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);
804 inode->i_mode = mode;
805 set_nlink(inode, nlink);
806 inode->i_uid = current_fsuid();
807 inode->i_rdev = rdev;
808 ip->i_d.di_projid = prid;
810 if (pip && XFS_INHERIT_GID(pip)) {
811 inode->i_gid = VFS_I(pip)->i_gid;
812 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
813 inode->i_mode |= S_ISGID;
815 inode->i_gid = current_fsgid();
819 * If the group ID of the new file does not match the effective group
820 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
821 * (and only if the irix_sgid_inherit compatibility variable is set).
823 if (irix_sgid_inherit &&
824 (inode->i_mode & S_ISGID) && !in_group_p(inode->i_gid))
825 inode->i_mode &= ~S_ISGID;
828 ip->i_d.di_nextents = 0;
829 ASSERT(ip->i_d.di_nblocks == 0);
831 tv = current_time(inode);
836 ip->i_d.di_extsize = 0;
837 ip->i_d.di_dmevmask = 0;
838 ip->i_d.di_dmstate = 0;
839 ip->i_d.di_flags = 0;
841 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
842 inode_set_iversion(inode, 1);
843 ip->i_d.di_flags2 = 0;
844 ip->i_d.di_cowextsize = 0;
845 ip->i_d.di_crtime = tv;
848 flags = XFS_ILOG_CORE;
849 switch (mode & S_IFMT) {
854 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
855 ip->i_df.if_flags = 0;
856 flags |= XFS_ILOG_DEV;
860 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
864 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
865 di_flags |= XFS_DIFLAG_RTINHERIT;
866 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
867 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
868 ip->i_d.di_extsize = pip->i_d.di_extsize;
870 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
871 di_flags |= XFS_DIFLAG_PROJINHERIT;
872 } else if (S_ISREG(mode)) {
873 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
874 di_flags |= XFS_DIFLAG_REALTIME;
875 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
876 di_flags |= XFS_DIFLAG_EXTSIZE;
877 ip->i_d.di_extsize = pip->i_d.di_extsize;
880 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
882 di_flags |= XFS_DIFLAG_NOATIME;
883 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
885 di_flags |= XFS_DIFLAG_NODUMP;
886 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
888 di_flags |= XFS_DIFLAG_SYNC;
889 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
890 xfs_inherit_nosymlinks)
891 di_flags |= XFS_DIFLAG_NOSYMLINKS;
892 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
893 xfs_inherit_nodefrag)
894 di_flags |= XFS_DIFLAG_NODEFRAG;
895 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
896 di_flags |= XFS_DIFLAG_FILESTREAM;
898 ip->i_d.di_flags |= di_flags;
900 if (pip && (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY)) {
901 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
902 ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
903 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
905 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
906 ip->i_d.di_flags2 |= XFS_DIFLAG2_DAX;
910 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
911 ip->i_df.if_flags = XFS_IFEXTENTS;
912 ip->i_df.if_bytes = 0;
913 ip->i_df.if_u1.if_root = NULL;
919 * Attribute fork settings for new inode.
921 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
922 ip->i_d.di_anextents = 0;
925 * Log the new values stuffed into the inode.
927 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
928 xfs_trans_log_inode(tp, ip, flags);
930 /* now that we have an i_mode we can setup the inode structure */
938 * Allocates a new inode from disk and return a pointer to the
939 * incore copy. This routine will internally commit the current
940 * transaction and allocate a new one if the Space Manager needed
941 * to do an allocation to replenish the inode free-list.
943 * This routine is designed to be called from xfs_create and
949 xfs_trans_t **tpp, /* input: current transaction;
950 output: may be a new transaction. */
951 xfs_inode_t *dp, /* directory within whose allocate
956 prid_t prid, /* project id */
957 xfs_inode_t **ipp) /* pointer to inode; it will be
962 xfs_buf_t *ialloc_context = NULL;
968 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
971 * xfs_ialloc will return a pointer to an incore inode if
972 * the Space Manager has an available inode on the free
973 * list. Otherwise, it will do an allocation and replenish
974 * the freelist. Since we can only do one allocation per
975 * transaction without deadlocks, we will need to commit the
976 * current transaction and start a new one. We will then
977 * need to call xfs_ialloc again to get the inode.
979 * If xfs_ialloc did an allocation to replenish the freelist,
980 * it returns the bp containing the head of the freelist as
981 * ialloc_context. We will hold a lock on it across the
982 * transaction commit so that no other process can steal
983 * the inode(s) that we've just allocated.
985 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
989 * Return an error if we were unable to allocate a new inode.
990 * This should only happen if we run out of space on disk or
991 * encounter a disk error.
997 if (!ialloc_context && !ip) {
1003 * If the AGI buffer is non-NULL, then we were unable to get an
1004 * inode in one operation. We need to commit the current
1005 * transaction and call xfs_ialloc() again. It is guaranteed
1006 * to succeed the second time.
1008 if (ialloc_context) {
1010 * Normally, xfs_trans_commit releases all the locks.
1011 * We call bhold to hang on to the ialloc_context across
1012 * the commit. Holding this buffer prevents any other
1013 * processes from doing any allocations in this
1016 xfs_trans_bhold(tp, ialloc_context);
1019 * We want the quota changes to be associated with the next
1020 * transaction, NOT this one. So, detach the dqinfo from this
1021 * and attach it to the next transaction.
1026 dqinfo = (void *)tp->t_dqinfo;
1027 tp->t_dqinfo = NULL;
1028 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1029 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1032 code = xfs_trans_roll(&tp);
1035 * Re-attach the quota info that we detached from prev trx.
1038 tp->t_dqinfo = dqinfo;
1039 tp->t_flags |= tflags;
1043 xfs_buf_relse(ialloc_context);
1048 xfs_trans_bjoin(tp, ialloc_context);
1051 * Call ialloc again. Since we've locked out all
1052 * other allocations in this allocation group,
1053 * this call should always succeed.
1055 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1056 &ialloc_context, &ip);
1059 * If we get an error at this point, return to the caller
1060 * so that the current transaction can be aborted.
1067 ASSERT(!ialloc_context && ip);
1078 * Decrement the link count on an inode & log the change. If this causes the
1079 * link count to go to zero, move the inode to AGI unlinked list so that it can
1080 * be freed when the last active reference goes away via xfs_inactive().
1082 static int /* error */
1087 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1089 drop_nlink(VFS_I(ip));
1090 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1092 if (VFS_I(ip)->i_nlink)
1095 return xfs_iunlink(tp, ip);
1099 * Increment the link count on an inode & log the change.
1106 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1108 inc_nlink(VFS_I(ip));
1109 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1115 struct xfs_name *name,
1120 int is_dir = S_ISDIR(mode);
1121 struct xfs_mount *mp = dp->i_mount;
1122 struct xfs_inode *ip = NULL;
1123 struct xfs_trans *tp = NULL;
1125 bool unlock_dp_on_error = false;
1127 struct xfs_dquot *udqp = NULL;
1128 struct xfs_dquot *gdqp = NULL;
1129 struct xfs_dquot *pdqp = NULL;
1130 struct xfs_trans_res *tres;
1133 trace_xfs_create(dp, name);
1135 if (XFS_FORCED_SHUTDOWN(mp))
1138 prid = xfs_get_initial_prid(dp);
1141 * Make sure that we have allocated dquot(s) on disk.
1143 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1144 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1145 &udqp, &gdqp, &pdqp);
1150 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1151 tres = &M_RES(mp)->tr_mkdir;
1153 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1154 tres = &M_RES(mp)->tr_create;
1158 * Initially assume that the file does not exist and
1159 * reserve the resources for that case. If that is not
1160 * the case we'll drop the one we have and get a more
1161 * appropriate transaction later.
1163 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1164 if (error == -ENOSPC) {
1165 /* flush outstanding delalloc blocks and retry */
1166 xfs_flush_inodes(mp);
1167 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1170 goto out_release_inode;
1172 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1173 unlock_dp_on_error = true;
1176 * Reserve disk quota and the inode.
1178 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1179 pdqp, resblks, 1, 0);
1181 goto out_trans_cancel;
1184 * A newly created regular or special file just has one directory
1185 * entry pointing to them, but a directory also the "." entry
1186 * pointing to itself.
1188 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
1190 goto out_trans_cancel;
1193 * Now we join the directory inode to the transaction. We do not do it
1194 * earlier because xfs_dir_ialloc might commit the previous transaction
1195 * (and release all the locks). An error from here on will result in
1196 * the transaction cancel unlocking dp so don't do it explicitly in the
1199 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1200 unlock_dp_on_error = false;
1202 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1203 resblks - XFS_IALLOC_SPACE_RES(mp));
1205 ASSERT(error != -ENOSPC);
1206 goto out_trans_cancel;
1208 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1209 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1212 error = xfs_dir_init(tp, ip, dp);
1214 goto out_trans_cancel;
1216 xfs_bumplink(tp, dp);
1220 * If this is a synchronous mount, make sure that the
1221 * create transaction goes to disk before returning to
1224 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1225 xfs_trans_set_sync(tp);
1228 * Attach the dquot(s) to the inodes and modify them incore.
1229 * These ids of the inode couldn't have changed since the new
1230 * inode has been locked ever since it was created.
1232 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1234 error = xfs_trans_commit(tp);
1236 goto out_release_inode;
1238 xfs_qm_dqrele(udqp);
1239 xfs_qm_dqrele(gdqp);
1240 xfs_qm_dqrele(pdqp);
1246 xfs_trans_cancel(tp);
1249 * Wait until after the current transaction is aborted to finish the
1250 * setup of the inode and release the inode. This prevents recursive
1251 * transactions and deadlocks from xfs_inactive.
1254 xfs_finish_inode_setup(ip);
1258 xfs_qm_dqrele(udqp);
1259 xfs_qm_dqrele(gdqp);
1260 xfs_qm_dqrele(pdqp);
1262 if (unlock_dp_on_error)
1263 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1269 struct xfs_inode *dp,
1271 struct xfs_inode **ipp)
1273 struct xfs_mount *mp = dp->i_mount;
1274 struct xfs_inode *ip = NULL;
1275 struct xfs_trans *tp = NULL;
1278 struct xfs_dquot *udqp = NULL;
1279 struct xfs_dquot *gdqp = NULL;
1280 struct xfs_dquot *pdqp = NULL;
1281 struct xfs_trans_res *tres;
1284 if (XFS_FORCED_SHUTDOWN(mp))
1287 prid = xfs_get_initial_prid(dp);
1290 * Make sure that we have allocated dquot(s) on disk.
1292 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1293 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1294 &udqp, &gdqp, &pdqp);
1298 resblks = XFS_IALLOC_SPACE_RES(mp);
1299 tres = &M_RES(mp)->tr_create_tmpfile;
1301 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1303 goto out_release_inode;
1305 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1306 pdqp, resblks, 1, 0);
1308 goto out_trans_cancel;
1310 error = xfs_dir_ialloc(&tp, dp, mode, 0, 0, prid, &ip);
1312 goto out_trans_cancel;
1314 if (mp->m_flags & XFS_MOUNT_WSYNC)
1315 xfs_trans_set_sync(tp);
1318 * Attach the dquot(s) to the inodes and modify them incore.
1319 * These ids of the inode couldn't have changed since the new
1320 * inode has been locked ever since it was created.
1322 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1324 error = xfs_iunlink(tp, ip);
1326 goto out_trans_cancel;
1328 error = xfs_trans_commit(tp);
1330 goto out_release_inode;
1332 xfs_qm_dqrele(udqp);
1333 xfs_qm_dqrele(gdqp);
1334 xfs_qm_dqrele(pdqp);
1340 xfs_trans_cancel(tp);
1343 * Wait until after the current transaction is aborted to finish the
1344 * setup of the inode and release the inode. This prevents recursive
1345 * transactions and deadlocks from xfs_inactive.
1348 xfs_finish_inode_setup(ip);
1352 xfs_qm_dqrele(udqp);
1353 xfs_qm_dqrele(gdqp);
1354 xfs_qm_dqrele(pdqp);
1363 struct xfs_name *target_name)
1365 xfs_mount_t *mp = tdp->i_mount;
1370 trace_xfs_link(tdp, target_name);
1372 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1374 if (XFS_FORCED_SHUTDOWN(mp))
1377 error = xfs_qm_dqattach(sip);
1381 error = xfs_qm_dqattach(tdp);
1385 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1386 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1387 if (error == -ENOSPC) {
1389 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1394 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1396 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1397 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1400 * If we are using project inheritance, we only allow hard link
1401 * creation in our tree when the project IDs are the same; else
1402 * the tree quota mechanism could be circumvented.
1404 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1405 tdp->i_d.di_projid != sip->i_d.di_projid)) {
1411 error = xfs_dir_canenter(tp, tdp, target_name);
1417 * Handle initial link state of O_TMPFILE inode
1419 if (VFS_I(sip)->i_nlink == 0) {
1420 error = xfs_iunlink_remove(tp, sip);
1425 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1429 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1430 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1432 xfs_bumplink(tp, sip);
1435 * If this is a synchronous mount, make sure that the
1436 * link transaction goes to disk before returning to
1439 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1440 xfs_trans_set_sync(tp);
1442 return xfs_trans_commit(tp);
1445 xfs_trans_cancel(tp);
1450 /* Clear the reflink flag and the cowblocks tag if possible. */
1452 xfs_itruncate_clear_reflink_flags(
1453 struct xfs_inode *ip)
1455 struct xfs_ifork *dfork;
1456 struct xfs_ifork *cfork;
1458 if (!xfs_is_reflink_inode(ip))
1460 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1461 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1462 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1463 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1464 if (cfork->if_bytes == 0)
1465 xfs_inode_clear_cowblocks_tag(ip);
1469 * Free up the underlying blocks past new_size. The new size must be smaller
1470 * than the current size. This routine can be used both for the attribute and
1471 * data fork, and does not modify the inode size, which is left to the caller.
1473 * The transaction passed to this routine must have made a permanent log
1474 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1475 * given transaction and start new ones, so make sure everything involved in
1476 * the transaction is tidy before calling here. Some transaction will be
1477 * returned to the caller to be committed. The incoming transaction must
1478 * already include the inode, and both inode locks must be held exclusively.
1479 * The inode must also be "held" within the transaction. On return the inode
1480 * will be "held" within the returned transaction. This routine does NOT
1481 * require any disk space to be reserved for it within the transaction.
1483 * If we get an error, we must return with the inode locked and linked into the
1484 * current transaction. This keeps things simple for the higher level code,
1485 * because it always knows that the inode is locked and held in the transaction
1486 * that returns to it whether errors occur or not. We don't mark the inode
1487 * dirty on error so that transactions can be easily aborted if possible.
1490 xfs_itruncate_extents_flags(
1491 struct xfs_trans **tpp,
1492 struct xfs_inode *ip,
1494 xfs_fsize_t new_size,
1497 struct xfs_mount *mp = ip->i_mount;
1498 struct xfs_trans *tp = *tpp;
1499 xfs_fileoff_t first_unmap_block;
1500 xfs_filblks_t unmap_len;
1503 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1504 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1505 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1506 ASSERT(new_size <= XFS_ISIZE(ip));
1507 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1508 ASSERT(ip->i_itemp != NULL);
1509 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1510 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1512 trace_xfs_itruncate_extents_start(ip, new_size);
1514 flags |= xfs_bmapi_aflag(whichfork);
1517 * Since it is possible for space to become allocated beyond
1518 * the end of the file (in a crash where the space is allocated
1519 * but the inode size is not yet updated), simply remove any
1520 * blocks which show up between the new EOF and the maximum
1521 * possible file size.
1523 * We have to free all the blocks to the bmbt maximum offset, even if
1524 * the page cache can't scale that far.
1526 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1527 if (first_unmap_block >= XFS_MAX_FILEOFF) {
1528 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1532 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1533 while (unmap_len > 0) {
1534 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1535 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1536 flags, XFS_ITRUNC_MAX_EXTENTS);
1541 * Duplicate the transaction that has the permanent
1542 * reservation and commit the old transaction.
1544 error = xfs_defer_finish(&tp);
1548 error = xfs_trans_roll_inode(&tp, ip);
1553 if (whichfork == XFS_DATA_FORK) {
1554 /* Remove all pending CoW reservations. */
1555 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1556 first_unmap_block, XFS_MAX_FILEOFF, true);
1560 xfs_itruncate_clear_reflink_flags(ip);
1564 * Always re-log the inode so that our permanent transaction can keep
1565 * on rolling it forward in the log.
1567 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1569 trace_xfs_itruncate_extents_end(ip, new_size);
1580 xfs_mount_t *mp = ip->i_mount;
1583 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1586 /* If this is a read-only mount, don't do this (would generate I/O) */
1587 if (mp->m_flags & XFS_MOUNT_RDONLY)
1590 if (!XFS_FORCED_SHUTDOWN(mp)) {
1594 * If we previously truncated this file and removed old data
1595 * in the process, we want to initiate "early" writeout on
1596 * the last close. This is an attempt to combat the notorious
1597 * NULL files problem which is particularly noticeable from a
1598 * truncate down, buffered (re-)write (delalloc), followed by
1599 * a crash. What we are effectively doing here is
1600 * significantly reducing the time window where we'd otherwise
1601 * be exposed to that problem.
1603 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1605 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1606 if (ip->i_delayed_blks > 0) {
1607 error = filemap_flush(VFS_I(ip)->i_mapping);
1614 if (VFS_I(ip)->i_nlink == 0)
1617 if (xfs_can_free_eofblocks(ip, false)) {
1620 * Check if the inode is being opened, written and closed
1621 * frequently and we have delayed allocation blocks outstanding
1622 * (e.g. streaming writes from the NFS server), truncating the
1623 * blocks past EOF will cause fragmentation to occur.
1625 * In this case don't do the truncation, but we have to be
1626 * careful how we detect this case. Blocks beyond EOF show up as
1627 * i_delayed_blks even when the inode is clean, so we need to
1628 * truncate them away first before checking for a dirty release.
1629 * Hence on the first dirty close we will still remove the
1630 * speculative allocation, but after that we will leave it in
1633 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1636 * If we can't get the iolock just skip truncating the blocks
1637 * past EOF because we could deadlock with the mmap_sem
1638 * otherwise. We'll get another chance to drop them once the
1639 * last reference to the inode is dropped, so we'll never leak
1640 * blocks permanently.
1642 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1643 error = xfs_free_eofblocks(ip);
1644 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1649 /* delalloc blocks after truncation means it really is dirty */
1650 if (ip->i_delayed_blks)
1651 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1657 * xfs_inactive_truncate
1659 * Called to perform a truncate when an inode becomes unlinked.
1662 xfs_inactive_truncate(
1663 struct xfs_inode *ip)
1665 struct xfs_mount *mp = ip->i_mount;
1666 struct xfs_trans *tp;
1669 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1671 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1674 xfs_ilock(ip, XFS_ILOCK_EXCL);
1675 xfs_trans_ijoin(tp, ip, 0);
1678 * Log the inode size first to prevent stale data exposure in the event
1679 * of a system crash before the truncate completes. See the related
1680 * comment in xfs_vn_setattr_size() for details.
1682 ip->i_d.di_size = 0;
1683 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1685 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1687 goto error_trans_cancel;
1689 ASSERT(ip->i_d.di_nextents == 0);
1691 error = xfs_trans_commit(tp);
1695 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1699 xfs_trans_cancel(tp);
1701 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1706 * xfs_inactive_ifree()
1708 * Perform the inode free when an inode is unlinked.
1712 struct xfs_inode *ip)
1714 struct xfs_mount *mp = ip->i_mount;
1715 struct xfs_trans *tp;
1719 * We try to use a per-AG reservation for any block needed by the finobt
1720 * tree, but as the finobt feature predates the per-AG reservation
1721 * support a degraded file system might not have enough space for the
1722 * reservation at mount time. In that case try to dip into the reserved
1725 * Send a warning if the reservation does happen to fail, as the inode
1726 * now remains allocated and sits on the unlinked list until the fs is
1729 if (unlikely(mp->m_finobt_nores)) {
1730 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1731 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1734 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1737 if (error == -ENOSPC) {
1738 xfs_warn_ratelimited(mp,
1739 "Failed to remove inode(s) from unlinked list. "
1740 "Please free space, unmount and run xfs_repair.");
1742 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1747 xfs_ilock(ip, XFS_ILOCK_EXCL);
1748 xfs_trans_ijoin(tp, ip, 0);
1750 error = xfs_ifree(tp, ip);
1753 * If we fail to free the inode, shut down. The cancel
1754 * might do that, we need to make sure. Otherwise the
1755 * inode might be lost for a long time or forever.
1757 if (!XFS_FORCED_SHUTDOWN(mp)) {
1758 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1760 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1762 xfs_trans_cancel(tp);
1763 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1768 * Credit the quota account(s). The inode is gone.
1770 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1773 * Just ignore errors at this point. There is nothing we can do except
1774 * to try to keep going. Make sure it's not a silent error.
1776 error = xfs_trans_commit(tp);
1778 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1781 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1788 * This is called when the vnode reference count for the vnode
1789 * goes to zero. If the file has been unlinked, then it must
1790 * now be truncated. Also, we clear all of the read-ahead state
1791 * kept for the inode here since the file is now closed.
1797 struct xfs_mount *mp;
1802 * If the inode is already free, then there can be nothing
1805 if (VFS_I(ip)->i_mode == 0) {
1806 ASSERT(ip->i_df.if_broot_bytes == 0);
1811 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1813 /* If this is a read-only mount, don't do this (would generate I/O) */
1814 if (mp->m_flags & XFS_MOUNT_RDONLY)
1817 /* Try to clean out the cow blocks if there are any. */
1818 if (xfs_inode_has_cow_data(ip))
1819 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1821 if (VFS_I(ip)->i_nlink != 0) {
1823 * force is true because we are evicting an inode from the
1824 * cache. Post-eof blocks must be freed, lest we end up with
1825 * broken free space accounting.
1827 * Note: don't bother with iolock here since lockdep complains
1828 * about acquiring it in reclaim context. We have the only
1829 * reference to the inode at this point anyways.
1831 if (xfs_can_free_eofblocks(ip, true))
1832 xfs_free_eofblocks(ip);
1837 if (S_ISREG(VFS_I(ip)->i_mode) &&
1838 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1839 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1842 error = xfs_qm_dqattach(ip);
1846 if (S_ISLNK(VFS_I(ip)->i_mode))
1847 error = xfs_inactive_symlink(ip);
1849 error = xfs_inactive_truncate(ip);
1854 * If there are attributes associated with the file then blow them away
1855 * now. The code calls a routine that recursively deconstructs the
1856 * attribute fork. If also blows away the in-core attribute fork.
1858 if (XFS_IFORK_Q(ip)) {
1859 error = xfs_attr_inactive(ip);
1865 ASSERT(ip->i_d.di_anextents == 0);
1866 ASSERT(ip->i_d.di_forkoff == 0);
1871 error = xfs_inactive_ifree(ip);
1876 * Release the dquots held by inode, if any.
1878 xfs_qm_dqdetach(ip);
1882 * In-Core Unlinked List Lookups
1883 * =============================
1885 * Every inode is supposed to be reachable from some other piece of metadata
1886 * with the exception of the root directory. Inodes with a connection to a
1887 * file descriptor but not linked from anywhere in the on-disk directory tree
1888 * are collectively known as unlinked inodes, though the filesystem itself
1889 * maintains links to these inodes so that on-disk metadata are consistent.
1891 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1892 * header contains a number of buckets that point to an inode, and each inode
1893 * record has a pointer to the next inode in the hash chain. This
1894 * singly-linked list causes scaling problems in the iunlink remove function
1895 * because we must walk that list to find the inode that points to the inode
1896 * being removed from the unlinked hash bucket list.
1898 * What if we modelled the unlinked list as a collection of records capturing
1899 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1900 * have a fast way to look up unlinked list predecessors, which avoids the
1901 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1904 * Because this is a backref cache, we ignore operational failures since the
1905 * iunlink code can fall back to the slow bucket walk. The only errors that
1906 * should bubble out are for obviously incorrect situations.
1908 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1909 * access or have otherwise provided for concurrency control.
1912 /* Capture a "X.next_unlinked = Y" relationship. */
1913 struct xfs_iunlink {
1914 struct rhash_head iu_rhash_head;
1915 xfs_agino_t iu_agino; /* X */
1916 xfs_agino_t iu_next_unlinked; /* Y */
1919 /* Unlinked list predecessor lookup hashtable construction */
1921 xfs_iunlink_obj_cmpfn(
1922 struct rhashtable_compare_arg *arg,
1925 const xfs_agino_t *key = arg->key;
1926 const struct xfs_iunlink *iu = obj;
1928 if (iu->iu_next_unlinked != *key)
1933 static const struct rhashtable_params xfs_iunlink_hash_params = {
1934 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1935 .key_len = sizeof(xfs_agino_t),
1936 .key_offset = offsetof(struct xfs_iunlink,
1938 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1939 .automatic_shrinking = true,
1940 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1944 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1945 * relation is found.
1948 xfs_iunlink_lookup_backref(
1949 struct xfs_perag *pag,
1952 struct xfs_iunlink *iu;
1954 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1955 xfs_iunlink_hash_params);
1956 return iu ? iu->iu_agino : NULLAGINO;
1960 * Take ownership of an iunlink cache entry and insert it into the hash table.
1961 * If successful, the entry will be owned by the cache; if not, it is freed.
1962 * Either way, the caller does not own @iu after this call.
1965 xfs_iunlink_insert_backref(
1966 struct xfs_perag *pag,
1967 struct xfs_iunlink *iu)
1971 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1972 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1974 * Fail loudly if there already was an entry because that's a sign of
1975 * corruption of in-memory data. Also fail loudly if we see an error
1976 * code we didn't anticipate from the rhashtable code. Currently we
1977 * only anticipate ENOMEM.
1980 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1984 * Absorb any runtime errors that aren't a result of corruption because
1985 * this is a cache and we can always fall back to bucket list scanning.
1987 if (error != 0 && error != -EEXIST)
1992 /* Remember that @prev_agino.next_unlinked = @this_agino. */
1994 xfs_iunlink_add_backref(
1995 struct xfs_perag *pag,
1996 xfs_agino_t prev_agino,
1997 xfs_agino_t this_agino)
1999 struct xfs_iunlink *iu;
2001 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
2004 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
2005 iu->iu_agino = prev_agino;
2006 iu->iu_next_unlinked = this_agino;
2008 return xfs_iunlink_insert_backref(pag, iu);
2012 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
2013 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
2014 * wasn't any such entry then we don't bother.
2017 xfs_iunlink_change_backref(
2018 struct xfs_perag *pag,
2020 xfs_agino_t next_unlinked)
2022 struct xfs_iunlink *iu;
2025 /* Look up the old entry; if there wasn't one then exit. */
2026 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
2027 xfs_iunlink_hash_params);
2032 * Remove the entry. This shouldn't ever return an error, but if we
2033 * couldn't remove the old entry we don't want to add it again to the
2034 * hash table, and if the entry disappeared on us then someone's
2035 * violated the locking rules and we need to fail loudly. Either way
2036 * we cannot remove the inode because internal state is or would have
2039 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
2040 &iu->iu_rhash_head, xfs_iunlink_hash_params);
2044 /* If there is no new next entry just free our item and return. */
2045 if (next_unlinked == NULLAGINO) {
2050 /* Update the entry and re-add it to the hash table. */
2051 iu->iu_next_unlinked = next_unlinked;
2052 return xfs_iunlink_insert_backref(pag, iu);
2055 /* Set up the in-core predecessor structures. */
2058 struct xfs_perag *pag)
2060 return rhashtable_init(&pag->pagi_unlinked_hash,
2061 &xfs_iunlink_hash_params);
2064 /* Free the in-core predecessor structures. */
2066 xfs_iunlink_free_item(
2070 struct xfs_iunlink *iu = ptr;
2071 bool *freed_anything = arg;
2073 *freed_anything = true;
2078 xfs_iunlink_destroy(
2079 struct xfs_perag *pag)
2081 bool freed_anything = false;
2083 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
2084 xfs_iunlink_free_item, &freed_anything);
2086 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
2090 * Point the AGI unlinked bucket at an inode and log the results. The caller
2091 * is responsible for validating the old value.
2094 xfs_iunlink_update_bucket(
2095 struct xfs_trans *tp,
2096 xfs_agnumber_t agno,
2097 struct xfs_buf *agibp,
2098 unsigned int bucket_index,
2099 xfs_agino_t new_agino)
2101 struct xfs_agi *agi = agibp->b_addr;
2102 xfs_agino_t old_value;
2105 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
2107 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2108 trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
2109 old_value, new_agino);
2112 * We should never find the head of the list already set to the value
2113 * passed in because either we're adding or removing ourselves from the
2116 if (old_value == new_agino) {
2117 xfs_buf_mark_corrupt(agibp);
2118 return -EFSCORRUPTED;
2121 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2122 offset = offsetof(struct xfs_agi, agi_unlinked) +
2123 (sizeof(xfs_agino_t) * bucket_index);
2124 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2128 /* Set an on-disk inode's next_unlinked pointer. */
2130 xfs_iunlink_update_dinode(
2131 struct xfs_trans *tp,
2132 xfs_agnumber_t agno,
2134 struct xfs_buf *ibp,
2135 struct xfs_dinode *dip,
2136 struct xfs_imap *imap,
2137 xfs_agino_t next_agino)
2139 struct xfs_mount *mp = tp->t_mountp;
2142 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2144 trace_xfs_iunlink_update_dinode(mp, agno, agino,
2145 be32_to_cpu(dip->di_next_unlinked), next_agino);
2147 dip->di_next_unlinked = cpu_to_be32(next_agino);
2148 offset = imap->im_boffset +
2149 offsetof(struct xfs_dinode, di_next_unlinked);
2151 /* need to recalc the inode CRC if appropriate */
2152 xfs_dinode_calc_crc(mp, dip);
2153 xfs_trans_inode_buf(tp, ibp);
2154 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2155 xfs_inobp_check(mp, ibp);
2158 /* Set an in-core inode's unlinked pointer and return the old value. */
2160 xfs_iunlink_update_inode(
2161 struct xfs_trans *tp,
2162 struct xfs_inode *ip,
2163 xfs_agnumber_t agno,
2164 xfs_agino_t next_agino,
2165 xfs_agino_t *old_next_agino)
2167 struct xfs_mount *mp = tp->t_mountp;
2168 struct xfs_dinode *dip;
2169 struct xfs_buf *ibp;
2170 xfs_agino_t old_value;
2173 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2175 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 0, 0);
2179 /* Make sure the old pointer isn't garbage. */
2180 old_value = be32_to_cpu(dip->di_next_unlinked);
2181 if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2182 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2183 sizeof(*dip), __this_address);
2184 error = -EFSCORRUPTED;
2189 * Since we're updating a linked list, we should never find that the
2190 * current pointer is the same as the new value, unless we're
2191 * terminating the list.
2193 *old_next_agino = old_value;
2194 if (old_value == next_agino) {
2195 if (next_agino != NULLAGINO) {
2196 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2197 dip, sizeof(*dip), __this_address);
2198 error = -EFSCORRUPTED;
2203 /* Ok, update the new pointer. */
2204 xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2205 ibp, dip, &ip->i_imap, next_agino);
2208 xfs_trans_brelse(tp, ibp);
2213 * This is called when the inode's link count has gone to 0 or we are creating
2214 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2216 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2217 * list when the inode is freed.
2221 struct xfs_trans *tp,
2222 struct xfs_inode *ip)
2224 struct xfs_mount *mp = tp->t_mountp;
2225 struct xfs_agi *agi;
2226 struct xfs_buf *agibp;
2227 xfs_agino_t next_agino;
2228 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2229 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2230 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2233 ASSERT(VFS_I(ip)->i_nlink == 0);
2234 ASSERT(VFS_I(ip)->i_mode != 0);
2235 trace_xfs_iunlink(ip);
2237 /* Get the agi buffer first. It ensures lock ordering on the list. */
2238 error = xfs_read_agi(mp, tp, agno, &agibp);
2241 agi = agibp->b_addr;
2244 * Get the index into the agi hash table for the list this inode will
2245 * go on. Make sure the pointer isn't garbage and that this inode
2246 * isn't already on the list.
2248 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2249 if (next_agino == agino ||
2250 !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2251 xfs_buf_mark_corrupt(agibp);
2252 return -EFSCORRUPTED;
2255 if (next_agino != NULLAGINO) {
2256 struct xfs_perag *pag;
2257 xfs_agino_t old_agino;
2260 * There is already another inode in the bucket, so point this
2261 * inode to the current head of the list.
2263 error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2267 ASSERT(old_agino == NULLAGINO);
2270 * agino has been unlinked, add a backref from the next inode
2273 pag = xfs_perag_get(mp, agno);
2274 error = xfs_iunlink_add_backref(pag, agino, next_agino);
2280 /* Point the head of the list to point to this inode. */
2281 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2284 /* Return the imap, dinode pointer, and buffer for an inode. */
2286 xfs_iunlink_map_ino(
2287 struct xfs_trans *tp,
2288 xfs_agnumber_t agno,
2290 struct xfs_imap *imap,
2291 struct xfs_dinode **dipp,
2292 struct xfs_buf **bpp)
2294 struct xfs_mount *mp = tp->t_mountp;
2298 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2300 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2305 error = xfs_imap_to_bp(mp, tp, imap, dipp, bpp, 0, 0);
2307 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2316 * Walk the unlinked chain from @head_agino until we find the inode that
2317 * points to @target_agino. Return the inode number, map, dinode pointer,
2318 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2320 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2321 * @agino, @imap, @dipp, and @bpp are all output parameters.
2323 * Do not call this function if @target_agino is the head of the list.
2326 xfs_iunlink_map_prev(
2327 struct xfs_trans *tp,
2328 xfs_agnumber_t agno,
2329 xfs_agino_t head_agino,
2330 xfs_agino_t target_agino,
2332 struct xfs_imap *imap,
2333 struct xfs_dinode **dipp,
2334 struct xfs_buf **bpp,
2335 struct xfs_perag *pag)
2337 struct xfs_mount *mp = tp->t_mountp;
2338 xfs_agino_t next_agino;
2341 ASSERT(head_agino != target_agino);
2344 /* See if our backref cache can find it faster. */
2345 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2346 if (*agino != NULLAGINO) {
2347 error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2351 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2355 * If we get here the cache contents were corrupt, so drop the
2356 * buffer and fall back to walking the bucket list.
2358 xfs_trans_brelse(tp, *bpp);
2363 trace_xfs_iunlink_map_prev_fallback(mp, agno);
2365 /* Otherwise, walk the entire bucket until we find it. */
2366 next_agino = head_agino;
2367 while (next_agino != target_agino) {
2368 xfs_agino_t unlinked_agino;
2371 xfs_trans_brelse(tp, *bpp);
2373 *agino = next_agino;
2374 error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2379 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2381 * Make sure this pointer is valid and isn't an obvious
2384 if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2385 next_agino == unlinked_agino) {
2386 XFS_CORRUPTION_ERROR(__func__,
2387 XFS_ERRLEVEL_LOW, mp,
2388 *dipp, sizeof(**dipp));
2389 error = -EFSCORRUPTED;
2392 next_agino = unlinked_agino;
2399 * Pull the on-disk inode from the AGI unlinked list.
2403 struct xfs_trans *tp,
2404 struct xfs_inode *ip)
2406 struct xfs_mount *mp = tp->t_mountp;
2407 struct xfs_agi *agi;
2408 struct xfs_buf *agibp;
2409 struct xfs_buf *last_ibp;
2410 struct xfs_dinode *last_dip = NULL;
2411 struct xfs_perag *pag = NULL;
2412 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2413 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2414 xfs_agino_t next_agino;
2415 xfs_agino_t head_agino;
2416 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2419 trace_xfs_iunlink_remove(ip);
2421 /* Get the agi buffer first. It ensures lock ordering on the list. */
2422 error = xfs_read_agi(mp, tp, agno, &agibp);
2425 agi = agibp->b_addr;
2428 * Get the index into the agi hash table for the list this inode will
2429 * go on. Make sure the head pointer isn't garbage.
2431 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2432 if (!xfs_verify_agino(mp, agno, head_agino)) {
2433 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2435 return -EFSCORRUPTED;
2439 * Set our inode's next_unlinked pointer to NULL and then return
2440 * the old pointer value so that we can update whatever was previous
2441 * to us in the list to point to whatever was next in the list.
2443 error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2448 * If there was a backref pointing from the next inode back to this
2449 * one, remove it because we've removed this inode from the list.
2451 * Later, if this inode was in the middle of the list we'll update
2452 * this inode's backref to point from the next inode.
2454 if (next_agino != NULLAGINO) {
2455 pag = xfs_perag_get(mp, agno);
2456 error = xfs_iunlink_change_backref(pag, next_agino,
2462 if (head_agino == agino) {
2463 /* Point the head of the list to the next unlinked inode. */
2464 error = xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2469 struct xfs_imap imap;
2470 xfs_agino_t prev_agino;
2473 pag = xfs_perag_get(mp, agno);
2475 /* We need to search the list for the inode being freed. */
2476 error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2477 &prev_agino, &imap, &last_dip, &last_ibp,
2482 /* Point the previous inode on the list to the next inode. */
2483 xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2484 last_dip, &imap, next_agino);
2487 * Now we deal with the backref for this inode. If this inode
2488 * pointed at a real inode, change the backref that pointed to
2489 * us to point to our old next. If this inode was the end of
2490 * the list, delete the backref that pointed to us. Note that
2491 * change_backref takes care of deleting the backref if
2492 * next_agino is NULLAGINO.
2494 error = xfs_iunlink_change_backref(pag, agino, next_agino);
2506 * Look up the inode number specified and mark it stale if it is found. If it is
2507 * dirty, return the inode so it can be attached to the cluster buffer so it can
2508 * be processed appropriately when the cluster free transaction completes.
2510 static struct xfs_inode *
2511 xfs_ifree_get_one_inode(
2512 struct xfs_perag *pag,
2513 struct xfs_inode *free_ip,
2516 struct xfs_mount *mp = pag->pag_mount;
2517 struct xfs_inode *ip;
2521 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2523 /* Inode not in memory, nothing to do */
2525 goto out_rcu_unlock;
2528 * because this is an RCU protected lookup, we could find a recently
2529 * freed or even reallocated inode during the lookup. We need to check
2530 * under the i_flags_lock for a valid inode here. Skip it if it is not
2531 * valid, the wrong inode or stale.
2533 spin_lock(&ip->i_flags_lock);
2534 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE)) {
2535 spin_unlock(&ip->i_flags_lock);
2536 goto out_rcu_unlock;
2538 spin_unlock(&ip->i_flags_lock);
2541 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2542 * other inodes that we did not find in the list attached to the buffer
2543 * and are not already marked stale. If we can't lock it, back off and
2546 if (ip != free_ip) {
2547 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2554 * Check the inode number again in case we're racing with
2555 * freeing in xfs_reclaim_inode(). See the comments in that
2556 * function for more information as to why the initial check is
2559 if (ip->i_ino != inum) {
2560 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2561 goto out_rcu_unlock;
2567 xfs_iflags_set(ip, XFS_ISTALE);
2570 * We don't need to attach clean inodes or those only with unlogged
2571 * changes (which we throw away, anyway).
2573 if (!ip->i_itemp || xfs_inode_clean(ip)) {
2574 ASSERT(ip != free_ip);
2576 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2588 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2589 * inodes that are in memory - they all must be marked stale and attached to
2590 * the cluster buffer.
2594 xfs_inode_t *free_ip,
2596 struct xfs_icluster *xic)
2598 xfs_mount_t *mp = free_ip->i_mount;
2605 xfs_inode_log_item_t *iip;
2606 struct xfs_log_item *lip;
2607 struct xfs_perag *pag;
2608 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2612 inum = xic->first_ino;
2613 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2614 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2616 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2618 * The allocation bitmap tells us which inodes of the chunk were
2619 * physically allocated. Skip the cluster if an inode falls into
2622 ioffset = inum - xic->first_ino;
2623 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2624 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2628 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2629 XFS_INO_TO_AGBNO(mp, inum));
2632 * We obtain and lock the backing buffer first in the process
2633 * here, as we have to ensure that any dirty inode that we
2634 * can't get the flush lock on is attached to the buffer.
2635 * If we scan the in-memory inodes first, then buffer IO can
2636 * complete before we get a lock on it, and hence we may fail
2637 * to mark all the active inodes on the buffer stale.
2639 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2640 mp->m_bsize * igeo->blocks_per_cluster,
2646 * This buffer may not have been correctly initialised as we
2647 * didn't read it from disk. That's not important because we are
2648 * only using to mark the buffer as stale in the log, and to
2649 * attach stale cached inodes on it. That means it will never be
2650 * dispatched for IO. If it is, we want to know about it, and we
2651 * want it to fail. We can acheive this by adding a write
2652 * verifier to the buffer.
2654 bp->b_ops = &xfs_inode_buf_ops;
2657 * Walk the inodes already attached to the buffer and mark them
2658 * stale. These will all have the flush locks held, so an
2659 * in-memory inode walk can't lock them. By marking them all
2660 * stale first, we will not attempt to lock them in the loop
2661 * below as the XFS_ISTALE flag will be set.
2663 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
2664 if (lip->li_type == XFS_LI_INODE) {
2665 iip = (xfs_inode_log_item_t *)lip;
2666 ASSERT(iip->ili_logged == 1);
2667 lip->li_cb = xfs_istale_done;
2668 xfs_trans_ail_copy_lsn(mp->m_ail,
2669 &iip->ili_flush_lsn,
2670 &iip->ili_item.li_lsn);
2671 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2677 * For each inode in memory attempt to add it to the inode
2678 * buffer and set it up for being staled on buffer IO
2679 * completion. This is safe as we've locked out tail pushing
2680 * and flushing by locking the buffer.
2682 * We have already marked every inode that was part of a
2683 * transaction stale above, which means there is no point in
2684 * even trying to lock them.
2686 for (i = 0; i < igeo->inodes_per_cluster; i++) {
2687 ip = xfs_ifree_get_one_inode(pag, free_ip, inum + i);
2692 iip->ili_last_fields = iip->ili_fields;
2693 iip->ili_fields = 0;
2694 iip->ili_fsync_fields = 0;
2695 iip->ili_logged = 1;
2696 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2697 &iip->ili_item.li_lsn);
2699 xfs_buf_attach_iodone(bp, xfs_istale_done,
2703 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2706 xfs_trans_stale_inode_buf(tp, bp);
2707 xfs_trans_binval(tp, bp);
2715 * Free any local-format buffers sitting around before we reset to
2719 xfs_ifree_local_data(
2720 struct xfs_inode *ip,
2723 struct xfs_ifork *ifp;
2725 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2728 ifp = XFS_IFORK_PTR(ip, whichfork);
2729 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2733 * This is called to return an inode to the inode free list.
2734 * The inode should already be truncated to 0 length and have
2735 * no pages associated with it. This routine also assumes that
2736 * the inode is already a part of the transaction.
2738 * The on-disk copy of the inode will have been added to the list
2739 * of unlinked inodes in the AGI. We need to remove the inode from
2740 * that list atomically with respect to freeing it here.
2744 struct xfs_trans *tp,
2745 struct xfs_inode *ip)
2748 struct xfs_icluster xic = { 0 };
2750 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2751 ASSERT(VFS_I(ip)->i_nlink == 0);
2752 ASSERT(ip->i_d.di_nextents == 0);
2753 ASSERT(ip->i_d.di_anextents == 0);
2754 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2755 ASSERT(ip->i_d.di_nblocks == 0);
2758 * Pull the on-disk inode from the AGI unlinked list.
2760 error = xfs_iunlink_remove(tp, ip);
2764 error = xfs_difree(tp, ip->i_ino, &xic);
2768 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2769 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2771 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2772 ip->i_d.di_flags = 0;
2773 ip->i_d.di_flags2 = 0;
2774 ip->i_d.di_dmevmask = 0;
2775 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2776 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2777 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2779 /* Don't attempt to replay owner changes for a deleted inode */
2780 ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER);
2783 * Bump the generation count so no one will be confused
2784 * by reincarnations of this inode.
2786 VFS_I(ip)->i_generation++;
2787 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2790 error = xfs_ifree_cluster(ip, tp, &xic);
2796 * This is called to unpin an inode. The caller must have the inode locked
2797 * in at least shared mode so that the buffer cannot be subsequently pinned
2798 * once someone is waiting for it to be unpinned.
2802 struct xfs_inode *ip)
2804 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2806 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2808 /* Give the log a push to start the unpinning I/O */
2809 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2815 struct xfs_inode *ip)
2817 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2818 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2823 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2824 if (xfs_ipincount(ip))
2826 } while (xfs_ipincount(ip));
2827 finish_wait(wq, &wait.wq_entry);
2832 struct xfs_inode *ip)
2834 if (xfs_ipincount(ip))
2835 __xfs_iunpin_wait(ip);
2839 * Removing an inode from the namespace involves removing the directory entry
2840 * and dropping the link count on the inode. Removing the directory entry can
2841 * result in locking an AGF (directory blocks were freed) and removing a link
2842 * count can result in placing the inode on an unlinked list which results in
2845 * The big problem here is that we have an ordering constraint on AGF and AGI
2846 * locking - inode allocation locks the AGI, then can allocate a new extent for
2847 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2848 * removes the inode from the unlinked list, requiring that we lock the AGI
2849 * first, and then freeing the inode can result in an inode chunk being freed
2850 * and hence freeing disk space requiring that we lock an AGF.
2852 * Hence the ordering that is imposed by other parts of the code is AGI before
2853 * AGF. This means we cannot remove the directory entry before we drop the inode
2854 * reference count and put it on the unlinked list as this results in a lock
2855 * order of AGF then AGI, and this can deadlock against inode allocation and
2856 * freeing. Therefore we must drop the link counts before we remove the
2859 * This is still safe from a transactional point of view - it is not until we
2860 * get to xfs_defer_finish() that we have the possibility of multiple
2861 * transactions in this operation. Hence as long as we remove the directory
2862 * entry and drop the link count in the first transaction of the remove
2863 * operation, there are no transactional constraints on the ordering here.
2868 struct xfs_name *name,
2871 xfs_mount_t *mp = dp->i_mount;
2872 xfs_trans_t *tp = NULL;
2873 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2877 trace_xfs_remove(dp, name);
2879 if (XFS_FORCED_SHUTDOWN(mp))
2882 error = xfs_qm_dqattach(dp);
2886 error = xfs_qm_dqattach(ip);
2891 * We try to get the real space reservation first,
2892 * allowing for directory btree deletion(s) implying
2893 * possible bmap insert(s). If we can't get the space
2894 * reservation then we use 0 instead, and avoid the bmap
2895 * btree insert(s) in the directory code by, if the bmap
2896 * insert tries to happen, instead trimming the LAST
2897 * block from the directory.
2899 resblks = XFS_REMOVE_SPACE_RES(mp);
2900 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2901 if (error == -ENOSPC) {
2903 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2907 ASSERT(error != -ENOSPC);
2911 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2913 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2914 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2917 * If we're removing a directory perform some additional validation.
2920 ASSERT(VFS_I(ip)->i_nlink >= 2);
2921 if (VFS_I(ip)->i_nlink != 2) {
2923 goto out_trans_cancel;
2925 if (!xfs_dir_isempty(ip)) {
2927 goto out_trans_cancel;
2930 /* Drop the link from ip's "..". */
2931 error = xfs_droplink(tp, dp);
2933 goto out_trans_cancel;
2935 /* Drop the "." link from ip to self. */
2936 error = xfs_droplink(tp, ip);
2938 goto out_trans_cancel;
2941 * When removing a non-directory we need to log the parent
2942 * inode here. For a directory this is done implicitly
2943 * by the xfs_droplink call for the ".." entry.
2945 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2947 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2949 /* Drop the link from dp to ip. */
2950 error = xfs_droplink(tp, ip);
2952 goto out_trans_cancel;
2954 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2956 ASSERT(error != -ENOENT);
2957 goto out_trans_cancel;
2961 * If this is a synchronous mount, make sure that the
2962 * remove transaction goes to disk before returning to
2965 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2966 xfs_trans_set_sync(tp);
2968 error = xfs_trans_commit(tp);
2972 if (is_dir && xfs_inode_is_filestream(ip))
2973 xfs_filestream_deassociate(ip);
2978 xfs_trans_cancel(tp);
2984 * Enter all inodes for a rename transaction into a sorted array.
2986 #define __XFS_SORT_INODES 5
2988 xfs_sort_for_rename(
2989 struct xfs_inode *dp1, /* in: old (source) directory inode */
2990 struct xfs_inode *dp2, /* in: new (target) directory inode */
2991 struct xfs_inode *ip1, /* in: inode of old entry */
2992 struct xfs_inode *ip2, /* in: inode of new entry */
2993 struct xfs_inode *wip, /* in: whiteout inode */
2994 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2995 int *num_inodes) /* in/out: inodes in array */
2999 ASSERT(*num_inodes == __XFS_SORT_INODES);
3000 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
3003 * i_tab contains a list of pointers to inodes. We initialize
3004 * the table here & we'll sort it. We will then use it to
3005 * order the acquisition of the inode locks.
3007 * Note that the table may contain duplicates. e.g., dp1 == dp2.
3020 * Sort the elements via bubble sort. (Remember, there are at
3021 * most 5 elements to sort, so this is adequate.)
3023 for (i = 0; i < *num_inodes; i++) {
3024 for (j = 1; j < *num_inodes; j++) {
3025 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
3026 struct xfs_inode *temp = i_tab[j];
3027 i_tab[j] = i_tab[j-1];
3036 struct xfs_trans *tp)
3039 * If this is a synchronous mount, make sure that the rename transaction
3040 * goes to disk before returning to the user.
3042 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
3043 xfs_trans_set_sync(tp);
3045 return xfs_trans_commit(tp);
3049 * xfs_cross_rename()
3051 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
3055 struct xfs_trans *tp,
3056 struct xfs_inode *dp1,
3057 struct xfs_name *name1,
3058 struct xfs_inode *ip1,
3059 struct xfs_inode *dp2,
3060 struct xfs_name *name2,
3061 struct xfs_inode *ip2,
3069 /* Swap inode number for dirent in first parent */
3070 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
3072 goto out_trans_abort;
3074 /* Swap inode number for dirent in second parent */
3075 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
3077 goto out_trans_abort;
3080 * If we're renaming one or more directories across different parents,
3081 * update the respective ".." entries (and link counts) to match the new
3085 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3087 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
3088 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
3089 dp1->i_ino, spaceres);
3091 goto out_trans_abort;
3093 /* transfer ip2 ".." reference to dp1 */
3094 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
3095 error = xfs_droplink(tp, dp2);
3097 goto out_trans_abort;
3098 xfs_bumplink(tp, dp1);
3102 * Although ip1 isn't changed here, userspace needs
3103 * to be warned about the change, so that applications
3104 * relying on it (like backup ones), will properly
3107 ip1_flags |= XFS_ICHGTIME_CHG;
3108 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3111 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
3112 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
3113 dp2->i_ino, spaceres);
3115 goto out_trans_abort;
3117 /* transfer ip1 ".." reference to dp2 */
3118 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
3119 error = xfs_droplink(tp, dp1);
3121 goto out_trans_abort;
3122 xfs_bumplink(tp, dp2);
3126 * Although ip2 isn't changed here, userspace needs
3127 * to be warned about the change, so that applications
3128 * relying on it (like backup ones), will properly
3131 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3132 ip2_flags |= XFS_ICHGTIME_CHG;
3137 xfs_trans_ichgtime(tp, ip1, ip1_flags);
3138 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
3141 xfs_trans_ichgtime(tp, ip2, ip2_flags);
3142 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
3145 xfs_trans_ichgtime(tp, dp2, dp2_flags);
3146 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
3148 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3149 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3150 return xfs_finish_rename(tp);
3153 xfs_trans_cancel(tp);
3158 * xfs_rename_alloc_whiteout()
3160 * Return a referenced, unlinked, unlocked inode that that can be used as a
3161 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3162 * crash between allocating the inode and linking it into the rename transaction
3163 * recovery will free the inode and we won't leak it.
3166 xfs_rename_alloc_whiteout(
3167 struct xfs_inode *dp,
3168 struct xfs_inode **wip)
3170 struct xfs_inode *tmpfile;
3173 error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
3178 * Prepare the tmpfile inode as if it were created through the VFS.
3179 * Complete the inode setup and flag it as linkable. nlink is already
3180 * zero, so we can skip the drop_nlink.
3182 xfs_setup_iops(tmpfile);
3183 xfs_finish_inode_setup(tmpfile);
3184 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3195 struct xfs_inode *src_dp,
3196 struct xfs_name *src_name,
3197 struct xfs_inode *src_ip,
3198 struct xfs_inode *target_dp,
3199 struct xfs_name *target_name,
3200 struct xfs_inode *target_ip,
3203 struct xfs_mount *mp = src_dp->i_mount;
3204 struct xfs_trans *tp;
3205 struct xfs_inode *wip = NULL; /* whiteout inode */
3206 struct xfs_inode *inodes[__XFS_SORT_INODES];
3207 struct xfs_buf *agibp;
3208 int num_inodes = __XFS_SORT_INODES;
3209 bool new_parent = (src_dp != target_dp);
3210 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3214 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3216 if ((flags & RENAME_EXCHANGE) && !target_ip)
3220 * If we are doing a whiteout operation, allocate the whiteout inode
3221 * we will be placing at the target and ensure the type is set
3224 if (flags & RENAME_WHITEOUT) {
3225 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3226 error = xfs_rename_alloc_whiteout(target_dp, &wip);
3230 /* setup target dirent info as whiteout */
3231 src_name->type = XFS_DIR3_FT_CHRDEV;
3234 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3235 inodes, &num_inodes);
3237 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3238 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3239 if (error == -ENOSPC) {
3241 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3245 goto out_release_wip;
3248 * Attach the dquots to the inodes
3250 error = xfs_qm_vop_rename_dqattach(inodes);
3252 goto out_trans_cancel;
3255 * Lock all the participating inodes. Depending upon whether
3256 * the target_name exists in the target directory, and
3257 * whether the target directory is the same as the source
3258 * directory, we can lock from 2 to 4 inodes.
3260 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3263 * Join all the inodes to the transaction. From this point on,
3264 * we can rely on either trans_commit or trans_cancel to unlock
3267 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3269 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3270 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3272 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3274 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3277 * If we are using project inheritance, we only allow renames
3278 * into our tree when the project IDs are the same; else the
3279 * tree quota mechanism would be circumvented.
3281 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3282 target_dp->i_d.di_projid != src_ip->i_d.di_projid)) {
3284 goto out_trans_cancel;
3287 /* RENAME_EXCHANGE is unique from here on. */
3288 if (flags & RENAME_EXCHANGE)
3289 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3290 target_dp, target_name, target_ip,
3294 * Check for expected errors before we dirty the transaction
3295 * so we can return an error without a transaction abort.
3297 if (target_ip == NULL) {
3299 * If there's no space reservation, check the entry will
3300 * fit before actually inserting it.
3303 error = xfs_dir_canenter(tp, target_dp, target_name);
3305 goto out_trans_cancel;
3309 * If target exists and it's a directory, check that whether
3310 * it can be destroyed.
3312 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3313 (!xfs_dir_isempty(target_ip) ||
3314 (VFS_I(target_ip)->i_nlink > 2))) {
3316 goto out_trans_cancel;
3321 * Directory entry creation below may acquire the AGF. Remove
3322 * the whiteout from the unlinked list first to preserve correct
3323 * AGI/AGF locking order. This dirties the transaction so failures
3324 * after this point will abort and log recovery will clean up the
3327 * For whiteouts, we need to bump the link count on the whiteout
3328 * inode. After this point, we have a real link, clear the tmpfile
3329 * state flag from the inode so it doesn't accidentally get misused
3333 ASSERT(VFS_I(wip)->i_nlink == 0);
3334 error = xfs_iunlink_remove(tp, wip);
3336 goto out_trans_cancel;
3338 xfs_bumplink(tp, wip);
3339 VFS_I(wip)->i_state &= ~I_LINKABLE;
3343 * Set up the target.
3345 if (target_ip == NULL) {
3347 * If target does not exist and the rename crosses
3348 * directories, adjust the target directory link count
3349 * to account for the ".." reference from the new entry.
3351 error = xfs_dir_createname(tp, target_dp, target_name,
3352 src_ip->i_ino, spaceres);
3354 goto out_trans_cancel;
3356 xfs_trans_ichgtime(tp, target_dp,
3357 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3359 if (new_parent && src_is_directory) {
3360 xfs_bumplink(tp, target_dp);
3362 } else { /* target_ip != NULL */
3364 * Link the source inode under the target name.
3365 * If the source inode is a directory and we are moving
3366 * it across directories, its ".." entry will be
3367 * inconsistent until we replace that down below.
3369 * In case there is already an entry with the same
3370 * name at the destination directory, remove it first.
3374 * Check whether the replace operation will need to allocate
3375 * blocks. This happens when the shortform directory lacks
3376 * space and we have to convert it to a block format directory.
3377 * When more blocks are necessary, we must lock the AGI first
3378 * to preserve locking order (AGI -> AGF).
3380 if (xfs_dir2_sf_replace_needblock(target_dp, src_ip->i_ino)) {
3381 error = xfs_read_agi(mp, tp,
3382 XFS_INO_TO_AGNO(mp, target_ip->i_ino),
3385 goto out_trans_cancel;
3388 error = xfs_dir_replace(tp, target_dp, target_name,
3389 src_ip->i_ino, spaceres);
3391 goto out_trans_cancel;
3393 xfs_trans_ichgtime(tp, target_dp,
3394 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3397 * Decrement the link count on the target since the target
3398 * dir no longer points to it.
3400 error = xfs_droplink(tp, target_ip);
3402 goto out_trans_cancel;
3404 if (src_is_directory) {
3406 * Drop the link from the old "." entry.
3408 error = xfs_droplink(tp, target_ip);
3410 goto out_trans_cancel;
3412 } /* target_ip != NULL */
3415 * Remove the source.
3417 if (new_parent && src_is_directory) {
3419 * Rewrite the ".." entry to point to the new
3422 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3423 target_dp->i_ino, spaceres);
3424 ASSERT(error != -EEXIST);
3426 goto out_trans_cancel;
3430 * We always want to hit the ctime on the source inode.
3432 * This isn't strictly required by the standards since the source
3433 * inode isn't really being changed, but old unix file systems did
3434 * it and some incremental backup programs won't work without it.
3436 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3437 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3440 * Adjust the link count on src_dp. This is necessary when
3441 * renaming a directory, either within one parent when
3442 * the target existed, or across two parent directories.
3444 if (src_is_directory && (new_parent || target_ip != NULL)) {
3447 * Decrement link count on src_directory since the
3448 * entry that's moved no longer points to it.
3450 error = xfs_droplink(tp, src_dp);
3452 goto out_trans_cancel;
3456 * For whiteouts, we only need to update the source dirent with the
3457 * inode number of the whiteout inode rather than removing it
3461 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3464 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3467 goto out_trans_cancel;
3469 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3470 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3472 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3474 error = xfs_finish_rename(tp);
3480 xfs_trans_cancel(tp);
3489 struct xfs_inode *ip,
3492 struct xfs_mount *mp = ip->i_mount;
3493 struct xfs_perag *pag;
3494 unsigned long first_index, mask;
3496 struct xfs_inode **cilist;
3497 struct xfs_inode *cip;
3498 struct xfs_ino_geometry *igeo = M_IGEO(mp);
3503 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3505 cilist_size = igeo->inodes_per_cluster * sizeof(struct xfs_inode *);
3506 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3510 mask = ~(igeo->inodes_per_cluster - 1);
3511 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3513 /* really need a gang lookup range call here */
3514 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3515 first_index, igeo->inodes_per_cluster);
3519 for (i = 0; i < nr_found; i++) {
3525 * because this is an RCU protected lookup, we could find a
3526 * recently freed or even reallocated inode during the lookup.
3527 * We need to check under the i_flags_lock for a valid inode
3528 * here. Skip it if it is not valid or the wrong inode.
3530 spin_lock(&cip->i_flags_lock);
3532 __xfs_iflags_test(cip, XFS_ISTALE)) {
3533 spin_unlock(&cip->i_flags_lock);
3538 * Once we fall off the end of the cluster, no point checking
3539 * any more inodes in the list because they will also all be
3540 * outside the cluster.
3542 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3543 spin_unlock(&cip->i_flags_lock);
3546 spin_unlock(&cip->i_flags_lock);
3549 * Do an un-protected check to see if the inode is dirty and
3550 * is a candidate for flushing. These checks will be repeated
3551 * later after the appropriate locks are acquired.
3553 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3557 * Try to get locks. If any are unavailable or it is pinned,
3558 * then this inode cannot be flushed and is skipped.
3561 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3563 if (!xfs_iflock_nowait(cip)) {
3564 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3567 if (xfs_ipincount(cip)) {
3569 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3575 * Check the inode number again, just to be certain we are not
3576 * racing with freeing in xfs_reclaim_inode(). See the comments
3577 * in that function for more information as to why the initial
3578 * check is not sufficient.
3582 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3587 * arriving here means that this inode can be flushed. First
3588 * re-check that it's dirty before flushing.
3590 if (!xfs_inode_clean(cip)) {
3592 error = xfs_iflush_int(cip, bp);
3594 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3595 goto cluster_corrupt_out;
3601 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3605 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3606 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3617 cluster_corrupt_out:
3619 * Corruption detected in the clustering loop. Invalidate the
3620 * inode buffer and shut down the filesystem.
3625 * We'll always have an inode attached to the buffer for completion
3626 * process by the time we are called from xfs_iflush(). Hence we have
3627 * always need to do IO completion processing to abort the inodes
3628 * attached to the buffer. handle them just like the shutdown case in
3631 ASSERT(bp->b_iodone);
3632 bp->b_flags |= XBF_ASYNC;
3633 bp->b_flags &= ~XBF_DONE;
3635 xfs_buf_ioerror(bp, -EIO);
3638 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3640 /* abort the corrupt inode, as it was not attached to the buffer */
3641 xfs_iflush_abort(cip, false);
3644 return -EFSCORRUPTED;
3648 * Flush dirty inode metadata into the backing buffer.
3650 * The caller must have the inode lock and the inode flush lock held. The
3651 * inode lock will still be held upon return to the caller, and the inode
3652 * flush lock will be released after the inode has reached the disk.
3654 * The caller must write out the buffer returned in *bpp and release it.
3658 struct xfs_inode *ip,
3659 struct xfs_buf **bpp)
3661 struct xfs_mount *mp = ip->i_mount;
3662 struct xfs_buf *bp = NULL;
3663 struct xfs_dinode *dip;
3666 XFS_STATS_INC(mp, xs_iflush_count);
3668 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3669 ASSERT(xfs_isiflocked(ip));
3670 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3671 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3675 xfs_iunpin_wait(ip);
3678 * For stale inodes we cannot rely on the backing buffer remaining
3679 * stale in cache for the remaining life of the stale inode and so
3680 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3681 * inodes below. We have to check this after ensuring the inode is
3682 * unpinned so that it is safe to reclaim the stale inode after the
3685 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3691 * This may have been unpinned because the filesystem is shutting
3692 * down forcibly. If that's the case we must not write this inode
3693 * to disk, because the log record didn't make it to disk.
3695 * We also have to remove the log item from the AIL in this case,
3696 * as we wait for an empty AIL as part of the unmount process.
3698 if (XFS_FORCED_SHUTDOWN(mp)) {
3704 * Get the buffer containing the on-disk inode. We are doing a try-lock
3705 * operation here, so we may get an EAGAIN error. In that case, we
3706 * simply want to return with the inode still dirty.
3708 * If we get any other error, we effectively have a corruption situation
3709 * and we cannot flush the inode, so we treat it the same as failing
3712 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3714 if (error == -EAGAIN) {
3722 * First flush out the inode that xfs_iflush was called with.
3724 error = xfs_iflush_int(ip, bp);
3729 * If the buffer is pinned then push on the log now so we won't
3730 * get stuck waiting in the write for too long.
3732 if (xfs_buf_ispinned(bp))
3733 xfs_log_force(mp, 0);
3736 * inode clustering: try to gather other inodes into this write
3738 * Note: Any error during clustering will result in the filesystem
3739 * being shut down and completion callbacks run on the cluster buffer.
3740 * As we have already flushed and attached this inode to the buffer,
3741 * it has already been aborted and released by xfs_iflush_cluster() and
3742 * so we have no further error handling to do here.
3744 error = xfs_iflush_cluster(ip, bp);
3754 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3756 /* abort the corrupt inode, as it was not attached to the buffer */
3757 xfs_iflush_abort(ip, false);
3762 * If there are inline format data / attr forks attached to this inode,
3763 * make sure they're not corrupt.
3766 xfs_inode_verify_forks(
3767 struct xfs_inode *ip)
3769 struct xfs_ifork *ifp;
3772 fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops);
3774 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3775 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
3776 ifp->if_u1.if_data, ifp->if_bytes, fa);
3780 fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops);
3782 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
3783 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
3784 ifp ? ifp->if_u1.if_data : NULL,
3785 ifp ? ifp->if_bytes : 0, fa);
3793 struct xfs_inode *ip,
3796 struct xfs_inode_log_item *iip = ip->i_itemp;
3797 struct xfs_dinode *dip;
3798 struct xfs_mount *mp = ip->i_mount;
3800 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3801 ASSERT(xfs_isiflocked(ip));
3802 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3803 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3804 ASSERT(iip != NULL && iip->ili_fields != 0);
3806 /* set *dip = inode's place in the buffer */
3807 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3809 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3810 mp, XFS_ERRTAG_IFLUSH_1)) {
3811 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3812 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3813 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3816 if (S_ISREG(VFS_I(ip)->i_mode)) {
3818 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3819 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3820 mp, XFS_ERRTAG_IFLUSH_3)) {
3821 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3822 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3823 __func__, ip->i_ino, ip);
3826 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3828 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3829 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3830 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3831 mp, XFS_ERRTAG_IFLUSH_4)) {
3832 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3833 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3834 __func__, ip->i_ino, ip);
3838 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3839 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3840 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3841 "%s: detected corrupt incore inode %Lu, "
3842 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3843 __func__, ip->i_ino,
3844 ip->i_d.di_nextents + ip->i_d.di_anextents,
3845 ip->i_d.di_nblocks, ip);
3848 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3849 mp, XFS_ERRTAG_IFLUSH_6)) {
3850 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3851 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3852 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3857 * Inode item log recovery for v2 inodes are dependent on the
3858 * di_flushiter count for correct sequencing. We bump the flush
3859 * iteration count so we can detect flushes which postdate a log record
3860 * during recovery. This is redundant as we now log every change and
3861 * hence this can't happen but we need to still do it to ensure
3862 * backwards compatibility with old kernels that predate logging all
3865 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3866 ip->i_d.di_flushiter++;
3868 /* Check the inline fork data before we write out. */
3869 if (!xfs_inode_verify_forks(ip))
3873 * Copy the dirty parts of the inode into the on-disk inode. We always
3874 * copy out the core of the inode, because if the inode is dirty at all
3877 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3879 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3880 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3881 ip->i_d.di_flushiter = 0;
3883 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3884 if (XFS_IFORK_Q(ip))
3885 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3886 xfs_inobp_check(mp, bp);
3889 * We've recorded everything logged in the inode, so we'd like to clear
3890 * the ili_fields bits so we don't log and flush things unnecessarily.
3891 * However, we can't stop logging all this information until the data
3892 * we've copied into the disk buffer is written to disk. If we did we
3893 * might overwrite the copy of the inode in the log with all the data
3894 * after re-logging only part of it, and in the face of a crash we
3895 * wouldn't have all the data we need to recover.
3897 * What we do is move the bits to the ili_last_fields field. When
3898 * logging the inode, these bits are moved back to the ili_fields field.
3899 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3900 * know that the information those bits represent is permanently on
3901 * disk. As long as the flush completes before the inode is logged
3902 * again, then both ili_fields and ili_last_fields will be cleared.
3904 * We can play with the ili_fields bits here, because the inode lock
3905 * must be held exclusively in order to set bits there and the flush
3906 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3907 * done routine can tell whether or not to look in the AIL. Also, store
3908 * the current LSN of the inode so that we can tell whether the item has
3909 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3910 * need the AIL lock, because it is a 64 bit value that cannot be read
3913 iip->ili_last_fields = iip->ili_fields;
3914 iip->ili_fields = 0;
3915 iip->ili_fsync_fields = 0;
3916 iip->ili_logged = 1;
3918 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3919 &iip->ili_item.li_lsn);
3922 * Attach the function xfs_iflush_done to the inode's
3923 * buffer. This will remove the inode from the AIL
3924 * and unlock the inode's flush lock when the inode is
3925 * completely written to disk.
3927 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3929 /* generate the checksum. */
3930 xfs_dinode_calc_crc(mp, dip);
3932 ASSERT(!list_empty(&bp->b_li_list));
3933 ASSERT(bp->b_iodone != NULL);
3937 return -EFSCORRUPTED;
3940 /* Release an inode. */
3943 struct xfs_inode *ip)
3945 trace_xfs_irele(ip, _RET_IP_);
3950 * Ensure all commited transactions touching the inode are written to the log.
3953 xfs_log_force_inode(
3954 struct xfs_inode *ip)
3958 xfs_ilock(ip, XFS_ILOCK_SHARED);
3959 if (xfs_ipincount(ip))
3960 lsn = ip->i_itemp->ili_last_lsn;
3961 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3965 return xfs_log_force_lsn(ip->i_mount, lsn, XFS_LOG_SYNC, NULL);