1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
6 #include <linux/iversion.h>
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
19 #include "xfs_trans_space.h"
20 #include "xfs_trans.h"
21 #include "xfs_buf_item.h"
22 #include "xfs_inode_item.h"
23 #include "xfs_ialloc.h"
25 #include "xfs_bmap_util.h"
26 #include "xfs_errortag.h"
27 #include "xfs_error.h"
28 #include "xfs_quota.h"
29 #include "xfs_filestream.h"
30 #include "xfs_trace.h"
31 #include "xfs_icache.h"
32 #include "xfs_symlink.h"
33 #include "xfs_trans_priv.h"
35 #include "xfs_bmap_btree.h"
36 #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 *tp, struct xfs_perag *pag,
52 * helper function to extract extent size hint from inode
59 * No point in aligning allocations if we need to COW to actually
62 if (xfs_is_always_cow_inode(ip))
64 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
66 if (XFS_IS_REALTIME_INODE(ip))
67 return ip->i_mount->m_sb.sb_rextsize;
72 * Helper function to extract CoW extent size hint from inode.
73 * Between the extent size hint and the CoW extent size hint, we
74 * return the greater of the two. If the value is zero (automatic),
75 * use the default size.
78 xfs_get_cowextsz_hint(
84 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
86 b = xfs_get_extsz_hint(ip);
90 return XFS_DEFAULT_COWEXTSZ_HINT;
95 * These two are wrapper routines around the xfs_ilock() routine used to
96 * centralize some grungy code. They are used in places that wish to lock the
97 * inode solely for reading the extents. The reason these places can't just
98 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
99 * bringing in of the extents from disk for a file in b-tree format. If the
100 * inode is in b-tree format, then we need to lock the inode exclusively until
101 * the extents are read in. Locking it exclusively all the time would limit
102 * our parallelism unnecessarily, though. What we do instead is check to see
103 * if the extents have been read in yet, and only lock the inode exclusively
106 * The functions return a value which should be given to the corresponding
107 * xfs_iunlock() call.
110 xfs_ilock_data_map_shared(
111 struct xfs_inode *ip)
113 uint lock_mode = XFS_ILOCK_SHARED;
115 if (xfs_need_iread_extents(&ip->i_df))
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;
127 if (ip->i_afp && xfs_need_iread_extents(ip->i_afp))
128 lock_mode = XFS_ILOCK_EXCL;
129 xfs_ilock(ip, lock_mode);
134 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
135 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
136 * various combinations of the locks to be obtained.
138 * The 3 locks should always be ordered so that the IO lock is obtained first,
139 * the mmap lock second and the ilock last in order to prevent deadlock.
141 * Basic locking order:
143 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
145 * mmap_lock locking order:
147 * i_rwsem -> page lock -> mmap_lock
148 * mmap_lock -> invalidate_lock -> page_lock
150 * The difference in mmap_lock locking order mean that we cannot hold the
151 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
152 * can fault in pages during copy in/out (for buffered IO) or require the
153 * mmap_lock in get_user_pages() to map the user pages into the kernel address
154 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
155 * fault because page faults already hold the mmap_lock.
157 * Hence to serialise fully against both syscall and mmap based IO, we need to
158 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
159 * both taken in places where we need to invalidate the page cache in a race
160 * free manner (e.g. truncate, hole punch and other extent manipulation
168 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
171 * You can't set both SHARED and EXCL for the same lock,
172 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
173 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
175 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
176 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
177 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
178 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
179 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
180 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
181 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
183 if (lock_flags & XFS_IOLOCK_EXCL) {
184 down_write_nested(&VFS_I(ip)->i_rwsem,
185 XFS_IOLOCK_DEP(lock_flags));
186 } else if (lock_flags & XFS_IOLOCK_SHARED) {
187 down_read_nested(&VFS_I(ip)->i_rwsem,
188 XFS_IOLOCK_DEP(lock_flags));
191 if (lock_flags & XFS_MMAPLOCK_EXCL) {
192 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
193 XFS_MMAPLOCK_DEP(lock_flags));
194 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
195 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
196 XFS_MMAPLOCK_DEP(lock_flags));
199 if (lock_flags & XFS_ILOCK_EXCL)
200 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
201 else if (lock_flags & XFS_ILOCK_SHARED)
202 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
206 * This is just like xfs_ilock(), except that the caller
207 * is guaranteed not to sleep. It returns 1 if it gets
208 * the requested locks and 0 otherwise. If the IO lock is
209 * obtained but the inode lock cannot be, then the IO lock
210 * is dropped before returning.
212 * ip -- the inode being locked
213 * lock_flags -- this parameter indicates the inode's locks to be
214 * to be locked. See the comment for xfs_ilock() for a list
222 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
225 * You can't set both SHARED and EXCL for the same lock,
226 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
227 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
229 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
230 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
231 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
232 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
233 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
234 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
235 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
237 if (lock_flags & XFS_IOLOCK_EXCL) {
238 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
240 } else if (lock_flags & XFS_IOLOCK_SHARED) {
241 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
245 if (lock_flags & XFS_MMAPLOCK_EXCL) {
246 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
247 goto out_undo_iolock;
248 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
249 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
250 goto out_undo_iolock;
253 if (lock_flags & XFS_ILOCK_EXCL) {
254 if (!mrtryupdate(&ip->i_lock))
255 goto out_undo_mmaplock;
256 } else if (lock_flags & XFS_ILOCK_SHARED) {
257 if (!mrtryaccess(&ip->i_lock))
258 goto out_undo_mmaplock;
263 if (lock_flags & XFS_MMAPLOCK_EXCL)
264 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
265 else if (lock_flags & XFS_MMAPLOCK_SHARED)
266 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
268 if (lock_flags & XFS_IOLOCK_EXCL)
269 up_write(&VFS_I(ip)->i_rwsem);
270 else if (lock_flags & XFS_IOLOCK_SHARED)
271 up_read(&VFS_I(ip)->i_rwsem);
277 * xfs_iunlock() is used to drop the inode locks acquired with
278 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
279 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
280 * that we know which locks to drop.
282 * ip -- the inode being unlocked
283 * lock_flags -- this parameter indicates the inode's locks to be
284 * to be unlocked. See the comment for xfs_ilock() for a list
285 * of valid values for this parameter.
294 * You can't set both SHARED and EXCL for the same lock,
295 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
296 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
298 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
299 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
300 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
301 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
302 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
303 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
304 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
305 ASSERT(lock_flags != 0);
307 if (lock_flags & XFS_IOLOCK_EXCL)
308 up_write(&VFS_I(ip)->i_rwsem);
309 else if (lock_flags & XFS_IOLOCK_SHARED)
310 up_read(&VFS_I(ip)->i_rwsem);
312 if (lock_flags & XFS_MMAPLOCK_EXCL)
313 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
314 else if (lock_flags & XFS_MMAPLOCK_SHARED)
315 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
317 if (lock_flags & XFS_ILOCK_EXCL)
318 mrunlock_excl(&ip->i_lock);
319 else if (lock_flags & XFS_ILOCK_SHARED)
320 mrunlock_shared(&ip->i_lock);
322 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
326 * give up write locks. the i/o lock cannot be held nested
327 * if it is being demoted.
334 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
336 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
338 if (lock_flags & XFS_ILOCK_EXCL)
339 mrdemote(&ip->i_lock);
340 if (lock_flags & XFS_MMAPLOCK_EXCL)
341 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
342 if (lock_flags & XFS_IOLOCK_EXCL)
343 downgrade_write(&VFS_I(ip)->i_rwsem);
345 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
348 #if defined(DEBUG) || defined(XFS_WARN)
350 __xfs_rwsem_islocked(
351 struct rw_semaphore *rwsem,
355 return rwsem_is_locked(rwsem);
358 return lockdep_is_held_type(rwsem, 0);
361 * We are checking that the lock is held at least in shared
362 * mode but don't care that it might be held exclusively
363 * (i.e. shared | excl). Hence we check if the lock is held
364 * in any mode rather than an explicit shared mode.
366 return lockdep_is_held_type(rwsem, -1);
371 struct xfs_inode *ip,
374 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
375 if (!(lock_flags & XFS_ILOCK_SHARED))
376 return !!ip->i_lock.mr_writer;
377 return rwsem_is_locked(&ip->i_lock.mr_lock);
380 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
381 return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
382 (lock_flags & XFS_IOLOCK_SHARED));
385 if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) {
386 return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
387 (lock_flags & XFS_IOLOCK_SHARED));
396 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
397 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
398 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
399 * errors and warnings.
401 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
403 xfs_lockdep_subclass_ok(
406 return subclass < MAX_LOCKDEP_SUBCLASSES;
409 #define xfs_lockdep_subclass_ok(subclass) (true)
413 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
414 * value. This can be called for any type of inode lock combination, including
415 * parent locking. Care must be taken to ensure we don't overrun the subclass
416 * storage fields in the class mask we build.
419 xfs_lock_inumorder(int lock_mode, int subclass)
423 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
425 ASSERT(xfs_lockdep_subclass_ok(subclass));
427 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
428 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
429 class += subclass << XFS_IOLOCK_SHIFT;
432 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
433 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
434 class += subclass << XFS_MMAPLOCK_SHIFT;
437 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
438 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
439 class += subclass << XFS_ILOCK_SHIFT;
442 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
446 * The following routine will lock n inodes in exclusive mode. We assume the
447 * caller calls us with the inodes in i_ino order.
449 * We need to detect deadlock where an inode that we lock is in the AIL and we
450 * start waiting for another inode that is locked by a thread in a long running
451 * transaction (such as truncate). This can result in deadlock since the long
452 * running trans might need to wait for the inode we just locked in order to
453 * push the tail and free space in the log.
455 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
456 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
457 * lock more than one at a time, lockdep will report false positives saying we
458 * have violated locking orders.
462 struct xfs_inode **ips,
466 int attempts = 0, i, j, try_lock;
467 struct xfs_log_item *lp;
470 * Currently supports between 2 and 5 inodes with exclusive locking. We
471 * support an arbitrary depth of locking here, but absolute limits on
472 * inodes depend on the type of locking and the limits placed by
473 * lockdep annotations in xfs_lock_inumorder. These are all checked by
476 ASSERT(ips && inodes >= 2 && inodes <= 5);
477 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
479 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
481 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
482 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
483 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
484 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
486 if (lock_mode & XFS_IOLOCK_EXCL) {
487 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
488 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
489 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
494 for (; i < inodes; i++) {
497 if (i && (ips[i] == ips[i - 1])) /* Already locked */
501 * If try_lock is not set yet, make sure all locked inodes are
502 * not in the AIL. If any are, set try_lock to be used later.
505 for (j = (i - 1); j >= 0 && !try_lock; j--) {
506 lp = &ips[j]->i_itemp->ili_item;
507 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
513 * If any of the previous locks we have locked is in the AIL,
514 * we must TRY to get the second and subsequent locks. If
515 * we can't get any, we must release all we have
519 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
523 /* try_lock means we have an inode locked that is in the AIL. */
525 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
529 * Unlock all previous guys and try again. xfs_iunlock will try
530 * to push the tail if the inode is in the AIL.
533 for (j = i - 1; j >= 0; j--) {
535 * Check to see if we've already unlocked this one. Not
536 * the first one going back, and the inode ptr is the
539 if (j != (i - 1) && ips[j] == ips[j + 1])
542 xfs_iunlock(ips[j], lock_mode);
545 if ((attempts % 5) == 0) {
546 delay(1); /* Don't just spin the CPU */
555 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
556 * mmaplock must be double-locked separately since we use i_rwsem and
557 * invalidate_lock for that. We now support taking one lock EXCL and the
562 struct xfs_inode *ip0,
564 struct xfs_inode *ip1,
567 struct xfs_inode *temp;
570 struct xfs_log_item *lp;
572 ASSERT(hweight32(ip0_mode) == 1);
573 ASSERT(hweight32(ip1_mode) == 1);
574 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
575 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
576 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
577 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
578 ASSERT(ip0->i_ino != ip1->i_ino);
580 if (ip0->i_ino > ip1->i_ino) {
584 mode_temp = ip0_mode;
586 ip1_mode = mode_temp;
590 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
593 * If the first lock we have locked is in the AIL, we must TRY to get
594 * the second lock. If we can't get it, we must release the first one
597 lp = &ip0->i_itemp->ili_item;
598 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
599 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
600 xfs_iunlock(ip0, ip0_mode);
601 if ((++attempts % 5) == 0)
602 delay(1); /* Don't just spin the CPU */
606 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
612 struct xfs_inode *ip)
616 if (ip->i_diflags & XFS_DIFLAG_ANY) {
617 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
618 flags |= FS_XFLAG_REALTIME;
619 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
620 flags |= FS_XFLAG_PREALLOC;
621 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
622 flags |= FS_XFLAG_IMMUTABLE;
623 if (ip->i_diflags & XFS_DIFLAG_APPEND)
624 flags |= FS_XFLAG_APPEND;
625 if (ip->i_diflags & XFS_DIFLAG_SYNC)
626 flags |= FS_XFLAG_SYNC;
627 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
628 flags |= FS_XFLAG_NOATIME;
629 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
630 flags |= FS_XFLAG_NODUMP;
631 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
632 flags |= FS_XFLAG_RTINHERIT;
633 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
634 flags |= FS_XFLAG_PROJINHERIT;
635 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
636 flags |= FS_XFLAG_NOSYMLINKS;
637 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
638 flags |= FS_XFLAG_EXTSIZE;
639 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
640 flags |= FS_XFLAG_EXTSZINHERIT;
641 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
642 flags |= FS_XFLAG_NODEFRAG;
643 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
644 flags |= FS_XFLAG_FILESTREAM;
647 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
648 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
649 flags |= FS_XFLAG_DAX;
650 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
651 flags |= FS_XFLAG_COWEXTSIZE;
655 flags |= FS_XFLAG_HASATTR;
660 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
661 * is allowed, otherwise it has to be an exact match. If a CI match is found,
662 * ci_name->name will point to a the actual name (caller must free) or
663 * will be set to NULL if an exact match is found.
668 struct xfs_name *name,
670 struct xfs_name *ci_name)
675 trace_xfs_lookup(dp, name);
677 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
680 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
684 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
692 kmem_free(ci_name->name);
698 /* Propagate di_flags from a parent inode to a child inode. */
700 xfs_inode_inherit_flags(
701 struct xfs_inode *ip,
702 const struct xfs_inode *pip)
704 unsigned int di_flags = 0;
705 xfs_failaddr_t failaddr;
706 umode_t mode = VFS_I(ip)->i_mode;
709 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
710 di_flags |= XFS_DIFLAG_RTINHERIT;
711 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
712 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
713 ip->i_extsize = pip->i_extsize;
715 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
716 di_flags |= XFS_DIFLAG_PROJINHERIT;
717 } else if (S_ISREG(mode)) {
718 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
719 xfs_sb_version_hasrealtime(&ip->i_mount->m_sb))
720 di_flags |= XFS_DIFLAG_REALTIME;
721 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
722 di_flags |= XFS_DIFLAG_EXTSIZE;
723 ip->i_extsize = pip->i_extsize;
726 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
728 di_flags |= XFS_DIFLAG_NOATIME;
729 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
731 di_flags |= XFS_DIFLAG_NODUMP;
732 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
734 di_flags |= XFS_DIFLAG_SYNC;
735 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
736 xfs_inherit_nosymlinks)
737 di_flags |= XFS_DIFLAG_NOSYMLINKS;
738 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
739 xfs_inherit_nodefrag)
740 di_flags |= XFS_DIFLAG_NODEFRAG;
741 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
742 di_flags |= XFS_DIFLAG_FILESTREAM;
744 ip->i_diflags |= di_flags;
747 * Inode verifiers on older kernels only check that the extent size
748 * hint is an integer multiple of the rt extent size on realtime files.
749 * They did not check the hint alignment on a directory with both
750 * rtinherit and extszinherit flags set. If the misaligned hint is
751 * propagated from a directory into a new realtime file, new file
752 * allocations will fail due to math errors in the rt allocator and/or
753 * trip the verifiers. Validate the hint settings in the new file so
754 * that we don't let broken hints propagate.
756 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
757 VFS_I(ip)->i_mode, ip->i_diflags);
759 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
760 XFS_DIFLAG_EXTSZINHERIT);
765 /* Propagate di_flags2 from a parent inode to a child inode. */
767 xfs_inode_inherit_flags2(
768 struct xfs_inode *ip,
769 const struct xfs_inode *pip)
771 xfs_failaddr_t failaddr;
773 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
774 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
775 ip->i_cowextsize = pip->i_cowextsize;
777 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
778 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
780 /* Don't let invalid cowextsize hints propagate. */
781 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
782 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
784 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
785 ip->i_cowextsize = 0;
790 * Initialise a newly allocated inode and return the in-core inode to the
791 * caller locked exclusively.
795 struct user_namespace *mnt_userns,
796 struct xfs_trans *tp,
797 struct xfs_inode *pip,
804 struct xfs_inode **ipp)
806 struct inode *dir = pip ? VFS_I(pip) : NULL;
807 struct xfs_mount *mp = tp->t_mountp;
808 struct xfs_inode *ip;
811 struct timespec64 tv;
815 * Protect against obviously corrupt allocation btree records. Later
816 * xfs_iget checks will catch re-allocation of other active in-memory
817 * and on-disk inodes. If we don't catch reallocating the parent inode
818 * here we will deadlock in xfs_iget() so we have to do these checks
821 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
822 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
823 return -EFSCORRUPTED;
827 * Get the in-core inode with the lock held exclusively to prevent
828 * others from looking at until we're done.
830 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
836 set_nlink(inode, nlink);
837 inode->i_rdev = rdev;
840 if (dir && !(dir->i_mode & S_ISGID) &&
841 (mp->m_flags & XFS_MOUNT_GRPID)) {
842 inode_fsuid_set(inode, mnt_userns);
843 inode->i_gid = dir->i_gid;
844 inode->i_mode = mode;
846 inode_init_owner(mnt_userns, inode, dir, mode);
850 * If the group ID of the new file does not match the effective group
851 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
852 * (and only if the irix_sgid_inherit compatibility variable is set).
854 if (irix_sgid_inherit &&
855 (inode->i_mode & S_ISGID) &&
856 !in_group_p(i_gid_into_mnt(mnt_userns, inode)))
857 inode->i_mode &= ~S_ISGID;
860 ip->i_df.if_nextents = 0;
861 ASSERT(ip->i_nblocks == 0);
863 tv = current_time(inode);
871 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
872 inode_set_iversion(inode, 1);
873 ip->i_cowextsize = 0;
877 flags = XFS_ILOG_CORE;
878 switch (mode & S_IFMT) {
883 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
884 flags |= XFS_ILOG_DEV;
888 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
889 xfs_inode_inherit_flags(ip, pip);
890 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
891 xfs_inode_inherit_flags2(ip, pip);
894 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
895 ip->i_df.if_bytes = 0;
896 ip->i_df.if_u1.if_root = NULL;
903 * If we need to create attributes immediately after allocating the
904 * inode, initialise an empty attribute fork right now. We use the
905 * default fork offset for attributes here as we don't know exactly what
906 * size or how many attributes we might be adding. We can do this
907 * safely here because we know the data fork is completely empty and
908 * this saves us from needing to run a separate transaction to set the
909 * fork offset in the immediate future.
911 if (init_xattrs && xfs_sb_version_hasattr(&mp->m_sb)) {
912 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
913 ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0);
917 * Log the new values stuffed into the inode.
919 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
920 xfs_trans_log_inode(tp, ip, flags);
922 /* now that we have an i_mode we can setup the inode structure */
930 * Decrement the link count on an inode & log the change. If this causes the
931 * link count to go to zero, move the inode to AGI unlinked list so that it can
932 * be freed when the last active reference goes away via xfs_inactive().
934 static int /* error */
939 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
941 drop_nlink(VFS_I(ip));
942 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
944 if (VFS_I(ip)->i_nlink)
947 return xfs_iunlink(tp, ip);
951 * Increment the link count on an inode & log the change.
958 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
960 inc_nlink(VFS_I(ip));
961 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
966 struct user_namespace *mnt_userns,
968 struct xfs_name *name,
974 int is_dir = S_ISDIR(mode);
975 struct xfs_mount *mp = dp->i_mount;
976 struct xfs_inode *ip = NULL;
977 struct xfs_trans *tp = NULL;
979 bool unlock_dp_on_error = false;
981 struct xfs_dquot *udqp = NULL;
982 struct xfs_dquot *gdqp = NULL;
983 struct xfs_dquot *pdqp = NULL;
984 struct xfs_trans_res *tres;
988 trace_xfs_create(dp, name);
990 if (XFS_FORCED_SHUTDOWN(mp))
993 prid = xfs_get_initial_prid(dp);
996 * Make sure that we have allocated dquot(s) on disk.
998 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns),
999 mapped_fsgid(mnt_userns), prid,
1000 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1001 &udqp, &gdqp, &pdqp);
1006 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1007 tres = &M_RES(mp)->tr_mkdir;
1009 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1010 tres = &M_RES(mp)->tr_create;
1014 * Initially assume that the file does not exist and
1015 * reserve the resources for that case. If that is not
1016 * the case we'll drop the one we have and get a more
1017 * appropriate transaction later.
1019 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1021 if (error == -ENOSPC) {
1022 /* flush outstanding delalloc blocks and retry */
1023 xfs_flush_inodes(mp);
1024 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1028 goto out_release_dquots;
1030 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1031 unlock_dp_on_error = true;
1033 error = xfs_iext_count_may_overflow(dp, XFS_DATA_FORK,
1034 XFS_IEXT_DIR_MANIP_CNT(mp));
1036 goto out_trans_cancel;
1039 * A newly created regular or special file just has one directory
1040 * entry pointing to them, but a directory also the "." entry
1041 * pointing to itself.
1043 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1045 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1046 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1048 goto out_trans_cancel;
1051 * Now we join the directory inode to the transaction. We do not do it
1052 * earlier because xfs_dialloc might commit the previous transaction
1053 * (and release all the locks). An error from here on will result in
1054 * the transaction cancel unlocking dp so don't do it explicitly in the
1057 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1058 unlock_dp_on_error = false;
1060 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1061 resblks - XFS_IALLOC_SPACE_RES(mp));
1063 ASSERT(error != -ENOSPC);
1064 goto out_trans_cancel;
1066 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1067 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1070 error = xfs_dir_init(tp, ip, dp);
1072 goto out_trans_cancel;
1074 xfs_bumplink(tp, dp);
1078 * If this is a synchronous mount, make sure that the
1079 * create transaction goes to disk before returning to
1082 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1083 xfs_trans_set_sync(tp);
1086 * Attach the dquot(s) to the inodes and modify them incore.
1087 * These ids of the inode couldn't have changed since the new
1088 * inode has been locked ever since it was created.
1090 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1092 error = xfs_trans_commit(tp);
1094 goto out_release_inode;
1096 xfs_qm_dqrele(udqp);
1097 xfs_qm_dqrele(gdqp);
1098 xfs_qm_dqrele(pdqp);
1104 xfs_trans_cancel(tp);
1107 * Wait until after the current transaction is aborted to finish the
1108 * setup of the inode and release the inode. This prevents recursive
1109 * transactions and deadlocks from xfs_inactive.
1112 xfs_finish_inode_setup(ip);
1116 xfs_qm_dqrele(udqp);
1117 xfs_qm_dqrele(gdqp);
1118 xfs_qm_dqrele(pdqp);
1120 if (unlock_dp_on_error)
1121 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1127 struct user_namespace *mnt_userns,
1128 struct xfs_inode *dp,
1130 struct xfs_inode **ipp)
1132 struct xfs_mount *mp = dp->i_mount;
1133 struct xfs_inode *ip = NULL;
1134 struct xfs_trans *tp = NULL;
1137 struct xfs_dquot *udqp = NULL;
1138 struct xfs_dquot *gdqp = NULL;
1139 struct xfs_dquot *pdqp = NULL;
1140 struct xfs_trans_res *tres;
1144 if (XFS_FORCED_SHUTDOWN(mp))
1147 prid = xfs_get_initial_prid(dp);
1150 * Make sure that we have allocated dquot(s) on disk.
1152 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns),
1153 mapped_fsgid(mnt_userns), prid,
1154 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1155 &udqp, &gdqp, &pdqp);
1159 resblks = XFS_IALLOC_SPACE_RES(mp);
1160 tres = &M_RES(mp)->tr_create_tmpfile;
1162 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1165 goto out_release_dquots;
1167 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1169 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1170 0, 0, prid, false, &ip);
1172 goto out_trans_cancel;
1174 if (mp->m_flags & XFS_MOUNT_WSYNC)
1175 xfs_trans_set_sync(tp);
1178 * Attach the dquot(s) to the inodes and modify them incore.
1179 * These ids of the inode couldn't have changed since the new
1180 * inode has been locked ever since it was created.
1182 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1184 error = xfs_iunlink(tp, ip);
1186 goto out_trans_cancel;
1188 error = xfs_trans_commit(tp);
1190 goto out_release_inode;
1192 xfs_qm_dqrele(udqp);
1193 xfs_qm_dqrele(gdqp);
1194 xfs_qm_dqrele(pdqp);
1200 xfs_trans_cancel(tp);
1203 * Wait until after the current transaction is aborted to finish the
1204 * setup of the inode and release the inode. This prevents recursive
1205 * transactions and deadlocks from xfs_inactive.
1208 xfs_finish_inode_setup(ip);
1212 xfs_qm_dqrele(udqp);
1213 xfs_qm_dqrele(gdqp);
1214 xfs_qm_dqrele(pdqp);
1223 struct xfs_name *target_name)
1225 xfs_mount_t *mp = tdp->i_mount;
1230 trace_xfs_link(tdp, target_name);
1232 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1234 if (XFS_FORCED_SHUTDOWN(mp))
1237 error = xfs_qm_dqattach(sip);
1241 error = xfs_qm_dqattach(tdp);
1245 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1246 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1247 if (error == -ENOSPC) {
1249 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1254 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1256 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1257 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1259 error = xfs_iext_count_may_overflow(tdp, XFS_DATA_FORK,
1260 XFS_IEXT_DIR_MANIP_CNT(mp));
1265 * If we are using project inheritance, we only allow hard link
1266 * creation in our tree when the project IDs are the same; else
1267 * the tree quota mechanism could be circumvented.
1269 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1270 tdp->i_projid != sip->i_projid)) {
1276 error = xfs_dir_canenter(tp, tdp, target_name);
1282 * Handle initial link state of O_TMPFILE inode
1284 if (VFS_I(sip)->i_nlink == 0) {
1285 struct xfs_perag *pag;
1287 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1288 error = xfs_iunlink_remove(tp, pag, sip);
1294 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1298 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1299 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1301 xfs_bumplink(tp, sip);
1304 * If this is a synchronous mount, make sure that the
1305 * link transaction goes to disk before returning to
1308 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1309 xfs_trans_set_sync(tp);
1311 return xfs_trans_commit(tp);
1314 xfs_trans_cancel(tp);
1319 /* Clear the reflink flag and the cowblocks tag if possible. */
1321 xfs_itruncate_clear_reflink_flags(
1322 struct xfs_inode *ip)
1324 struct xfs_ifork *dfork;
1325 struct xfs_ifork *cfork;
1327 if (!xfs_is_reflink_inode(ip))
1329 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1330 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1331 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1332 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1333 if (cfork->if_bytes == 0)
1334 xfs_inode_clear_cowblocks_tag(ip);
1338 * Free up the underlying blocks past new_size. The new size must be smaller
1339 * than the current size. This routine can be used both for the attribute and
1340 * data fork, and does not modify the inode size, which is left to the caller.
1342 * The transaction passed to this routine must have made a permanent log
1343 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1344 * given transaction and start new ones, so make sure everything involved in
1345 * the transaction is tidy before calling here. Some transaction will be
1346 * returned to the caller to be committed. The incoming transaction must
1347 * already include the inode, and both inode locks must be held exclusively.
1348 * The inode must also be "held" within the transaction. On return the inode
1349 * will be "held" within the returned transaction. This routine does NOT
1350 * require any disk space to be reserved for it within the transaction.
1352 * If we get an error, we must return with the inode locked and linked into the
1353 * current transaction. This keeps things simple for the higher level code,
1354 * because it always knows that the inode is locked and held in the transaction
1355 * that returns to it whether errors occur or not. We don't mark the inode
1356 * dirty on error so that transactions can be easily aborted if possible.
1359 xfs_itruncate_extents_flags(
1360 struct xfs_trans **tpp,
1361 struct xfs_inode *ip,
1363 xfs_fsize_t new_size,
1366 struct xfs_mount *mp = ip->i_mount;
1367 struct xfs_trans *tp = *tpp;
1368 xfs_fileoff_t first_unmap_block;
1369 xfs_filblks_t unmap_len;
1372 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1373 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1374 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1375 ASSERT(new_size <= XFS_ISIZE(ip));
1376 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1377 ASSERT(ip->i_itemp != NULL);
1378 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1379 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1381 trace_xfs_itruncate_extents_start(ip, new_size);
1383 flags |= xfs_bmapi_aflag(whichfork);
1386 * Since it is possible for space to become allocated beyond
1387 * the end of the file (in a crash where the space is allocated
1388 * but the inode size is not yet updated), simply remove any
1389 * blocks which show up between the new EOF and the maximum
1390 * possible file size.
1392 * We have to free all the blocks to the bmbt maximum offset, even if
1393 * the page cache can't scale that far.
1395 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1396 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1397 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1401 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1402 while (unmap_len > 0) {
1403 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1404 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1405 flags, XFS_ITRUNC_MAX_EXTENTS);
1409 /* free the just unmapped extents */
1410 error = xfs_defer_finish(&tp);
1415 if (whichfork == XFS_DATA_FORK) {
1416 /* Remove all pending CoW reservations. */
1417 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1418 first_unmap_block, XFS_MAX_FILEOFF, true);
1422 xfs_itruncate_clear_reflink_flags(ip);
1426 * Always re-log the inode so that our permanent transaction can keep
1427 * on rolling it forward in the log.
1429 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1431 trace_xfs_itruncate_extents_end(ip, new_size);
1442 xfs_mount_t *mp = ip->i_mount;
1445 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1448 /* If this is a read-only mount, don't do this (would generate I/O) */
1449 if (mp->m_flags & XFS_MOUNT_RDONLY)
1452 if (!XFS_FORCED_SHUTDOWN(mp)) {
1456 * If we previously truncated this file and removed old data
1457 * in the process, we want to initiate "early" writeout on
1458 * the last close. This is an attempt to combat the notorious
1459 * NULL files problem which is particularly noticeable from a
1460 * truncate down, buffered (re-)write (delalloc), followed by
1461 * a crash. What we are effectively doing here is
1462 * significantly reducing the time window where we'd otherwise
1463 * be exposed to that problem.
1465 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1467 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1468 if (ip->i_delayed_blks > 0) {
1469 error = filemap_flush(VFS_I(ip)->i_mapping);
1476 if (VFS_I(ip)->i_nlink == 0)
1480 * If we can't get the iolock just skip truncating the blocks past EOF
1481 * because we could deadlock with the mmap_lock otherwise. We'll get
1482 * another chance to drop them once the last reference to the inode is
1483 * dropped, so we'll never leak blocks permanently.
1485 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1488 if (xfs_can_free_eofblocks(ip, false)) {
1490 * Check if the inode is being opened, written and closed
1491 * frequently and we have delayed allocation blocks outstanding
1492 * (e.g. streaming writes from the NFS server), truncating the
1493 * blocks past EOF will cause fragmentation to occur.
1495 * In this case don't do the truncation, but we have to be
1496 * careful how we detect this case. Blocks beyond EOF show up as
1497 * i_delayed_blks even when the inode is clean, so we need to
1498 * truncate them away first before checking for a dirty release.
1499 * Hence on the first dirty close we will still remove the
1500 * speculative allocation, but after that we will leave it in
1503 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1506 error = xfs_free_eofblocks(ip);
1510 /* delalloc blocks after truncation means it really is dirty */
1511 if (ip->i_delayed_blks)
1512 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1516 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1521 * xfs_inactive_truncate
1523 * Called to perform a truncate when an inode becomes unlinked.
1526 xfs_inactive_truncate(
1527 struct xfs_inode *ip)
1529 struct xfs_mount *mp = ip->i_mount;
1530 struct xfs_trans *tp;
1533 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1535 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1538 xfs_ilock(ip, XFS_ILOCK_EXCL);
1539 xfs_trans_ijoin(tp, ip, 0);
1542 * Log the inode size first to prevent stale data exposure in the event
1543 * of a system crash before the truncate completes. See the related
1544 * comment in xfs_vn_setattr_size() for details.
1546 ip->i_disk_size = 0;
1547 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1549 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1551 goto error_trans_cancel;
1553 ASSERT(ip->i_df.if_nextents == 0);
1555 error = xfs_trans_commit(tp);
1559 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1563 xfs_trans_cancel(tp);
1565 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1570 * xfs_inactive_ifree()
1572 * Perform the inode free when an inode is unlinked.
1576 struct xfs_inode *ip)
1578 struct xfs_mount *mp = ip->i_mount;
1579 struct xfs_trans *tp;
1583 * We try to use a per-AG reservation for any block needed by the finobt
1584 * tree, but as the finobt feature predates the per-AG reservation
1585 * support a degraded file system might not have enough space for the
1586 * reservation at mount time. In that case try to dip into the reserved
1589 * Send a warning if the reservation does happen to fail, as the inode
1590 * now remains allocated and sits on the unlinked list until the fs is
1593 if (unlikely(mp->m_finobt_nores)) {
1594 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1595 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1598 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1601 if (error == -ENOSPC) {
1602 xfs_warn_ratelimited(mp,
1603 "Failed to remove inode(s) from unlinked list. "
1604 "Please free space, unmount and run xfs_repair.");
1606 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1612 * We do not hold the inode locked across the entire rolling transaction
1613 * here. We only need to hold it for the first transaction that
1614 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1615 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1616 * here breaks the relationship between cluster buffer invalidation and
1617 * stale inode invalidation on cluster buffer item journal commit
1618 * completion, and can result in leaving dirty stale inodes hanging
1621 * We have no need for serialising this inode operation against other
1622 * operations - we freed the inode and hence reallocation is required
1623 * and that will serialise on reallocating the space the deferops need
1624 * to free. Hence we can unlock the inode on the first commit of
1625 * the transaction rather than roll it right through the deferops. This
1626 * avoids relogging the XFS_ISTALE inode.
1628 * We check that xfs_ifree() hasn't grown an internal transaction roll
1629 * by asserting that the inode is still locked when it returns.
1631 xfs_ilock(ip, XFS_ILOCK_EXCL);
1632 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1634 error = xfs_ifree(tp, ip);
1635 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1638 * If we fail to free the inode, shut down. The cancel
1639 * might do that, we need to make sure. Otherwise the
1640 * inode might be lost for a long time or forever.
1642 if (!XFS_FORCED_SHUTDOWN(mp)) {
1643 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1645 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1647 xfs_trans_cancel(tp);
1652 * Credit the quota account(s). The inode is gone.
1654 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1657 * Just ignore errors at this point. There is nothing we can do except
1658 * to try to keep going. Make sure it's not a silent error.
1660 error = xfs_trans_commit(tp);
1662 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1671 * This is called when the vnode reference count for the vnode
1672 * goes to zero. If the file has been unlinked, then it must
1673 * now be truncated. Also, we clear all of the read-ahead state
1674 * kept for the inode here since the file is now closed.
1680 struct xfs_mount *mp;
1685 * If the inode is already free, then there can be nothing
1688 if (VFS_I(ip)->i_mode == 0) {
1689 ASSERT(ip->i_df.if_broot_bytes == 0);
1694 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1696 /* If this is a read-only mount, don't do this (would generate I/O) */
1697 if (mp->m_flags & XFS_MOUNT_RDONLY)
1700 /* Metadata inodes require explicit resource cleanup. */
1701 if (xfs_is_metadata_inode(ip))
1704 /* Try to clean out the cow blocks if there are any. */
1705 if (xfs_inode_has_cow_data(ip))
1706 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1708 if (VFS_I(ip)->i_nlink != 0) {
1710 * force is true because we are evicting an inode from the
1711 * cache. Post-eof blocks must be freed, lest we end up with
1712 * broken free space accounting.
1714 * Note: don't bother with iolock here since lockdep complains
1715 * about acquiring it in reclaim context. We have the only
1716 * reference to the inode at this point anyways.
1718 if (xfs_can_free_eofblocks(ip, true))
1719 xfs_free_eofblocks(ip);
1724 if (S_ISREG(VFS_I(ip)->i_mode) &&
1725 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1726 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1729 error = xfs_qm_dqattach(ip);
1733 if (S_ISLNK(VFS_I(ip)->i_mode))
1734 error = xfs_inactive_symlink(ip);
1736 error = xfs_inactive_truncate(ip);
1741 * If there are attributes associated with the file then blow them away
1742 * now. The code calls a routine that recursively deconstructs the
1743 * attribute fork. If also blows away the in-core attribute fork.
1745 if (XFS_IFORK_Q(ip)) {
1746 error = xfs_attr_inactive(ip);
1752 ASSERT(ip->i_forkoff == 0);
1757 xfs_inactive_ifree(ip);
1761 * We're done making metadata updates for this inode, so we can release
1762 * the attached dquots.
1764 xfs_qm_dqdetach(ip);
1768 * In-Core Unlinked List Lookups
1769 * =============================
1771 * Every inode is supposed to be reachable from some other piece of metadata
1772 * with the exception of the root directory. Inodes with a connection to a
1773 * file descriptor but not linked from anywhere in the on-disk directory tree
1774 * are collectively known as unlinked inodes, though the filesystem itself
1775 * maintains links to these inodes so that on-disk metadata are consistent.
1777 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1778 * header contains a number of buckets that point to an inode, and each inode
1779 * record has a pointer to the next inode in the hash chain. This
1780 * singly-linked list causes scaling problems in the iunlink remove function
1781 * because we must walk that list to find the inode that points to the inode
1782 * being removed from the unlinked hash bucket list.
1784 * What if we modelled the unlinked list as a collection of records capturing
1785 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1786 * have a fast way to look up unlinked list predecessors, which avoids the
1787 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1790 * Because this is a backref cache, we ignore operational failures since the
1791 * iunlink code can fall back to the slow bucket walk. The only errors that
1792 * should bubble out are for obviously incorrect situations.
1794 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1795 * access or have otherwise provided for concurrency control.
1798 /* Capture a "X.next_unlinked = Y" relationship. */
1799 struct xfs_iunlink {
1800 struct rhash_head iu_rhash_head;
1801 xfs_agino_t iu_agino; /* X */
1802 xfs_agino_t iu_next_unlinked; /* Y */
1805 /* Unlinked list predecessor lookup hashtable construction */
1807 xfs_iunlink_obj_cmpfn(
1808 struct rhashtable_compare_arg *arg,
1811 const xfs_agino_t *key = arg->key;
1812 const struct xfs_iunlink *iu = obj;
1814 if (iu->iu_next_unlinked != *key)
1819 static const struct rhashtable_params xfs_iunlink_hash_params = {
1820 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1821 .key_len = sizeof(xfs_agino_t),
1822 .key_offset = offsetof(struct xfs_iunlink,
1824 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1825 .automatic_shrinking = true,
1826 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1830 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1831 * relation is found.
1834 xfs_iunlink_lookup_backref(
1835 struct xfs_perag *pag,
1838 struct xfs_iunlink *iu;
1840 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1841 xfs_iunlink_hash_params);
1842 return iu ? iu->iu_agino : NULLAGINO;
1846 * Take ownership of an iunlink cache entry and insert it into the hash table.
1847 * If successful, the entry will be owned by the cache; if not, it is freed.
1848 * Either way, the caller does not own @iu after this call.
1851 xfs_iunlink_insert_backref(
1852 struct xfs_perag *pag,
1853 struct xfs_iunlink *iu)
1857 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1858 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1860 * Fail loudly if there already was an entry because that's a sign of
1861 * corruption of in-memory data. Also fail loudly if we see an error
1862 * code we didn't anticipate from the rhashtable code. Currently we
1863 * only anticipate ENOMEM.
1866 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1870 * Absorb any runtime errors that aren't a result of corruption because
1871 * this is a cache and we can always fall back to bucket list scanning.
1873 if (error != 0 && error != -EEXIST)
1878 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1880 xfs_iunlink_add_backref(
1881 struct xfs_perag *pag,
1882 xfs_agino_t prev_agino,
1883 xfs_agino_t this_agino)
1885 struct xfs_iunlink *iu;
1887 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
1890 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
1891 iu->iu_agino = prev_agino;
1892 iu->iu_next_unlinked = this_agino;
1894 return xfs_iunlink_insert_backref(pag, iu);
1898 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
1899 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
1900 * wasn't any such entry then we don't bother.
1903 xfs_iunlink_change_backref(
1904 struct xfs_perag *pag,
1906 xfs_agino_t next_unlinked)
1908 struct xfs_iunlink *iu;
1911 /* Look up the old entry; if there wasn't one then exit. */
1912 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1913 xfs_iunlink_hash_params);
1918 * Remove the entry. This shouldn't ever return an error, but if we
1919 * couldn't remove the old entry we don't want to add it again to the
1920 * hash table, and if the entry disappeared on us then someone's
1921 * violated the locking rules and we need to fail loudly. Either way
1922 * we cannot remove the inode because internal state is or would have
1925 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
1926 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1930 /* If there is no new next entry just free our item and return. */
1931 if (next_unlinked == NULLAGINO) {
1936 /* Update the entry and re-add it to the hash table. */
1937 iu->iu_next_unlinked = next_unlinked;
1938 return xfs_iunlink_insert_backref(pag, iu);
1941 /* Set up the in-core predecessor structures. */
1944 struct xfs_perag *pag)
1946 return rhashtable_init(&pag->pagi_unlinked_hash,
1947 &xfs_iunlink_hash_params);
1950 /* Free the in-core predecessor structures. */
1952 xfs_iunlink_free_item(
1956 struct xfs_iunlink *iu = ptr;
1957 bool *freed_anything = arg;
1959 *freed_anything = true;
1964 xfs_iunlink_destroy(
1965 struct xfs_perag *pag)
1967 bool freed_anything = false;
1969 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
1970 xfs_iunlink_free_item, &freed_anything);
1972 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
1976 * Point the AGI unlinked bucket at an inode and log the results. The caller
1977 * is responsible for validating the old value.
1980 xfs_iunlink_update_bucket(
1981 struct xfs_trans *tp,
1982 struct xfs_perag *pag,
1983 struct xfs_buf *agibp,
1984 unsigned int bucket_index,
1985 xfs_agino_t new_agino)
1987 struct xfs_agi *agi = agibp->b_addr;
1988 xfs_agino_t old_value;
1991 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, pag->pag_agno, new_agino));
1993 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1994 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1995 old_value, new_agino);
1998 * We should never find the head of the list already set to the value
1999 * passed in because either we're adding or removing ourselves from the
2002 if (old_value == new_agino) {
2003 xfs_buf_mark_corrupt(agibp);
2004 return -EFSCORRUPTED;
2007 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2008 offset = offsetof(struct xfs_agi, agi_unlinked) +
2009 (sizeof(xfs_agino_t) * bucket_index);
2010 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2014 /* Set an on-disk inode's next_unlinked pointer. */
2016 xfs_iunlink_update_dinode(
2017 struct xfs_trans *tp,
2018 struct xfs_perag *pag,
2020 struct xfs_buf *ibp,
2021 struct xfs_dinode *dip,
2022 struct xfs_imap *imap,
2023 xfs_agino_t next_agino)
2025 struct xfs_mount *mp = tp->t_mountp;
2028 ASSERT(xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino));
2030 trace_xfs_iunlink_update_dinode(mp, pag->pag_agno, agino,
2031 be32_to_cpu(dip->di_next_unlinked), next_agino);
2033 dip->di_next_unlinked = cpu_to_be32(next_agino);
2034 offset = imap->im_boffset +
2035 offsetof(struct xfs_dinode, di_next_unlinked);
2037 /* need to recalc the inode CRC if appropriate */
2038 xfs_dinode_calc_crc(mp, dip);
2039 xfs_trans_inode_buf(tp, ibp);
2040 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2043 /* Set an in-core inode's unlinked pointer and return the old value. */
2045 xfs_iunlink_update_inode(
2046 struct xfs_trans *tp,
2047 struct xfs_inode *ip,
2048 struct xfs_perag *pag,
2049 xfs_agino_t next_agino,
2050 xfs_agino_t *old_next_agino)
2052 struct xfs_mount *mp = tp->t_mountp;
2053 struct xfs_dinode *dip;
2054 struct xfs_buf *ibp;
2055 xfs_agino_t old_value;
2058 ASSERT(xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino));
2060 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &ibp);
2063 dip = xfs_buf_offset(ibp, ip->i_imap.im_boffset);
2065 /* Make sure the old pointer isn't garbage. */
2066 old_value = be32_to_cpu(dip->di_next_unlinked);
2067 if (!xfs_verify_agino_or_null(mp, pag->pag_agno, old_value)) {
2068 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2069 sizeof(*dip), __this_address);
2070 error = -EFSCORRUPTED;
2075 * Since we're updating a linked list, we should never find that the
2076 * current pointer is the same as the new value, unless we're
2077 * terminating the list.
2079 *old_next_agino = old_value;
2080 if (old_value == next_agino) {
2081 if (next_agino != NULLAGINO) {
2082 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2083 dip, sizeof(*dip), __this_address);
2084 error = -EFSCORRUPTED;
2089 /* Ok, update the new pointer. */
2090 xfs_iunlink_update_dinode(tp, pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
2091 ibp, dip, &ip->i_imap, next_agino);
2094 xfs_trans_brelse(tp, ibp);
2099 * This is called when the inode's link count has gone to 0 or we are creating
2100 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2102 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2103 * list when the inode is freed.
2107 struct xfs_trans *tp,
2108 struct xfs_inode *ip)
2110 struct xfs_mount *mp = tp->t_mountp;
2111 struct xfs_perag *pag;
2112 struct xfs_agi *agi;
2113 struct xfs_buf *agibp;
2114 xfs_agino_t next_agino;
2115 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2116 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2119 ASSERT(VFS_I(ip)->i_nlink == 0);
2120 ASSERT(VFS_I(ip)->i_mode != 0);
2121 trace_xfs_iunlink(ip);
2123 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2125 /* Get the agi buffer first. It ensures lock ordering on the list. */
2126 error = xfs_read_agi(mp, tp, pag->pag_agno, &agibp);
2129 agi = agibp->b_addr;
2132 * Get the index into the agi hash table for the list this inode will
2133 * go on. Make sure the pointer isn't garbage and that this inode
2134 * isn't already on the list.
2136 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2137 if (next_agino == agino ||
2138 !xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino)) {
2139 xfs_buf_mark_corrupt(agibp);
2140 error = -EFSCORRUPTED;
2144 if (next_agino != NULLAGINO) {
2145 xfs_agino_t old_agino;
2148 * There is already another inode in the bucket, so point this
2149 * inode to the current head of the list.
2151 error = xfs_iunlink_update_inode(tp, ip, pag, next_agino,
2155 ASSERT(old_agino == NULLAGINO);
2158 * agino has been unlinked, add a backref from the next inode
2161 error = xfs_iunlink_add_backref(pag, agino, next_agino);
2166 /* Point the head of the list to point to this inode. */
2167 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
2173 /* Return the imap, dinode pointer, and buffer for an inode. */
2175 xfs_iunlink_map_ino(
2176 struct xfs_trans *tp,
2177 xfs_agnumber_t agno,
2179 struct xfs_imap *imap,
2180 struct xfs_dinode **dipp,
2181 struct xfs_buf **bpp)
2183 struct xfs_mount *mp = tp->t_mountp;
2187 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2189 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2194 error = xfs_imap_to_bp(mp, tp, imap, bpp);
2196 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2201 *dipp = xfs_buf_offset(*bpp, imap->im_boffset);
2206 * Walk the unlinked chain from @head_agino until we find the inode that
2207 * points to @target_agino. Return the inode number, map, dinode pointer,
2208 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2210 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2211 * @agino, @imap, @dipp, and @bpp are all output parameters.
2213 * Do not call this function if @target_agino is the head of the list.
2216 xfs_iunlink_map_prev(
2217 struct xfs_trans *tp,
2218 struct xfs_perag *pag,
2219 xfs_agino_t head_agino,
2220 xfs_agino_t target_agino,
2222 struct xfs_imap *imap,
2223 struct xfs_dinode **dipp,
2224 struct xfs_buf **bpp)
2226 struct xfs_mount *mp = tp->t_mountp;
2227 xfs_agino_t next_agino;
2230 ASSERT(head_agino != target_agino);
2233 /* See if our backref cache can find it faster. */
2234 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2235 if (*agino != NULLAGINO) {
2236 error = xfs_iunlink_map_ino(tp, pag->pag_agno, *agino, imap,
2241 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2245 * If we get here the cache contents were corrupt, so drop the
2246 * buffer and fall back to walking the bucket list.
2248 xfs_trans_brelse(tp, *bpp);
2253 trace_xfs_iunlink_map_prev_fallback(mp, pag->pag_agno);
2255 /* Otherwise, walk the entire bucket until we find it. */
2256 next_agino = head_agino;
2257 while (next_agino != target_agino) {
2258 xfs_agino_t unlinked_agino;
2261 xfs_trans_brelse(tp, *bpp);
2263 *agino = next_agino;
2264 error = xfs_iunlink_map_ino(tp, pag->pag_agno, next_agino, imap,
2269 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2271 * Make sure this pointer is valid and isn't an obvious
2274 if (!xfs_verify_agino(mp, pag->pag_agno, unlinked_agino) ||
2275 next_agino == unlinked_agino) {
2276 XFS_CORRUPTION_ERROR(__func__,
2277 XFS_ERRLEVEL_LOW, mp,
2278 *dipp, sizeof(**dipp));
2279 error = -EFSCORRUPTED;
2282 next_agino = unlinked_agino;
2289 * Pull the on-disk inode from the AGI unlinked list.
2293 struct xfs_trans *tp,
2294 struct xfs_perag *pag,
2295 struct xfs_inode *ip)
2297 struct xfs_mount *mp = tp->t_mountp;
2298 struct xfs_agi *agi;
2299 struct xfs_buf *agibp;
2300 struct xfs_buf *last_ibp;
2301 struct xfs_dinode *last_dip = NULL;
2302 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2303 xfs_agino_t next_agino;
2304 xfs_agino_t head_agino;
2305 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2308 trace_xfs_iunlink_remove(ip);
2310 /* Get the agi buffer first. It ensures lock ordering on the list. */
2311 error = xfs_read_agi(mp, tp, pag->pag_agno, &agibp);
2314 agi = agibp->b_addr;
2317 * Get the index into the agi hash table for the list this inode will
2318 * go on. Make sure the head pointer isn't garbage.
2320 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2321 if (!xfs_verify_agino(mp, pag->pag_agno, head_agino)) {
2322 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2324 return -EFSCORRUPTED;
2328 * Set our inode's next_unlinked pointer to NULL and then return
2329 * the old pointer value so that we can update whatever was previous
2330 * to us in the list to point to whatever was next in the list.
2332 error = xfs_iunlink_update_inode(tp, ip, pag, NULLAGINO, &next_agino);
2337 * If there was a backref pointing from the next inode back to this
2338 * one, remove it because we've removed this inode from the list.
2340 * Later, if this inode was in the middle of the list we'll update
2341 * this inode's backref to point from the next inode.
2343 if (next_agino != NULLAGINO) {
2344 error = xfs_iunlink_change_backref(pag, next_agino, NULLAGINO);
2349 if (head_agino != agino) {
2350 struct xfs_imap imap;
2351 xfs_agino_t prev_agino;
2353 /* We need to search the list for the inode being freed. */
2354 error = xfs_iunlink_map_prev(tp, pag, head_agino, agino,
2355 &prev_agino, &imap, &last_dip, &last_ibp);
2359 /* Point the previous inode on the list to the next inode. */
2360 xfs_iunlink_update_dinode(tp, pag, prev_agino, last_ibp,
2361 last_dip, &imap, next_agino);
2364 * Now we deal with the backref for this inode. If this inode
2365 * pointed at a real inode, change the backref that pointed to
2366 * us to point to our old next. If this inode was the end of
2367 * the list, delete the backref that pointed to us. Note that
2368 * change_backref takes care of deleting the backref if
2369 * next_agino is NULLAGINO.
2371 return xfs_iunlink_change_backref(agibp->b_pag, agino,
2375 /* Point the head of the list to the next unlinked inode. */
2376 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2381 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2382 * mark it stale. We should only find clean inodes in this lookup that aren't
2386 xfs_ifree_mark_inode_stale(
2387 struct xfs_perag *pag,
2388 struct xfs_inode *free_ip,
2391 struct xfs_mount *mp = pag->pag_mount;
2392 struct xfs_inode_log_item *iip;
2393 struct xfs_inode *ip;
2397 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2399 /* Inode not in memory, nothing to do */
2406 * because this is an RCU protected lookup, we could find a recently
2407 * freed or even reallocated inode during the lookup. We need to check
2408 * under the i_flags_lock for a valid inode here. Skip it if it is not
2409 * valid, the wrong inode or stale.
2411 spin_lock(&ip->i_flags_lock);
2412 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2413 goto out_iflags_unlock;
2416 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2417 * other inodes that we did not find in the list attached to the buffer
2418 * and are not already marked stale. If we can't lock it, back off and
2421 if (ip != free_ip) {
2422 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2423 spin_unlock(&ip->i_flags_lock);
2429 ip->i_flags |= XFS_ISTALE;
2432 * If the inode is flushing, it is already attached to the buffer. All
2433 * we needed to do here is mark the inode stale so buffer IO completion
2434 * will remove it from the AIL.
2437 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2438 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2439 ASSERT(iip->ili_last_fields);
2444 * Inodes not attached to the buffer can be released immediately.
2445 * Everything else has to go through xfs_iflush_abort() on journal
2446 * commit as the flock synchronises removal of the inode from the
2447 * cluster buffer against inode reclaim.
2449 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2452 __xfs_iflags_set(ip, XFS_IFLUSHING);
2453 spin_unlock(&ip->i_flags_lock);
2456 /* we have a dirty inode in memory that has not yet been flushed. */
2457 spin_lock(&iip->ili_lock);
2458 iip->ili_last_fields = iip->ili_fields;
2459 iip->ili_fields = 0;
2460 iip->ili_fsync_fields = 0;
2461 spin_unlock(&iip->ili_lock);
2462 ASSERT(iip->ili_last_fields);
2465 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2470 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2472 spin_unlock(&ip->i_flags_lock);
2477 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2478 * inodes that are in memory - they all must be marked stale and attached to
2479 * the cluster buffer.
2483 struct xfs_trans *tp,
2484 struct xfs_perag *pag,
2485 struct xfs_inode *free_ip,
2486 struct xfs_icluster *xic)
2488 struct xfs_mount *mp = free_ip->i_mount;
2489 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2492 xfs_ino_t inum = xic->first_ino;
2498 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2500 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2502 * The allocation bitmap tells us which inodes of the chunk were
2503 * physically allocated. Skip the cluster if an inode falls into
2506 ioffset = inum - xic->first_ino;
2507 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2508 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2512 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2513 XFS_INO_TO_AGBNO(mp, inum));
2516 * We obtain and lock the backing buffer first in the process
2517 * here to ensure dirty inodes attached to the buffer remain in
2518 * the flushing state while we mark them stale.
2520 * If we scan the in-memory inodes first, then buffer IO can
2521 * complete before we get a lock on it, and hence we may fail
2522 * to mark all the active inodes on the buffer stale.
2524 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2525 mp->m_bsize * igeo->blocks_per_cluster,
2531 * This buffer may not have been correctly initialised as we
2532 * didn't read it from disk. That's not important because we are
2533 * only using to mark the buffer as stale in the log, and to
2534 * attach stale cached inodes on it. That means it will never be
2535 * dispatched for IO. If it is, we want to know about it, and we
2536 * want it to fail. We can acheive this by adding a write
2537 * verifier to the buffer.
2539 bp->b_ops = &xfs_inode_buf_ops;
2542 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2543 * too. This requires lookups, and will skip inodes that we've
2544 * already marked XFS_ISTALE.
2546 for (i = 0; i < igeo->inodes_per_cluster; i++)
2547 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2549 xfs_trans_stale_inode_buf(tp, bp);
2550 xfs_trans_binval(tp, bp);
2556 * This is called to return an inode to the inode free list.
2557 * The inode should already be truncated to 0 length and have
2558 * no pages associated with it. This routine also assumes that
2559 * the inode is already a part of the transaction.
2561 * The on-disk copy of the inode will have been added to the list
2562 * of unlinked inodes in the AGI. We need to remove the inode from
2563 * that list atomically with respect to freeing it here.
2567 struct xfs_trans *tp,
2568 struct xfs_inode *ip)
2570 struct xfs_mount *mp = ip->i_mount;
2571 struct xfs_perag *pag;
2572 struct xfs_icluster xic = { 0 };
2573 struct xfs_inode_log_item *iip = ip->i_itemp;
2576 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2577 ASSERT(VFS_I(ip)->i_nlink == 0);
2578 ASSERT(ip->i_df.if_nextents == 0);
2579 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2580 ASSERT(ip->i_nblocks == 0);
2582 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2585 * Pull the on-disk inode from the AGI unlinked list.
2587 error = xfs_iunlink_remove(tp, pag, ip);
2591 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2596 * Free any local-format data sitting around before we reset the
2597 * data fork to extents format. Note that the attr fork data has
2598 * already been freed by xfs_attr_inactive.
2600 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2601 kmem_free(ip->i_df.if_u1.if_data);
2602 ip->i_df.if_u1.if_data = NULL;
2603 ip->i_df.if_bytes = 0;
2606 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2608 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2609 ip->i_forkoff = 0; /* mark the attr fork not in use */
2610 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2611 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2612 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2614 /* Don't attempt to replay owner changes for a deleted inode */
2615 spin_lock(&iip->ili_lock);
2616 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2617 spin_unlock(&iip->ili_lock);
2620 * Bump the generation count so no one will be confused
2621 * by reincarnations of this inode.
2623 VFS_I(ip)->i_generation++;
2624 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2627 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2634 * This is called to unpin an inode. The caller must have the inode locked
2635 * in at least shared mode so that the buffer cannot be subsequently pinned
2636 * once someone is waiting for it to be unpinned.
2640 struct xfs_inode *ip)
2642 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2644 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2646 /* Give the log a push to start the unpinning I/O */
2647 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2653 struct xfs_inode *ip)
2655 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2656 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2661 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2662 if (xfs_ipincount(ip))
2664 } while (xfs_ipincount(ip));
2665 finish_wait(wq, &wait.wq_entry);
2670 struct xfs_inode *ip)
2672 if (xfs_ipincount(ip))
2673 __xfs_iunpin_wait(ip);
2677 * Removing an inode from the namespace involves removing the directory entry
2678 * and dropping the link count on the inode. Removing the directory entry can
2679 * result in locking an AGF (directory blocks were freed) and removing a link
2680 * count can result in placing the inode on an unlinked list which results in
2683 * The big problem here is that we have an ordering constraint on AGF and AGI
2684 * locking - inode allocation locks the AGI, then can allocate a new extent for
2685 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2686 * removes the inode from the unlinked list, requiring that we lock the AGI
2687 * first, and then freeing the inode can result in an inode chunk being freed
2688 * and hence freeing disk space requiring that we lock an AGF.
2690 * Hence the ordering that is imposed by other parts of the code is AGI before
2691 * AGF. This means we cannot remove the directory entry before we drop the inode
2692 * reference count and put it on the unlinked list as this results in a lock
2693 * order of AGF then AGI, and this can deadlock against inode allocation and
2694 * freeing. Therefore we must drop the link counts before we remove the
2697 * This is still safe from a transactional point of view - it is not until we
2698 * get to xfs_defer_finish() that we have the possibility of multiple
2699 * transactions in this operation. Hence as long as we remove the directory
2700 * entry and drop the link count in the first transaction of the remove
2701 * operation, there are no transactional constraints on the ordering here.
2706 struct xfs_name *name,
2709 xfs_mount_t *mp = dp->i_mount;
2710 xfs_trans_t *tp = NULL;
2711 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2715 trace_xfs_remove(dp, name);
2717 if (XFS_FORCED_SHUTDOWN(mp))
2720 error = xfs_qm_dqattach(dp);
2724 error = xfs_qm_dqattach(ip);
2729 * We try to get the real space reservation first,
2730 * allowing for directory btree deletion(s) implying
2731 * possible bmap insert(s). If we can't get the space
2732 * reservation then we use 0 instead, and avoid the bmap
2733 * btree insert(s) in the directory code by, if the bmap
2734 * insert tries to happen, instead trimming the LAST
2735 * block from the directory.
2737 resblks = XFS_REMOVE_SPACE_RES(mp);
2738 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2739 if (error == -ENOSPC) {
2741 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2745 ASSERT(error != -ENOSPC);
2749 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2751 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2752 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2755 * If we're removing a directory perform some additional validation.
2758 ASSERT(VFS_I(ip)->i_nlink >= 2);
2759 if (VFS_I(ip)->i_nlink != 2) {
2761 goto out_trans_cancel;
2763 if (!xfs_dir_isempty(ip)) {
2765 goto out_trans_cancel;
2768 /* Drop the link from ip's "..". */
2769 error = xfs_droplink(tp, dp);
2771 goto out_trans_cancel;
2773 /* Drop the "." link from ip to self. */
2774 error = xfs_droplink(tp, ip);
2776 goto out_trans_cancel;
2779 * Point the unlinked child directory's ".." entry to the root
2780 * directory to eliminate back-references to inodes that may
2781 * get freed before the child directory is closed. If the fs
2782 * gets shrunk, this can lead to dirent inode validation errors.
2784 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2785 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2786 tp->t_mountp->m_sb.sb_rootino, 0);
2792 * When removing a non-directory we need to log the parent
2793 * inode here. For a directory this is done implicitly
2794 * by the xfs_droplink call for the ".." entry.
2796 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2798 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2800 /* Drop the link from dp to ip. */
2801 error = xfs_droplink(tp, ip);
2803 goto out_trans_cancel;
2805 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2807 ASSERT(error != -ENOENT);
2808 goto out_trans_cancel;
2812 * If this is a synchronous mount, make sure that the
2813 * remove transaction goes to disk before returning to
2816 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2817 xfs_trans_set_sync(tp);
2819 error = xfs_trans_commit(tp);
2823 if (is_dir && xfs_inode_is_filestream(ip))
2824 xfs_filestream_deassociate(ip);
2829 xfs_trans_cancel(tp);
2835 * Enter all inodes for a rename transaction into a sorted array.
2837 #define __XFS_SORT_INODES 5
2839 xfs_sort_for_rename(
2840 struct xfs_inode *dp1, /* in: old (source) directory inode */
2841 struct xfs_inode *dp2, /* in: new (target) directory inode */
2842 struct xfs_inode *ip1, /* in: inode of old entry */
2843 struct xfs_inode *ip2, /* in: inode of new entry */
2844 struct xfs_inode *wip, /* in: whiteout inode */
2845 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2846 int *num_inodes) /* in/out: inodes in array */
2850 ASSERT(*num_inodes == __XFS_SORT_INODES);
2851 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2854 * i_tab contains a list of pointers to inodes. We initialize
2855 * the table here & we'll sort it. We will then use it to
2856 * order the acquisition of the inode locks.
2858 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2871 * Sort the elements via bubble sort. (Remember, there are at
2872 * most 5 elements to sort, so this is adequate.)
2874 for (i = 0; i < *num_inodes; i++) {
2875 for (j = 1; j < *num_inodes; j++) {
2876 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2877 struct xfs_inode *temp = i_tab[j];
2878 i_tab[j] = i_tab[j-1];
2887 struct xfs_trans *tp)
2890 * If this is a synchronous mount, make sure that the rename transaction
2891 * goes to disk before returning to the user.
2893 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2894 xfs_trans_set_sync(tp);
2896 return xfs_trans_commit(tp);
2900 * xfs_cross_rename()
2902 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2906 struct xfs_trans *tp,
2907 struct xfs_inode *dp1,
2908 struct xfs_name *name1,
2909 struct xfs_inode *ip1,
2910 struct xfs_inode *dp2,
2911 struct xfs_name *name2,
2912 struct xfs_inode *ip2,
2920 /* Swap inode number for dirent in first parent */
2921 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2923 goto out_trans_abort;
2925 /* Swap inode number for dirent in second parent */
2926 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2928 goto out_trans_abort;
2931 * If we're renaming one or more directories across different parents,
2932 * update the respective ".." entries (and link counts) to match the new
2936 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2938 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2939 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2940 dp1->i_ino, spaceres);
2942 goto out_trans_abort;
2944 /* transfer ip2 ".." reference to dp1 */
2945 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2946 error = xfs_droplink(tp, dp2);
2948 goto out_trans_abort;
2949 xfs_bumplink(tp, dp1);
2953 * Although ip1 isn't changed here, userspace needs
2954 * to be warned about the change, so that applications
2955 * relying on it (like backup ones), will properly
2958 ip1_flags |= XFS_ICHGTIME_CHG;
2959 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2962 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2963 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2964 dp2->i_ino, spaceres);
2966 goto out_trans_abort;
2968 /* transfer ip1 ".." reference to dp2 */
2969 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2970 error = xfs_droplink(tp, dp1);
2972 goto out_trans_abort;
2973 xfs_bumplink(tp, dp2);
2977 * Although ip2 isn't changed here, userspace needs
2978 * to be warned about the change, so that applications
2979 * relying on it (like backup ones), will properly
2982 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2983 ip2_flags |= XFS_ICHGTIME_CHG;
2988 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2989 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2992 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2993 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2996 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2997 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2999 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3000 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3001 return xfs_finish_rename(tp);
3004 xfs_trans_cancel(tp);
3009 * xfs_rename_alloc_whiteout()
3011 * Return a referenced, unlinked, unlocked inode that can be used as a
3012 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3013 * crash between allocating the inode and linking it into the rename transaction
3014 * recovery will free the inode and we won't leak it.
3017 xfs_rename_alloc_whiteout(
3018 struct user_namespace *mnt_userns,
3019 struct xfs_inode *dp,
3020 struct xfs_inode **wip)
3022 struct xfs_inode *tmpfile;
3025 error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
3031 * Prepare the tmpfile inode as if it were created through the VFS.
3032 * Complete the inode setup and flag it as linkable. nlink is already
3033 * zero, so we can skip the drop_nlink.
3035 xfs_setup_iops(tmpfile);
3036 xfs_finish_inode_setup(tmpfile);
3037 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3048 struct user_namespace *mnt_userns,
3049 struct xfs_inode *src_dp,
3050 struct xfs_name *src_name,
3051 struct xfs_inode *src_ip,
3052 struct xfs_inode *target_dp,
3053 struct xfs_name *target_name,
3054 struct xfs_inode *target_ip,
3057 struct xfs_mount *mp = src_dp->i_mount;
3058 struct xfs_trans *tp;
3059 struct xfs_inode *wip = NULL; /* whiteout inode */
3060 struct xfs_inode *inodes[__XFS_SORT_INODES];
3062 int num_inodes = __XFS_SORT_INODES;
3063 bool new_parent = (src_dp != target_dp);
3064 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3068 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3070 if ((flags & RENAME_EXCHANGE) && !target_ip)
3074 * If we are doing a whiteout operation, allocate the whiteout inode
3075 * we will be placing at the target and ensure the type is set
3078 if (flags & RENAME_WHITEOUT) {
3079 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3080 error = xfs_rename_alloc_whiteout(mnt_userns, target_dp, &wip);
3084 /* setup target dirent info as whiteout */
3085 src_name->type = XFS_DIR3_FT_CHRDEV;
3088 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3089 inodes, &num_inodes);
3091 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3092 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3093 if (error == -ENOSPC) {
3095 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3099 goto out_release_wip;
3102 * Attach the dquots to the inodes
3104 error = xfs_qm_vop_rename_dqattach(inodes);
3106 goto out_trans_cancel;
3109 * Lock all the participating inodes. Depending upon whether
3110 * the target_name exists in the target directory, and
3111 * whether the target directory is the same as the source
3112 * directory, we can lock from 2 to 4 inodes.
3114 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3117 * Join all the inodes to the transaction. From this point on,
3118 * we can rely on either trans_commit or trans_cancel to unlock
3121 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3123 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3124 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3126 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3128 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3131 * If we are using project inheritance, we only allow renames
3132 * into our tree when the project IDs are the same; else the
3133 * tree quota mechanism would be circumvented.
3135 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
3136 target_dp->i_projid != src_ip->i_projid)) {
3138 goto out_trans_cancel;
3141 /* RENAME_EXCHANGE is unique from here on. */
3142 if (flags & RENAME_EXCHANGE)
3143 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3144 target_dp, target_name, target_ip,
3148 * Check for expected errors before we dirty the transaction
3149 * so we can return an error without a transaction abort.
3151 * Extent count overflow check:
3153 * From the perspective of src_dp, a rename operation is essentially a
3154 * directory entry remove operation. Hence the only place where we check
3155 * for extent count overflow for src_dp is in
3156 * xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns
3157 * -ENOSPC when it detects a possible extent count overflow and in
3158 * response, the higher layers of directory handling code do the
3160 * 1. Data/Free blocks: XFS lets these blocks linger until a
3161 * future remove operation removes them.
3162 * 2. Dabtree blocks: XFS swaps the blocks with the last block in the
3163 * Leaf space and unmaps the last block.
3165 * For target_dp, there are two cases depending on whether the
3166 * destination directory entry exists or not.
3168 * When destination directory entry does not exist (i.e. target_ip ==
3169 * NULL), extent count overflow check is performed only when transaction
3170 * has a non-zero sized space reservation associated with it. With a
3171 * zero-sized space reservation, XFS allows a rename operation to
3172 * continue only when the directory has sufficient free space in its
3173 * data/leaf/free space blocks to hold the new entry.
3175 * When destination directory entry exists (i.e. target_ip != NULL), all
3176 * we need to do is change the inode number associated with the already
3177 * existing entry. Hence there is no need to perform an extent count
3180 if (target_ip == NULL) {
3182 * If there's no space reservation, check the entry will
3183 * fit before actually inserting it.
3186 error = xfs_dir_canenter(tp, target_dp, target_name);
3188 goto out_trans_cancel;
3190 error = xfs_iext_count_may_overflow(target_dp,
3192 XFS_IEXT_DIR_MANIP_CNT(mp));
3194 goto out_trans_cancel;
3198 * If target exists and it's a directory, check that whether
3199 * it can be destroyed.
3201 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3202 (!xfs_dir_isempty(target_ip) ||
3203 (VFS_I(target_ip)->i_nlink > 2))) {
3205 goto out_trans_cancel;
3210 * Lock the AGI buffers we need to handle bumping the nlink of the
3211 * whiteout inode off the unlinked list and to handle dropping the
3212 * nlink of the target inode. Per locking order rules, do this in
3213 * increasing AG order and before directory block allocation tries to
3214 * grab AGFs because we grab AGIs before AGFs.
3216 * The (vfs) caller must ensure that if src is a directory then
3217 * target_ip is either null or an empty directory.
3219 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3220 if (inodes[i] == wip ||
3221 (inodes[i] == target_ip &&
3222 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3224 xfs_agnumber_t agno;
3226 agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
3227 error = xfs_read_agi(mp, tp, agno, &bp);
3229 goto out_trans_cancel;
3234 * Directory entry creation below may acquire the AGF. Remove
3235 * the whiteout from the unlinked list first to preserve correct
3236 * AGI/AGF locking order. This dirties the transaction so failures
3237 * after this point will abort and log recovery will clean up the
3240 * For whiteouts, we need to bump the link count on the whiteout
3241 * inode. After this point, we have a real link, clear the tmpfile
3242 * state flag from the inode so it doesn't accidentally get misused
3246 struct xfs_perag *pag;
3248 ASSERT(VFS_I(wip)->i_nlink == 0);
3250 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
3251 error = xfs_iunlink_remove(tp, pag, wip);
3254 goto out_trans_cancel;
3256 xfs_bumplink(tp, wip);
3257 VFS_I(wip)->i_state &= ~I_LINKABLE;
3261 * Set up the target.
3263 if (target_ip == NULL) {
3265 * If target does not exist and the rename crosses
3266 * directories, adjust the target directory link count
3267 * to account for the ".." reference from the new entry.
3269 error = xfs_dir_createname(tp, target_dp, target_name,
3270 src_ip->i_ino, spaceres);
3272 goto out_trans_cancel;
3274 xfs_trans_ichgtime(tp, target_dp,
3275 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3277 if (new_parent && src_is_directory) {
3278 xfs_bumplink(tp, target_dp);
3280 } else { /* target_ip != NULL */
3282 * Link the source inode under the target name.
3283 * If the source inode is a directory and we are moving
3284 * it across directories, its ".." entry will be
3285 * inconsistent until we replace that down below.
3287 * In case there is already an entry with the same
3288 * name at the destination directory, remove it first.
3290 error = xfs_dir_replace(tp, target_dp, target_name,
3291 src_ip->i_ino, spaceres);
3293 goto out_trans_cancel;
3295 xfs_trans_ichgtime(tp, target_dp,
3296 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3299 * Decrement the link count on the target since the target
3300 * dir no longer points to it.
3302 error = xfs_droplink(tp, target_ip);
3304 goto out_trans_cancel;
3306 if (src_is_directory) {
3308 * Drop the link from the old "." entry.
3310 error = xfs_droplink(tp, target_ip);
3312 goto out_trans_cancel;
3314 } /* target_ip != NULL */
3317 * Remove the source.
3319 if (new_parent && src_is_directory) {
3321 * Rewrite the ".." entry to point to the new
3324 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3325 target_dp->i_ino, spaceres);
3326 ASSERT(error != -EEXIST);
3328 goto out_trans_cancel;
3332 * We always want to hit the ctime on the source inode.
3334 * This isn't strictly required by the standards since the source
3335 * inode isn't really being changed, but old unix file systems did
3336 * it and some incremental backup programs won't work without it.
3338 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3339 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3342 * Adjust the link count on src_dp. This is necessary when
3343 * renaming a directory, either within one parent when
3344 * the target existed, or across two parent directories.
3346 if (src_is_directory && (new_parent || target_ip != NULL)) {
3349 * Decrement link count on src_directory since the
3350 * entry that's moved no longer points to it.
3352 error = xfs_droplink(tp, src_dp);
3354 goto out_trans_cancel;
3358 * For whiteouts, we only need to update the source dirent with the
3359 * inode number of the whiteout inode rather than removing it
3363 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3367 * NOTE: We don't need to check for extent count overflow here
3368 * because the dir remove name code will leave the dir block in
3369 * place if the extent count would overflow.
3371 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3376 goto out_trans_cancel;
3378 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3379 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3381 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3383 error = xfs_finish_rename(tp);
3389 xfs_trans_cancel(tp);
3398 struct xfs_inode *ip,
3401 struct xfs_inode_log_item *iip = ip->i_itemp;
3402 struct xfs_dinode *dip;
3403 struct xfs_mount *mp = ip->i_mount;
3406 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3407 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3408 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3409 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3410 ASSERT(iip->ili_item.li_buf == bp);
3412 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3415 * We don't flush the inode if any of the following checks fail, but we
3416 * do still update the log item and attach to the backing buffer as if
3417 * the flush happened. This is a formality to facilitate predictable
3418 * error handling as the caller will shutdown and fail the buffer.
3420 error = -EFSCORRUPTED;
3421 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3422 mp, XFS_ERRTAG_IFLUSH_1)) {
3423 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3424 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3425 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3428 if (S_ISREG(VFS_I(ip)->i_mode)) {
3430 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3431 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3432 mp, XFS_ERRTAG_IFLUSH_3)) {
3433 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3434 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3435 __func__, ip->i_ino, ip);
3438 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3440 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3441 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3442 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3443 mp, XFS_ERRTAG_IFLUSH_4)) {
3444 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3445 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3446 __func__, ip->i_ino, ip);
3450 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3451 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3452 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3453 "%s: detected corrupt incore inode %Lu, "
3454 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3455 __func__, ip->i_ino,
3456 ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3460 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3461 mp, XFS_ERRTAG_IFLUSH_6)) {
3462 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3463 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3464 __func__, ip->i_ino, ip->i_forkoff, ip);
3469 * Inode item log recovery for v2 inodes are dependent on the flushiter
3470 * count for correct sequencing. We bump the flush iteration count so
3471 * we can detect flushes which postdate a log record during recovery.
3472 * This is redundant as we now log every change and hence this can't
3473 * happen but we need to still do it to ensure backwards compatibility
3474 * with old kernels that predate logging all inode changes.
3476 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3480 * If there are inline format data / attr forks attached to this inode,
3481 * make sure they are not corrupt.
3483 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3484 xfs_ifork_verify_local_data(ip))
3486 if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3487 xfs_ifork_verify_local_attr(ip))
3491 * Copy the dirty parts of the inode into the on-disk inode. We always
3492 * copy out the core of the inode, because if the inode is dirty at all
3495 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3497 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3498 if (!xfs_sb_version_has_v3inode(&mp->m_sb)) {
3499 if (ip->i_flushiter == DI_MAX_FLUSH)
3500 ip->i_flushiter = 0;
3503 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3504 if (XFS_IFORK_Q(ip))
3505 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3508 * We've recorded everything logged in the inode, so we'd like to clear
3509 * the ili_fields bits so we don't log and flush things unnecessarily.
3510 * However, we can't stop logging all this information until the data
3511 * we've copied into the disk buffer is written to disk. If we did we
3512 * might overwrite the copy of the inode in the log with all the data
3513 * after re-logging only part of it, and in the face of a crash we
3514 * wouldn't have all the data we need to recover.
3516 * What we do is move the bits to the ili_last_fields field. When
3517 * logging the inode, these bits are moved back to the ili_fields field.
3518 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3519 * we know that the information those bits represent is permanently on
3520 * disk. As long as the flush completes before the inode is logged
3521 * again, then both ili_fields and ili_last_fields will be cleared.
3525 spin_lock(&iip->ili_lock);
3526 iip->ili_last_fields = iip->ili_fields;
3527 iip->ili_fields = 0;
3528 iip->ili_fsync_fields = 0;
3529 spin_unlock(&iip->ili_lock);
3532 * Store the current LSN of the inode so that we can tell whether the
3533 * item has moved in the AIL from xfs_buf_inode_iodone().
3535 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3536 &iip->ili_item.li_lsn);
3538 /* generate the checksum. */
3539 xfs_dinode_calc_crc(mp, dip);
3544 * Non-blocking flush of dirty inode metadata into the backing buffer.
3546 * The caller must have a reference to the inode and hold the cluster buffer
3547 * locked. The function will walk across all the inodes on the cluster buffer it
3548 * can find and lock without blocking, and flush them to the cluster buffer.
3550 * On successful flushing of at least one inode, the caller must write out the
3551 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3552 * the caller needs to release the buffer. On failure, the filesystem will be
3553 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3560 struct xfs_mount *mp = bp->b_mount;
3561 struct xfs_log_item *lip, *n;
3562 struct xfs_inode *ip;
3563 struct xfs_inode_log_item *iip;
3568 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3569 * can remove itself from the list.
3571 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3572 iip = (struct xfs_inode_log_item *)lip;
3573 ip = iip->ili_inode;
3576 * Quick and dirty check to avoid locks if possible.
3578 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3580 if (xfs_ipincount(ip))
3584 * The inode is still attached to the buffer, which means it is
3585 * dirty but reclaim might try to grab it. Check carefully for
3586 * that, and grab the ilock while still holding the i_flags_lock
3587 * to guarantee reclaim will not be able to reclaim this inode
3588 * once we drop the i_flags_lock.
3590 spin_lock(&ip->i_flags_lock);
3591 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3592 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3593 spin_unlock(&ip->i_flags_lock);
3598 * ILOCK will pin the inode against reclaim and prevent
3599 * concurrent transactions modifying the inode while we are
3600 * flushing the inode. If we get the lock, set the flushing
3601 * state before we drop the i_flags_lock.
3603 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3604 spin_unlock(&ip->i_flags_lock);
3607 __xfs_iflags_set(ip, XFS_IFLUSHING);
3608 spin_unlock(&ip->i_flags_lock);
3611 * Abort flushing this inode if we are shut down because the
3612 * inode may not currently be in the AIL. This can occur when
3613 * log I/O failure unpins the inode without inserting into the
3614 * AIL, leaving a dirty/unpinned inode attached to the buffer
3615 * that otherwise looks like it should be flushed.
3617 if (XFS_FORCED_SHUTDOWN(mp)) {
3618 xfs_iunpin_wait(ip);
3619 xfs_iflush_abort(ip);
3620 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3625 /* don't block waiting on a log force to unpin dirty inodes */
3626 if (xfs_ipincount(ip)) {
3627 xfs_iflags_clear(ip, XFS_IFLUSHING);
3628 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3632 if (!xfs_inode_clean(ip))
3633 error = xfs_iflush(ip, bp);
3635 xfs_iflags_clear(ip, XFS_IFLUSHING);
3636 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3643 bp->b_flags |= XBF_ASYNC;
3644 xfs_buf_ioend_fail(bp);
3645 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3652 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3653 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3658 /* Release an inode. */
3661 struct xfs_inode *ip)
3663 trace_xfs_irele(ip, _RET_IP_);
3668 * Ensure all commited transactions touching the inode are written to the log.
3671 xfs_log_force_inode(
3672 struct xfs_inode *ip)
3676 xfs_ilock(ip, XFS_ILOCK_SHARED);
3677 if (xfs_ipincount(ip))
3678 seq = ip->i_itemp->ili_commit_seq;
3679 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3683 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3687 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3688 * abide vfs locking order (lowest pointer value goes first) and breaking the
3689 * layout leases before proceeding. The loop is needed because we cannot call
3690 * the blocking break_layout() with the iolocks held, and therefore have to
3691 * back out both locks.
3694 xfs_iolock_two_inodes_and_break_layout(
3704 /* Wait to break both inodes' layouts before we start locking. */
3705 error = break_layout(src, true);
3709 error = break_layout(dest, true);
3714 /* Lock one inode and make sure nobody got in and leased it. */
3716 error = break_layout(src, false);
3719 if (error == -EWOULDBLOCK)
3727 /* Lock the other inode and make sure nobody got in and leased it. */
3728 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3729 error = break_layout(dest, false);
3733 if (error == -EWOULDBLOCK)
3742 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3747 struct xfs_inode *ip1,
3748 struct xfs_inode *ip2)
3752 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3755 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3756 VFS_I(ip2)->i_mapping);
3760 /* Unlock both inodes to allow IO and mmap activity. */
3762 xfs_iunlock2_io_mmap(
3763 struct xfs_inode *ip1,
3764 struct xfs_inode *ip2)
3766 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3767 VFS_I(ip2)->i_mapping);
3768 inode_unlock(VFS_I(ip2));
3770 inode_unlock(VFS_I(ip1));