the hash chain. This
1902 * singly-linked list causes scaling problems in the iunlink remove function
1903 * because we must walk that list to find the inode that points to the inode
1904 * being removed from the unlinked hash bucket list.
1906 * What if we modelled the unlinked list as a collection of records capturing
1907 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1908 * have a fast way to look up unlinked list predecessors, which avoids the
1909 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1912 * Because this is a backref cache, we ignore operational failures since the
1913 * iunlink code can fall back to the slow bucket walk. The only errors that
1914 * should bubble out are for obviously incorrect situations.
1916 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1917 * access or have otherwise provided for concurrency control.
1920 /* Capture a "X.next_unlinked = Y" relationship. */
1921 struct xfs_iunlink {
1922 struct rhash_head iu_rhash_head;
1923 xfs_agino_t iu_agino; /* X */
1924 xfs_agino_t iu_next_unlinked; /* Y */
1927 /* Unlinked list predecessor lookup hashtable construction */
1929 xfs_iunlink_obj_cmpfn(
1930 struct rhashtable_compare_arg *arg,
1933 const xfs_agino_t *key = arg->key;
1934 const struct xfs_iunlink *iu = obj;
1936 if (iu->iu_next_unlinked != *key)
1941 static const struct rhashtable_params xfs_iunlink_hash_params = {
1942 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1943 .key_len = sizeof(xfs_agino_t),
1944 .key_offset = offsetof(struct xfs_iunlink,
1946 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1947 .automatic_shrinking = true,
1948 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1952 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1953 * relation is found.
1956 xfs_iunlink_lookup_backref(
1957 struct xfs_perag *pag,
1960 struct xfs_iunlink *iu;
1962 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1963 xfs_iunlink_hash_params);
1964 return iu ? iu->iu_agino : NULLAGINO;
1968 * Take ownership of an iunlink cache entry and insert it into the hash table.
1969 * If successful, the entry will be owned by the cache; if not, it is freed.
1970 * Either way, the caller does not own @iu after this call.
1973 xfs_iunlink_insert_backref(
1974 struct xfs_perag *pag,
1975 struct xfs_iunlink *iu)
1979 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1980 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1982 * Fail loudly if there already was an entry because that's a sign of
1983 * corruption of in-memory data. Also fail loudly if we see an error
1984 * code we didn't anticipate from the rhashtable code. Currently we
1985 * only anticipate ENOMEM.
1988 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1992 * Absorb any runtime errors that aren't a result of corruption because
1993 * this is a cache and we can always fall back to bucket list scanning.
1995 if (error != 0 && error != -EEXIST)
2000 /* Remember that @prev_agino.next_unlinked = @this_agino. */
2002 xfs_iunlink_add_backref(
2003 struct xfs_perag *pag,
2004 xfs_agino_t prev_agino,
2005 xfs_agino_t this_agino)
2007 struct xfs_iunlink *iu;
2009 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
2012 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
2013 iu->iu_agino = prev_agino;
2014 iu->iu_next_unlinked = this_agino;
2016 return xfs_iunlink_insert_backref(pag, iu);
2020 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
2021 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
2022 * wasn't any such entry then we don't bother.
2025 xfs_iunlink_change_backref(
2026 struct xfs_perag *pag,
2028 xfs_agino_t next_unlinked)
2030 struct xfs_iunlink *iu;
2033 /* Look up the old entry; if there wasn't one then exit. */
2034 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
2035 xfs_iunlink_hash_params);
2040 * Remove the entry. This shouldn't ever return an error, but if we
2041 * couldn't remove the old entry we don't want to add it again to the
2042 * hash table, and if the entry disappeared on us then someone's
2043 * violated the locking rules and we need to fail loudly. Either way
2044 * we cannot remove the inode because internal state is or would have
2047 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
2048 &iu->iu_rhash_head, xfs_iunlink_hash_params);
2052 /* If there is no new next entry just free our item and return. */
2053 if (next_unlinked == NULLAGINO) {
2058 /* Update the entry and re-add it to the hash table. */
2059 iu->iu_next_unlinked = next_unlinked;
2060 return xfs_iunlink_insert_backref(pag, iu);
2063 /* Set up the in-core predecessor structures. */
2066 struct xfs_perag *pag)
2068 return rhashtable_init(&pag->pagi_unlinked_hash,
2069 &xfs_iunlink_hash_params);
2072 /* Free the in-core predecessor structures. */
2074 xfs_iunlink_free_item(
2078 struct xfs_iunlink *iu = ptr;
2079 bool *freed_anything = arg;
2081 *freed_anything = true;
2086 xfs_iunlink_destroy(
2087 struct xfs_perag *pag)
2089 bool freed_anything = false;
2091 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
2092 xfs_iunlink_free_item, &freed_anything);
2094 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
2098 * Point the AGI unlinked bucket at an inode and log the results. The caller
2099 * is responsible for validating the old value.
2102 xfs_iunlink_update_bucket(
2103 struct xfs_trans *tp,
2104 xfs_agnumber_t agno,
2105 struct xfs_buf *agibp,
2106 unsigned int bucket_index,
2107 xfs_agino_t new_agino)
2109 struct xfs_agi *agi = agibp->b_addr;
2110 xfs_agino_t old_value;
2113 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
2115 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2116 trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
2117 old_value, new_agino);
2120 * We should never find the head of the list already set to the value
2121 * passed in because either we're adding or removing ourselves from the
2124 if (old_value == new_agino) {
2125 xfs_buf_mark_corrupt(agibp);
2126 return -EFSCORRUPTED;
2129 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2130 offset = offsetof(struct xfs_agi, agi_unlinked) +
2131 (sizeof(xfs_agino_t) * bucket_index);
2132 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2136 /* Set an on-disk inode's next_unlinked pointer. */
2138 xfs_iunlink_update_dinode(
2139 struct xfs_trans *tp,
2140 xfs_agnumber_t agno,
2142 struct xfs_buf *ibp,
2143 struct xfs_dinode *dip,
2144 struct xfs_imap *imap,
2145 xfs_agino_t next_agino)
2147 struct xfs_mount *mp = tp->t_mountp;
2150 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2152 trace_xfs_iunlink_update_dinode(mp, agno, agino,
2153 be32_to_cpu(dip->di_next_unlinked), next_agino);
2155 dip->di_next_unlinked = cpu_to_be32(next_agino);
2156 offset = imap->im_boffset +
2157 offsetof(struct xfs_dinode, di_next_unlinked);
2159 /* need to recalc the inode CRC if appropriate */
2160 xfs_dinode_calc_crc(mp, dip);
2161 xfs_trans_inode_buf(tp, ibp);
2162 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2165 /* Set an in-core inode's unlinked pointer and return the old value. */
2167 xfs_iunlink_update_inode(
2168 struct xfs_trans *tp,
2169 struct xfs_inode *ip,
2170 xfs_agnumber_t agno,
2171 xfs_agino_t next_agino,
2172 xfs_agino_t *old_next_agino)
2174 struct xfs_mount *mp = tp->t_mountp;
2175 struct xfs_dinode *dip;
2176 struct xfs_buf *ibp;
2177 xfs_agino_t old_value;
2180 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2182 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 0);
2186 /* Make sure the old pointer isn't garbage. */
2187 old_value = be32_to_cpu(dip->di_next_unlinked);
2188 if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2189 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2190 sizeof(*dip), __this_address);
2191 error = -EFSCORRUPTED;
2196 * Since we're updating a linked list, we should never find that the
2197 * current pointer is the same as the new value, unless we're
2198 * terminating the list.
2200 *old_next_agino = old_value;
2201 if (old_value == next_agino) {
2202 if (next_agino != NULLAGINO) {
2203 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2204 dip, sizeof(*dip), __this_address);
2205 error = -EFSCORRUPTED;
2210 /* Ok, update the new pointer. */
2211 xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2212 ibp, dip, &ip->i_imap, next_agino);
2215 xfs_trans_brelse(tp, ibp);
2220 * This is called when the inode's link count has gone to 0 or we are creating
2221 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2223 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2224 * list when the inode is freed.
2228 struct xfs_trans *tp,
2229 struct xfs_inode *ip)
2231 struct xfs_mount *mp = tp->t_mountp;
2232 struct xfs_agi *agi;
2233 struct xfs_buf *agibp;
2234 xfs_agino_t next_agino;
2235 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2236 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2237 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2240 ASSERT(VFS_I(ip)->i_nlink == 0);
2241 ASSERT(VFS_I(ip)->i_mode != 0);
2242 trace_xfs_iunlink(ip);
2244 /* Get the agi buffer first. It ensures lock ordering on the list. */
2245 error = xfs_read_agi(mp, tp, agno, &agibp);
2248 agi = agibp->b_addr;
2251 * Get the index into the agi hash table for the list this inode will
2252 * go on. Make sure the pointer isn't garbage and that this inode
2253 * isn't already on the list.
2255 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2256 if (next_agino == agino ||
2257 !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2258 xfs_buf_mark_corrupt(agibp);
2259 return -EFSCORRUPTED;
2262 if (next_agino != NULLAGINO) {
2263 xfs_agino_t old_agino;
2266 * There is already another inode in the bucket, so point this
2267 * inode to the current head of the list.
2269 error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2273 ASSERT(old_agino == NULLAGINO);
2276 * agino has been unlinked, add a backref from the next inode
2279 error = xfs_iunlink_add_backref(agibp->b_pag, agino, next_agino);
2284 /* Point the head of the list to point to this inode. */
2285 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2288 /* Return the imap, dinode pointer, and buffer for an inode. */
2290 xfs_iunlink_map_ino(
2291 struct xfs_trans *tp,
2292 xfs_agnumber_t agno,
2294 struct xfs_imap *imap,
2295 struct xfs_dinode **dipp,
2296 struct xfs_buf **bpp)
2298 struct xfs_mount *mp = tp->t_mountp;
2302 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2304 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2309 error = xfs_imap_to_bp(mp, tp, imap, dipp, bpp, 0);
2311 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2320 * Walk the unlinked chain from @head_agino until we find the inode that
2321 * points to @target_agino. Return the inode number, map, dinode pointer,
2322 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2324 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2325 * @agino, @imap, @dipp, and @bpp are all output parameters.
2327 * Do not call this function if @target_agino is the head of the list.
2330 xfs_iunlink_map_prev(
2331 struct xfs_trans *tp,
2332 xfs_agnumber_t agno,
2333 xfs_agino_t head_agino,
2334 xfs_agino_t target_agino,
2336 struct xfs_imap *imap,
2337 struct xfs_dinode **dipp,
2338 struct xfs_buf **bpp,
2339 struct xfs_perag *pag)
2341 struct xfs_mount *mp = tp->t_mountp;
2342 xfs_agino_t next_agino;
2345 ASSERT(head_agino != target_agino);
2348 /* See if our backref cache can find it faster. */
2349 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2350 if (*agino != NULLAGINO) {
2351 error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2355 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2359 * If we get here the cache contents were corrupt, so drop the
2360 * buffer and fall back to walking the bucket list.
2362 xfs_trans_brelse(tp, *bpp);
2367 trace_xfs_iunlink_map_prev_fallback(mp, agno);
2369 /* Otherwise, walk the entire bucket until we find it. */
2370 next_agino = head_agino;
2371 while (next_agino != target_agino) {
2372 xfs_agino_t unlinked_agino;
2375 xfs_trans_brelse(tp, *bpp);
2377 *agino = next_agino;
2378 error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2383 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2385 * Make sure this pointer is valid and isn't an obvious
2388 if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2389 next_agino == unlinked_agino) {
2390 XFS_CORRUPTION_ERROR(__func__,
2391 XFS_ERRLEVEL_LOW, mp,
2392 *dipp, sizeof(**dipp));
2393 error = -EFSCORRUPTED;
2396 next_agino = unlinked_agino;
2403 * Pull the on-disk inode from the AGI unlinked list.
2407 struct xfs_trans *tp,
2408 struct xfs_inode *ip)
2410 struct xfs_mount *mp = tp->t_mountp;
2411 struct xfs_agi *agi;
2412 struct xfs_buf *agibp;
2413 struct xfs_buf *last_ibp;
2414 struct xfs_dinode *last_dip = NULL;
2415 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2416 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2417 xfs_agino_t next_agino;
2418 xfs_agino_t head_agino;
2419 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2422 trace_xfs_iunlink_remove(ip);
2424 /* Get the agi buffer first. It ensures lock ordering on the list. */
2425 error = xfs_read_agi(mp, tp, agno, &agibp);
2428 agi = agibp->b_addr;
2431 * Get the index into the agi hash table for the list this inode will
2432 * go on. Make sure the head pointer isn't garbage.
2434 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2435 if (!xfs_verify_agino(mp, agno, head_agino)) {
2436 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2438 return -EFSCORRUPTED;
2442 * Set our inode's next_unlinked pointer to NULL and then return
2443 * the old pointer value so that we can update whatever was previous
2444 * to us in the list to point to whatever was next in the list.
2446 error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2451 * If there was a backref pointing from the next inode back to this
2452 * one, remove it because we've removed this inode from the list.
2454 * Later, if this inode was in the middle of the list we'll update
2455 * this inode's backref to point from the next inode.
2457 if (next_agino != NULLAGINO) {
2458 error = xfs_iunlink_change_backref(agibp->b_pag, next_agino,
2464 if (head_agino != agino) {
2465 struct xfs_imap imap;
2466 xfs_agino_t prev_agino;
2468 /* We need to search the list for the inode being freed. */
2469 error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2470 &prev_agino, &imap, &last_dip, &last_ibp,
2475 /* Point the previous inode on the list to the next inode. */
2476 xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2477 last_dip, &imap, next_agino);
2480 * Now we deal with the backref for this inode. If this inode
2481 * pointed at a real inode, change the backref that pointed to
2482 * us to point to our old next. If this inode was the end of
2483 * the list, delete the backref that pointed to us. Note that
2484 * change_backref takes care of deleting the backref if
2485 * next_agino is NULLAGINO.
2487 return xfs_iunlink_change_backref(agibp->b_pag, agino,
2491 /* Point the head of the list to the next unlinked inode. */
2492 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2497 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2498 * mark it stale. We should only find clean inodes in this lookup that aren't
2502 xfs_ifree_mark_inode_stale(
2504 struct xfs_inode *free_ip,
2507 struct xfs_mount *mp = bp->b_mount;
2508 struct xfs_perag *pag = bp->b_pag;
2509 struct xfs_inode_log_item *iip;
2510 struct xfs_inode *ip;
2514 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2516 /* Inode not in memory, nothing to do */
2523 * because this is an RCU protected lookup, we could find a recently
2524 * freed or even reallocated inode during the lookup. We need to check
2525 * under the i_flags_lock for a valid inode here. Skip it if it is not
2526 * valid, the wrong inode or stale.
2528 spin_lock(&ip->i_flags_lock);
2529 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2530 goto out_iflags_unlock;
2533 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2534 * other inodes that we did not find in the list attached to the buffer
2535 * and are not already marked stale. If we can't lock it, back off and
2538 if (ip != free_ip) {
2539 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2540 spin_unlock(&ip->i_flags_lock);
2546 ip->i_flags |= XFS_ISTALE;
2549 * If the inode is flushing, it is already attached to the buffer. All
2550 * we needed to do here is mark the inode stale so buffer IO completion
2551 * will remove it from the AIL.
2554 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2555 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2556 ASSERT(iip->ili_last_fields);
2561 * Inodes not attached to the buffer can be released immediately.
2562 * Everything else has to go through xfs_iflush_abort() on journal
2563 * commit as the flock synchronises removal of the inode from the
2564 * cluster buffer against inode reclaim.
2566 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2569 __xfs_iflags_set(ip, XFS_IFLUSHING);
2570 spin_unlock(&ip->i_flags_lock);
2573 /* we have a dirty inode in memory that has not yet been flushed. */
2574 spin_lock(&iip->ili_lock);
2575 iip->ili_last_fields = iip->ili_fields;
2576 iip->ili_fields = 0;
2577 iip->ili_fsync_fields = 0;
2578 spin_unlock(&iip->ili_lock);
2579 ASSERT(iip->ili_last_fields);
2582 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2587 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2589 spin_unlock(&ip->i_flags_lock);
2594 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2595 * inodes that are in memory - they all must be marked stale and attached to
2596 * the cluster buffer.
2600 struct xfs_inode *free_ip,
2601 struct xfs_trans *tp,
2602 struct xfs_icluster *xic)
2604 struct xfs_mount *mp = free_ip->i_mount;
2605 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2608 xfs_ino_t inum = xic->first_ino;
2614 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2616 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2618 * The allocation bitmap tells us which inodes of the chunk were
2619 * physically allocated. Skip the cluster if an inode falls into
2622 ioffset = inum - xic->first_ino;
2623 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2624 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2628 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2629 XFS_INO_TO_AGBNO(mp, inum));
2632 * We obtain and lock the backing buffer first in the process
2633 * here to ensure dirty inodes attached to the buffer remain in
2634 * the flushing state while we mark them stale.
2636 * If we scan the in-memory inodes first, then buffer IO can
2637 * complete before we get a lock on it, and hence we may fail
2638 * to mark all the active inodes on the buffer stale.
2640 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2641 mp->m_bsize * igeo->blocks_per_cluster,
2647 * This buffer may not have been correctly initialised as we
2648 * didn't read it from disk. That's not important because we are
2649 * only using to mark the buffer as stale in the log, and to
2650 * attach stale cached inodes on it. That means it will never be
2651 * dispatched for IO. If it is, we want to know about it, and we
2652 * want it to fail. We can acheive this by adding a write
2653 * verifier to the buffer.
2655 bp->b_ops = &xfs_inode_buf_ops;
2658 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2659 * too. This requires lookups, and will skip inodes that we've
2660 * already marked XFS_ISTALE.
2662 for (i = 0; i < igeo->inodes_per_cluster; i++)
2663 xfs_ifree_mark_inode_stale(bp, free_ip, inum + i);
2665 xfs_trans_stale_inode_buf(tp, bp);
2666 xfs_trans_binval(tp, bp);
2672 * This is called to return an inode to the inode free list. The inode should
2673 * already be truncated to 0 length and have no pages associated with it. This
2674 * routine also assumes that the inode is already a part of the transaction.
2676 * The on-disk copy of the inode will have been added to the list of unlinked
2677 * inodes in the AGI. We need to remove the inode from that list atomically with
2678 * respect to freeing it here.
2682 struct xfs_trans *tp,
2683 struct xfs_inode *ip)
2686 struct xfs_icluster xic = { 0 };
2687 struct xfs_inode_log_item *iip = ip->i_itemp;
2689 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2690 ASSERT(VFS_I(ip)->i_nlink == 0);
2691 ASSERT(ip->i_df.if_nextents == 0);
2692 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2693 ASSERT(ip->i_d.di_nblocks == 0);
2696 * Free the inode first so that we guarantee that the AGI lock is going
2697 * to be taken before we remove the inode from the unlinked list. This
2698 * makes the AGI lock -> unlinked list modification order the same as
2699 * used in O_TMPFILE creation.
2701 error = xfs_difree(tp, ip->i_ino, &xic);
2705 error = xfs_iunlink_remove(tp, ip);
2710 * Free any local-format data sitting around before we reset the
2711 * data fork to extents format. Note that the attr fork data has
2712 * already been freed by xfs_attr_inactive.
2714 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2715 kmem_free(ip->i_df.if_u1.if_data);
2716 ip->i_df.if_u1.if_data = NULL;
2717 ip->i_df.if_bytes = 0;
2720 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2721 ip->i_d.di_flags = 0;
2722 ip->i_d.di_flags2 = ip->i_mount->m_ino_geo.new_diflags2;
2723 ip->i_d.di_dmevmask = 0;
2724 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2725 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2727 /* Don't attempt to replay owner changes for a deleted inode */
2728 spin_lock(&iip->ili_lock);
2729 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2730 spin_unlock(&iip->ili_lock);
2733 * Bump the generation count so no one will be confused
2734 * by reincarnations of this inode.
2736 VFS_I(ip)->i_generation++;
2737 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2740 error = xfs_ifree_cluster(ip, tp, &xic);
2746 * This is called to unpin an inode. The caller must have the inode locked
2747 * in at least shared mode so that the buffer cannot be subsequently pinned
2748 * once someone is waiting for it to be unpinned.
2752 struct xfs_inode *ip)
2754 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2756 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2758 /* Give the log a push to start the unpinning I/O */
2759 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2765 struct xfs_inode *ip)
2767 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2768 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2773 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2774 if (xfs_ipincount(ip))
2776 } while (xfs_ipincount(ip));
2777 finish_wait(wq, &wait.wq_entry);
2782 struct xfs_inode *ip)
2784 if (xfs_ipincount(ip))
2785 __xfs_iunpin_wait(ip);
2789 * Removing an inode from the namespace involves removing the directory entry
2790 * and dropping the link count on the inode. Removing the directory entry can
2791 * result in locking an AGF (directory blocks were freed) and removing a link
2792 * count can result in placing the inode on an unlinked list which results in
2795 * The big problem here is that we have an ordering constraint on AGF and AGI
2796 * locking - inode allocation locks the AGI, then can allocate a new extent for
2797 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2798 * removes the inode from the unlinked list, requiring that we lock the AGI
2799 * first, and then freeing the inode can result in an inode chunk being freed
2800 * and hence freeing disk space requiring that we lock an AGF.
2802 * Hence the ordering that is imposed by other parts of the code is AGI before
2803 * AGF. This means we cannot remove the directory entry before we drop the inode
2804 * reference count and put it on the unlinked list as this results in a lock
2805 * order of AGF then AGI, and this can deadlock against inode allocation and
2806 * freeing. Therefore we must drop the link counts before we remove the
2809 * This is still safe from a transactional point of view - it is not until we
2810 * get to xfs_defer_finish() that we have the possibility of multiple
2811 * transactions in this operation. Hence as long as we remove the directory
2812 * entry and drop the link count in the first transaction of the remove
2813 * operation, there are no transactional constraints on the ordering here.
2818 struct xfs_name *name,
2821 xfs_mount_t *mp = dp->i_mount;
2822 xfs_trans_t *tp = NULL;
2823 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2827 trace_xfs_remove(dp, name);
2829 if (XFS_FORCED_SHUTDOWN(mp))
2832 error = xfs_qm_dqattach(dp);
2836 error = xfs_qm_dqattach(ip);
2841 * We try to get the real space reservation first,
2842 * allowing for directory btree deletion(s) implying
2843 * possible bmap insert(s). If we can't get the space
2844 * reservation then we use 0 instead, and avoid the bmap
2845 * btree insert(s) in the directory code by, if the bmap
2846 * insert tries to happen, instead trimming the LAST
2847 * block from the directory.
2849 resblks = XFS_REMOVE_SPACE_RES(mp);
2850 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2851 if (error == -ENOSPC) {
2853 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2857 ASSERT(error != -ENOSPC);
2861 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2863 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2864 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2867 * If we're removing a directory perform some additional validation.
2870 ASSERT(VFS_I(ip)->i_nlink >= 2);
2871 if (VFS_I(ip)->i_nlink != 2) {
2873 goto out_trans_cancel;
2875 if (!xfs_dir_isempty(ip)) {
2877 goto out_trans_cancel;
2880 /* Drop the link from ip's "..". */
2881 error = xfs_droplink(tp, dp);
2883 goto out_trans_cancel;
2885 /* Drop the "." link from ip to self. */
2886 error = xfs_droplink(tp, ip);
2888 goto out_trans_cancel;
2891 * When removing a non-directory we need to log the parent
2892 * inode here. For a directory this is done implicitly
2893 * by the xfs_droplink call for the ".." entry.
2895 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2897 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2899 /* Drop the link from dp to ip. */
2900 error = xfs_droplink(tp, ip);
2902 goto out_trans_cancel;
2904 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2906 ASSERT(error != -ENOENT);
2907 goto out_trans_cancel;
2911 * If this is a synchronous mount, make sure that the
2912 * remove transaction goes to disk before returning to
2915 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2916 xfs_trans_set_sync(tp);
2918 error = xfs_trans_commit(tp);
2922 if (is_dir && xfs_inode_is_filestream(ip))
2923 xfs_filestream_deassociate(ip);
2928 xfs_trans_cancel(tp);
2934 * Enter all inodes for a rename transaction into a sorted array.
2936 #define __XFS_SORT_INODES 5
2938 xfs_sort_for_rename(
2939 struct xfs_inode *dp1, /* in: old (source) directory inode */
2940 struct xfs_inode *dp2, /* in: new (target) directory inode */
2941 struct xfs_inode *ip1, /* in: inode of old entry */
2942 struct xfs_inode *ip2, /* in: inode of new entry */
2943 struct xfs_inode *wip, /* in: whiteout inode */
2944 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2945 int *num_inodes) /* in/out: inodes in array */
2949 ASSERT(*num_inodes == __XFS_SORT_INODES);
2950 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2953 * i_tab contains a list of pointers to inodes. We initialize
2954 * the table here & we'll sort it. We will then use it to
2955 * order the acquisition of the inode locks.
2957 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2970 * Sort the elements via bubble sort. (Remember, there are at
2971 * most 5 elements to sort, so this is adequate.)
2973 for (i = 0; i < *num_inodes; i++) {
2974 for (j = 1; j < *num_inodes; j++) {
2975 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2976 struct xfs_inode *temp = i_tab[j];
2977 i_tab[j] = i_tab[j-1];
2986 struct xfs_trans *tp)
2989 * If this is a synchronous mount, make sure that the rename transaction
2990 * goes to disk before returning to the user.
2992 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2993 xfs_trans_set_sync(tp);
2995 return xfs_trans_commit(tp);
2999 * xfs_cross_rename()
3001 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
3005 struct xfs_trans *tp,
3006 struct xfs_inode *dp1,
3007 struct xfs_name *name1,
3008 struct xfs_inode *ip1,
3009 struct xfs_inode *dp2,
3010 struct xfs_name *name2,
3011 struct xfs_inode *ip2,
3019 /* Swap inode number for dirent in first parent */
3020 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
3022 goto out_trans_abort;
3024 /* Swap inode number for dirent in second parent */
3025 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
3027 goto out_trans_abort;
3030 * If we're renaming one or more directories across different parents,
3031 * update the respective ".." entries (and link counts) to match the new
3035 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3037 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
3038 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
3039 dp1->i_ino, spaceres);
3041 goto out_trans_abort;
3043 /* transfer ip2 ".." reference to dp1 */
3044 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
3045 error = xfs_droplink(tp, dp2);
3047 goto out_trans_abort;
3048 xfs_bumplink(tp, dp1);
3052 * Although ip1 isn't changed here, userspace needs
3053 * to be warned about the change, so that applications
3054 * relying on it (like backup ones), will properly
3057 ip1_flags |= XFS_ICHGTIME_CHG;
3058 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3061 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
3062 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
3063 dp2->i_ino, spaceres);
3065 goto out_trans_abort;
3067 /* transfer ip1 ".." reference to dp2 */
3068 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
3069 error = xfs_droplink(tp, dp1);
3071 goto out_trans_abort;
3072 xfs_bumplink(tp, dp2);
3076 * Although ip2 isn't changed here, userspace needs
3077 * to be warned about the change, so that applications
3078 * relying on it (like backup ones), will properly
3081 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3082 ip2_flags |= XFS_ICHGTIME_CHG;
3087 xfs_trans_ichgtime(tp, ip1, ip1_flags);
3088 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
3091 xfs_trans_ichgtime(tp, ip2, ip2_flags);
3092 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
3095 xfs_trans_ichgtime(tp, dp2, dp2_flags);
3096 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
3098 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3099 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3100 return xfs_finish_rename(tp);
3103 xfs_trans_cancel(tp);
3108 * xfs_rename_alloc_whiteout()
3110 * Return a referenced, unlinked, unlocked inode that can be used as a
3111 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3112 * crash between allocating the inode and linking it into the rename transaction
3113 * recovery will free the inode and we won't leak it.
3116 xfs_rename_alloc_whiteout(
3117 struct xfs_inode *dp,
3118 struct xfs_inode **wip)
3120 struct xfs_inode *tmpfile;
3123 error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
3128 * Prepare the tmpfile inode as if it were created through the VFS.
3129 * Complete the inode setup and flag it as linkable. nlink is already
3130 * zero, so we can skip the drop_nlink.
3132 xfs_setup_iops(tmpfile);
3133 xfs_finish_inode_setup(tmpfile);
3134 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3145 struct xfs_inode *src_dp,
3146 struct xfs_name *src_name,
3147 struct xfs_inode *src_ip,
3148 struct xfs_inode *target_dp,
3149 struct xfs_name *target_name,
3150 struct xfs_inode *target_ip,
3153 struct xfs_mount *mp = src_dp->i_mount;
3154 struct xfs_trans *tp;
3155 struct xfs_inode *wip = NULL; /* whiteout inode */
3156 struct xfs_inode *inodes[__XFS_SORT_INODES];
3158 int num_inodes = __XFS_SORT_INODES;
3159 bool new_parent = (src_dp != target_dp);
3160 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3164 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3166 if ((flags & RENAME_EXCHANGE) && !target_ip)
3170 * If we are doing a whiteout operation, allocate the whiteout inode
3171 * we will be placing at the target and ensure the type is set
3174 if (flags & RENAME_WHITEOUT) {
3175 error = xfs_rename_alloc_whiteout(target_dp, &wip);
3179 /* setup target dirent info as whiteout */
3180 src_name->type = XFS_DIR3_FT_CHRDEV;
3183 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3184 inodes, &num_inodes);
3186 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3187 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3188 if (error == -ENOSPC) {
3190 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3194 goto out_release_wip;
3197 * Attach the dquots to the inodes
3199 error = xfs_qm_vop_rename_dqattach(inodes);
3201 goto out_trans_cancel;
3204 * Lock all the participating inodes. Depending upon whether
3205 * the target_name exists in the target directory, and
3206 * whether the target directory is the same as the source
3207 * directory, we can lock from 2 to 4 inodes.
3209 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3212 * Join all the inodes to the transaction. From this point on,
3213 * we can rely on either trans_commit or trans_cancel to unlock
3216 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3218 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3219 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3221 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3223 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3226 * If we are using project inheritance, we only allow renames
3227 * into our tree when the project IDs are the same; else the
3228 * tree quota mechanism would be circumvented.
3230 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3231 target_dp->i_d.di_projid != src_ip->i_d.di_projid)) {
3233 goto out_trans_cancel;
3236 /* RENAME_EXCHANGE is unique from here on. */
3237 if (flags & RENAME_EXCHANGE)
3238 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3239 target_dp, target_name, target_ip,
3243 * Check for expected errors before we dirty the transaction
3244 * so we can return an error without a transaction abort.
3246 if (target_ip == NULL) {
3248 * If there's no space reservation, check the entry will
3249 * fit before actually inserting it.
3252 error = xfs_dir_canenter(tp, target_dp, target_name);
3254 goto out_trans_cancel;
3258 * If target exists and it's a directory, check that whether
3259 * it can be destroyed.
3261 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3262 (!xfs_dir_isempty(target_ip) ||
3263 (VFS_I(target_ip)->i_nlink > 2))) {
3265 goto out_trans_cancel;
3270 * Lock the AGI buffers we need to handle bumping the nlink of the
3271 * whiteout inode off the unlinked list and to handle dropping the
3272 * nlink of the target inode. Per locking order rules, do this in
3273 * increasing AG order and before directory block allocation tries to
3274 * grab AGFs because we grab AGIs before AGFs.
3276 * The (vfs) caller must ensure that if src is a directory then
3277 * target_ip is either null or an empty directory.
3279 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3280 if (inodes[i] == wip ||
3281 (inodes[i] == target_ip &&
3282 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3284 xfs_agnumber_t agno;
3286 agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
3287 error = xfs_read_agi(mp, tp, agno, &bp);
3289 goto out_trans_cancel;
3294 * Directory entry creation below may acquire the AGF. Remove
3295 * the whiteout from the unlinked list first to preserve correct
3296 * AGI/AGF locking order. This dirties the transaction so failures
3297 * after this point will abort and log recovery will clean up the
3300 * For whiteouts, we need to bump the link count on the whiteout
3301 * inode. After this point, we have a real link, clear the tmpfile
3302 * state flag from the inode so it doesn't accidentally get misused
3306 ASSERT(VFS_I(wip)->i_nlink == 0);
3307 error = xfs_iunlink_remove(tp, wip);
3309 goto out_trans_cancel;
3311 xfs_bumplink(tp, wip);
3312 VFS_I(wip)->i_state &= ~I_LINKABLE;
3316 * Set up the target.
3318 if (target_ip == NULL) {
3320 * If target does not exist and the rename crosses
3321 * directories, adjust the target directory link count
3322 * to account for the ".." reference from the new entry.
3324 error = xfs_dir_createname(tp, target_dp, target_name,
3325 src_ip->i_ino, spaceres);
3327 goto out_trans_cancel;
3329 xfs_trans_ichgtime(tp, target_dp,
3330 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3332 if (new_parent && src_is_directory) {
3333 xfs_bumplink(tp, target_dp);
3335 } else { /* target_ip != NULL */
3337 * Link the source inode under the target name.
3338 * If the source inode is a directory and we are moving
3339 * it across directories, its ".." entry will be
3340 * inconsistent until we replace that down below.
3342 * In case there is already an entry with the same
3343 * name at the destination directory, remove it first.
3345 error = xfs_dir_replace(tp, target_dp, target_name,
3346 src_ip->i_ino, spaceres);
3348 goto out_trans_cancel;
3350 xfs_trans_ichgtime(tp, target_dp,
3351 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3354 * Decrement the link count on the target since the target
3355 * dir no longer points to it.
3357 error = xfs_droplink(tp, target_ip);
3359 goto out_trans_cancel;
3361 if (src_is_directory) {
3363 * Drop the link from the old "." entry.
3365 error = xfs_droplink(tp, target_ip);
3367 goto out_trans_cancel;
3369 } /* target_ip != NULL */
3372 * Remove the source.
3374 if (new_parent && src_is_directory) {
3376 * Rewrite the ".." entry to point to the new
3379 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3380 target_dp->i_ino, spaceres);
3381 ASSERT(error != -EEXIST);
3383 goto out_trans_cancel;
3387 * We always want to hit the ctime on the source inode.
3389 * This isn't strictly required by the standards since the source
3390 * inode isn't really being changed, but old unix file systems did
3391 * it and some incremental backup programs won't work without it.
3393 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3394 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3397 * Adjust the link count on src_dp. This is necessary when
3398 * renaming a directory, either within one parent when
3399 * the target existed, or across two parent directories.
3401 if (src_is_directory && (new_parent || target_ip != NULL)) {
3404 * Decrement link count on src_directory since the
3405 * entry that's moved no longer points to it.
3407 error = xfs_droplink(tp, src_dp);
3409 goto out_trans_cancel;
3413 * For whiteouts, we only need to update the source dirent with the
3414 * inode number of the whiteout inode rather than removing it
3418 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3421 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3424 goto out_trans_cancel;
3426 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3427 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3429 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3431 error = xfs_finish_rename(tp);
3437 xfs_trans_cancel(tp);
3446 struct xfs_inode *ip,
3449 struct xfs_inode_log_item *iip = ip->i_itemp;
3450 struct xfs_dinode *dip;
3451 struct xfs_mount *mp = ip->i_mount;
3454 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3455 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3456 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3457 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3458 ASSERT(iip->ili_item.li_buf == bp);
3460 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3463 * We don't flush the inode if any of the following checks fail, but we
3464 * do still update the log item and attach to the backing buffer as if
3465 * the flush happened. This is a formality to facilitate predictable
3466 * error handling as the caller will shutdown and fail the buffer.
3468 error = -EFSCORRUPTED;
3469 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3470 mp, XFS_ERRTAG_IFLUSH_1)) {
3471 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3472 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3473 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3476 if (S_ISREG(VFS_I(ip)->i_mode)) {
3478 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3479 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3480 mp, XFS_ERRTAG_IFLUSH_3)) {
3481 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3482 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3483 __func__, ip->i_ino, ip);
3486 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3488 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3489 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3490 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3491 mp, XFS_ERRTAG_IFLUSH_4)) {
3492 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3493 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3494 __func__, ip->i_ino, ip);
3498 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3499 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3500 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3501 "%s: detected corrupt incore inode %Lu, "
3502 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3503 __func__, ip->i_ino,
3504 ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3505 ip->i_d.di_nblocks, ip);
3508 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3509 mp, XFS_ERRTAG_IFLUSH_6)) {
3510 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3511 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3512 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3517 * Inode item log recovery for v2 inodes are dependent on the
3518 * di_flushiter count for correct sequencing. We bump the flush
3519 * iteration count so we can detect flushes which postdate a log record
3520 * during recovery. This is redundant as we now log every change and
3521 * hence this can't happen but we need to still do it to ensure
3522 * backwards compatibility with old kernels that predate logging all
3525 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3526 ip->i_d.di_flushiter++;
3529 * If there are inline format data / attr forks attached to this inode,
3530 * make sure they are not corrupt.
3532 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3533 xfs_ifork_verify_local_data(ip))
3535 if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3536 xfs_ifork_verify_local_attr(ip))
3540 * Copy the dirty parts of the inode into the on-disk inode. We always
3541 * copy out the core of the inode, because if the inode is dirty at all
3544 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3546 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3547 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3548 ip->i_d.di_flushiter = 0;
3550 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3551 if (XFS_IFORK_Q(ip))
3552 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3555 * We've recorded everything logged in the inode, so we'd like to clear
3556 * the ili_fields bits so we don't log and flush things unnecessarily.
3557 * However, we can't stop logging all this information until the data
3558 * we've copied into the disk buffer is written to disk. If we did we
3559 * might overwrite the copy of the inode in the log with all the data
3560 * after re-logging only part of it, and in the face of a crash we
3561 * wouldn't have all the data we need to recover.
3563 * What we do is move the bits to the ili_last_fields field. When
3564 * logging the inode, these bits are moved back to the ili_fields field.
3565 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3566 * we know that the information those bits represent is permanently on
3567 * disk. As long as the flush completes before the inode is logged
3568 * again, then both ili_fields and ili_last_fields will be cleared.
3572 spin_lock(&iip->ili_lock);
3573 iip->ili_last_fields = iip->ili_fields;
3574 iip->ili_fields = 0;
3575 iip->ili_fsync_fields = 0;
3576 spin_unlock(&iip->ili_lock);
3579 * Store the current LSN of the inode so that we can tell whether the
3580 * item has moved in the AIL from xfs_buf_inode_iodone().
3582 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3583 &iip->ili_item.li_lsn);
3585 /* generate the checksum. */
3586 xfs_dinode_calc_crc(mp, dip);
3591 * Non-blocking flush of dirty inode metadata into the backing buffer.
3593 * The caller must have a reference to the inode and hold the cluster buffer
3594 * locked. The function will walk across all the inodes on the cluster buffer it
3595 * can find and lock without blocking, and flush them to the cluster buffer.
3597 * On successful flushing of at least one inode, the caller must write out the
3598 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3599 * the caller needs to release the buffer. On failure, the filesystem will be
3600 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3607 struct xfs_mount *mp = bp->b_mount;
3608 struct xfs_log_item *lip, *n;
3609 struct xfs_inode *ip;
3610 struct xfs_inode_log_item *iip;
3615 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3616 * can remove itself from the list.
3618 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3619 iip = (struct xfs_inode_log_item *)lip;
3620 ip = iip->ili_inode;
3623 * Quick and dirty check to avoid locks if possible.
3625 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3627 if (xfs_ipincount(ip))
3631 * The inode is still attached to the buffer, which means it is
3632 * dirty but reclaim might try to grab it. Check carefully for
3633 * that, and grab the ilock while still holding the i_flags_lock
3634 * to guarantee reclaim will not be able to reclaim this inode
3635 * once we drop the i_flags_lock.
3637 spin_lock(&ip->i_flags_lock);
3638 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3639 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3640 spin_unlock(&ip->i_flags_lock);
3645 * ILOCK will pin the inode against reclaim and prevent
3646 * concurrent transactions modifying the inode while we are
3647 * flushing the inode. If we get the lock, set the flushing
3648 * state before we drop the i_flags_lock.
3650 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3651 spin_unlock(&ip->i_flags_lock);
3654 __xfs_iflags_set(ip, XFS_IFLUSHING);
3655 spin_unlock(&ip->i_flags_lock);
3658 * Abort flushing this inode if we are shut down because the
3659 * inode may not currently be in the AIL. This can occur when
3660 * log I/O failure unpins the inode without inserting into the
3661 * AIL, leaving a dirty/unpinned inode attached to the buffer
3662 * that otherwise looks like it should be flushed.
3664 if (XFS_FORCED_SHUTDOWN(mp)) {
3665 xfs_iunpin_wait(ip);
3666 xfs_iflush_abort(ip);
3667 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3672 /* don't block waiting on a log force to unpin dirty inodes */
3673 if (xfs_ipincount(ip)) {
3674 xfs_iflags_clear(ip, XFS_IFLUSHING);
3675 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3679 if (!xfs_inode_clean(ip))
3680 error = xfs_iflush(ip, bp);
3682 xfs_iflags_clear(ip, XFS_IFLUSHING);
3683 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3690 bp->b_flags |= XBF_ASYNC;
3691 xfs_buf_ioend_fail(bp);
3692 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3699 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3700 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3705 /* Release an inode. */
3708 struct xfs_inode *ip)
3710 trace_xfs_irele(ip, _RET_IP_);
3715 * Ensure all commited transactions touching the inode are written to the log.
3718 xfs_log_force_inode(
3719 struct xfs_inode *ip)
3723 xfs_ilock(ip, XFS_ILOCK_SHARED);
3724 if (xfs_ipincount(ip))
3725 seq = ip->i_itemp->ili_commit_seq;
3726 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3730 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3734 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3735 * abide vfs locking order (lowest pointer value goes first) and breaking the
3736 * layout leases before proceeding. The loop is needed because we cannot call
3737 * the blocking break_layout() with the iolocks held, and therefore have to
3738 * back out both locks.
3741 xfs_iolock_two_inodes_and_break_layout(
3751 /* Wait to break both inodes' layouts before we start locking. */
3752 error = break_layout(src, true);
3756 error = break_layout(dest, true);
3761 /* Lock one inode and make sure nobody got in and leased it. */
3763 error = break_layout(src, false);
3766 if (error == -EWOULDBLOCK)
3774 /* Lock the other inode and make sure nobody got in and leased it. */
3775 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3776 error = break_layout(dest, false);
3780 if (error == -EWOULDBLOCK)
3789 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3794 struct xfs_inode *ip1,
3795 struct xfs_inode *ip2)
3799 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3803 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3805 xfs_lock_two_inodes(ip1, XFS_MMAPLOCK_EXCL,
3806 ip2, XFS_MMAPLOCK_EXCL);
3810 /* Unlock both inodes to allow IO and mmap activity. */
3812 xfs_iunlock2_io_mmap(
3813 struct xfs_inode *ip1,
3814 struct xfs_inode *ip2)
3816 bool same_inode = (ip1 == ip2);
3818 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3820 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3821 inode_unlock(VFS_I(ip2));
3823 inode_unlock(VFS_I(ip1));