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_is_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_has_realtime(ip->i_mount))
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) && xfs_has_grpid(mp)) {
841 inode_fsuid_set(inode, mnt_userns);
842 inode->i_gid = dir->i_gid;
843 inode->i_mode = mode;
845 inode_init_owner(mnt_userns, inode, dir, mode);
849 * If the group ID of the new file does not match the effective group
850 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
851 * (and only if the irix_sgid_inherit compatibility variable is set).
853 if (irix_sgid_inherit &&
854 (inode->i_mode & S_ISGID) &&
855 !in_group_p(i_gid_into_mnt(mnt_userns, inode)))
856 inode->i_mode &= ~S_ISGID;
859 ip->i_df.if_nextents = 0;
860 ASSERT(ip->i_nblocks == 0);
862 tv = current_time(inode);
870 if (xfs_has_v3inodes(mp)) {
871 inode_set_iversion(inode, 1);
872 ip->i_cowextsize = 0;
876 flags = XFS_ILOG_CORE;
877 switch (mode & S_IFMT) {
882 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
883 flags |= XFS_ILOG_DEV;
887 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
888 xfs_inode_inherit_flags(ip, pip);
889 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
890 xfs_inode_inherit_flags2(ip, pip);
893 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
894 ip->i_df.if_bytes = 0;
895 ip->i_df.if_u1.if_root = NULL;
902 * If we need to create attributes immediately after allocating the
903 * inode, initialise an empty attribute fork right now. We use the
904 * default fork offset for attributes here as we don't know exactly what
905 * size or how many attributes we might be adding. We can do this
906 * safely here because we know the data fork is completely empty and
907 * this saves us from needing to run a separate transaction to set the
908 * fork offset in the immediate future.
910 if (init_xattrs && xfs_has_attr(mp)) {
911 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
912 ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0);
916 * Log the new values stuffed into the inode.
918 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
919 xfs_trans_log_inode(tp, ip, flags);
921 /* now that we have an i_mode we can setup the inode structure */
929 * Decrement the link count on an inode & log the change. If this causes the
930 * link count to go to zero, move the inode to AGI unlinked list so that it can
931 * be freed when the last active reference goes away via xfs_inactive().
933 static int /* error */
938 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
940 drop_nlink(VFS_I(ip));
941 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
943 if (VFS_I(ip)->i_nlink)
946 return xfs_iunlink(tp, ip);
950 * Increment the link count on an inode & log the change.
957 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
959 inc_nlink(VFS_I(ip));
960 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
965 struct user_namespace *mnt_userns,
967 struct xfs_name *name,
973 int is_dir = S_ISDIR(mode);
974 struct xfs_mount *mp = dp->i_mount;
975 struct xfs_inode *ip = NULL;
976 struct xfs_trans *tp = NULL;
978 bool unlock_dp_on_error = false;
980 struct xfs_dquot *udqp = NULL;
981 struct xfs_dquot *gdqp = NULL;
982 struct xfs_dquot *pdqp = NULL;
983 struct xfs_trans_res *tres;
987 trace_xfs_create(dp, name);
989 if (xfs_is_shutdown(mp))
992 prid = xfs_get_initial_prid(dp);
995 * Make sure that we have allocated dquot(s) on disk.
997 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns),
998 mapped_fsgid(mnt_userns), prid,
999 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1000 &udqp, &gdqp, &pdqp);
1005 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1006 tres = &M_RES(mp)->tr_mkdir;
1008 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1009 tres = &M_RES(mp)->tr_create;
1013 * Initially assume that the file does not exist and
1014 * reserve the resources for that case. If that is not
1015 * the case we'll drop the one we have and get a more
1016 * appropriate transaction later.
1018 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1020 if (error == -ENOSPC) {
1021 /* flush outstanding delalloc blocks and retry */
1022 xfs_flush_inodes(mp);
1023 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1027 goto out_release_dquots;
1029 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1030 unlock_dp_on_error = true;
1032 error = xfs_iext_count_may_overflow(dp, XFS_DATA_FORK,
1033 XFS_IEXT_DIR_MANIP_CNT(mp));
1035 goto out_trans_cancel;
1038 * A newly created regular or special file just has one directory
1039 * entry pointing to them, but a directory also the "." entry
1040 * pointing to itself.
1042 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1044 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1045 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1047 goto out_trans_cancel;
1050 * Now we join the directory inode to the transaction. We do not do it
1051 * earlier because xfs_dialloc might commit the previous transaction
1052 * (and release all the locks). An error from here on will result in
1053 * the transaction cancel unlocking dp so don't do it explicitly in the
1056 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1057 unlock_dp_on_error = false;
1059 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1060 resblks - XFS_IALLOC_SPACE_RES(mp));
1062 ASSERT(error != -ENOSPC);
1063 goto out_trans_cancel;
1065 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1066 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1069 error = xfs_dir_init(tp, ip, dp);
1071 goto out_trans_cancel;
1073 xfs_bumplink(tp, dp);
1077 * If this is a synchronous mount, make sure that the
1078 * create transaction goes to disk before returning to
1081 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1082 xfs_trans_set_sync(tp);
1085 * Attach the dquot(s) to the inodes and modify them incore.
1086 * These ids of the inode couldn't have changed since the new
1087 * inode has been locked ever since it was created.
1089 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1091 error = xfs_trans_commit(tp);
1093 goto out_release_inode;
1095 xfs_qm_dqrele(udqp);
1096 xfs_qm_dqrele(gdqp);
1097 xfs_qm_dqrele(pdqp);
1103 xfs_trans_cancel(tp);
1106 * Wait until after the current transaction is aborted to finish the
1107 * setup of the inode and release the inode. This prevents recursive
1108 * transactions and deadlocks from xfs_inactive.
1111 xfs_finish_inode_setup(ip);
1115 xfs_qm_dqrele(udqp);
1116 xfs_qm_dqrele(gdqp);
1117 xfs_qm_dqrele(pdqp);
1119 if (unlock_dp_on_error)
1120 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1126 struct user_namespace *mnt_userns,
1127 struct xfs_inode *dp,
1129 struct xfs_inode **ipp)
1131 struct xfs_mount *mp = dp->i_mount;
1132 struct xfs_inode *ip = NULL;
1133 struct xfs_trans *tp = NULL;
1136 struct xfs_dquot *udqp = NULL;
1137 struct xfs_dquot *gdqp = NULL;
1138 struct xfs_dquot *pdqp = NULL;
1139 struct xfs_trans_res *tres;
1143 if (xfs_is_shutdown(mp))
1146 prid = xfs_get_initial_prid(dp);
1149 * Make sure that we have allocated dquot(s) on disk.
1151 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns),
1152 mapped_fsgid(mnt_userns), prid,
1153 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1154 &udqp, &gdqp, &pdqp);
1158 resblks = XFS_IALLOC_SPACE_RES(mp);
1159 tres = &M_RES(mp)->tr_create_tmpfile;
1161 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1164 goto out_release_dquots;
1166 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1168 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1169 0, 0, prid, false, &ip);
1171 goto out_trans_cancel;
1173 if (xfs_has_wsync(mp))
1174 xfs_trans_set_sync(tp);
1177 * Attach the dquot(s) to the inodes and modify them incore.
1178 * These ids of the inode couldn't have changed since the new
1179 * inode has been locked ever since it was created.
1181 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1183 error = xfs_iunlink(tp, ip);
1185 goto out_trans_cancel;
1187 error = xfs_trans_commit(tp);
1189 goto out_release_inode;
1191 xfs_qm_dqrele(udqp);
1192 xfs_qm_dqrele(gdqp);
1193 xfs_qm_dqrele(pdqp);
1199 xfs_trans_cancel(tp);
1202 * Wait until after the current transaction is aborted to finish the
1203 * setup of the inode and release the inode. This prevents recursive
1204 * transactions and deadlocks from xfs_inactive.
1207 xfs_finish_inode_setup(ip);
1211 xfs_qm_dqrele(udqp);
1212 xfs_qm_dqrele(gdqp);
1213 xfs_qm_dqrele(pdqp);
1222 struct xfs_name *target_name)
1224 xfs_mount_t *mp = tdp->i_mount;
1229 trace_xfs_link(tdp, target_name);
1231 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1233 if (xfs_is_shutdown(mp))
1236 error = xfs_qm_dqattach(sip);
1240 error = xfs_qm_dqattach(tdp);
1244 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1245 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1246 if (error == -ENOSPC) {
1248 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1253 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1255 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1256 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1258 error = xfs_iext_count_may_overflow(tdp, XFS_DATA_FORK,
1259 XFS_IEXT_DIR_MANIP_CNT(mp));
1264 * If we are using project inheritance, we only allow hard link
1265 * creation in our tree when the project IDs are the same; else
1266 * the tree quota mechanism could be circumvented.
1268 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1269 tdp->i_projid != sip->i_projid)) {
1275 error = xfs_dir_canenter(tp, tdp, target_name);
1281 * Handle initial link state of O_TMPFILE inode
1283 if (VFS_I(sip)->i_nlink == 0) {
1284 struct xfs_perag *pag;
1286 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1287 error = xfs_iunlink_remove(tp, pag, sip);
1293 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1297 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1298 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1300 xfs_bumplink(tp, sip);
1303 * If this is a synchronous mount, make sure that the
1304 * link transaction goes to disk before returning to
1307 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1308 xfs_trans_set_sync(tp);
1310 return xfs_trans_commit(tp);
1313 xfs_trans_cancel(tp);
1318 /* Clear the reflink flag and the cowblocks tag if possible. */
1320 xfs_itruncate_clear_reflink_flags(
1321 struct xfs_inode *ip)
1323 struct xfs_ifork *dfork;
1324 struct xfs_ifork *cfork;
1326 if (!xfs_is_reflink_inode(ip))
1328 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1329 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1330 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1331 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1332 if (cfork->if_bytes == 0)
1333 xfs_inode_clear_cowblocks_tag(ip);
1337 * Free up the underlying blocks past new_size. The new size must be smaller
1338 * than the current size. This routine can be used both for the attribute and
1339 * data fork, and does not modify the inode size, which is left to the caller.
1341 * The transaction passed to this routine must have made a permanent log
1342 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1343 * given transaction and start new ones, so make sure everything involved in
1344 * the transaction is tidy before calling here. Some transaction will be
1345 * returned to the caller to be committed. The incoming transaction must
1346 * already include the inode, and both inode locks must be held exclusively.
1347 * The inode must also be "held" within the transaction. On return the inode
1348 * will be "held" within the returned transaction. This routine does NOT
1349 * require any disk space to be reserved for it within the transaction.
1351 * If we get an error, we must return with the inode locked and linked into the
1352 * current transaction. This keeps things simple for the higher level code,
1353 * because it always knows that the inode is locked and held in the transaction
1354 * that returns to it whether errors occur or not. We don't mark the inode
1355 * dirty on error so that transactions can be easily aborted if possible.
1358 xfs_itruncate_extents_flags(
1359 struct xfs_trans **tpp,
1360 struct xfs_inode *ip,
1362 xfs_fsize_t new_size,
1365 struct xfs_mount *mp = ip->i_mount;
1366 struct xfs_trans *tp = *tpp;
1367 xfs_fileoff_t first_unmap_block;
1368 xfs_filblks_t unmap_len;
1371 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1372 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1373 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1374 ASSERT(new_size <= XFS_ISIZE(ip));
1375 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1376 ASSERT(ip->i_itemp != NULL);
1377 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1378 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1380 trace_xfs_itruncate_extents_start(ip, new_size);
1382 flags |= xfs_bmapi_aflag(whichfork);
1385 * Since it is possible for space to become allocated beyond
1386 * the end of the file (in a crash where the space is allocated
1387 * but the inode size is not yet updated), simply remove any
1388 * blocks which show up between the new EOF and the maximum
1389 * possible file size.
1391 * We have to free all the blocks to the bmbt maximum offset, even if
1392 * the page cache can't scale that far.
1394 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1395 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1396 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1400 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1401 while (unmap_len > 0) {
1402 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1403 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1404 flags, XFS_ITRUNC_MAX_EXTENTS);
1408 /* free the just unmapped extents */
1409 error = xfs_defer_finish(&tp);
1414 if (whichfork == XFS_DATA_FORK) {
1415 /* Remove all pending CoW reservations. */
1416 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1417 first_unmap_block, XFS_MAX_FILEOFF, true);
1421 xfs_itruncate_clear_reflink_flags(ip);
1425 * Always re-log the inode so that our permanent transaction can keep
1426 * on rolling it forward in the log.
1428 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1430 trace_xfs_itruncate_extents_end(ip, new_size);
1441 xfs_mount_t *mp = ip->i_mount;
1444 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1447 /* If this is a read-only mount, don't do this (would generate I/O) */
1448 if (xfs_is_readonly(mp))
1451 if (!xfs_is_shutdown(mp)) {
1455 * If we previously truncated this file and removed old data
1456 * in the process, we want to initiate "early" writeout on
1457 * the last close. This is an attempt to combat the notorious
1458 * NULL files problem which is particularly noticeable from a
1459 * truncate down, buffered (re-)write (delalloc), followed by
1460 * a crash. What we are effectively doing here is
1461 * significantly reducing the time window where we'd otherwise
1462 * be exposed to that problem.
1464 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1466 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1467 if (ip->i_delayed_blks > 0) {
1468 error = filemap_flush(VFS_I(ip)->i_mapping);
1475 if (VFS_I(ip)->i_nlink == 0)
1479 * If we can't get the iolock just skip truncating the blocks past EOF
1480 * because we could deadlock with the mmap_lock otherwise. We'll get
1481 * another chance to drop them once the last reference to the inode is
1482 * dropped, so we'll never leak blocks permanently.
1484 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1487 if (xfs_can_free_eofblocks(ip, false)) {
1489 * Check if the inode is being opened, written and closed
1490 * frequently and we have delayed allocation blocks outstanding
1491 * (e.g. streaming writes from the NFS server), truncating the
1492 * blocks past EOF will cause fragmentation to occur.
1494 * In this case don't do the truncation, but we have to be
1495 * careful how we detect this case. Blocks beyond EOF show up as
1496 * i_delayed_blks even when the inode is clean, so we need to
1497 * truncate them away first before checking for a dirty release.
1498 * Hence on the first dirty close we will still remove the
1499 * speculative allocation, but after that we will leave it in
1502 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1505 error = xfs_free_eofblocks(ip);
1509 /* delalloc blocks after truncation means it really is dirty */
1510 if (ip->i_delayed_blks)
1511 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1515 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1520 * xfs_inactive_truncate
1522 * Called to perform a truncate when an inode becomes unlinked.
1525 xfs_inactive_truncate(
1526 struct xfs_inode *ip)
1528 struct xfs_mount *mp = ip->i_mount;
1529 struct xfs_trans *tp;
1532 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1534 ASSERT(xfs_is_shutdown(mp));
1537 xfs_ilock(ip, XFS_ILOCK_EXCL);
1538 xfs_trans_ijoin(tp, ip, 0);
1541 * Log the inode size first to prevent stale data exposure in the event
1542 * of a system crash before the truncate completes. See the related
1543 * comment in xfs_vn_setattr_size() for details.
1545 ip->i_disk_size = 0;
1546 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1548 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1550 goto error_trans_cancel;
1552 ASSERT(ip->i_df.if_nextents == 0);
1554 error = xfs_trans_commit(tp);
1558 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1562 xfs_trans_cancel(tp);
1564 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1569 * xfs_inactive_ifree()
1571 * Perform the inode free when an inode is unlinked.
1575 struct xfs_inode *ip)
1577 struct xfs_mount *mp = ip->i_mount;
1578 struct xfs_trans *tp;
1582 * We try to use a per-AG reservation for any block needed by the finobt
1583 * tree, but as the finobt feature predates the per-AG reservation
1584 * support a degraded file system might not have enough space for the
1585 * reservation at mount time. In that case try to dip into the reserved
1588 * Send a warning if the reservation does happen to fail, as the inode
1589 * now remains allocated and sits on the unlinked list until the fs is
1592 if (unlikely(mp->m_finobt_nores)) {
1593 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1594 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1597 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1600 if (error == -ENOSPC) {
1601 xfs_warn_ratelimited(mp,
1602 "Failed to remove inode(s) from unlinked list. "
1603 "Please free space, unmount and run xfs_repair.");
1605 ASSERT(xfs_is_shutdown(mp));
1611 * We do not hold the inode locked across the entire rolling transaction
1612 * here. We only need to hold it for the first transaction that
1613 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1614 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1615 * here breaks the relationship between cluster buffer invalidation and
1616 * stale inode invalidation on cluster buffer item journal commit
1617 * completion, and can result in leaving dirty stale inodes hanging
1620 * We have no need for serialising this inode operation against other
1621 * operations - we freed the inode and hence reallocation is required
1622 * and that will serialise on reallocating the space the deferops need
1623 * to free. Hence we can unlock the inode on the first commit of
1624 * the transaction rather than roll it right through the deferops. This
1625 * avoids relogging the XFS_ISTALE inode.
1627 * We check that xfs_ifree() hasn't grown an internal transaction roll
1628 * by asserting that the inode is still locked when it returns.
1630 xfs_ilock(ip, XFS_ILOCK_EXCL);
1631 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1633 error = xfs_ifree(tp, ip);
1634 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1637 * If we fail to free the inode, shut down. The cancel
1638 * might do that, we need to make sure. Otherwise the
1639 * inode might be lost for a long time or forever.
1641 if (!xfs_is_shutdown(mp)) {
1642 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1644 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1646 xfs_trans_cancel(tp);
1651 * Credit the quota account(s). The inode is gone.
1653 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1656 * Just ignore errors at this point. There is nothing we can do except
1657 * to try to keep going. Make sure it's not a silent error.
1659 error = xfs_trans_commit(tp);
1661 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1668 * Returns true if we need to update the on-disk metadata before we can free
1669 * the memory used by this inode. Updates include freeing post-eof
1670 * preallocations; freeing COW staging extents; and marking the inode free in
1671 * the inobt if it is on the unlinked list.
1674 xfs_inode_needs_inactive(
1675 struct xfs_inode *ip)
1677 struct xfs_mount *mp = ip->i_mount;
1678 struct xfs_ifork *cow_ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1681 * If the inode is already free, then there can be nothing
1684 if (VFS_I(ip)->i_mode == 0)
1687 /* If this is a read-only mount, don't do this (would generate I/O) */
1688 if (xfs_is_readonly(mp))
1691 /* If the log isn't running, push inodes straight to reclaim. */
1692 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1695 /* Metadata inodes require explicit resource cleanup. */
1696 if (xfs_is_metadata_inode(ip))
1699 /* Want to clean out the cow blocks if there are any. */
1700 if (cow_ifp && cow_ifp->if_bytes > 0)
1703 /* Unlinked files must be freed. */
1704 if (VFS_I(ip)->i_nlink == 0)
1708 * This file isn't being freed, so check if there are post-eof blocks
1709 * to free. @force is true because we are evicting an inode from the
1710 * cache. Post-eof blocks must be freed, lest we end up with broken
1711 * free space accounting.
1713 * Note: don't bother with iolock here since lockdep complains about
1714 * acquiring it in reclaim context. We have the only reference to the
1715 * inode at this point anyways.
1717 return xfs_can_free_eofblocks(ip, true);
1723 * This is called when the vnode reference count for the vnode
1724 * goes to zero. If the file has been unlinked, then it must
1725 * now be truncated. Also, we clear all of the read-ahead state
1726 * kept for the inode here since the file is now closed.
1732 struct xfs_mount *mp;
1737 * If the inode is already free, then there can be nothing
1740 if (VFS_I(ip)->i_mode == 0) {
1741 ASSERT(ip->i_df.if_broot_bytes == 0);
1746 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1748 /* If this is a read-only mount, don't do this (would generate I/O) */
1749 if (xfs_is_readonly(mp))
1752 /* Metadata inodes require explicit resource cleanup. */
1753 if (xfs_is_metadata_inode(ip))
1756 /* Try to clean out the cow blocks if there are any. */
1757 if (xfs_inode_has_cow_data(ip))
1758 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1760 if (VFS_I(ip)->i_nlink != 0) {
1762 * force is true because we are evicting an inode from the
1763 * cache. Post-eof blocks must be freed, lest we end up with
1764 * broken free space accounting.
1766 * Note: don't bother with iolock here since lockdep complains
1767 * about acquiring it in reclaim context. We have the only
1768 * reference to the inode at this point anyways.
1770 if (xfs_can_free_eofblocks(ip, true))
1771 xfs_free_eofblocks(ip);
1776 if (S_ISREG(VFS_I(ip)->i_mode) &&
1777 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1778 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1781 error = xfs_qm_dqattach(ip);
1785 if (S_ISLNK(VFS_I(ip)->i_mode))
1786 error = xfs_inactive_symlink(ip);
1788 error = xfs_inactive_truncate(ip);
1793 * If there are attributes associated with the file then blow them away
1794 * now. The code calls a routine that recursively deconstructs the
1795 * attribute fork. If also blows away the in-core attribute fork.
1797 if (XFS_IFORK_Q(ip)) {
1798 error = xfs_attr_inactive(ip);
1804 ASSERT(ip->i_forkoff == 0);
1809 xfs_inactive_ifree(ip);
1813 * We're done making metadata updates for this inode, so we can release
1814 * the attached dquots.
1816 xfs_qm_dqdetach(ip);
1820 * In-Core Unlinked List Lookups
1821 * =============================
1823 * Every inode is supposed to be reachable from some other piece of metadata
1824 * with the exception of the root directory. Inodes with a connection to a
1825 * file descriptor but not linked from anywhere in the on-disk directory tree
1826 * are collectively known as unlinked inodes, though the filesystem itself
1827 * maintains links to these inodes so that on-disk metadata are consistent.
1829 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1830 * header contains a number of buckets that point to an inode, and each inode
1831 * record has a pointer to the next inode in the hash chain. This
1832 * singly-linked list causes scaling problems in the iunlink remove function
1833 * because we must walk that list to find the inode that points to the inode
1834 * being removed from the unlinked hash bucket list.
1836 * What if we modelled the unlinked list as a collection of records capturing
1837 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1838 * have a fast way to look up unlinked list predecessors, which avoids the
1839 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1842 * Because this is a backref cache, we ignore operational failures since the
1843 * iunlink code can fall back to the slow bucket walk. The only errors that
1844 * should bubble out are for obviously incorrect situations.
1846 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1847 * access or have otherwise provided for concurrency control.
1850 /* Capture a "X.next_unlinked = Y" relationship. */
1851 struct xfs_iunlink {
1852 struct rhash_head iu_rhash_head;
1853 xfs_agino_t iu_agino; /* X */
1854 xfs_agino_t iu_next_unlinked; /* Y */
1857 /* Unlinked list predecessor lookup hashtable construction */
1859 xfs_iunlink_obj_cmpfn(
1860 struct rhashtable_compare_arg *arg,
1863 const xfs_agino_t *key = arg->key;
1864 const struct xfs_iunlink *iu = obj;
1866 if (iu->iu_next_unlinked != *key)
1871 static const struct rhashtable_params xfs_iunlink_hash_params = {
1872 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1873 .key_len = sizeof(xfs_agino_t),
1874 .key_offset = offsetof(struct xfs_iunlink,
1876 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1877 .automatic_shrinking = true,
1878 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1882 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1883 * relation is found.
1886 xfs_iunlink_lookup_backref(
1887 struct xfs_perag *pag,
1890 struct xfs_iunlink *iu;
1892 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1893 xfs_iunlink_hash_params);
1894 return iu ? iu->iu_agino : NULLAGINO;
1898 * Take ownership of an iunlink cache entry and insert it into the hash table.
1899 * If successful, the entry will be owned by the cache; if not, it is freed.
1900 * Either way, the caller does not own @iu after this call.
1903 xfs_iunlink_insert_backref(
1904 struct xfs_perag *pag,
1905 struct xfs_iunlink *iu)
1909 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1910 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1912 * Fail loudly if there already was an entry because that's a sign of
1913 * corruption of in-memory data. Also fail loudly if we see an error
1914 * code we didn't anticipate from the rhashtable code. Currently we
1915 * only anticipate ENOMEM.
1918 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1922 * Absorb any runtime errors that aren't a result of corruption because
1923 * this is a cache and we can always fall back to bucket list scanning.
1925 if (error != 0 && error != -EEXIST)
1930 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1932 xfs_iunlink_add_backref(
1933 struct xfs_perag *pag,
1934 xfs_agino_t prev_agino,
1935 xfs_agino_t this_agino)
1937 struct xfs_iunlink *iu;
1939 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
1942 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
1943 iu->iu_agino = prev_agino;
1944 iu->iu_next_unlinked = this_agino;
1946 return xfs_iunlink_insert_backref(pag, iu);
1950 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
1951 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
1952 * wasn't any such entry then we don't bother.
1955 xfs_iunlink_change_backref(
1956 struct xfs_perag *pag,
1958 xfs_agino_t next_unlinked)
1960 struct xfs_iunlink *iu;
1963 /* Look up the old entry; if there wasn't one then exit. */
1964 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1965 xfs_iunlink_hash_params);
1970 * Remove the entry. This shouldn't ever return an error, but if we
1971 * couldn't remove the old entry we don't want to add it again to the
1972 * hash table, and if the entry disappeared on us then someone's
1973 * violated the locking rules and we need to fail loudly. Either way
1974 * we cannot remove the inode because internal state is or would have
1977 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
1978 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1982 /* If there is no new next entry just free our item and return. */
1983 if (next_unlinked == NULLAGINO) {
1988 /* Update the entry and re-add it to the hash table. */
1989 iu->iu_next_unlinked = next_unlinked;
1990 return xfs_iunlink_insert_backref(pag, iu);
1993 /* Set up the in-core predecessor structures. */
1996 struct xfs_perag *pag)
1998 return rhashtable_init(&pag->pagi_unlinked_hash,
1999 &xfs_iunlink_hash_params);
2002 /* Free the in-core predecessor structures. */
2004 xfs_iunlink_free_item(
2008 struct xfs_iunlink *iu = ptr;
2009 bool *freed_anything = arg;
2011 *freed_anything = true;
2016 xfs_iunlink_destroy(
2017 struct xfs_perag *pag)
2019 bool freed_anything = false;
2021 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
2022 xfs_iunlink_free_item, &freed_anything);
2024 ASSERT(freed_anything == false || xfs_is_shutdown(pag->pag_mount));
2028 * Point the AGI unlinked bucket at an inode and log the results. The caller
2029 * is responsible for validating the old value.
2032 xfs_iunlink_update_bucket(
2033 struct xfs_trans *tp,
2034 struct xfs_perag *pag,
2035 struct xfs_buf *agibp,
2036 unsigned int bucket_index,
2037 xfs_agino_t new_agino)
2039 struct xfs_agi *agi = agibp->b_addr;
2040 xfs_agino_t old_value;
2043 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, pag->pag_agno, new_agino));
2045 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2046 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
2047 old_value, new_agino);
2050 * We should never find the head of the list already set to the value
2051 * passed in because either we're adding or removing ourselves from the
2054 if (old_value == new_agino) {
2055 xfs_buf_mark_corrupt(agibp);
2056 return -EFSCORRUPTED;
2059 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2060 offset = offsetof(struct xfs_agi, agi_unlinked) +
2061 (sizeof(xfs_agino_t) * bucket_index);
2062 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2066 /* Set an on-disk inode's next_unlinked pointer. */
2068 xfs_iunlink_update_dinode(
2069 struct xfs_trans *tp,
2070 struct xfs_perag *pag,
2072 struct xfs_buf *ibp,
2073 struct xfs_dinode *dip,
2074 struct xfs_imap *imap,
2075 xfs_agino_t next_agino)
2077 struct xfs_mount *mp = tp->t_mountp;
2080 ASSERT(xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino));
2082 trace_xfs_iunlink_update_dinode(mp, pag->pag_agno, agino,
2083 be32_to_cpu(dip->di_next_unlinked), next_agino);
2085 dip->di_next_unlinked = cpu_to_be32(next_agino);
2086 offset = imap->im_boffset +
2087 offsetof(struct xfs_dinode, di_next_unlinked);
2089 /* need to recalc the inode CRC if appropriate */
2090 xfs_dinode_calc_crc(mp, dip);
2091 xfs_trans_inode_buf(tp, ibp);
2092 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2095 /* Set an in-core inode's unlinked pointer and return the old value. */
2097 xfs_iunlink_update_inode(
2098 struct xfs_trans *tp,
2099 struct xfs_inode *ip,
2100 struct xfs_perag *pag,
2101 xfs_agino_t next_agino,
2102 xfs_agino_t *old_next_agino)
2104 struct xfs_mount *mp = tp->t_mountp;
2105 struct xfs_dinode *dip;
2106 struct xfs_buf *ibp;
2107 xfs_agino_t old_value;
2110 ASSERT(xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino));
2112 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &ibp);
2115 dip = xfs_buf_offset(ibp, ip->i_imap.im_boffset);
2117 /* Make sure the old pointer isn't garbage. */
2118 old_value = be32_to_cpu(dip->di_next_unlinked);
2119 if (!xfs_verify_agino_or_null(mp, pag->pag_agno, old_value)) {
2120 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2121 sizeof(*dip), __this_address);
2122 error = -EFSCORRUPTED;
2127 * Since we're updating a linked list, we should never find that the
2128 * current pointer is the same as the new value, unless we're
2129 * terminating the list.
2131 *old_next_agino = old_value;
2132 if (old_value == next_agino) {
2133 if (next_agino != NULLAGINO) {
2134 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2135 dip, sizeof(*dip), __this_address);
2136 error = -EFSCORRUPTED;
2141 /* Ok, update the new pointer. */
2142 xfs_iunlink_update_dinode(tp, pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
2143 ibp, dip, &ip->i_imap, next_agino);
2146 xfs_trans_brelse(tp, ibp);
2151 * This is called when the inode's link count has gone to 0 or we are creating
2152 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2154 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2155 * list when the inode is freed.
2159 struct xfs_trans *tp,
2160 struct xfs_inode *ip)
2162 struct xfs_mount *mp = tp->t_mountp;
2163 struct xfs_perag *pag;
2164 struct xfs_agi *agi;
2165 struct xfs_buf *agibp;
2166 xfs_agino_t next_agino;
2167 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2168 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2171 ASSERT(VFS_I(ip)->i_nlink == 0);
2172 ASSERT(VFS_I(ip)->i_mode != 0);
2173 trace_xfs_iunlink(ip);
2175 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2177 /* Get the agi buffer first. It ensures lock ordering on the list. */
2178 error = xfs_read_agi(mp, tp, pag->pag_agno, &agibp);
2181 agi = agibp->b_addr;
2184 * Get the index into the agi hash table for the list this inode will
2185 * go on. Make sure the pointer isn't garbage and that this inode
2186 * isn't already on the list.
2188 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2189 if (next_agino == agino ||
2190 !xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino)) {
2191 xfs_buf_mark_corrupt(agibp);
2192 error = -EFSCORRUPTED;
2196 if (next_agino != NULLAGINO) {
2197 xfs_agino_t old_agino;
2200 * There is already another inode in the bucket, so point this
2201 * inode to the current head of the list.
2203 error = xfs_iunlink_update_inode(tp, ip, pag, next_agino,
2207 ASSERT(old_agino == NULLAGINO);
2210 * agino has been unlinked, add a backref from the next inode
2213 error = xfs_iunlink_add_backref(pag, agino, next_agino);
2218 /* Point the head of the list to point to this inode. */
2219 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
2225 /* Return the imap, dinode pointer, and buffer for an inode. */
2227 xfs_iunlink_map_ino(
2228 struct xfs_trans *tp,
2229 xfs_agnumber_t agno,
2231 struct xfs_imap *imap,
2232 struct xfs_dinode **dipp,
2233 struct xfs_buf **bpp)
2235 struct xfs_mount *mp = tp->t_mountp;
2239 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2241 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2246 error = xfs_imap_to_bp(mp, tp, imap, bpp);
2248 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2253 *dipp = xfs_buf_offset(*bpp, imap->im_boffset);
2258 * Walk the unlinked chain from @head_agino until we find the inode that
2259 * points to @target_agino. Return the inode number, map, dinode pointer,
2260 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2262 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2263 * @agino, @imap, @dipp, and @bpp are all output parameters.
2265 * Do not call this function if @target_agino is the head of the list.
2268 xfs_iunlink_map_prev(
2269 struct xfs_trans *tp,
2270 struct xfs_perag *pag,
2271 xfs_agino_t head_agino,
2272 xfs_agino_t target_agino,
2274 struct xfs_imap *imap,
2275 struct xfs_dinode **dipp,
2276 struct xfs_buf **bpp)
2278 struct xfs_mount *mp = tp->t_mountp;
2279 xfs_agino_t next_agino;
2282 ASSERT(head_agino != target_agino);
2285 /* See if our backref cache can find it faster. */
2286 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2287 if (*agino != NULLAGINO) {
2288 error = xfs_iunlink_map_ino(tp, pag->pag_agno, *agino, imap,
2293 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2297 * If we get here the cache contents were corrupt, so drop the
2298 * buffer and fall back to walking the bucket list.
2300 xfs_trans_brelse(tp, *bpp);
2305 trace_xfs_iunlink_map_prev_fallback(mp, pag->pag_agno);
2307 /* Otherwise, walk the entire bucket until we find it. */
2308 next_agino = head_agino;
2309 while (next_agino != target_agino) {
2310 xfs_agino_t unlinked_agino;
2313 xfs_trans_brelse(tp, *bpp);
2315 *agino = next_agino;
2316 error = xfs_iunlink_map_ino(tp, pag->pag_agno, next_agino, imap,
2321 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2323 * Make sure this pointer is valid and isn't an obvious
2326 if (!xfs_verify_agino(mp, pag->pag_agno, unlinked_agino) ||
2327 next_agino == unlinked_agino) {
2328 XFS_CORRUPTION_ERROR(__func__,
2329 XFS_ERRLEVEL_LOW, mp,
2330 *dipp, sizeof(**dipp));
2331 error = -EFSCORRUPTED;
2334 next_agino = unlinked_agino;
2341 * Pull the on-disk inode from the AGI unlinked list.
2345 struct xfs_trans *tp,
2346 struct xfs_perag *pag,
2347 struct xfs_inode *ip)
2349 struct xfs_mount *mp = tp->t_mountp;
2350 struct xfs_agi *agi;
2351 struct xfs_buf *agibp;
2352 struct xfs_buf *last_ibp;
2353 struct xfs_dinode *last_dip = NULL;
2354 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2355 xfs_agino_t next_agino;
2356 xfs_agino_t head_agino;
2357 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2360 trace_xfs_iunlink_remove(ip);
2362 /* Get the agi buffer first. It ensures lock ordering on the list. */
2363 error = xfs_read_agi(mp, tp, pag->pag_agno, &agibp);
2366 agi = agibp->b_addr;
2369 * Get the index into the agi hash table for the list this inode will
2370 * go on. Make sure the head pointer isn't garbage.
2372 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2373 if (!xfs_verify_agino(mp, pag->pag_agno, head_agino)) {
2374 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2376 return -EFSCORRUPTED;
2380 * Set our inode's next_unlinked pointer to NULL and then return
2381 * the old pointer value so that we can update whatever was previous
2382 * to us in the list to point to whatever was next in the list.
2384 error = xfs_iunlink_update_inode(tp, ip, pag, NULLAGINO, &next_agino);
2389 * If there was a backref pointing from the next inode back to this
2390 * one, remove it because we've removed this inode from the list.
2392 * Later, if this inode was in the middle of the list we'll update
2393 * this inode's backref to point from the next inode.
2395 if (next_agino != NULLAGINO) {
2396 error = xfs_iunlink_change_backref(pag, next_agino, NULLAGINO);
2401 if (head_agino != agino) {
2402 struct xfs_imap imap;
2403 xfs_agino_t prev_agino;
2405 /* We need to search the list for the inode being freed. */
2406 error = xfs_iunlink_map_prev(tp, pag, head_agino, agino,
2407 &prev_agino, &imap, &last_dip, &last_ibp);
2411 /* Point the previous inode on the list to the next inode. */
2412 xfs_iunlink_update_dinode(tp, pag, prev_agino, last_ibp,
2413 last_dip, &imap, next_agino);
2416 * Now we deal with the backref for this inode. If this inode
2417 * pointed at a real inode, change the backref that pointed to
2418 * us to point to our old next. If this inode was the end of
2419 * the list, delete the backref that pointed to us. Note that
2420 * change_backref takes care of deleting the backref if
2421 * next_agino is NULLAGINO.
2423 return xfs_iunlink_change_backref(agibp->b_pag, agino,
2427 /* Point the head of the list to the next unlinked inode. */
2428 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2433 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2434 * mark it stale. We should only find clean inodes in this lookup that aren't
2438 xfs_ifree_mark_inode_stale(
2439 struct xfs_perag *pag,
2440 struct xfs_inode *free_ip,
2443 struct xfs_mount *mp = pag->pag_mount;
2444 struct xfs_inode_log_item *iip;
2445 struct xfs_inode *ip;
2449 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2451 /* Inode not in memory, nothing to do */
2458 * because this is an RCU protected lookup, we could find a recently
2459 * freed or even reallocated inode during the lookup. We need to check
2460 * under the i_flags_lock for a valid inode here. Skip it if it is not
2461 * valid, the wrong inode or stale.
2463 spin_lock(&ip->i_flags_lock);
2464 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2465 goto out_iflags_unlock;
2468 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2469 * other inodes that we did not find in the list attached to the buffer
2470 * and are not already marked stale. If we can't lock it, back off and
2473 if (ip != free_ip) {
2474 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2475 spin_unlock(&ip->i_flags_lock);
2481 ip->i_flags |= XFS_ISTALE;
2484 * If the inode is flushing, it is already attached to the buffer. All
2485 * we needed to do here is mark the inode stale so buffer IO completion
2486 * will remove it from the AIL.
2489 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2490 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2491 ASSERT(iip->ili_last_fields);
2496 * Inodes not attached to the buffer can be released immediately.
2497 * Everything else has to go through xfs_iflush_abort() on journal
2498 * commit as the flock synchronises removal of the inode from the
2499 * cluster buffer against inode reclaim.
2501 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2504 __xfs_iflags_set(ip, XFS_IFLUSHING);
2505 spin_unlock(&ip->i_flags_lock);
2508 /* we have a dirty inode in memory that has not yet been flushed. */
2509 spin_lock(&iip->ili_lock);
2510 iip->ili_last_fields = iip->ili_fields;
2511 iip->ili_fields = 0;
2512 iip->ili_fsync_fields = 0;
2513 spin_unlock(&iip->ili_lock);
2514 ASSERT(iip->ili_last_fields);
2517 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2522 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2524 spin_unlock(&ip->i_flags_lock);
2529 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2530 * inodes that are in memory - they all must be marked stale and attached to
2531 * the cluster buffer.
2535 struct xfs_trans *tp,
2536 struct xfs_perag *pag,
2537 struct xfs_inode *free_ip,
2538 struct xfs_icluster *xic)
2540 struct xfs_mount *mp = free_ip->i_mount;
2541 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2544 xfs_ino_t inum = xic->first_ino;
2550 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2552 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2554 * The allocation bitmap tells us which inodes of the chunk were
2555 * physically allocated. Skip the cluster if an inode falls into
2558 ioffset = inum - xic->first_ino;
2559 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2560 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2564 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2565 XFS_INO_TO_AGBNO(mp, inum));
2568 * We obtain and lock the backing buffer first in the process
2569 * here to ensure dirty inodes attached to the buffer remain in
2570 * the flushing state while we mark them stale.
2572 * If we scan the in-memory inodes first, then buffer IO can
2573 * complete before we get a lock on it, and hence we may fail
2574 * to mark all the active inodes on the buffer stale.
2576 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2577 mp->m_bsize * igeo->blocks_per_cluster,
2583 * This buffer may not have been correctly initialised as we
2584 * didn't read it from disk. That's not important because we are
2585 * only using to mark the buffer as stale in the log, and to
2586 * attach stale cached inodes on it. That means it will never be
2587 * dispatched for IO. If it is, we want to know about it, and we
2588 * want it to fail. We can acheive this by adding a write
2589 * verifier to the buffer.
2591 bp->b_ops = &xfs_inode_buf_ops;
2594 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2595 * too. This requires lookups, and will skip inodes that we've
2596 * already marked XFS_ISTALE.
2598 for (i = 0; i < igeo->inodes_per_cluster; i++)
2599 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2601 xfs_trans_stale_inode_buf(tp, bp);
2602 xfs_trans_binval(tp, bp);
2608 * This is called to return an inode to the inode free list.
2609 * The inode should already be truncated to 0 length and have
2610 * no pages associated with it. This routine also assumes that
2611 * the inode is already a part of the transaction.
2613 * The on-disk copy of the inode will have been added to the list
2614 * of unlinked inodes in the AGI. We need to remove the inode from
2615 * that list atomically with respect to freeing it here.
2619 struct xfs_trans *tp,
2620 struct xfs_inode *ip)
2622 struct xfs_mount *mp = ip->i_mount;
2623 struct xfs_perag *pag;
2624 struct xfs_icluster xic = { 0 };
2625 struct xfs_inode_log_item *iip = ip->i_itemp;
2628 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2629 ASSERT(VFS_I(ip)->i_nlink == 0);
2630 ASSERT(ip->i_df.if_nextents == 0);
2631 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2632 ASSERT(ip->i_nblocks == 0);
2634 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2637 * Pull the on-disk inode from the AGI unlinked list.
2639 error = xfs_iunlink_remove(tp, pag, ip);
2643 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2648 * Free any local-format data sitting around before we reset the
2649 * data fork to extents format. Note that the attr fork data has
2650 * already been freed by xfs_attr_inactive.
2652 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2653 kmem_free(ip->i_df.if_u1.if_data);
2654 ip->i_df.if_u1.if_data = NULL;
2655 ip->i_df.if_bytes = 0;
2658 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2660 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2661 ip->i_forkoff = 0; /* mark the attr fork not in use */
2662 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2663 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2664 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2666 /* Don't attempt to replay owner changes for a deleted inode */
2667 spin_lock(&iip->ili_lock);
2668 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2669 spin_unlock(&iip->ili_lock);
2672 * Bump the generation count so no one will be confused
2673 * by reincarnations of this inode.
2675 VFS_I(ip)->i_generation++;
2676 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2679 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2686 * This is called to unpin an inode. The caller must have the inode locked
2687 * in at least shared mode so that the buffer cannot be subsequently pinned
2688 * once someone is waiting for it to be unpinned.
2692 struct xfs_inode *ip)
2694 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2696 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2698 /* Give the log a push to start the unpinning I/O */
2699 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2705 struct xfs_inode *ip)
2707 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2708 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2713 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2714 if (xfs_ipincount(ip))
2716 } while (xfs_ipincount(ip));
2717 finish_wait(wq, &wait.wq_entry);
2722 struct xfs_inode *ip)
2724 if (xfs_ipincount(ip))
2725 __xfs_iunpin_wait(ip);
2729 * Removing an inode from the namespace involves removing the directory entry
2730 * and dropping the link count on the inode. Removing the directory entry can
2731 * result in locking an AGF (directory blocks were freed) and removing a link
2732 * count can result in placing the inode on an unlinked list which results in
2735 * The big problem here is that we have an ordering constraint on AGF and AGI
2736 * locking - inode allocation locks the AGI, then can allocate a new extent for
2737 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2738 * removes the inode from the unlinked list, requiring that we lock the AGI
2739 * first, and then freeing the inode can result in an inode chunk being freed
2740 * and hence freeing disk space requiring that we lock an AGF.
2742 * Hence the ordering that is imposed by other parts of the code is AGI before
2743 * AGF. This means we cannot remove the directory entry before we drop the inode
2744 * reference count and put it on the unlinked list as this results in a lock
2745 * order of AGF then AGI, and this can deadlock against inode allocation and
2746 * freeing. Therefore we must drop the link counts before we remove the
2749 * This is still safe from a transactional point of view - it is not until we
2750 * get to xfs_defer_finish() that we have the possibility of multiple
2751 * transactions in this operation. Hence as long as we remove the directory
2752 * entry and drop the link count in the first transaction of the remove
2753 * operation, there are no transactional constraints on the ordering here.
2758 struct xfs_name *name,
2761 xfs_mount_t *mp = dp->i_mount;
2762 xfs_trans_t *tp = NULL;
2763 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2767 trace_xfs_remove(dp, name);
2769 if (xfs_is_shutdown(mp))
2772 error = xfs_qm_dqattach(dp);
2776 error = xfs_qm_dqattach(ip);
2781 * We try to get the real space reservation first,
2782 * allowing for directory btree deletion(s) implying
2783 * possible bmap insert(s). If we can't get the space
2784 * reservation then we use 0 instead, and avoid the bmap
2785 * btree insert(s) in the directory code by, if the bmap
2786 * insert tries to happen, instead trimming the LAST
2787 * block from the directory.
2789 resblks = XFS_REMOVE_SPACE_RES(mp);
2790 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2791 if (error == -ENOSPC) {
2793 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2797 ASSERT(error != -ENOSPC);
2801 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2803 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2804 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2807 * If we're removing a directory perform some additional validation.
2810 ASSERT(VFS_I(ip)->i_nlink >= 2);
2811 if (VFS_I(ip)->i_nlink != 2) {
2813 goto out_trans_cancel;
2815 if (!xfs_dir_isempty(ip)) {
2817 goto out_trans_cancel;
2820 /* Drop the link from ip's "..". */
2821 error = xfs_droplink(tp, dp);
2823 goto out_trans_cancel;
2825 /* Drop the "." link from ip to self. */
2826 error = xfs_droplink(tp, ip);
2828 goto out_trans_cancel;
2831 * Point the unlinked child directory's ".." entry to the root
2832 * directory to eliminate back-references to inodes that may
2833 * get freed before the child directory is closed. If the fs
2834 * gets shrunk, this can lead to dirent inode validation errors.
2836 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2837 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2838 tp->t_mountp->m_sb.sb_rootino, 0);
2844 * When removing a non-directory we need to log the parent
2845 * inode here. For a directory this is done implicitly
2846 * by the xfs_droplink call for the ".." entry.
2848 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2850 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2852 /* Drop the link from dp to ip. */
2853 error = xfs_droplink(tp, ip);
2855 goto out_trans_cancel;
2857 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2859 ASSERT(error != -ENOENT);
2860 goto out_trans_cancel;
2864 * If this is a synchronous mount, make sure that the
2865 * remove transaction goes to disk before returning to
2868 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2869 xfs_trans_set_sync(tp);
2871 error = xfs_trans_commit(tp);
2875 if (is_dir && xfs_inode_is_filestream(ip))
2876 xfs_filestream_deassociate(ip);
2881 xfs_trans_cancel(tp);
2887 * Enter all inodes for a rename transaction into a sorted array.
2889 #define __XFS_SORT_INODES 5
2891 xfs_sort_for_rename(
2892 struct xfs_inode *dp1, /* in: old (source) directory inode */
2893 struct xfs_inode *dp2, /* in: new (target) directory inode */
2894 struct xfs_inode *ip1, /* in: inode of old entry */
2895 struct xfs_inode *ip2, /* in: inode of new entry */
2896 struct xfs_inode *wip, /* in: whiteout inode */
2897 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2898 int *num_inodes) /* in/out: inodes in array */
2902 ASSERT(*num_inodes == __XFS_SORT_INODES);
2903 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2906 * i_tab contains a list of pointers to inodes. We initialize
2907 * the table here & we'll sort it. We will then use it to
2908 * order the acquisition of the inode locks.
2910 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2923 * Sort the elements via bubble sort. (Remember, there are at
2924 * most 5 elements to sort, so this is adequate.)
2926 for (i = 0; i < *num_inodes; i++) {
2927 for (j = 1; j < *num_inodes; j++) {
2928 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2929 struct xfs_inode *temp = i_tab[j];
2930 i_tab[j] = i_tab[j-1];
2939 struct xfs_trans *tp)
2942 * If this is a synchronous mount, make sure that the rename transaction
2943 * goes to disk before returning to the user.
2945 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2946 xfs_trans_set_sync(tp);
2948 return xfs_trans_commit(tp);
2952 * xfs_cross_rename()
2954 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2958 struct xfs_trans *tp,
2959 struct xfs_inode *dp1,
2960 struct xfs_name *name1,
2961 struct xfs_inode *ip1,
2962 struct xfs_inode *dp2,
2963 struct xfs_name *name2,
2964 struct xfs_inode *ip2,
2972 /* Swap inode number for dirent in first parent */
2973 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2975 goto out_trans_abort;
2977 /* Swap inode number for dirent in second parent */
2978 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2980 goto out_trans_abort;
2983 * If we're renaming one or more directories across different parents,
2984 * update the respective ".." entries (and link counts) to match the new
2988 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2990 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2991 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2992 dp1->i_ino, spaceres);
2994 goto out_trans_abort;
2996 /* transfer ip2 ".." reference to dp1 */
2997 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2998 error = xfs_droplink(tp, dp2);
3000 goto out_trans_abort;
3001 xfs_bumplink(tp, dp1);
3005 * Although ip1 isn't changed here, userspace needs
3006 * to be warned about the change, so that applications
3007 * relying on it (like backup ones), will properly
3010 ip1_flags |= XFS_ICHGTIME_CHG;
3011 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3014 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
3015 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
3016 dp2->i_ino, spaceres);
3018 goto out_trans_abort;
3020 /* transfer ip1 ".." reference to dp2 */
3021 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
3022 error = xfs_droplink(tp, dp1);
3024 goto out_trans_abort;
3025 xfs_bumplink(tp, dp2);
3029 * Although ip2 isn't changed here, userspace needs
3030 * to be warned about the change, so that applications
3031 * relying on it (like backup ones), will properly
3034 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3035 ip2_flags |= XFS_ICHGTIME_CHG;
3040 xfs_trans_ichgtime(tp, ip1, ip1_flags);
3041 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
3044 xfs_trans_ichgtime(tp, ip2, ip2_flags);
3045 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
3048 xfs_trans_ichgtime(tp, dp2, dp2_flags);
3049 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
3051 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3052 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3053 return xfs_finish_rename(tp);
3056 xfs_trans_cancel(tp);
3061 * xfs_rename_alloc_whiteout()
3063 * Return a referenced, unlinked, unlocked inode that can be used as a
3064 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3065 * crash between allocating the inode and linking it into the rename transaction
3066 * recovery will free the inode and we won't leak it.
3069 xfs_rename_alloc_whiteout(
3070 struct user_namespace *mnt_userns,
3071 struct xfs_inode *dp,
3072 struct xfs_inode **wip)
3074 struct xfs_inode *tmpfile;
3077 error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
3083 * Prepare the tmpfile inode as if it were created through the VFS.
3084 * Complete the inode setup and flag it as linkable. nlink is already
3085 * zero, so we can skip the drop_nlink.
3087 xfs_setup_iops(tmpfile);
3088 xfs_finish_inode_setup(tmpfile);
3089 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3100 struct user_namespace *mnt_userns,
3101 struct xfs_inode *src_dp,
3102 struct xfs_name *src_name,
3103 struct xfs_inode *src_ip,
3104 struct xfs_inode *target_dp,
3105 struct xfs_name *target_name,
3106 struct xfs_inode *target_ip,
3109 struct xfs_mount *mp = src_dp->i_mount;
3110 struct xfs_trans *tp;
3111 struct xfs_inode *wip = NULL; /* whiteout inode */
3112 struct xfs_inode *inodes[__XFS_SORT_INODES];
3114 int num_inodes = __XFS_SORT_INODES;
3115 bool new_parent = (src_dp != target_dp);
3116 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3120 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3122 if ((flags & RENAME_EXCHANGE) && !target_ip)
3126 * If we are doing a whiteout operation, allocate the whiteout inode
3127 * we will be placing at the target and ensure the type is set
3130 if (flags & RENAME_WHITEOUT) {
3131 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3132 error = xfs_rename_alloc_whiteout(mnt_userns, target_dp, &wip);
3136 /* setup target dirent info as whiteout */
3137 src_name->type = XFS_DIR3_FT_CHRDEV;
3140 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3141 inodes, &num_inodes);
3143 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3144 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3145 if (error == -ENOSPC) {
3147 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3151 goto out_release_wip;
3154 * Attach the dquots to the inodes
3156 error = xfs_qm_vop_rename_dqattach(inodes);
3158 goto out_trans_cancel;
3161 * Lock all the participating inodes. Depending upon whether
3162 * the target_name exists in the target directory, and
3163 * whether the target directory is the same as the source
3164 * directory, we can lock from 2 to 4 inodes.
3166 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3169 * Join all the inodes to the transaction. From this point on,
3170 * we can rely on either trans_commit or trans_cancel to unlock
3173 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3175 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3176 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3178 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3180 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3183 * If we are using project inheritance, we only allow renames
3184 * into our tree when the project IDs are the same; else the
3185 * tree quota mechanism would be circumvented.
3187 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
3188 target_dp->i_projid != src_ip->i_projid)) {
3190 goto out_trans_cancel;
3193 /* RENAME_EXCHANGE is unique from here on. */
3194 if (flags & RENAME_EXCHANGE)
3195 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3196 target_dp, target_name, target_ip,
3200 * Check for expected errors before we dirty the transaction
3201 * so we can return an error without a transaction abort.
3203 * Extent count overflow check:
3205 * From the perspective of src_dp, a rename operation is essentially a
3206 * directory entry remove operation. Hence the only place where we check
3207 * for extent count overflow for src_dp is in
3208 * xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns
3209 * -ENOSPC when it detects a possible extent count overflow and in
3210 * response, the higher layers of directory handling code do the
3212 * 1. Data/Free blocks: XFS lets these blocks linger until a
3213 * future remove operation removes them.
3214 * 2. Dabtree blocks: XFS swaps the blocks with the last block in the
3215 * Leaf space and unmaps the last block.
3217 * For target_dp, there are two cases depending on whether the
3218 * destination directory entry exists or not.
3220 * When destination directory entry does not exist (i.e. target_ip ==
3221 * NULL), extent count overflow check is performed only when transaction
3222 * has a non-zero sized space reservation associated with it. With a
3223 * zero-sized space reservation, XFS allows a rename operation to
3224 * continue only when the directory has sufficient free space in its
3225 * data/leaf/free space blocks to hold the new entry.
3227 * When destination directory entry exists (i.e. target_ip != NULL), all
3228 * we need to do is change the inode number associated with the already
3229 * existing entry. Hence there is no need to perform an extent count
3232 if (target_ip == NULL) {
3234 * If there's no space reservation, check the entry will
3235 * fit before actually inserting it.
3238 error = xfs_dir_canenter(tp, target_dp, target_name);
3240 goto out_trans_cancel;
3242 error = xfs_iext_count_may_overflow(target_dp,
3244 XFS_IEXT_DIR_MANIP_CNT(mp));
3246 goto out_trans_cancel;
3250 * If target exists and it's a directory, check that whether
3251 * it can be destroyed.
3253 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3254 (!xfs_dir_isempty(target_ip) ||
3255 (VFS_I(target_ip)->i_nlink > 2))) {
3257 goto out_trans_cancel;
3262 * Lock the AGI buffers we need to handle bumping the nlink of the
3263 * whiteout inode off the unlinked list and to handle dropping the
3264 * nlink of the target inode. Per locking order rules, do this in
3265 * increasing AG order and before directory block allocation tries to
3266 * grab AGFs because we grab AGIs before AGFs.
3268 * The (vfs) caller must ensure that if src is a directory then
3269 * target_ip is either null or an empty directory.
3271 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3272 if (inodes[i] == wip ||
3273 (inodes[i] == target_ip &&
3274 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3276 xfs_agnumber_t agno;
3278 agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
3279 error = xfs_read_agi(mp, tp, agno, &bp);
3281 goto out_trans_cancel;
3286 * Directory entry creation below may acquire the AGF. Remove
3287 * the whiteout from the unlinked list first to preserve correct
3288 * AGI/AGF locking order. This dirties the transaction so failures
3289 * after this point will abort and log recovery will clean up the
3292 * For whiteouts, we need to bump the link count on the whiteout
3293 * inode. After this point, we have a real link, clear the tmpfile
3294 * state flag from the inode so it doesn't accidentally get misused
3298 struct xfs_perag *pag;
3300 ASSERT(VFS_I(wip)->i_nlink == 0);
3302 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
3303 error = xfs_iunlink_remove(tp, pag, wip);
3306 goto out_trans_cancel;
3308 xfs_bumplink(tp, wip);
3309 VFS_I(wip)->i_state &= ~I_LINKABLE;
3313 * Set up the target.
3315 if (target_ip == NULL) {
3317 * If target does not exist and the rename crosses
3318 * directories, adjust the target directory link count
3319 * to account for the ".." reference from the new entry.
3321 error = xfs_dir_createname(tp, target_dp, target_name,
3322 src_ip->i_ino, spaceres);
3324 goto out_trans_cancel;
3326 xfs_trans_ichgtime(tp, target_dp,
3327 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3329 if (new_parent && src_is_directory) {
3330 xfs_bumplink(tp, target_dp);
3332 } else { /* target_ip != NULL */
3334 * Link the source inode under the target name.
3335 * If the source inode is a directory and we are moving
3336 * it across directories, its ".." entry will be
3337 * inconsistent until we replace that down below.
3339 * In case there is already an entry with the same
3340 * name at the destination directory, remove it first.
3342 error = xfs_dir_replace(tp, target_dp, target_name,
3343 src_ip->i_ino, spaceres);
3345 goto out_trans_cancel;
3347 xfs_trans_ichgtime(tp, target_dp,
3348 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3351 * Decrement the link count on the target since the target
3352 * dir no longer points to it.
3354 error = xfs_droplink(tp, target_ip);
3356 goto out_trans_cancel;
3358 if (src_is_directory) {
3360 * Drop the link from the old "." entry.
3362 error = xfs_droplink(tp, target_ip);
3364 goto out_trans_cancel;
3366 } /* target_ip != NULL */
3369 * Remove the source.
3371 if (new_parent && src_is_directory) {
3373 * Rewrite the ".." entry to point to the new
3376 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3377 target_dp->i_ino, spaceres);
3378 ASSERT(error != -EEXIST);
3380 goto out_trans_cancel;
3384 * We always want to hit the ctime on the source inode.
3386 * This isn't strictly required by the standards since the source
3387 * inode isn't really being changed, but old unix file systems did
3388 * it and some incremental backup programs won't work without it.
3390 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3391 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3394 * Adjust the link count on src_dp. This is necessary when
3395 * renaming a directory, either within one parent when
3396 * the target existed, or across two parent directories.
3398 if (src_is_directory && (new_parent || target_ip != NULL)) {
3401 * Decrement link count on src_directory since the
3402 * entry that's moved no longer points to it.
3404 error = xfs_droplink(tp, src_dp);
3406 goto out_trans_cancel;
3410 * For whiteouts, we only need to update the source dirent with the
3411 * inode number of the whiteout inode rather than removing it
3415 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3419 * NOTE: We don't need to check for extent count overflow here
3420 * because the dir remove name code will leave the dir block in
3421 * place if the extent count would overflow.
3423 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3428 goto out_trans_cancel;
3430 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3431 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3433 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3435 error = xfs_finish_rename(tp);
3441 xfs_trans_cancel(tp);
3450 struct xfs_inode *ip,
3453 struct xfs_inode_log_item *iip = ip->i_itemp;
3454 struct xfs_dinode *dip;
3455 struct xfs_mount *mp = ip->i_mount;
3458 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3459 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3460 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3461 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3462 ASSERT(iip->ili_item.li_buf == bp);
3464 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3467 * We don't flush the inode if any of the following checks fail, but we
3468 * do still update the log item and attach to the backing buffer as if
3469 * the flush happened. This is a formality to facilitate predictable
3470 * error handling as the caller will shutdown and fail the buffer.
3472 error = -EFSCORRUPTED;
3473 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3474 mp, XFS_ERRTAG_IFLUSH_1)) {
3475 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3476 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3477 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3480 if (S_ISREG(VFS_I(ip)->i_mode)) {
3482 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3483 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3484 mp, XFS_ERRTAG_IFLUSH_3)) {
3485 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3486 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3487 __func__, ip->i_ino, ip);
3490 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3492 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3493 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3494 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3495 mp, XFS_ERRTAG_IFLUSH_4)) {
3496 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3497 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3498 __func__, ip->i_ino, ip);
3502 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3503 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3504 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3505 "%s: detected corrupt incore inode %Lu, "
3506 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3507 __func__, ip->i_ino,
3508 ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3512 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3513 mp, XFS_ERRTAG_IFLUSH_6)) {
3514 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3515 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3516 __func__, ip->i_ino, ip->i_forkoff, ip);
3521 * Inode item log recovery for v2 inodes are dependent on the flushiter
3522 * count for correct sequencing. We bump the flush iteration count so
3523 * we can detect flushes which postdate a log record during recovery.
3524 * This is redundant as we now log every change and hence this can't
3525 * happen but we need to still do it to ensure backwards compatibility
3526 * with old kernels that predate logging all inode changes.
3528 if (!xfs_has_v3inodes(mp))
3532 * If there are inline format data / attr forks attached to this inode,
3533 * make sure they are not corrupt.
3535 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3536 xfs_ifork_verify_local_data(ip))
3538 if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3539 xfs_ifork_verify_local_attr(ip))
3543 * Copy the dirty parts of the inode into the on-disk inode. We always
3544 * copy out the core of the inode, because if the inode is dirty at all
3547 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3549 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3550 if (!xfs_has_v3inodes(mp)) {
3551 if (ip->i_flushiter == DI_MAX_FLUSH)
3552 ip->i_flushiter = 0;
3555 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3556 if (XFS_IFORK_Q(ip))
3557 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3560 * We've recorded everything logged in the inode, so we'd like to clear
3561 * the ili_fields bits so we don't log and flush things unnecessarily.
3562 * However, we can't stop logging all this information until the data
3563 * we've copied into the disk buffer is written to disk. If we did we
3564 * might overwrite the copy of the inode in the log with all the data
3565 * after re-logging only part of it, and in the face of a crash we
3566 * wouldn't have all the data we need to recover.
3568 * What we do is move the bits to the ili_last_fields field. When
3569 * logging the inode, these bits are moved back to the ili_fields field.
3570 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3571 * we know that the information those bits represent is permanently on
3572 * disk. As long as the flush completes before the inode is logged
3573 * again, then both ili_fields and ili_last_fields will be cleared.
3577 spin_lock(&iip->ili_lock);
3578 iip->ili_last_fields = iip->ili_fields;
3579 iip->ili_fields = 0;
3580 iip->ili_fsync_fields = 0;
3581 spin_unlock(&iip->ili_lock);
3584 * Store the current LSN of the inode so that we can tell whether the
3585 * item has moved in the AIL from xfs_buf_inode_iodone().
3587 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3588 &iip->ili_item.li_lsn);
3590 /* generate the checksum. */
3591 xfs_dinode_calc_crc(mp, dip);
3596 * Non-blocking flush of dirty inode metadata into the backing buffer.
3598 * The caller must have a reference to the inode and hold the cluster buffer
3599 * locked. The function will walk across all the inodes on the cluster buffer it
3600 * can find and lock without blocking, and flush them to the cluster buffer.
3602 * On successful flushing of at least one inode, the caller must write out the
3603 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3604 * the caller needs to release the buffer. On failure, the filesystem will be
3605 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3612 struct xfs_mount *mp = bp->b_mount;
3613 struct xfs_log_item *lip, *n;
3614 struct xfs_inode *ip;
3615 struct xfs_inode_log_item *iip;
3620 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3621 * can remove itself from the list.
3623 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3624 iip = (struct xfs_inode_log_item *)lip;
3625 ip = iip->ili_inode;
3628 * Quick and dirty check to avoid locks if possible.
3630 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3632 if (xfs_ipincount(ip))
3636 * The inode is still attached to the buffer, which means it is
3637 * dirty but reclaim might try to grab it. Check carefully for
3638 * that, and grab the ilock while still holding the i_flags_lock
3639 * to guarantee reclaim will not be able to reclaim this inode
3640 * once we drop the i_flags_lock.
3642 spin_lock(&ip->i_flags_lock);
3643 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3644 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3645 spin_unlock(&ip->i_flags_lock);
3650 * ILOCK will pin the inode against reclaim and prevent
3651 * concurrent transactions modifying the inode while we are
3652 * flushing the inode. If we get the lock, set the flushing
3653 * state before we drop the i_flags_lock.
3655 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3656 spin_unlock(&ip->i_flags_lock);
3659 __xfs_iflags_set(ip, XFS_IFLUSHING);
3660 spin_unlock(&ip->i_flags_lock);
3663 * Abort flushing this inode if we are shut down because the
3664 * inode may not currently be in the AIL. This can occur when
3665 * log I/O failure unpins the inode without inserting into the
3666 * AIL, leaving a dirty/unpinned inode attached to the buffer
3667 * that otherwise looks like it should be flushed.
3669 if (xfs_is_shutdown(mp)) {
3670 xfs_iunpin_wait(ip);
3671 xfs_iflush_abort(ip);
3672 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3677 /* don't block waiting on a log force to unpin dirty inodes */
3678 if (xfs_ipincount(ip)) {
3679 xfs_iflags_clear(ip, XFS_IFLUSHING);
3680 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3684 if (!xfs_inode_clean(ip))
3685 error = xfs_iflush(ip, bp);
3687 xfs_iflags_clear(ip, XFS_IFLUSHING);
3688 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3695 bp->b_flags |= XBF_ASYNC;
3696 xfs_buf_ioend_fail(bp);
3697 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3704 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3705 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3710 /* Release an inode. */
3713 struct xfs_inode *ip)
3715 trace_xfs_irele(ip, _RET_IP_);
3720 * Ensure all commited transactions touching the inode are written to the log.
3723 xfs_log_force_inode(
3724 struct xfs_inode *ip)
3728 xfs_ilock(ip, XFS_ILOCK_SHARED);
3729 if (xfs_ipincount(ip))
3730 seq = ip->i_itemp->ili_commit_seq;
3731 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3735 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3739 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3740 * abide vfs locking order (lowest pointer value goes first) and breaking the
3741 * layout leases before proceeding. The loop is needed because we cannot call
3742 * the blocking break_layout() with the iolocks held, and therefore have to
3743 * back out both locks.
3746 xfs_iolock_two_inodes_and_break_layout(
3756 /* Wait to break both inodes' layouts before we start locking. */
3757 error = break_layout(src, true);
3761 error = break_layout(dest, true);
3766 /* Lock one inode and make sure nobody got in and leased it. */
3768 error = break_layout(src, false);
3771 if (error == -EWOULDBLOCK)
3779 /* Lock the other inode and make sure nobody got in and leased it. */
3780 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3781 error = break_layout(dest, false);
3785 if (error == -EWOULDBLOCK)
3794 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3799 struct xfs_inode *ip1,
3800 struct xfs_inode *ip2)
3804 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3807 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3808 VFS_I(ip2)->i_mapping);
3812 /* Unlock both inodes to allow IO and mmap activity. */
3814 xfs_iunlock2_io_mmap(
3815 struct xfs_inode *ip1,
3816 struct xfs_inode *ip2)
3818 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3819 VFS_I(ip2)->i_mapping);
3820 inode_unlock(VFS_I(ip2));
3822 inode_unlock(VFS_I(ip1));