1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
6 #include <linux/iversion.h>
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans_resv.h"
15 #include "xfs_mount.h"
16 #include "xfs_defer.h"
17 #include "xfs_inode.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_ialloc.h"
26 #include "xfs_bmap_util.h"
27 #include "xfs_errortag.h"
28 #include "xfs_error.h"
29 #include "xfs_quota.h"
30 #include "xfs_filestream.h"
31 #include "xfs_trace.h"
32 #include "xfs_icache.h"
33 #include "xfs_symlink.h"
34 #include "xfs_trans_priv.h"
36 #include "xfs_bmap_btree.h"
37 #include "xfs_reflink.h"
39 kmem_zone_t *xfs_inode_zone;
42 * Used in xfs_itruncate_extents(). This is the maximum number of extents
43 * freed from a file in a single transaction.
45 #define XFS_ITRUNC_MAX_EXTENTS 2
47 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
48 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
49 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
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_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
65 return ip->i_d.di_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_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
85 a = ip->i_d.di_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 (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
116 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
117 lock_mode = XFS_ILOCK_EXCL;
118 xfs_ilock(ip, lock_mode);
123 xfs_ilock_attr_map_shared(
124 struct xfs_inode *ip)
126 uint lock_mode = XFS_ILOCK_SHARED;
128 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
129 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
136 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
137 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
138 * various combinations of the locks to be obtained.
140 * The 3 locks should always be ordered so that the IO lock is obtained first,
141 * the mmap lock second and the ilock last in order to prevent deadlock.
143 * Basic locking order:
145 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
147 * mmap_sem locking order:
149 * i_rwsem -> page lock -> mmap_sem
150 * mmap_sem -> i_mmap_lock -> page_lock
152 * The difference in mmap_sem locking order mean that we cannot hold the
153 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
154 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
155 * in get_user_pages() to map the user pages into the kernel address space for
156 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
157 * page faults already hold the mmap_sem.
159 * Hence to serialise fully against both syscall and mmap based IO, we need to
160 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
161 * taken in places where we need to invalidate the page cache in a race
162 * free manner (e.g. truncate, hole punch and other extent manipulation
170 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
173 * You can't set both SHARED and EXCL for the same lock,
174 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
175 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
177 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
178 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
179 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
180 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
181 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
182 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
183 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
185 if (lock_flags & XFS_IOLOCK_EXCL) {
186 down_write_nested(&VFS_I(ip)->i_rwsem,
187 XFS_IOLOCK_DEP(lock_flags));
188 } else if (lock_flags & XFS_IOLOCK_SHARED) {
189 down_read_nested(&VFS_I(ip)->i_rwsem,
190 XFS_IOLOCK_DEP(lock_flags));
193 if (lock_flags & XFS_MMAPLOCK_EXCL)
194 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
195 else if (lock_flags & XFS_MMAPLOCK_SHARED)
196 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
198 if (lock_flags & XFS_ILOCK_EXCL)
199 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
200 else if (lock_flags & XFS_ILOCK_SHARED)
201 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
205 * This is just like xfs_ilock(), except that the caller
206 * is guaranteed not to sleep. It returns 1 if it gets
207 * the requested locks and 0 otherwise. If the IO lock is
208 * obtained but the inode lock cannot be, then the IO lock
209 * is dropped before returning.
211 * ip -- the inode being locked
212 * lock_flags -- this parameter indicates the inode's locks to be
213 * to be locked. See the comment for xfs_ilock() for a list
221 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
224 * You can't set both SHARED and EXCL for the same lock,
225 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
226 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
228 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
229 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
230 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
231 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
232 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
233 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
234 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
236 if (lock_flags & XFS_IOLOCK_EXCL) {
237 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
239 } else if (lock_flags & XFS_IOLOCK_SHARED) {
240 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
244 if (lock_flags & XFS_MMAPLOCK_EXCL) {
245 if (!mrtryupdate(&ip->i_mmaplock))
246 goto out_undo_iolock;
247 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
248 if (!mrtryaccess(&ip->i_mmaplock))
249 goto out_undo_iolock;
252 if (lock_flags & XFS_ILOCK_EXCL) {
253 if (!mrtryupdate(&ip->i_lock))
254 goto out_undo_mmaplock;
255 } else if (lock_flags & XFS_ILOCK_SHARED) {
256 if (!mrtryaccess(&ip->i_lock))
257 goto out_undo_mmaplock;
262 if (lock_flags & XFS_MMAPLOCK_EXCL)
263 mrunlock_excl(&ip->i_mmaplock);
264 else if (lock_flags & XFS_MMAPLOCK_SHARED)
265 mrunlock_shared(&ip->i_mmaplock);
267 if (lock_flags & XFS_IOLOCK_EXCL)
268 up_write(&VFS_I(ip)->i_rwsem);
269 else if (lock_flags & XFS_IOLOCK_SHARED)
270 up_read(&VFS_I(ip)->i_rwsem);
276 * xfs_iunlock() is used to drop the inode locks acquired with
277 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
278 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
279 * that we know which locks to drop.
281 * ip -- the inode being unlocked
282 * lock_flags -- this parameter indicates the inode's locks to be
283 * to be unlocked. See the comment for xfs_ilock() for a list
284 * of valid values for this parameter.
293 * You can't set both SHARED and EXCL for the same lock,
294 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
295 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
297 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
298 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
299 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
300 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
301 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
302 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
303 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
304 ASSERT(lock_flags != 0);
306 if (lock_flags & XFS_IOLOCK_EXCL)
307 up_write(&VFS_I(ip)->i_rwsem);
308 else if (lock_flags & XFS_IOLOCK_SHARED)
309 up_read(&VFS_I(ip)->i_rwsem);
311 if (lock_flags & XFS_MMAPLOCK_EXCL)
312 mrunlock_excl(&ip->i_mmaplock);
313 else if (lock_flags & XFS_MMAPLOCK_SHARED)
314 mrunlock_shared(&ip->i_mmaplock);
316 if (lock_flags & XFS_ILOCK_EXCL)
317 mrunlock_excl(&ip->i_lock);
318 else if (lock_flags & XFS_ILOCK_SHARED)
319 mrunlock_shared(&ip->i_lock);
321 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
325 * give up write locks. the i/o lock cannot be held nested
326 * if it is being demoted.
333 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
335 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
337 if (lock_flags & XFS_ILOCK_EXCL)
338 mrdemote(&ip->i_lock);
339 if (lock_flags & XFS_MMAPLOCK_EXCL)
340 mrdemote(&ip->i_mmaplock);
341 if (lock_flags & XFS_IOLOCK_EXCL)
342 downgrade_write(&VFS_I(ip)->i_rwsem);
344 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
347 #if defined(DEBUG) || defined(XFS_WARN)
353 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
354 if (!(lock_flags & XFS_ILOCK_SHARED))
355 return !!ip->i_lock.mr_writer;
356 return rwsem_is_locked(&ip->i_lock.mr_lock);
359 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
360 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
361 return !!ip->i_mmaplock.mr_writer;
362 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
365 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
366 if (!(lock_flags & XFS_IOLOCK_SHARED))
367 return !debug_locks ||
368 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
369 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
378 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
379 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
380 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
381 * errors and warnings.
383 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
385 xfs_lockdep_subclass_ok(
388 return subclass < MAX_LOCKDEP_SUBCLASSES;
391 #define xfs_lockdep_subclass_ok(subclass) (true)
395 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
396 * value. This can be called for any type of inode lock combination, including
397 * parent locking. Care must be taken to ensure we don't overrun the subclass
398 * storage fields in the class mask we build.
401 xfs_lock_inumorder(int lock_mode, int subclass)
405 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
407 ASSERT(xfs_lockdep_subclass_ok(subclass));
409 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
410 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
411 class += subclass << XFS_IOLOCK_SHIFT;
414 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
415 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
416 class += subclass << XFS_MMAPLOCK_SHIFT;
419 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
420 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
421 class += subclass << XFS_ILOCK_SHIFT;
424 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
428 * The following routine will lock n inodes in exclusive mode. We assume the
429 * caller calls us with the inodes in i_ino order.
431 * We need to detect deadlock where an inode that we lock is in the AIL and we
432 * start waiting for another inode that is locked by a thread in a long running
433 * transaction (such as truncate). This can result in deadlock since the long
434 * running trans might need to wait for the inode we just locked in order to
435 * push the tail and free space in the log.
437 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
438 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
439 * lock more than one at a time, lockdep will report false positives saying we
440 * have violated locking orders.
444 struct xfs_inode **ips,
448 int attempts = 0, i, j, try_lock;
449 struct xfs_log_item *lp;
452 * Currently supports between 2 and 5 inodes with exclusive locking. We
453 * support an arbitrary depth of locking here, but absolute limits on
454 * inodes depend on the the type of locking and the limits placed by
455 * lockdep annotations in xfs_lock_inumorder. These are all checked by
458 ASSERT(ips && inodes >= 2 && inodes <= 5);
459 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
461 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
463 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
464 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
466 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
468 if (lock_mode & XFS_IOLOCK_EXCL) {
469 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
470 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
471 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
476 for (; i < inodes; i++) {
479 if (i && (ips[i] == ips[i - 1])) /* Already locked */
483 * If try_lock is not set yet, make sure all locked inodes are
484 * not in the AIL. If any are, set try_lock to be used later.
487 for (j = (i - 1); j >= 0 && !try_lock; j--) {
488 lp = &ips[j]->i_itemp->ili_item;
489 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
495 * If any of the previous locks we have locked is in the AIL,
496 * we must TRY to get the second and subsequent locks. If
497 * we can't get any, we must release all we have
501 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
505 /* try_lock means we have an inode locked that is in the AIL. */
507 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
511 * Unlock all previous guys and try again. xfs_iunlock will try
512 * to push the tail if the inode is in the AIL.
515 for (j = i - 1; j >= 0; j--) {
517 * Check to see if we've already unlocked this one. Not
518 * the first one going back, and the inode ptr is the
521 if (j != (i - 1) && ips[j] == ips[j + 1])
524 xfs_iunlock(ips[j], lock_mode);
527 if ((attempts % 5) == 0) {
528 delay(1); /* Don't just spin the CPU */
537 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
538 * the mmaplock or the ilock, but not more than one type at a time. If we lock
539 * more than one at a time, lockdep will report false positives saying we have
540 * violated locking orders. The iolock must be double-locked separately since
541 * we use i_rwsem for that. We now support taking one lock EXCL and the other
546 struct xfs_inode *ip0,
548 struct xfs_inode *ip1,
551 struct xfs_inode *temp;
554 struct xfs_log_item *lp;
556 ASSERT(hweight32(ip0_mode) == 1);
557 ASSERT(hweight32(ip1_mode) == 1);
558 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
559 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
560 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
561 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
563 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
565 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
566 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
567 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
569 ASSERT(ip0->i_ino != ip1->i_ino);
571 if (ip0->i_ino > ip1->i_ino) {
575 mode_temp = ip0_mode;
577 ip1_mode = mode_temp;
581 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
584 * If the first lock we have locked is in the AIL, we must TRY to get
585 * the second lock. If we can't get it, we must release the first one
588 lp = &ip0->i_itemp->ili_item;
589 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
590 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
591 xfs_iunlock(ip0, ip0_mode);
592 if ((++attempts % 5) == 0)
593 delay(1); /* Don't just spin the CPU */
597 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
603 struct xfs_inode *ip)
605 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
606 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
609 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
610 if (xfs_isiflocked(ip))
612 } while (!xfs_iflock_nowait(ip));
614 finish_wait(wq, &wait.wq_entry);
625 if (di_flags & XFS_DIFLAG_ANY) {
626 if (di_flags & XFS_DIFLAG_REALTIME)
627 flags |= FS_XFLAG_REALTIME;
628 if (di_flags & XFS_DIFLAG_PREALLOC)
629 flags |= FS_XFLAG_PREALLOC;
630 if (di_flags & XFS_DIFLAG_IMMUTABLE)
631 flags |= FS_XFLAG_IMMUTABLE;
632 if (di_flags & XFS_DIFLAG_APPEND)
633 flags |= FS_XFLAG_APPEND;
634 if (di_flags & XFS_DIFLAG_SYNC)
635 flags |= FS_XFLAG_SYNC;
636 if (di_flags & XFS_DIFLAG_NOATIME)
637 flags |= FS_XFLAG_NOATIME;
638 if (di_flags & XFS_DIFLAG_NODUMP)
639 flags |= FS_XFLAG_NODUMP;
640 if (di_flags & XFS_DIFLAG_RTINHERIT)
641 flags |= FS_XFLAG_RTINHERIT;
642 if (di_flags & XFS_DIFLAG_PROJINHERIT)
643 flags |= FS_XFLAG_PROJINHERIT;
644 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
645 flags |= FS_XFLAG_NOSYMLINKS;
646 if (di_flags & XFS_DIFLAG_EXTSIZE)
647 flags |= FS_XFLAG_EXTSIZE;
648 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
649 flags |= FS_XFLAG_EXTSZINHERIT;
650 if (di_flags & XFS_DIFLAG_NODEFRAG)
651 flags |= FS_XFLAG_NODEFRAG;
652 if (di_flags & XFS_DIFLAG_FILESTREAM)
653 flags |= FS_XFLAG_FILESTREAM;
656 if (di_flags2 & XFS_DIFLAG2_ANY) {
657 if (di_flags2 & XFS_DIFLAG2_DAX)
658 flags |= FS_XFLAG_DAX;
659 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
660 flags |= FS_XFLAG_COWEXTSIZE;
664 flags |= FS_XFLAG_HASATTR;
671 struct xfs_inode *ip)
673 struct xfs_icdinode *dic = &ip->i_d;
675 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
680 * is allowed, otherwise it has to be an exact match. If a CI match is found,
681 * ci_name->name will point to a the actual name (caller must free) or
682 * will be set to NULL if an exact match is found.
687 struct xfs_name *name,
689 struct xfs_name *ci_name)
694 trace_xfs_lookup(dp, name);
696 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
699 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
703 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
711 kmem_free(ci_name->name);
718 * Allocate an inode on disk and return a copy of its in-core version.
719 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
720 * appropriately within the inode. The uid and gid for the inode are
721 * set according to the contents of the given cred structure.
723 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
724 * has a free inode available, call xfs_iget() to obtain the in-core
725 * version of the allocated inode. Finally, fill in the inode and
726 * log its initial contents. In this case, ialloc_context would be
729 * If xfs_dialloc() does not have an available inode, it will replenish
730 * its supply by doing an allocation. Since we can only do one
731 * allocation within a transaction without deadlocks, we must commit
732 * the current transaction before returning the inode itself.
733 * In this case, therefore, we will set ialloc_context and return.
734 * The caller should then commit the current transaction, start a new
735 * transaction, and call xfs_ialloc() again to actually get the inode.
737 * To ensure that some other process does not grab the inode that
738 * was allocated during the first call to xfs_ialloc(), this routine
739 * also returns the [locked] bp pointing to the head of the freelist
740 * as ialloc_context. The caller should hold this buffer across
741 * the commit and pass it back into this routine on the second call.
743 * If we are allocating quota inodes, we do not have a parent inode
744 * to attach to or associate with (i.e. pip == NULL) because they
745 * are not linked into the directory structure - they are attached
746 * directly to the superblock - and so have no parent.
756 xfs_buf_t **ialloc_context,
759 struct xfs_mount *mp = tp->t_mountp;
764 struct timespec64 tv;
768 * Call the space management code to pick
769 * the on-disk inode to be allocated.
771 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
772 ialloc_context, &ino);
775 if (*ialloc_context || ino == NULLFSINO) {
779 ASSERT(*ialloc_context == NULL);
782 * Protect against obviously corrupt allocation btree records. Later
783 * xfs_iget checks will catch re-allocation of other active in-memory
784 * and on-disk inodes. If we don't catch reallocating the parent inode
785 * here we will deadlock in xfs_iget() so we have to do these checks
788 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
789 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
790 return -EFSCORRUPTED;
794 * Get the in-core inode with the lock held exclusively.
795 * This is because we're setting fields here we need
796 * to prevent others from looking at until we're done.
798 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
799 XFS_ILOCK_EXCL, &ip);
806 * We always convert v1 inodes to v2 now - we only support filesystems
807 * with >= v2 inode capability, so there is no reason for ever leaving
808 * an inode in v1 format.
810 if (ip->i_d.di_version == 1)
811 ip->i_d.di_version = 2;
813 inode->i_mode = mode;
814 set_nlink(inode, nlink);
815 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
816 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
817 inode->i_rdev = rdev;
818 ip->i_d.di_projid = prid;
820 if (pip && XFS_INHERIT_GID(pip)) {
821 ip->i_d.di_gid = pip->i_d.di_gid;
822 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
823 inode->i_mode |= S_ISGID;
827 * If the group ID of the new file does not match the effective group
828 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
829 * (and only if the irix_sgid_inherit compatibility variable is set).
831 if ((irix_sgid_inherit) &&
832 (inode->i_mode & S_ISGID) &&
833 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
834 inode->i_mode &= ~S_ISGID;
837 ip->i_d.di_nextents = 0;
838 ASSERT(ip->i_d.di_nblocks == 0);
840 tv = current_time(inode);
845 ip->i_d.di_extsize = 0;
846 ip->i_d.di_dmevmask = 0;
847 ip->i_d.di_dmstate = 0;
848 ip->i_d.di_flags = 0;
850 if (ip->i_d.di_version == 3) {
851 inode_set_iversion(inode, 1);
852 ip->i_d.di_flags2 = 0;
853 ip->i_d.di_cowextsize = 0;
854 ip->i_d.di_crtime = tv;
858 flags = XFS_ILOG_CORE;
859 switch (mode & S_IFMT) {
864 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
865 ip->i_df.if_flags = 0;
866 flags |= XFS_ILOG_DEV;
870 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
874 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
875 di_flags |= XFS_DIFLAG_RTINHERIT;
876 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
877 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
878 ip->i_d.di_extsize = pip->i_d.di_extsize;
880 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
881 di_flags |= XFS_DIFLAG_PROJINHERIT;
882 } else if (S_ISREG(mode)) {
883 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
884 di_flags |= XFS_DIFLAG_REALTIME;
885 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
886 di_flags |= XFS_DIFLAG_EXTSIZE;
887 ip->i_d.di_extsize = pip->i_d.di_extsize;
890 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
892 di_flags |= XFS_DIFLAG_NOATIME;
893 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
895 di_flags |= XFS_DIFLAG_NODUMP;
896 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
898 di_flags |= XFS_DIFLAG_SYNC;
899 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
900 xfs_inherit_nosymlinks)
901 di_flags |= XFS_DIFLAG_NOSYMLINKS;
902 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
903 xfs_inherit_nodefrag)
904 di_flags |= XFS_DIFLAG_NODEFRAG;
905 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
906 di_flags |= XFS_DIFLAG_FILESTREAM;
908 ip->i_d.di_flags |= di_flags;
911 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
912 pip->i_d.di_version == 3 &&
913 ip->i_d.di_version == 3) {
914 uint64_t di_flags2 = 0;
916 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
917 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
918 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
920 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
921 di_flags2 |= XFS_DIFLAG2_DAX;
923 ip->i_d.di_flags2 |= di_flags2;
927 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
928 ip->i_df.if_flags = XFS_IFEXTENTS;
929 ip->i_df.if_bytes = 0;
930 ip->i_df.if_u1.if_root = NULL;
936 * Attribute fork settings for new inode.
938 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
939 ip->i_d.di_anextents = 0;
942 * Log the new values stuffed into the inode.
944 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
945 xfs_trans_log_inode(tp, ip, flags);
947 /* now that we have an i_mode we can setup the inode structure */
955 * Allocates a new inode from disk and return a pointer to the
956 * incore copy. This routine will internally commit the current
957 * transaction and allocate a new one if the Space Manager needed
958 * to do an allocation to replenish the inode free-list.
960 * This routine is designed to be called from xfs_create and
966 xfs_trans_t **tpp, /* input: current transaction;
967 output: may be a new transaction. */
968 xfs_inode_t *dp, /* directory within whose allocate
973 prid_t prid, /* project id */
974 xfs_inode_t **ipp) /* pointer to inode; it will be
979 xfs_buf_t *ialloc_context = NULL;
985 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
988 * xfs_ialloc will return a pointer to an incore inode if
989 * the Space Manager has an available inode on the free
990 * list. Otherwise, it will do an allocation and replenish
991 * the freelist. Since we can only do one allocation per
992 * transaction without deadlocks, we will need to commit the
993 * current transaction and start a new one. We will then
994 * need to call xfs_ialloc again to get the inode.
996 * If xfs_ialloc did an allocation to replenish the freelist,
997 * it returns the bp containing the head of the freelist as
998 * ialloc_context. We will hold a lock on it across the
999 * transaction commit so that no other process can steal
1000 * the inode(s) that we've just allocated.
1002 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
1006 * Return an error if we were unable to allocate a new inode.
1007 * This should only happen if we run out of space on disk or
1008 * encounter a disk error.
1014 if (!ialloc_context && !ip) {
1020 * If the AGI buffer is non-NULL, then we were unable to get an
1021 * inode in one operation. We need to commit the current
1022 * transaction and call xfs_ialloc() again. It is guaranteed
1023 * to succeed the second time.
1025 if (ialloc_context) {
1027 * Normally, xfs_trans_commit releases all the locks.
1028 * We call bhold to hang on to the ialloc_context across
1029 * the commit. Holding this buffer prevents any other
1030 * processes from doing any allocations in this
1033 xfs_trans_bhold(tp, ialloc_context);
1036 * We want the quota changes to be associated with the next
1037 * transaction, NOT this one. So, detach the dqinfo from this
1038 * and attach it to the next transaction.
1043 dqinfo = (void *)tp->t_dqinfo;
1044 tp->t_dqinfo = NULL;
1045 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1046 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1049 code = xfs_trans_roll(&tp);
1052 * Re-attach the quota info that we detached from prev trx.
1055 tp->t_dqinfo = dqinfo;
1056 tp->t_flags |= tflags;
1060 xfs_buf_relse(ialloc_context);
1065 xfs_trans_bjoin(tp, ialloc_context);
1068 * Call ialloc again. Since we've locked out all
1069 * other allocations in this allocation group,
1070 * this call should always succeed.
1072 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1073 &ialloc_context, &ip);
1076 * If we get an error at this point, return to the caller
1077 * so that the current transaction can be aborted.
1084 ASSERT(!ialloc_context && ip);
1095 * Decrement the link count on an inode & log the change. If this causes the
1096 * link count to go to zero, move the inode to AGI unlinked list so that it can
1097 * be freed when the last active reference goes away via xfs_inactive().
1099 static int /* error */
1104 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1106 drop_nlink(VFS_I(ip));
1107 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1109 if (VFS_I(ip)->i_nlink)
1112 return xfs_iunlink(tp, ip);
1116 * Increment the link count on an inode & log the change.
1123 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1125 ASSERT(ip->i_d.di_version > 1);
1126 inc_nlink(VFS_I(ip));
1127 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1133 struct xfs_name *name,
1138 int is_dir = S_ISDIR(mode);
1139 struct xfs_mount *mp = dp->i_mount;
1140 struct xfs_inode *ip = NULL;
1141 struct xfs_trans *tp = NULL;
1143 bool unlock_dp_on_error = false;
1145 struct xfs_dquot *udqp = NULL;
1146 struct xfs_dquot *gdqp = NULL;
1147 struct xfs_dquot *pdqp = NULL;
1148 struct xfs_trans_res *tres;
1151 trace_xfs_create(dp, name);
1153 if (XFS_FORCED_SHUTDOWN(mp))
1156 prid = xfs_get_initial_prid(dp);
1159 * Make sure that we have allocated dquot(s) on disk.
1161 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1162 xfs_kgid_to_gid(current_fsgid()), prid,
1163 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1164 &udqp, &gdqp, &pdqp);
1169 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1170 tres = &M_RES(mp)->tr_mkdir;
1172 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1173 tres = &M_RES(mp)->tr_create;
1177 * Initially assume that the file does not exist and
1178 * reserve the resources for that case. If that is not
1179 * the case we'll drop the one we have and get a more
1180 * appropriate transaction later.
1182 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1183 if (error == -ENOSPC) {
1184 /* flush outstanding delalloc blocks and retry */
1185 xfs_flush_inodes(mp);
1186 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1189 goto out_release_inode;
1191 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1192 unlock_dp_on_error = true;
1195 * Reserve disk quota and the inode.
1197 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1198 pdqp, resblks, 1, 0);
1200 goto out_trans_cancel;
1203 * A newly created regular or special file just has one directory
1204 * entry pointing to them, but a directory also the "." entry
1205 * pointing to itself.
1207 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
1209 goto out_trans_cancel;
1212 * Now we join the directory inode to the transaction. We do not do it
1213 * earlier because xfs_dir_ialloc might commit the previous transaction
1214 * (and release all the locks). An error from here on will result in
1215 * the transaction cancel unlocking dp so don't do it explicitly in the
1218 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1219 unlock_dp_on_error = false;
1221 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1223 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1225 ASSERT(error != -ENOSPC);
1226 goto out_trans_cancel;
1228 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1229 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1232 error = xfs_dir_init(tp, ip, dp);
1234 goto out_trans_cancel;
1236 xfs_bumplink(tp, dp);
1240 * If this is a synchronous mount, make sure that the
1241 * create transaction goes to disk before returning to
1244 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1245 xfs_trans_set_sync(tp);
1248 * Attach the dquot(s) to the inodes and modify them incore.
1249 * These ids of the inode couldn't have changed since the new
1250 * inode has been locked ever since it was created.
1252 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1254 error = xfs_trans_commit(tp);
1256 goto out_release_inode;
1258 xfs_qm_dqrele(udqp);
1259 xfs_qm_dqrele(gdqp);
1260 xfs_qm_dqrele(pdqp);
1266 xfs_trans_cancel(tp);
1269 * Wait until after the current transaction is aborted to finish the
1270 * setup of the inode and release the inode. This prevents recursive
1271 * transactions and deadlocks from xfs_inactive.
1274 xfs_finish_inode_setup(ip);
1278 xfs_qm_dqrele(udqp);
1279 xfs_qm_dqrele(gdqp);
1280 xfs_qm_dqrele(pdqp);
1282 if (unlock_dp_on_error)
1283 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1289 struct xfs_inode *dp,
1291 struct xfs_inode **ipp)
1293 struct xfs_mount *mp = dp->i_mount;
1294 struct xfs_inode *ip = NULL;
1295 struct xfs_trans *tp = NULL;
1298 struct xfs_dquot *udqp = NULL;
1299 struct xfs_dquot *gdqp = NULL;
1300 struct xfs_dquot *pdqp = NULL;
1301 struct xfs_trans_res *tres;
1304 if (XFS_FORCED_SHUTDOWN(mp))
1307 prid = xfs_get_initial_prid(dp);
1310 * Make sure that we have allocated dquot(s) on disk.
1312 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1313 xfs_kgid_to_gid(current_fsgid()), prid,
1314 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1315 &udqp, &gdqp, &pdqp);
1319 resblks = XFS_IALLOC_SPACE_RES(mp);
1320 tres = &M_RES(mp)->tr_create_tmpfile;
1322 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1324 goto out_release_inode;
1326 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1327 pdqp, resblks, 1, 0);
1329 goto out_trans_cancel;
1331 error = xfs_dir_ialloc(&tp, dp, mode, 0, 0, prid, &ip);
1333 goto out_trans_cancel;
1335 if (mp->m_flags & XFS_MOUNT_WSYNC)
1336 xfs_trans_set_sync(tp);
1339 * Attach the dquot(s) to the inodes and modify them incore.
1340 * These ids of the inode couldn't have changed since the new
1341 * inode has been locked ever since it was created.
1343 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1345 error = xfs_iunlink(tp, ip);
1347 goto out_trans_cancel;
1349 error = xfs_trans_commit(tp);
1351 goto out_release_inode;
1353 xfs_qm_dqrele(udqp);
1354 xfs_qm_dqrele(gdqp);
1355 xfs_qm_dqrele(pdqp);
1361 xfs_trans_cancel(tp);
1364 * Wait until after the current transaction is aborted to finish the
1365 * setup of the inode and release the inode. This prevents recursive
1366 * transactions and deadlocks from xfs_inactive.
1369 xfs_finish_inode_setup(ip);
1373 xfs_qm_dqrele(udqp);
1374 xfs_qm_dqrele(gdqp);
1375 xfs_qm_dqrele(pdqp);
1384 struct xfs_name *target_name)
1386 xfs_mount_t *mp = tdp->i_mount;
1391 trace_xfs_link(tdp, target_name);
1393 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1395 if (XFS_FORCED_SHUTDOWN(mp))
1398 error = xfs_qm_dqattach(sip);
1402 error = xfs_qm_dqattach(tdp);
1406 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1407 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1408 if (error == -ENOSPC) {
1410 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1415 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1417 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1418 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1421 * If we are using project inheritance, we only allow hard link
1422 * creation in our tree when the project IDs are the same; else
1423 * the tree quota mechanism could be circumvented.
1425 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1426 tdp->i_d.di_projid != sip->i_d.di_projid)) {
1432 error = xfs_dir_canenter(tp, tdp, target_name);
1438 * Handle initial link state of O_TMPFILE inode
1440 if (VFS_I(sip)->i_nlink == 0) {
1441 error = xfs_iunlink_remove(tp, sip);
1446 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1450 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1451 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1453 xfs_bumplink(tp, sip);
1456 * If this is a synchronous mount, make sure that the
1457 * link transaction goes to disk before returning to
1460 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1461 xfs_trans_set_sync(tp);
1463 return xfs_trans_commit(tp);
1466 xfs_trans_cancel(tp);
1471 /* Clear the reflink flag and the cowblocks tag if possible. */
1473 xfs_itruncate_clear_reflink_flags(
1474 struct xfs_inode *ip)
1476 struct xfs_ifork *dfork;
1477 struct xfs_ifork *cfork;
1479 if (!xfs_is_reflink_inode(ip))
1481 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1482 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1483 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1484 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1485 if (cfork->if_bytes == 0)
1486 xfs_inode_clear_cowblocks_tag(ip);
1490 * Free up the underlying blocks past new_size. The new size must be smaller
1491 * than the current size. This routine can be used both for the attribute and
1492 * data fork, and does not modify the inode size, which is left to the caller.
1494 * The transaction passed to this routine must have made a permanent log
1495 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1496 * given transaction and start new ones, so make sure everything involved in
1497 * the transaction is tidy before calling here. Some transaction will be
1498 * returned to the caller to be committed. The incoming transaction must
1499 * already include the inode, and both inode locks must be held exclusively.
1500 * The inode must also be "held" within the transaction. On return the inode
1501 * will be "held" within the returned transaction. This routine does NOT
1502 * require any disk space to be reserved for it within the transaction.
1504 * If we get an error, we must return with the inode locked and linked into the
1505 * current transaction. This keeps things simple for the higher level code,
1506 * because it always knows that the inode is locked and held in the transaction
1507 * that returns to it whether errors occur or not. We don't mark the inode
1508 * dirty on error so that transactions can be easily aborted if possible.
1511 xfs_itruncate_extents_flags(
1512 struct xfs_trans **tpp,
1513 struct xfs_inode *ip,
1515 xfs_fsize_t new_size,
1518 struct xfs_mount *mp = ip->i_mount;
1519 struct xfs_trans *tp = *tpp;
1520 xfs_fileoff_t first_unmap_block;
1521 xfs_fileoff_t last_block;
1522 xfs_filblks_t unmap_len;
1526 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1527 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1528 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1529 ASSERT(new_size <= XFS_ISIZE(ip));
1530 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1531 ASSERT(ip->i_itemp != NULL);
1532 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1533 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1535 trace_xfs_itruncate_extents_start(ip, new_size);
1537 flags |= xfs_bmapi_aflag(whichfork);
1540 * Since it is possible for space to become allocated beyond
1541 * the end of the file (in a crash where the space is allocated
1542 * but the inode size is not yet updated), simply remove any
1543 * blocks which show up between the new EOF and the maximum
1544 * possible file size. If the first block to be removed is
1545 * beyond the maximum file size (ie it is the same as last_block),
1546 * then there is nothing to do.
1548 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1549 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1550 if (first_unmap_block == last_block)
1553 ASSERT(first_unmap_block < last_block);
1554 unmap_len = last_block - first_unmap_block + 1;
1556 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1557 error = xfs_bunmapi(tp, ip, first_unmap_block, unmap_len, flags,
1558 XFS_ITRUNC_MAX_EXTENTS, &done);
1563 * Duplicate the transaction that has the permanent
1564 * reservation and commit the old transaction.
1566 error = xfs_defer_finish(&tp);
1570 error = xfs_trans_roll_inode(&tp, ip);
1575 if (whichfork == XFS_DATA_FORK) {
1576 /* Remove all pending CoW reservations. */
1577 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1578 first_unmap_block, last_block, true);
1582 xfs_itruncate_clear_reflink_flags(ip);
1586 * Always re-log the inode so that our permanent transaction can keep
1587 * on rolling it forward in the log.
1589 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1591 trace_xfs_itruncate_extents_end(ip, new_size);
1602 xfs_mount_t *mp = ip->i_mount;
1605 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1608 /* If this is a read-only mount, don't do this (would generate I/O) */
1609 if (mp->m_flags & XFS_MOUNT_RDONLY)
1612 if (!XFS_FORCED_SHUTDOWN(mp)) {
1616 * If we previously truncated this file and removed old data
1617 * in the process, we want to initiate "early" writeout on
1618 * the last close. This is an attempt to combat the notorious
1619 * NULL files problem which is particularly noticeable from a
1620 * truncate down, buffered (re-)write (delalloc), followed by
1621 * a crash. What we are effectively doing here is
1622 * significantly reducing the time window where we'd otherwise
1623 * be exposed to that problem.
1625 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1627 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1628 if (ip->i_delayed_blks > 0) {
1629 error = filemap_flush(VFS_I(ip)->i_mapping);
1636 if (VFS_I(ip)->i_nlink == 0)
1639 if (xfs_can_free_eofblocks(ip, false)) {
1642 * Check if the inode is being opened, written and closed
1643 * frequently and we have delayed allocation blocks outstanding
1644 * (e.g. streaming writes from the NFS server), truncating the
1645 * blocks past EOF will cause fragmentation to occur.
1647 * In this case don't do the truncation, but we have to be
1648 * careful how we detect this case. Blocks beyond EOF show up as
1649 * i_delayed_blks even when the inode is clean, so we need to
1650 * truncate them away first before checking for a dirty release.
1651 * Hence on the first dirty close we will still remove the
1652 * speculative allocation, but after that we will leave it in
1655 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1658 * If we can't get the iolock just skip truncating the blocks
1659 * past EOF because we could deadlock with the mmap_sem
1660 * otherwise. We'll get another chance to drop them once the
1661 * last reference to the inode is dropped, so we'll never leak
1662 * blocks permanently.
1664 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1665 error = xfs_free_eofblocks(ip);
1666 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1671 /* delalloc blocks after truncation means it really is dirty */
1672 if (ip->i_delayed_blks)
1673 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1679 * xfs_inactive_truncate
1681 * Called to perform a truncate when an inode becomes unlinked.
1684 xfs_inactive_truncate(
1685 struct xfs_inode *ip)
1687 struct xfs_mount *mp = ip->i_mount;
1688 struct xfs_trans *tp;
1691 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1693 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1696 xfs_ilock(ip, XFS_ILOCK_EXCL);
1697 xfs_trans_ijoin(tp, ip, 0);
1700 * Log the inode size first to prevent stale data exposure in the event
1701 * of a system crash before the truncate completes. See the related
1702 * comment in xfs_vn_setattr_size() for details.
1704 ip->i_d.di_size = 0;
1705 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1707 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1709 goto error_trans_cancel;
1711 ASSERT(ip->i_d.di_nextents == 0);
1713 error = xfs_trans_commit(tp);
1717 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1721 xfs_trans_cancel(tp);
1723 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1728 * xfs_inactive_ifree()
1730 * Perform the inode free when an inode is unlinked.
1734 struct xfs_inode *ip)
1736 struct xfs_mount *mp = ip->i_mount;
1737 struct xfs_trans *tp;
1741 * We try to use a per-AG reservation for any block needed by the finobt
1742 * tree, but as the finobt feature predates the per-AG reservation
1743 * support a degraded file system might not have enough space for the
1744 * reservation at mount time. In that case try to dip into the reserved
1747 * Send a warning if the reservation does happen to fail, as the inode
1748 * now remains allocated and sits on the unlinked list until the fs is
1751 if (unlikely(mp->m_finobt_nores)) {
1752 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1753 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1756 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1759 if (error == -ENOSPC) {
1760 xfs_warn_ratelimited(mp,
1761 "Failed to remove inode(s) from unlinked list. "
1762 "Please free space, unmount and run xfs_repair.");
1764 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1769 xfs_ilock(ip, XFS_ILOCK_EXCL);
1770 xfs_trans_ijoin(tp, ip, 0);
1772 error = xfs_ifree(tp, ip);
1775 * If we fail to free the inode, shut down. The cancel
1776 * might do that, we need to make sure. Otherwise the
1777 * inode might be lost for a long time or forever.
1779 if (!XFS_FORCED_SHUTDOWN(mp)) {
1780 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1782 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1784 xfs_trans_cancel(tp);
1785 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1790 * Credit the quota account(s). The inode is gone.
1792 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1795 * Just ignore errors at this point. There is nothing we can do except
1796 * to try to keep going. Make sure it's not a silent error.
1798 error = xfs_trans_commit(tp);
1800 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1803 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1810 * This is called when the vnode reference count for the vnode
1811 * goes to zero. If the file has been unlinked, then it must
1812 * now be truncated. Also, we clear all of the read-ahead state
1813 * kept for the inode here since the file is now closed.
1819 struct xfs_mount *mp;
1824 * If the inode is already free, then there can be nothing
1827 if (VFS_I(ip)->i_mode == 0) {
1828 ASSERT(ip->i_df.if_broot_bytes == 0);
1833 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1835 /* If this is a read-only mount, don't do this (would generate I/O) */
1836 if (mp->m_flags & XFS_MOUNT_RDONLY)
1839 /* Try to clean out the cow blocks if there are any. */
1840 if (xfs_inode_has_cow_data(ip))
1841 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1843 if (VFS_I(ip)->i_nlink != 0) {
1845 * force is true because we are evicting an inode from the
1846 * cache. Post-eof blocks must be freed, lest we end up with
1847 * broken free space accounting.
1849 * Note: don't bother with iolock here since lockdep complains
1850 * about acquiring it in reclaim context. We have the only
1851 * reference to the inode at this point anyways.
1853 if (xfs_can_free_eofblocks(ip, true))
1854 xfs_free_eofblocks(ip);
1859 if (S_ISREG(VFS_I(ip)->i_mode) &&
1860 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1861 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1864 error = xfs_qm_dqattach(ip);
1868 if (S_ISLNK(VFS_I(ip)->i_mode))
1869 error = xfs_inactive_symlink(ip);
1871 error = xfs_inactive_truncate(ip);
1876 * If there are attributes associated with the file then blow them away
1877 * now. The code calls a routine that recursively deconstructs the
1878 * attribute fork. If also blows away the in-core attribute fork.
1880 if (XFS_IFORK_Q(ip)) {
1881 error = xfs_attr_inactive(ip);
1887 ASSERT(ip->i_d.di_anextents == 0);
1888 ASSERT(ip->i_d.di_forkoff == 0);
1893 error = xfs_inactive_ifree(ip);
1898 * Release the dquots held by inode, if any.
1900 xfs_qm_dqdetach(ip);
1904 * In-Core Unlinked List Lookups
1905 * =============================
1907 * Every inode is supposed to be reachable from some other piece of metadata
1908 * with the exception of the root directory. Inodes with a connection to a
1909 * file descriptor but not linked from anywhere in the on-disk directory tree
1910 * are collectively known as unlinked inodes, though the filesystem itself
1911 * maintains links to these inodes so that on-disk metadata are consistent.
1913 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1914 * header contains a number of buckets that point to an inode, and each inode
1915 * record has a pointer to the next inode in the hash chain. This
1916 * singly-linked list causes scaling problems in the iunlink remove function
1917 * because we must walk that list to find the inode that points to the inode
1918 * being removed from the unlinked hash bucket list.
1920 * What if we modelled the unlinked list as a collection of records capturing
1921 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1922 * have a fast way to look up unlinked list predecessors, which avoids the
1923 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1926 * Because this is a backref cache, we ignore operational failures since the
1927 * iunlink code can fall back to the slow bucket walk. The only errors that
1928 * should bubble out are for obviously incorrect situations.
1930 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1931 * access or have otherwise provided for concurrency control.
1934 /* Capture a "X.next_unlinked = Y" relationship. */
1935 struct xfs_iunlink {
1936 struct rhash_head iu_rhash_head;
1937 xfs_agino_t iu_agino; /* X */
1938 xfs_agino_t iu_next_unlinked; /* Y */
1941 /* Unlinked list predecessor lookup hashtable construction */
1943 xfs_iunlink_obj_cmpfn(
1944 struct rhashtable_compare_arg *arg,
1947 const xfs_agino_t *key = arg->key;
1948 const struct xfs_iunlink *iu = obj;
1950 if (iu->iu_next_unlinked != *key)
1955 static const struct rhashtable_params xfs_iunlink_hash_params = {
1956 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1957 .key_len = sizeof(xfs_agino_t),
1958 .key_offset = offsetof(struct xfs_iunlink,
1960 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1961 .automatic_shrinking = true,
1962 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1966 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1967 * relation is found.
1970 xfs_iunlink_lookup_backref(
1971 struct xfs_perag *pag,
1974 struct xfs_iunlink *iu;
1976 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1977 xfs_iunlink_hash_params);
1978 return iu ? iu->iu_agino : NULLAGINO;
1982 * Take ownership of an iunlink cache entry and insert it into the hash table.
1983 * If successful, the entry will be owned by the cache; if not, it is freed.
1984 * Either way, the caller does not own @iu after this call.
1987 xfs_iunlink_insert_backref(
1988 struct xfs_perag *pag,
1989 struct xfs_iunlink *iu)
1993 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1994 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1996 * Fail loudly if there already was an entry because that's a sign of
1997 * corruption of in-memory data. Also fail loudly if we see an error
1998 * code we didn't anticipate from the rhashtable code. Currently we
1999 * only anticipate ENOMEM.
2002 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
2006 * Absorb any runtime errors that aren't a result of corruption because
2007 * this is a cache and we can always fall back to bucket list scanning.
2009 if (error != 0 && error != -EEXIST)
2014 /* Remember that @prev_agino.next_unlinked = @this_agino. */
2016 xfs_iunlink_add_backref(
2017 struct xfs_perag *pag,
2018 xfs_agino_t prev_agino,
2019 xfs_agino_t this_agino)
2021 struct xfs_iunlink *iu;
2023 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
2026 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
2027 iu->iu_agino = prev_agino;
2028 iu->iu_next_unlinked = this_agino;
2030 return xfs_iunlink_insert_backref(pag, iu);
2034 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
2035 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
2036 * wasn't any such entry then we don't bother.
2039 xfs_iunlink_change_backref(
2040 struct xfs_perag *pag,
2042 xfs_agino_t next_unlinked)
2044 struct xfs_iunlink *iu;
2047 /* Look up the old entry; if there wasn't one then exit. */
2048 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
2049 xfs_iunlink_hash_params);
2054 * Remove the entry. This shouldn't ever return an error, but if we
2055 * couldn't remove the old entry we don't want to add it again to the
2056 * hash table, and if the entry disappeared on us then someone's
2057 * violated the locking rules and we need to fail loudly. Either way
2058 * we cannot remove the inode because internal state is or would have
2061 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
2062 &iu->iu_rhash_head, xfs_iunlink_hash_params);
2066 /* If there is no new next entry just free our item and return. */
2067 if (next_unlinked == NULLAGINO) {
2072 /* Update the entry and re-add it to the hash table. */
2073 iu->iu_next_unlinked = next_unlinked;
2074 return xfs_iunlink_insert_backref(pag, iu);
2077 /* Set up the in-core predecessor structures. */
2080 struct xfs_perag *pag)
2082 return rhashtable_init(&pag->pagi_unlinked_hash,
2083 &xfs_iunlink_hash_params);
2086 /* Free the in-core predecessor structures. */
2088 xfs_iunlink_free_item(
2092 struct xfs_iunlink *iu = ptr;
2093 bool *freed_anything = arg;
2095 *freed_anything = true;
2100 xfs_iunlink_destroy(
2101 struct xfs_perag *pag)
2103 bool freed_anything = false;
2105 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
2106 xfs_iunlink_free_item, &freed_anything);
2108 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
2112 * Point the AGI unlinked bucket at an inode and log the results. The caller
2113 * is responsible for validating the old value.
2116 xfs_iunlink_update_bucket(
2117 struct xfs_trans *tp,
2118 xfs_agnumber_t agno,
2119 struct xfs_buf *agibp,
2120 unsigned int bucket_index,
2121 xfs_agino_t new_agino)
2123 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
2124 xfs_agino_t old_value;
2127 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
2129 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2130 trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
2131 old_value, new_agino);
2134 * We should never find the head of the list already set to the value
2135 * passed in because either we're adding or removing ourselves from the
2138 if (old_value == new_agino) {
2139 xfs_buf_corruption_error(agibp);
2140 return -EFSCORRUPTED;
2143 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2144 offset = offsetof(struct xfs_agi, agi_unlinked) +
2145 (sizeof(xfs_agino_t) * bucket_index);
2146 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2150 /* Set an on-disk inode's next_unlinked pointer. */
2152 xfs_iunlink_update_dinode(
2153 struct xfs_trans *tp,
2154 xfs_agnumber_t agno,
2156 struct xfs_buf *ibp,
2157 struct xfs_dinode *dip,
2158 struct xfs_imap *imap,
2159 xfs_agino_t next_agino)
2161 struct xfs_mount *mp = tp->t_mountp;
2164 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2166 trace_xfs_iunlink_update_dinode(mp, agno, agino,
2167 be32_to_cpu(dip->di_next_unlinked), next_agino);
2169 dip->di_next_unlinked = cpu_to_be32(next_agino);
2170 offset = imap->im_boffset +
2171 offsetof(struct xfs_dinode, di_next_unlinked);
2173 /* need to recalc the inode CRC if appropriate */
2174 xfs_dinode_calc_crc(mp, dip);
2175 xfs_trans_inode_buf(tp, ibp);
2176 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2177 xfs_inobp_check(mp, ibp);
2180 /* Set an in-core inode's unlinked pointer and return the old value. */
2182 xfs_iunlink_update_inode(
2183 struct xfs_trans *tp,
2184 struct xfs_inode *ip,
2185 xfs_agnumber_t agno,
2186 xfs_agino_t next_agino,
2187 xfs_agino_t *old_next_agino)
2189 struct xfs_mount *mp = tp->t_mountp;
2190 struct xfs_dinode *dip;
2191 struct xfs_buf *ibp;
2192 xfs_agino_t old_value;
2195 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2197 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 0, 0);
2201 /* Make sure the old pointer isn't garbage. */
2202 old_value = be32_to_cpu(dip->di_next_unlinked);
2203 if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2204 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2205 sizeof(*dip), __this_address);
2206 error = -EFSCORRUPTED;
2211 * Since we're updating a linked list, we should never find that the
2212 * current pointer is the same as the new value, unless we're
2213 * terminating the list.
2215 *old_next_agino = old_value;
2216 if (old_value == next_agino) {
2217 if (next_agino != NULLAGINO) {
2218 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2219 dip, sizeof(*dip), __this_address);
2220 error = -EFSCORRUPTED;
2225 /* Ok, update the new pointer. */
2226 xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2227 ibp, dip, &ip->i_imap, next_agino);
2230 xfs_trans_brelse(tp, ibp);
2235 * This is called when the inode's link count has gone to 0 or we are creating
2236 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2238 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2239 * list when the inode is freed.
2243 struct xfs_trans *tp,
2244 struct xfs_inode *ip)
2246 struct xfs_mount *mp = tp->t_mountp;
2247 struct xfs_agi *agi;
2248 struct xfs_buf *agibp;
2249 xfs_agino_t next_agino;
2250 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2251 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2252 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2255 ASSERT(VFS_I(ip)->i_nlink == 0);
2256 ASSERT(VFS_I(ip)->i_mode != 0);
2257 trace_xfs_iunlink(ip);
2259 /* Get the agi buffer first. It ensures lock ordering on the list. */
2260 error = xfs_read_agi(mp, tp, agno, &agibp);
2263 agi = XFS_BUF_TO_AGI(agibp);
2266 * Get the index into the agi hash table for the list this inode will
2267 * go on. Make sure the pointer isn't garbage and that this inode
2268 * isn't already on the list.
2270 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2271 if (next_agino == agino ||
2272 !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2273 xfs_buf_corruption_error(agibp);
2274 return -EFSCORRUPTED;
2277 if (next_agino != NULLAGINO) {
2278 struct xfs_perag *pag;
2279 xfs_agino_t old_agino;
2282 * There is already another inode in the bucket, so point this
2283 * inode to the current head of the list.
2285 error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2289 ASSERT(old_agino == NULLAGINO);
2292 * agino has been unlinked, add a backref from the next inode
2295 pag = xfs_perag_get(mp, agno);
2296 error = xfs_iunlink_add_backref(pag, agino, next_agino);
2302 /* Point the head of the list to point to this inode. */
2303 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2306 /* Return the imap, dinode pointer, and buffer for an inode. */
2308 xfs_iunlink_map_ino(
2309 struct xfs_trans *tp,
2310 xfs_agnumber_t agno,
2312 struct xfs_imap *imap,
2313 struct xfs_dinode **dipp,
2314 struct xfs_buf **bpp)
2316 struct xfs_mount *mp = tp->t_mountp;
2320 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2322 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2327 error = xfs_imap_to_bp(mp, tp, imap, dipp, bpp, 0, 0);
2329 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2338 * Walk the unlinked chain from @head_agino until we find the inode that
2339 * points to @target_agino. Return the inode number, map, dinode pointer,
2340 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2342 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2343 * @agino, @imap, @dipp, and @bpp are all output parameters.
2345 * Do not call this function if @target_agino is the head of the list.
2348 xfs_iunlink_map_prev(
2349 struct xfs_trans *tp,
2350 xfs_agnumber_t agno,
2351 xfs_agino_t head_agino,
2352 xfs_agino_t target_agino,
2354 struct xfs_imap *imap,
2355 struct xfs_dinode **dipp,
2356 struct xfs_buf **bpp,
2357 struct xfs_perag *pag)
2359 struct xfs_mount *mp = tp->t_mountp;
2360 xfs_agino_t next_agino;
2363 ASSERT(head_agino != target_agino);
2366 /* See if our backref cache can find it faster. */
2367 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2368 if (*agino != NULLAGINO) {
2369 error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2373 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2377 * If we get here the cache contents were corrupt, so drop the
2378 * buffer and fall back to walking the bucket list.
2380 xfs_trans_brelse(tp, *bpp);
2385 trace_xfs_iunlink_map_prev_fallback(mp, agno);
2387 /* Otherwise, walk the entire bucket until we find it. */
2388 next_agino = head_agino;
2389 while (next_agino != target_agino) {
2390 xfs_agino_t unlinked_agino;
2393 xfs_trans_brelse(tp, *bpp);
2395 *agino = next_agino;
2396 error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2401 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2403 * Make sure this pointer is valid and isn't an obvious
2406 if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2407 next_agino == unlinked_agino) {
2408 XFS_CORRUPTION_ERROR(__func__,
2409 XFS_ERRLEVEL_LOW, mp,
2410 *dipp, sizeof(**dipp));
2411 error = -EFSCORRUPTED;
2414 next_agino = unlinked_agino;
2421 * Pull the on-disk inode from the AGI unlinked list.
2425 struct xfs_trans *tp,
2426 struct xfs_inode *ip)
2428 struct xfs_mount *mp = tp->t_mountp;
2429 struct xfs_agi *agi;
2430 struct xfs_buf *agibp;
2431 struct xfs_buf *last_ibp;
2432 struct xfs_dinode *last_dip = NULL;
2433 struct xfs_perag *pag = NULL;
2434 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2435 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2436 xfs_agino_t next_agino;
2437 xfs_agino_t head_agino;
2438 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2441 trace_xfs_iunlink_remove(ip);
2443 /* Get the agi buffer first. It ensures lock ordering on the list. */
2444 error = xfs_read_agi(mp, tp, agno, &agibp);
2447 agi = XFS_BUF_TO_AGI(agibp);
2450 * Get the index into the agi hash table for the list this inode will
2451 * go on. Make sure the head pointer isn't garbage.
2453 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2454 if (!xfs_verify_agino(mp, agno, head_agino)) {
2455 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2457 return -EFSCORRUPTED;
2461 * Set our inode's next_unlinked pointer to NULL and then return
2462 * the old pointer value so that we can update whatever was previous
2463 * to us in the list to point to whatever was next in the list.
2465 error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2470 * If there was a backref pointing from the next inode back to this
2471 * one, remove it because we've removed this inode from the list.
2473 * Later, if this inode was in the middle of the list we'll update
2474 * this inode's backref to point from the next inode.
2476 if (next_agino != NULLAGINO) {
2477 pag = xfs_perag_get(mp, agno);
2478 error = xfs_iunlink_change_backref(pag, next_agino,
2484 if (head_agino == agino) {
2485 /* Point the head of the list to the next unlinked inode. */
2486 error = xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2491 struct xfs_imap imap;
2492 xfs_agino_t prev_agino;
2495 pag = xfs_perag_get(mp, agno);
2497 /* We need to search the list for the inode being freed. */
2498 error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2499 &prev_agino, &imap, &last_dip, &last_ibp,
2504 /* Point the previous inode on the list to the next inode. */
2505 xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2506 last_dip, &imap, next_agino);
2509 * Now we deal with the backref for this inode. If this inode
2510 * pointed at a real inode, change the backref that pointed to
2511 * us to point to our old next. If this inode was the end of
2512 * the list, delete the backref that pointed to us. Note that
2513 * change_backref takes care of deleting the backref if
2514 * next_agino is NULLAGINO.
2516 error = xfs_iunlink_change_backref(pag, agino, next_agino);
2528 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2529 * inodes that are in memory - they all must be marked stale and attached to
2530 * the cluster buffer.
2534 xfs_inode_t *free_ip,
2536 struct xfs_icluster *xic)
2538 xfs_mount_t *mp = free_ip->i_mount;
2545 xfs_inode_log_item_t *iip;
2546 struct xfs_log_item *lip;
2547 struct xfs_perag *pag;
2548 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2551 inum = xic->first_ino;
2552 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2553 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2555 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2557 * The allocation bitmap tells us which inodes of the chunk were
2558 * physically allocated. Skip the cluster if an inode falls into
2561 ioffset = inum - xic->first_ino;
2562 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2563 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2567 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2568 XFS_INO_TO_AGBNO(mp, inum));
2571 * We obtain and lock the backing buffer first in the process
2572 * here, as we have to ensure that any dirty inode that we
2573 * can't get the flush lock on is attached to the buffer.
2574 * If we scan the in-memory inodes first, then buffer IO can
2575 * complete before we get a lock on it, and hence we may fail
2576 * to mark all the active inodes on the buffer stale.
2578 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2579 mp->m_bsize * igeo->blocks_per_cluster,
2586 * This buffer may not have been correctly initialised as we
2587 * didn't read it from disk. That's not important because we are
2588 * only using to mark the buffer as stale in the log, and to
2589 * attach stale cached inodes on it. That means it will never be
2590 * dispatched for IO. If it is, we want to know about it, and we
2591 * want it to fail. We can acheive this by adding a write
2592 * verifier to the buffer.
2594 bp->b_ops = &xfs_inode_buf_ops;
2597 * Walk the inodes already attached to the buffer and mark them
2598 * stale. These will all have the flush locks held, so an
2599 * in-memory inode walk can't lock them. By marking them all
2600 * stale first, we will not attempt to lock them in the loop
2601 * below as the XFS_ISTALE flag will be set.
2603 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
2604 if (lip->li_type == XFS_LI_INODE) {
2605 iip = (xfs_inode_log_item_t *)lip;
2606 ASSERT(iip->ili_logged == 1);
2607 lip->li_cb = xfs_istale_done;
2608 xfs_trans_ail_copy_lsn(mp->m_ail,
2609 &iip->ili_flush_lsn,
2610 &iip->ili_item.li_lsn);
2611 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2617 * For each inode in memory attempt to add it to the inode
2618 * buffer and set it up for being staled on buffer IO
2619 * completion. This is safe as we've locked out tail pushing
2620 * and flushing by locking the buffer.
2622 * We have already marked every inode that was part of a
2623 * transaction stale above, which means there is no point in
2624 * even trying to lock them.
2626 for (i = 0; i < igeo->inodes_per_cluster; i++) {
2629 ip = radix_tree_lookup(&pag->pag_ici_root,
2630 XFS_INO_TO_AGINO(mp, (inum + i)));
2632 /* Inode not in memory, nothing to do */
2639 * because this is an RCU protected lookup, we could
2640 * find a recently freed or even reallocated inode
2641 * during the lookup. We need to check under the
2642 * i_flags_lock for a valid inode here. Skip it if it
2643 * is not valid, the wrong inode or stale.
2645 spin_lock(&ip->i_flags_lock);
2646 if (ip->i_ino != inum + i ||
2647 __xfs_iflags_test(ip, XFS_ISTALE)) {
2648 spin_unlock(&ip->i_flags_lock);
2652 spin_unlock(&ip->i_flags_lock);
2655 * Don't try to lock/unlock the current inode, but we
2656 * _cannot_ skip the other inodes that we did not find
2657 * in the list attached to the buffer and are not
2658 * already marked stale. If we can't lock it, back off
2661 if (ip != free_ip) {
2662 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2669 * Check the inode number again in case we're
2670 * racing with freeing in xfs_reclaim_inode().
2671 * See the comments in that function for more
2672 * information as to why the initial check is
2675 if (ip->i_ino != inum + i) {
2676 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2684 xfs_iflags_set(ip, XFS_ISTALE);
2687 * we don't need to attach clean inodes or those only
2688 * with unlogged changes (which we throw away, anyway).
2691 if (!iip || xfs_inode_clean(ip)) {
2692 ASSERT(ip != free_ip);
2694 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2698 iip->ili_last_fields = iip->ili_fields;
2699 iip->ili_fields = 0;
2700 iip->ili_fsync_fields = 0;
2701 iip->ili_logged = 1;
2702 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2703 &iip->ili_item.li_lsn);
2705 xfs_buf_attach_iodone(bp, xfs_istale_done,
2709 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2712 xfs_trans_stale_inode_buf(tp, bp);
2713 xfs_trans_binval(tp, bp);
2721 * Free any local-format buffers sitting around before we reset to
2725 xfs_ifree_local_data(
2726 struct xfs_inode *ip,
2729 struct xfs_ifork *ifp;
2731 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2734 ifp = XFS_IFORK_PTR(ip, whichfork);
2735 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2739 * This is called to return an inode to the inode free list.
2740 * The inode should already be truncated to 0 length and have
2741 * no pages associated with it. This routine also assumes that
2742 * the inode is already a part of the transaction.
2744 * The on-disk copy of the inode will have been added to the list
2745 * of unlinked inodes in the AGI. We need to remove the inode from
2746 * that list atomically with respect to freeing it here.
2750 struct xfs_trans *tp,
2751 struct xfs_inode *ip)
2754 struct xfs_icluster xic = { 0 };
2756 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2757 ASSERT(VFS_I(ip)->i_nlink == 0);
2758 ASSERT(ip->i_d.di_nextents == 0);
2759 ASSERT(ip->i_d.di_anextents == 0);
2760 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2761 ASSERT(ip->i_d.di_nblocks == 0);
2764 * Pull the on-disk inode from the AGI unlinked list.
2766 error = xfs_iunlink_remove(tp, ip);
2770 error = xfs_difree(tp, ip->i_ino, &xic);
2774 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2775 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2777 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2778 ip->i_d.di_flags = 0;
2779 ip->i_d.di_flags2 = 0;
2780 ip->i_d.di_dmevmask = 0;
2781 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2782 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2783 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2785 /* Don't attempt to replay owner changes for a deleted inode */
2786 ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER);
2789 * Bump the generation count so no one will be confused
2790 * by reincarnations of this inode.
2792 VFS_I(ip)->i_generation++;
2793 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2796 error = xfs_ifree_cluster(ip, tp, &xic);
2802 * This is called to unpin an inode. The caller must have the inode locked
2803 * in at least shared mode so that the buffer cannot be subsequently pinned
2804 * once someone is waiting for it to be unpinned.
2808 struct xfs_inode *ip)
2810 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2812 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2814 /* Give the log a push to start the unpinning I/O */
2815 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2821 struct xfs_inode *ip)
2823 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2824 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2829 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2830 if (xfs_ipincount(ip))
2832 } while (xfs_ipincount(ip));
2833 finish_wait(wq, &wait.wq_entry);
2838 struct xfs_inode *ip)
2840 if (xfs_ipincount(ip))
2841 __xfs_iunpin_wait(ip);
2845 * Removing an inode from the namespace involves removing the directory entry
2846 * and dropping the link count on the inode. Removing the directory entry can
2847 * result in locking an AGF (directory blocks were freed) and removing a link
2848 * count can result in placing the inode on an unlinked list which results in
2851 * The big problem here is that we have an ordering constraint on AGF and AGI
2852 * locking - inode allocation locks the AGI, then can allocate a new extent for
2853 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2854 * removes the inode from the unlinked list, requiring that we lock the AGI
2855 * first, and then freeing the inode can result in an inode chunk being freed
2856 * and hence freeing disk space requiring that we lock an AGF.
2858 * Hence the ordering that is imposed by other parts of the code is AGI before
2859 * AGF. This means we cannot remove the directory entry before we drop the inode
2860 * reference count and put it on the unlinked list as this results in a lock
2861 * order of AGF then AGI, and this can deadlock against inode allocation and
2862 * freeing. Therefore we must drop the link counts before we remove the
2865 * This is still safe from a transactional point of view - it is not until we
2866 * get to xfs_defer_finish() that we have the possibility of multiple
2867 * transactions in this operation. Hence as long as we remove the directory
2868 * entry and drop the link count in the first transaction of the remove
2869 * operation, there are no transactional constraints on the ordering here.
2874 struct xfs_name *name,
2877 xfs_mount_t *mp = dp->i_mount;
2878 xfs_trans_t *tp = NULL;
2879 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2883 trace_xfs_remove(dp, name);
2885 if (XFS_FORCED_SHUTDOWN(mp))
2888 error = xfs_qm_dqattach(dp);
2892 error = xfs_qm_dqattach(ip);
2897 * We try to get the real space reservation first,
2898 * allowing for directory btree deletion(s) implying
2899 * possible bmap insert(s). If we can't get the space
2900 * reservation then we use 0 instead, and avoid the bmap
2901 * btree insert(s) in the directory code by, if the bmap
2902 * insert tries to happen, instead trimming the LAST
2903 * block from the directory.
2905 resblks = XFS_REMOVE_SPACE_RES(mp);
2906 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2907 if (error == -ENOSPC) {
2909 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2913 ASSERT(error != -ENOSPC);
2917 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2919 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2920 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2923 * If we're removing a directory perform some additional validation.
2926 ASSERT(VFS_I(ip)->i_nlink >= 2);
2927 if (VFS_I(ip)->i_nlink != 2) {
2929 goto out_trans_cancel;
2931 if (!xfs_dir_isempty(ip)) {
2933 goto out_trans_cancel;
2936 /* Drop the link from ip's "..". */
2937 error = xfs_droplink(tp, dp);
2939 goto out_trans_cancel;
2941 /* Drop the "." link from ip to self. */
2942 error = xfs_droplink(tp, ip);
2944 goto out_trans_cancel;
2947 * When removing a non-directory we need to log the parent
2948 * inode here. For a directory this is done implicitly
2949 * by the xfs_droplink call for the ".." entry.
2951 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2953 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2955 /* Drop the link from dp to ip. */
2956 error = xfs_droplink(tp, ip);
2958 goto out_trans_cancel;
2960 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2962 ASSERT(error != -ENOENT);
2963 goto out_trans_cancel;
2967 * If this is a synchronous mount, make sure that the
2968 * remove transaction goes to disk before returning to
2971 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2972 xfs_trans_set_sync(tp);
2974 error = xfs_trans_commit(tp);
2978 if (is_dir && xfs_inode_is_filestream(ip))
2979 xfs_filestream_deassociate(ip);
2984 xfs_trans_cancel(tp);
2990 * Enter all inodes for a rename transaction into a sorted array.
2992 #define __XFS_SORT_INODES 5
2994 xfs_sort_for_rename(
2995 struct xfs_inode *dp1, /* in: old (source) directory inode */
2996 struct xfs_inode *dp2, /* in: new (target) directory inode */
2997 struct xfs_inode *ip1, /* in: inode of old entry */
2998 struct xfs_inode *ip2, /* in: inode of new entry */
2999 struct xfs_inode *wip, /* in: whiteout inode */
3000 struct xfs_inode **i_tab,/* out: sorted array of inodes */
3001 int *num_inodes) /* in/out: inodes in array */
3005 ASSERT(*num_inodes == __XFS_SORT_INODES);
3006 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
3009 * i_tab contains a list of pointers to inodes. We initialize
3010 * the table here & we'll sort it. We will then use it to
3011 * order the acquisition of the inode locks.
3013 * Note that the table may contain duplicates. e.g., dp1 == dp2.
3026 * Sort the elements via bubble sort. (Remember, there are at
3027 * most 5 elements to sort, so this is adequate.)
3029 for (i = 0; i < *num_inodes; i++) {
3030 for (j = 1; j < *num_inodes; j++) {
3031 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
3032 struct xfs_inode *temp = i_tab[j];
3033 i_tab[j] = i_tab[j-1];
3042 struct xfs_trans *tp)
3045 * If this is a synchronous mount, make sure that the rename transaction
3046 * goes to disk before returning to the user.
3048 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
3049 xfs_trans_set_sync(tp);
3051 return xfs_trans_commit(tp);
3055 * xfs_cross_rename()
3057 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
3061 struct xfs_trans *tp,
3062 struct xfs_inode *dp1,
3063 struct xfs_name *name1,
3064 struct xfs_inode *ip1,
3065 struct xfs_inode *dp2,
3066 struct xfs_name *name2,
3067 struct xfs_inode *ip2,
3075 /* Swap inode number for dirent in first parent */
3076 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
3078 goto out_trans_abort;
3080 /* Swap inode number for dirent in second parent */
3081 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
3083 goto out_trans_abort;
3086 * If we're renaming one or more directories across different parents,
3087 * update the respective ".." entries (and link counts) to match the new
3091 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3093 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
3094 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
3095 dp1->i_ino, spaceres);
3097 goto out_trans_abort;
3099 /* transfer ip2 ".." reference to dp1 */
3100 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
3101 error = xfs_droplink(tp, dp2);
3103 goto out_trans_abort;
3104 xfs_bumplink(tp, dp1);
3108 * Although ip1 isn't changed here, userspace needs
3109 * to be warned about the change, so that applications
3110 * relying on it (like backup ones), will properly
3113 ip1_flags |= XFS_ICHGTIME_CHG;
3114 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3117 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
3118 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
3119 dp2->i_ino, spaceres);
3121 goto out_trans_abort;
3123 /* transfer ip1 ".." reference to dp2 */
3124 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
3125 error = xfs_droplink(tp, dp1);
3127 goto out_trans_abort;
3128 xfs_bumplink(tp, dp2);
3132 * Although ip2 isn't changed here, userspace needs
3133 * to be warned about the change, so that applications
3134 * relying on it (like backup ones), will properly
3137 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3138 ip2_flags |= XFS_ICHGTIME_CHG;
3143 xfs_trans_ichgtime(tp, ip1, ip1_flags);
3144 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
3147 xfs_trans_ichgtime(tp, ip2, ip2_flags);
3148 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
3151 xfs_trans_ichgtime(tp, dp2, dp2_flags);
3152 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
3154 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3155 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3156 return xfs_finish_rename(tp);
3159 xfs_trans_cancel(tp);
3164 * xfs_rename_alloc_whiteout()
3166 * Return a referenced, unlinked, unlocked inode that that can be used as a
3167 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3168 * crash between allocating the inode and linking it into the rename transaction
3169 * recovery will free the inode and we won't leak it.
3172 xfs_rename_alloc_whiteout(
3173 struct xfs_inode *dp,
3174 struct xfs_inode **wip)
3176 struct xfs_inode *tmpfile;
3179 error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
3184 * Prepare the tmpfile inode as if it were created through the VFS.
3185 * Complete the inode setup and flag it as linkable. nlink is already
3186 * zero, so we can skip the drop_nlink.
3188 xfs_setup_iops(tmpfile);
3189 xfs_finish_inode_setup(tmpfile);
3190 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3201 struct xfs_inode *src_dp,
3202 struct xfs_name *src_name,
3203 struct xfs_inode *src_ip,
3204 struct xfs_inode *target_dp,
3205 struct xfs_name *target_name,
3206 struct xfs_inode *target_ip,
3209 struct xfs_mount *mp = src_dp->i_mount;
3210 struct xfs_trans *tp;
3211 struct xfs_inode *wip = NULL; /* whiteout inode */
3212 struct xfs_inode *inodes[__XFS_SORT_INODES];
3213 struct xfs_buf *agibp;
3214 int num_inodes = __XFS_SORT_INODES;
3215 bool new_parent = (src_dp != target_dp);
3216 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3220 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3222 if ((flags & RENAME_EXCHANGE) && !target_ip)
3226 * If we are doing a whiteout operation, allocate the whiteout inode
3227 * we will be placing at the target and ensure the type is set
3230 if (flags & RENAME_WHITEOUT) {
3231 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3232 error = xfs_rename_alloc_whiteout(target_dp, &wip);
3236 /* setup target dirent info as whiteout */
3237 src_name->type = XFS_DIR3_FT_CHRDEV;
3240 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3241 inodes, &num_inodes);
3243 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3244 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3245 if (error == -ENOSPC) {
3247 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3251 goto out_release_wip;
3254 * Attach the dquots to the inodes
3256 error = xfs_qm_vop_rename_dqattach(inodes);
3258 goto out_trans_cancel;
3261 * Lock all the participating inodes. Depending upon whether
3262 * the target_name exists in the target directory, and
3263 * whether the target directory is the same as the source
3264 * directory, we can lock from 2 to 4 inodes.
3266 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3269 * Join all the inodes to the transaction. From this point on,
3270 * we can rely on either trans_commit or trans_cancel to unlock
3273 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3275 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3276 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3278 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3280 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3283 * If we are using project inheritance, we only allow renames
3284 * into our tree when the project IDs are the same; else the
3285 * tree quota mechanism would be circumvented.
3287 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3288 target_dp->i_d.di_projid != src_ip->i_d.di_projid)) {
3290 goto out_trans_cancel;
3293 /* RENAME_EXCHANGE is unique from here on. */
3294 if (flags & RENAME_EXCHANGE)
3295 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3296 target_dp, target_name, target_ip,
3300 * Check for expected errors before we dirty the transaction
3301 * so we can return an error without a transaction abort.
3303 if (target_ip == NULL) {
3305 * If there's no space reservation, check the entry will
3306 * fit before actually inserting it.
3309 error = xfs_dir_canenter(tp, target_dp, target_name);
3311 goto out_trans_cancel;
3315 * If target exists and it's a directory, check that whether
3316 * it can be destroyed.
3318 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3319 (!xfs_dir_isempty(target_ip) ||
3320 (VFS_I(target_ip)->i_nlink > 2))) {
3322 goto out_trans_cancel;
3327 * Directory entry creation below may acquire the AGF. Remove
3328 * the whiteout from the unlinked list first to preserve correct
3329 * AGI/AGF locking order. This dirties the transaction so failures
3330 * after this point will abort and log recovery will clean up the
3333 * For whiteouts, we need to bump the link count on the whiteout
3334 * inode. After this point, we have a real link, clear the tmpfile
3335 * state flag from the inode so it doesn't accidentally get misused
3339 ASSERT(VFS_I(wip)->i_nlink == 0);
3340 error = xfs_iunlink_remove(tp, wip);
3342 goto out_trans_cancel;
3344 xfs_bumplink(tp, wip);
3345 VFS_I(wip)->i_state &= ~I_LINKABLE;
3349 * Set up the target.
3351 if (target_ip == NULL) {
3353 * If target does not exist and the rename crosses
3354 * directories, adjust the target directory link count
3355 * to account for the ".." reference from the new entry.
3357 error = xfs_dir_createname(tp, target_dp, target_name,
3358 src_ip->i_ino, spaceres);
3360 goto out_trans_cancel;
3362 xfs_trans_ichgtime(tp, target_dp,
3363 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3365 if (new_parent && src_is_directory) {
3366 xfs_bumplink(tp, target_dp);
3368 } else { /* target_ip != NULL */
3370 * Link the source inode under the target name.
3371 * If the source inode is a directory and we are moving
3372 * it across directories, its ".." entry will be
3373 * inconsistent until we replace that down below.
3375 * In case there is already an entry with the same
3376 * name at the destination directory, remove it first.
3380 * Check whether the replace operation will need to allocate
3381 * blocks. This happens when the shortform directory lacks
3382 * space and we have to convert it to a block format directory.
3383 * When more blocks are necessary, we must lock the AGI first
3384 * to preserve locking order (AGI -> AGF).
3386 if (xfs_dir2_sf_replace_needblock(target_dp, src_ip->i_ino)) {
3387 error = xfs_read_agi(mp, tp,
3388 XFS_INO_TO_AGNO(mp, target_ip->i_ino),
3391 goto out_trans_cancel;
3394 error = xfs_dir_replace(tp, target_dp, target_name,
3395 src_ip->i_ino, spaceres);
3397 goto out_trans_cancel;
3399 xfs_trans_ichgtime(tp, target_dp,
3400 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3403 * Decrement the link count on the target since the target
3404 * dir no longer points to it.
3406 error = xfs_droplink(tp, target_ip);
3408 goto out_trans_cancel;
3410 if (src_is_directory) {
3412 * Drop the link from the old "." entry.
3414 error = xfs_droplink(tp, target_ip);
3416 goto out_trans_cancel;
3418 } /* target_ip != NULL */
3421 * Remove the source.
3423 if (new_parent && src_is_directory) {
3425 * Rewrite the ".." entry to point to the new
3428 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3429 target_dp->i_ino, spaceres);
3430 ASSERT(error != -EEXIST);
3432 goto out_trans_cancel;
3436 * We always want to hit the ctime on the source inode.
3438 * This isn't strictly required by the standards since the source
3439 * inode isn't really being changed, but old unix file systems did
3440 * it and some incremental backup programs won't work without it.
3442 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3443 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3446 * Adjust the link count on src_dp. This is necessary when
3447 * renaming a directory, either within one parent when
3448 * the target existed, or across two parent directories.
3450 if (src_is_directory && (new_parent || target_ip != NULL)) {
3453 * Decrement link count on src_directory since the
3454 * entry that's moved no longer points to it.
3456 error = xfs_droplink(tp, src_dp);
3458 goto out_trans_cancel;
3462 * For whiteouts, we only need to update the source dirent with the
3463 * inode number of the whiteout inode rather than removing it
3467 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3470 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3473 goto out_trans_cancel;
3475 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3476 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3478 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3480 error = xfs_finish_rename(tp);
3486 xfs_trans_cancel(tp);
3495 struct xfs_inode *ip,
3498 struct xfs_mount *mp = ip->i_mount;
3499 struct xfs_perag *pag;
3500 unsigned long first_index, mask;
3502 struct xfs_inode **cilist;
3503 struct xfs_inode *cip;
3504 struct xfs_ino_geometry *igeo = M_IGEO(mp);
3509 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3511 cilist_size = igeo->inodes_per_cluster * sizeof(struct xfs_inode *);
3512 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3516 mask = ~(igeo->inodes_per_cluster - 1);
3517 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3519 /* really need a gang lookup range call here */
3520 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3521 first_index, igeo->inodes_per_cluster);
3525 for (i = 0; i < nr_found; i++) {
3531 * because this is an RCU protected lookup, we could find a
3532 * recently freed or even reallocated inode during the lookup.
3533 * We need to check under the i_flags_lock for a valid inode
3534 * here. Skip it if it is not valid or the wrong inode.
3536 spin_lock(&cip->i_flags_lock);
3538 __xfs_iflags_test(cip, XFS_ISTALE)) {
3539 spin_unlock(&cip->i_flags_lock);
3544 * Once we fall off the end of the cluster, no point checking
3545 * any more inodes in the list because they will also all be
3546 * outside the cluster.
3548 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3549 spin_unlock(&cip->i_flags_lock);
3552 spin_unlock(&cip->i_flags_lock);
3555 * Do an un-protected check to see if the inode is dirty and
3556 * is a candidate for flushing. These checks will be repeated
3557 * later after the appropriate locks are acquired.
3559 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3563 * Try to get locks. If any are unavailable or it is pinned,
3564 * then this inode cannot be flushed and is skipped.
3567 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3569 if (!xfs_iflock_nowait(cip)) {
3570 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3573 if (xfs_ipincount(cip)) {
3575 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3581 * Check the inode number again, just to be certain we are not
3582 * racing with freeing in xfs_reclaim_inode(). See the comments
3583 * in that function for more information as to why the initial
3584 * check is not sufficient.
3588 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3593 * arriving here means that this inode can be flushed. First
3594 * re-check that it's dirty before flushing.
3596 if (!xfs_inode_clean(cip)) {
3598 error = xfs_iflush_int(cip, bp);
3600 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3601 goto cluster_corrupt_out;
3607 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3611 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3612 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3623 cluster_corrupt_out:
3625 * Corruption detected in the clustering loop. Invalidate the
3626 * inode buffer and shut down the filesystem.
3631 * We'll always have an inode attached to the buffer for completion
3632 * process by the time we are called from xfs_iflush(). Hence we have
3633 * always need to do IO completion processing to abort the inodes
3634 * attached to the buffer. handle them just like the shutdown case in
3637 ASSERT(bp->b_iodone);
3638 bp->b_flags |= XBF_ASYNC;
3639 bp->b_flags &= ~XBF_DONE;
3641 xfs_buf_ioerror(bp, -EIO);
3644 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3646 /* abort the corrupt inode, as it was not attached to the buffer */
3647 xfs_iflush_abort(cip, false);
3650 return -EFSCORRUPTED;
3654 * Flush dirty inode metadata into the backing buffer.
3656 * The caller must have the inode lock and the inode flush lock held. The
3657 * inode lock will still be held upon return to the caller, and the inode
3658 * flush lock will be released after the inode has reached the disk.
3660 * The caller must write out the buffer returned in *bpp and release it.
3664 struct xfs_inode *ip,
3665 struct xfs_buf **bpp)
3667 struct xfs_mount *mp = ip->i_mount;
3668 struct xfs_buf *bp = NULL;
3669 struct xfs_dinode *dip;
3672 XFS_STATS_INC(mp, xs_iflush_count);
3674 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3675 ASSERT(xfs_isiflocked(ip));
3676 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3677 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3681 xfs_iunpin_wait(ip);
3684 * For stale inodes we cannot rely on the backing buffer remaining
3685 * stale in cache for the remaining life of the stale inode and so
3686 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3687 * inodes below. We have to check this after ensuring the inode is
3688 * unpinned so that it is safe to reclaim the stale inode after the
3691 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3697 * This may have been unpinned because the filesystem is shutting
3698 * down forcibly. If that's the case we must not write this inode
3699 * to disk, because the log record didn't make it to disk.
3701 * We also have to remove the log item from the AIL in this case,
3702 * as we wait for an empty AIL as part of the unmount process.
3704 if (XFS_FORCED_SHUTDOWN(mp)) {
3710 * Get the buffer containing the on-disk inode. We are doing a try-lock
3711 * operation here, so we may get an EAGAIN error. In that case, we
3712 * simply want to return with the inode still dirty.
3714 * If we get any other error, we effectively have a corruption situation
3715 * and we cannot flush the inode, so we treat it the same as failing
3718 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3720 if (error == -EAGAIN) {
3728 * First flush out the inode that xfs_iflush was called with.
3730 error = xfs_iflush_int(ip, bp);
3735 * If the buffer is pinned then push on the log now so we won't
3736 * get stuck waiting in the write for too long.
3738 if (xfs_buf_ispinned(bp))
3739 xfs_log_force(mp, 0);
3742 * inode clustering: try to gather other inodes into this write
3744 * Note: Any error during clustering will result in the filesystem
3745 * being shut down and completion callbacks run on the cluster buffer.
3746 * As we have already flushed and attached this inode to the buffer,
3747 * it has already been aborted and released by xfs_iflush_cluster() and
3748 * so we have no further error handling to do here.
3750 error = xfs_iflush_cluster(ip, bp);
3760 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3762 /* abort the corrupt inode, as it was not attached to the buffer */
3763 xfs_iflush_abort(ip, false);
3768 * If there are inline format data / attr forks attached to this inode,
3769 * make sure they're not corrupt.
3772 xfs_inode_verify_forks(
3773 struct xfs_inode *ip)
3775 struct xfs_ifork *ifp;
3778 fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops);
3780 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3781 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
3782 ifp->if_u1.if_data, ifp->if_bytes, fa);
3786 fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops);
3788 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
3789 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
3790 ifp ? ifp->if_u1.if_data : NULL,
3791 ifp ? ifp->if_bytes : 0, fa);
3799 struct xfs_inode *ip,
3802 struct xfs_inode_log_item *iip = ip->i_itemp;
3803 struct xfs_dinode *dip;
3804 struct xfs_mount *mp = ip->i_mount;
3806 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3807 ASSERT(xfs_isiflocked(ip));
3808 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3809 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3810 ASSERT(iip != NULL && iip->ili_fields != 0);
3811 ASSERT(ip->i_d.di_version > 1);
3813 /* set *dip = inode's place in the buffer */
3814 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3816 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3817 mp, XFS_ERRTAG_IFLUSH_1)) {
3818 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3819 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3820 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3823 if (S_ISREG(VFS_I(ip)->i_mode)) {
3825 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3826 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3827 mp, XFS_ERRTAG_IFLUSH_3)) {
3828 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3829 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3830 __func__, ip->i_ino, ip);
3833 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3835 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3836 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3837 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3838 mp, XFS_ERRTAG_IFLUSH_4)) {
3839 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3840 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3841 __func__, ip->i_ino, ip);
3845 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3846 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3847 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3848 "%s: detected corrupt incore inode %Lu, "
3849 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3850 __func__, ip->i_ino,
3851 ip->i_d.di_nextents + ip->i_d.di_anextents,
3852 ip->i_d.di_nblocks, ip);
3855 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3856 mp, XFS_ERRTAG_IFLUSH_6)) {
3857 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3858 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3859 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3864 * Inode item log recovery for v2 inodes are dependent on the
3865 * di_flushiter count for correct sequencing. We bump the flush
3866 * iteration count so we can detect flushes which postdate a log record
3867 * during recovery. This is redundant as we now log every change and
3868 * hence this can't happen but we need to still do it to ensure
3869 * backwards compatibility with old kernels that predate logging all
3872 if (ip->i_d.di_version < 3)
3873 ip->i_d.di_flushiter++;
3875 /* Check the inline fork data before we write out. */
3876 if (!xfs_inode_verify_forks(ip))
3880 * Copy the dirty parts of the inode into the on-disk inode. We always
3881 * copy out the core of the inode, because if the inode is dirty at all
3884 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3886 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3887 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3888 ip->i_d.di_flushiter = 0;
3890 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3891 if (XFS_IFORK_Q(ip))
3892 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3893 xfs_inobp_check(mp, bp);
3896 * We've recorded everything logged in the inode, so we'd like to clear
3897 * the ili_fields bits so we don't log and flush things unnecessarily.
3898 * However, we can't stop logging all this information until the data
3899 * we've copied into the disk buffer is written to disk. If we did we
3900 * might overwrite the copy of the inode in the log with all the data
3901 * after re-logging only part of it, and in the face of a crash we
3902 * wouldn't have all the data we need to recover.
3904 * What we do is move the bits to the ili_last_fields field. When
3905 * logging the inode, these bits are moved back to the ili_fields field.
3906 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3907 * know that the information those bits represent is permanently on
3908 * disk. As long as the flush completes before the inode is logged
3909 * again, then both ili_fields and ili_last_fields will be cleared.
3911 * We can play with the ili_fields bits here, because the inode lock
3912 * must be held exclusively in order to set bits there and the flush
3913 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3914 * done routine can tell whether or not to look in the AIL. Also, store
3915 * the current LSN of the inode so that we can tell whether the item has
3916 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3917 * need the AIL lock, because it is a 64 bit value that cannot be read
3920 iip->ili_last_fields = iip->ili_fields;
3921 iip->ili_fields = 0;
3922 iip->ili_fsync_fields = 0;
3923 iip->ili_logged = 1;
3925 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3926 &iip->ili_item.li_lsn);
3929 * Attach the function xfs_iflush_done to the inode's
3930 * buffer. This will remove the inode from the AIL
3931 * and unlock the inode's flush lock when the inode is
3932 * completely written to disk.
3934 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3936 /* generate the checksum. */
3937 xfs_dinode_calc_crc(mp, dip);
3939 ASSERT(!list_empty(&bp->b_li_list));
3940 ASSERT(bp->b_iodone != NULL);
3944 return -EFSCORRUPTED;
3947 /* Release an inode. */
3950 struct xfs_inode *ip)
3952 trace_xfs_irele(ip, _RET_IP_);