2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
19 #include <linux/iversion.h>
23 #include "xfs_shared.h"
24 #include "xfs_format.h"
25 #include "xfs_log_format.h"
26 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_defer.h"
30 #include "xfs_inode.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
34 #include "xfs_attr_sf.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_bmap_util.h"
43 #include "xfs_errortag.h"
44 #include "xfs_error.h"
45 #include "xfs_quota.h"
46 #include "xfs_filestream.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
50 #include "xfs_symlink.h"
51 #include "xfs_trans_priv.h"
53 #include "xfs_bmap_btree.h"
54 #include "xfs_reflink.h"
55 #include "xfs_dir2_priv.h"
57 kmem_zone_t *xfs_inode_zone;
60 * Used in xfs_itruncate_extents(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
66 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
67 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
70 * helper function to extract extent size hint from inode
76 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
77 return ip->i_d.di_extsize;
78 if (XFS_IS_REALTIME_INODE(ip))
79 return ip->i_mount->m_sb.sb_rextsize;
84 * Helper function to extract CoW extent size hint from inode.
85 * Between the extent size hint and the CoW extent size hint, we
86 * return the greater of the two. If the value is zero (automatic),
87 * use the default size.
90 xfs_get_cowextsz_hint(
96 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
97 a = ip->i_d.di_cowextsize;
98 b = xfs_get_extsz_hint(ip);
102 return XFS_DEFAULT_COWEXTSZ_HINT;
107 * These two are wrapper routines around the xfs_ilock() routine used to
108 * centralize some grungy code. They are used in places that wish to lock the
109 * inode solely for reading the extents. The reason these places can't just
110 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
111 * bringing in of the extents from disk for a file in b-tree format. If the
112 * inode is in b-tree format, then we need to lock the inode exclusively until
113 * the extents are read in. Locking it exclusively all the time would limit
114 * our parallelism unnecessarily, though. What we do instead is check to see
115 * if the extents have been read in yet, and only lock the inode exclusively
118 * The functions return a value which should be given to the corresponding
119 * xfs_iunlock() call.
122 xfs_ilock_data_map_shared(
123 struct xfs_inode *ip)
125 uint lock_mode = XFS_ILOCK_SHARED;
127 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
128 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
129 lock_mode = XFS_ILOCK_EXCL;
130 xfs_ilock(ip, lock_mode);
135 xfs_ilock_attr_map_shared(
136 struct xfs_inode *ip)
138 uint lock_mode = XFS_ILOCK_SHARED;
140 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
141 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
142 lock_mode = XFS_ILOCK_EXCL;
143 xfs_ilock(ip, lock_mode);
148 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
149 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
150 * various combinations of the locks to be obtained.
152 * The 3 locks should always be ordered so that the IO lock is obtained first,
153 * the mmap lock second and the ilock last in order to prevent deadlock.
155 * Basic locking order:
157 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
159 * mmap_sem locking order:
161 * i_rwsem -> page lock -> mmap_sem
162 * mmap_sem -> i_mmap_lock -> page_lock
164 * The difference in mmap_sem locking order mean that we cannot hold the
165 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
166 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
167 * in get_user_pages() to map the user pages into the kernel address space for
168 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
169 * page faults already hold the mmap_sem.
171 * Hence to serialise fully against both syscall and mmap based IO, we need to
172 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
173 * taken in places where we need to invalidate the page cache in a race
174 * free manner (e.g. truncate, hole punch and other extent manipulation
182 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
185 * You can't set both SHARED and EXCL for the same lock,
186 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
187 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
189 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
190 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
191 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
192 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
193 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
194 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
195 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
197 if (lock_flags & XFS_IOLOCK_EXCL) {
198 down_write_nested(&VFS_I(ip)->i_rwsem,
199 XFS_IOLOCK_DEP(lock_flags));
200 } else if (lock_flags & XFS_IOLOCK_SHARED) {
201 down_read_nested(&VFS_I(ip)->i_rwsem,
202 XFS_IOLOCK_DEP(lock_flags));
205 if (lock_flags & XFS_MMAPLOCK_EXCL)
206 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
207 else if (lock_flags & XFS_MMAPLOCK_SHARED)
208 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
210 if (lock_flags & XFS_ILOCK_EXCL)
211 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 else if (lock_flags & XFS_ILOCK_SHARED)
213 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
217 * This is just like xfs_ilock(), except that the caller
218 * is guaranteed not to sleep. It returns 1 if it gets
219 * the requested locks and 0 otherwise. If the IO lock is
220 * obtained but the inode lock cannot be, then the IO lock
221 * is dropped before returning.
223 * ip -- the inode being locked
224 * lock_flags -- this parameter indicates the inode's locks to be
225 * to be locked. See the comment for xfs_ilock() for a list
233 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
236 * You can't set both SHARED and EXCL for the same lock,
237 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
238 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
240 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
241 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
242 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
243 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
244 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
245 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
246 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
248 if (lock_flags & XFS_IOLOCK_EXCL) {
249 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
251 } else if (lock_flags & XFS_IOLOCK_SHARED) {
252 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
256 if (lock_flags & XFS_MMAPLOCK_EXCL) {
257 if (!mrtryupdate(&ip->i_mmaplock))
258 goto out_undo_iolock;
259 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
260 if (!mrtryaccess(&ip->i_mmaplock))
261 goto out_undo_iolock;
264 if (lock_flags & XFS_ILOCK_EXCL) {
265 if (!mrtryupdate(&ip->i_lock))
266 goto out_undo_mmaplock;
267 } else if (lock_flags & XFS_ILOCK_SHARED) {
268 if (!mrtryaccess(&ip->i_lock))
269 goto out_undo_mmaplock;
274 if (lock_flags & XFS_MMAPLOCK_EXCL)
275 mrunlock_excl(&ip->i_mmaplock);
276 else if (lock_flags & XFS_MMAPLOCK_SHARED)
277 mrunlock_shared(&ip->i_mmaplock);
279 if (lock_flags & XFS_IOLOCK_EXCL)
280 up_write(&VFS_I(ip)->i_rwsem);
281 else if (lock_flags & XFS_IOLOCK_SHARED)
282 up_read(&VFS_I(ip)->i_rwsem);
288 * xfs_iunlock() is used to drop the inode locks acquired with
289 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
290 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
291 * that we know which locks to drop.
293 * ip -- the inode being unlocked
294 * lock_flags -- this parameter indicates the inode's locks to be
295 * to be unlocked. See the comment for xfs_ilock() for a list
296 * of valid values for this parameter.
305 * You can't set both SHARED and EXCL for the same lock,
306 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
307 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
309 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
310 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
311 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
312 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
313 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
314 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
315 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
316 ASSERT(lock_flags != 0);
318 if (lock_flags & XFS_IOLOCK_EXCL)
319 up_write(&VFS_I(ip)->i_rwsem);
320 else if (lock_flags & XFS_IOLOCK_SHARED)
321 up_read(&VFS_I(ip)->i_rwsem);
323 if (lock_flags & XFS_MMAPLOCK_EXCL)
324 mrunlock_excl(&ip->i_mmaplock);
325 else if (lock_flags & XFS_MMAPLOCK_SHARED)
326 mrunlock_shared(&ip->i_mmaplock);
328 if (lock_flags & XFS_ILOCK_EXCL)
329 mrunlock_excl(&ip->i_lock);
330 else if (lock_flags & XFS_ILOCK_SHARED)
331 mrunlock_shared(&ip->i_lock);
333 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
337 * give up write locks. the i/o lock cannot be held nested
338 * if it is being demoted.
345 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
347 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
349 if (lock_flags & XFS_ILOCK_EXCL)
350 mrdemote(&ip->i_lock);
351 if (lock_flags & XFS_MMAPLOCK_EXCL)
352 mrdemote(&ip->i_mmaplock);
353 if (lock_flags & XFS_IOLOCK_EXCL)
354 downgrade_write(&VFS_I(ip)->i_rwsem);
356 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
359 #if defined(DEBUG) || defined(XFS_WARN)
365 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
366 if (!(lock_flags & XFS_ILOCK_SHARED))
367 return !!ip->i_lock.mr_writer;
368 return rwsem_is_locked(&ip->i_lock.mr_lock);
371 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
372 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
373 return !!ip->i_mmaplock.mr_writer;
374 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
377 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
378 if (!(lock_flags & XFS_IOLOCK_SHARED))
379 return !debug_locks ||
380 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
381 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
390 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
391 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
392 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
393 * errors and warnings.
395 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
397 xfs_lockdep_subclass_ok(
400 return subclass < MAX_LOCKDEP_SUBCLASSES;
403 #define xfs_lockdep_subclass_ok(subclass) (true)
407 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
408 * value. This can be called for any type of inode lock combination, including
409 * parent locking. Care must be taken to ensure we don't overrun the subclass
410 * storage fields in the class mask we build.
413 xfs_lock_inumorder(int lock_mode, int subclass)
417 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
419 ASSERT(xfs_lockdep_subclass_ok(subclass));
421 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
422 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
423 class += subclass << XFS_IOLOCK_SHIFT;
426 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
427 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
428 class += subclass << XFS_MMAPLOCK_SHIFT;
431 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
432 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
433 class += subclass << XFS_ILOCK_SHIFT;
436 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
440 * The following routine will lock n inodes in exclusive mode. We assume the
441 * caller calls us with the inodes in i_ino order.
443 * We need to detect deadlock where an inode that we lock is in the AIL and we
444 * start waiting for another inode that is locked by a thread in a long running
445 * transaction (such as truncate). This can result in deadlock since the long
446 * running trans might need to wait for the inode we just locked in order to
447 * push the tail and free space in the log.
449 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
450 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
451 * lock more than one at a time, lockdep will report false positives saying we
452 * have violated locking orders.
460 int attempts = 0, i, j, try_lock;
464 * Currently supports between 2 and 5 inodes with exclusive locking. We
465 * support an arbitrary depth of locking here, but absolute limits on
466 * inodes depend on the the type of locking and the limits placed by
467 * lockdep annotations in xfs_lock_inumorder. These are all checked by
470 ASSERT(ips && inodes >= 2 && inodes <= 5);
471 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
473 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
475 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
476 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
477 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
478 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
480 if (lock_mode & XFS_IOLOCK_EXCL) {
481 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
482 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
483 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
488 for (; i < inodes; i++) {
491 if (i && (ips[i] == ips[i - 1])) /* Already locked */
495 * If try_lock is not set yet, make sure all locked inodes are
496 * not in the AIL. If any are, set try_lock to be used later.
499 for (j = (i - 1); j >= 0 && !try_lock; j--) {
500 lp = (xfs_log_item_t *)ips[j]->i_itemp;
501 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
507 * If any of the previous locks we have locked is in the AIL,
508 * we must TRY to get the second and subsequent locks. If
509 * we can't get any, we must release all we have
513 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
517 /* try_lock means we have an inode locked that is in the AIL. */
519 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
523 * Unlock all previous guys and try again. xfs_iunlock will try
524 * to push the tail if the inode is in the AIL.
527 for (j = i - 1; j >= 0; j--) {
529 * Check to see if we've already unlocked this one. Not
530 * the first one going back, and the inode ptr is the
533 if (j != (i - 1) && ips[j] == ips[j + 1])
536 xfs_iunlock(ips[j], lock_mode);
539 if ((attempts % 5) == 0) {
540 delay(1); /* Don't just spin the CPU */
549 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
550 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
551 * lock more than one at a time, lockdep will report false positives saying we
552 * have violated locking orders.
564 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
565 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
566 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
568 ASSERT(ip0->i_ino != ip1->i_ino);
570 if (ip0->i_ino > ip1->i_ino) {
577 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
580 * If the first lock we have locked is in the AIL, we must TRY to get
581 * the second lock. If we can't get it, we must release the first one
584 lp = (xfs_log_item_t *)ip0->i_itemp;
585 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
586 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
587 xfs_iunlock(ip0, lock_mode);
588 if ((++attempts % 5) == 0)
589 delay(1); /* Don't just spin the CPU */
593 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
600 struct xfs_inode *ip)
602 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
603 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
606 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
607 if (xfs_isiflocked(ip))
609 } while (!xfs_iflock_nowait(ip));
611 finish_wait(wq, &wait.wq_entry);
622 if (di_flags & XFS_DIFLAG_ANY) {
623 if (di_flags & XFS_DIFLAG_REALTIME)
624 flags |= FS_XFLAG_REALTIME;
625 if (di_flags & XFS_DIFLAG_PREALLOC)
626 flags |= FS_XFLAG_PREALLOC;
627 if (di_flags & XFS_DIFLAG_IMMUTABLE)
628 flags |= FS_XFLAG_IMMUTABLE;
629 if (di_flags & XFS_DIFLAG_APPEND)
630 flags |= FS_XFLAG_APPEND;
631 if (di_flags & XFS_DIFLAG_SYNC)
632 flags |= FS_XFLAG_SYNC;
633 if (di_flags & XFS_DIFLAG_NOATIME)
634 flags |= FS_XFLAG_NOATIME;
635 if (di_flags & XFS_DIFLAG_NODUMP)
636 flags |= FS_XFLAG_NODUMP;
637 if (di_flags & XFS_DIFLAG_RTINHERIT)
638 flags |= FS_XFLAG_RTINHERIT;
639 if (di_flags & XFS_DIFLAG_PROJINHERIT)
640 flags |= FS_XFLAG_PROJINHERIT;
641 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
642 flags |= FS_XFLAG_NOSYMLINKS;
643 if (di_flags & XFS_DIFLAG_EXTSIZE)
644 flags |= FS_XFLAG_EXTSIZE;
645 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
646 flags |= FS_XFLAG_EXTSZINHERIT;
647 if (di_flags & XFS_DIFLAG_NODEFRAG)
648 flags |= FS_XFLAG_NODEFRAG;
649 if (di_flags & XFS_DIFLAG_FILESTREAM)
650 flags |= FS_XFLAG_FILESTREAM;
653 if (di_flags2 & XFS_DIFLAG2_ANY) {
654 if (di_flags2 & XFS_DIFLAG2_DAX)
655 flags |= FS_XFLAG_DAX;
656 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
657 flags |= FS_XFLAG_COWEXTSIZE;
661 flags |= FS_XFLAG_HASATTR;
668 struct xfs_inode *ip)
670 struct xfs_icdinode *dic = &ip->i_d;
672 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
676 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
677 * is allowed, otherwise it has to be an exact match. If a CI match is found,
678 * ci_name->name will point to a the actual name (caller must free) or
679 * will be set to NULL if an exact match is found.
684 struct xfs_name *name,
686 struct xfs_name *ci_name)
691 trace_xfs_lookup(dp, name);
693 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
696 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
700 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
708 kmem_free(ci_name->name);
715 * Allocate an inode on disk and return a copy of its in-core version.
716 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
717 * appropriately within the inode. The uid and gid for the inode are
718 * set according to the contents of the given cred structure.
720 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
721 * has a free inode available, call xfs_iget() to obtain the in-core
722 * version of the allocated inode. Finally, fill in the inode and
723 * log its initial contents. In this case, ialloc_context would be
726 * If xfs_dialloc() does not have an available inode, it will replenish
727 * its supply by doing an allocation. Since we can only do one
728 * allocation within a transaction without deadlocks, we must commit
729 * the current transaction before returning the inode itself.
730 * In this case, therefore, we will set ialloc_context and return.
731 * The caller should then commit the current transaction, start a new
732 * transaction, and call xfs_ialloc() again to actually get the inode.
734 * To ensure that some other process does not grab the inode that
735 * was allocated during the first call to xfs_ialloc(), this routine
736 * also returns the [locked] bp pointing to the head of the freelist
737 * as ialloc_context. The caller should hold this buffer across
738 * the commit and pass it back into this routine on the second call.
740 * If we are allocating quota inodes, we do not have a parent inode
741 * to attach to or associate with (i.e. pip == NULL) because they
742 * are not linked into the directory structure - they are attached
743 * directly to the superblock - and so have no parent.
754 xfs_buf_t **ialloc_context,
757 struct xfs_mount *mp = tp->t_mountp;
766 * Call the space management code to pick
767 * the on-disk inode to be allocated.
769 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
770 ialloc_context, &ino);
773 if (*ialloc_context || ino == NULLFSINO) {
777 ASSERT(*ialloc_context == NULL);
780 * Get the in-core inode with the lock held exclusively.
781 * This is because we're setting fields here we need
782 * to prevent others from looking at until we're done.
784 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
785 XFS_ILOCK_EXCL, &ip);
792 * We always convert v1 inodes to v2 now - we only support filesystems
793 * with >= v2 inode capability, so there is no reason for ever leaving
794 * an inode in v1 format.
796 if (ip->i_d.di_version == 1)
797 ip->i_d.di_version = 2;
799 inode->i_mode = mode;
800 set_nlink(inode, nlink);
801 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
802 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
803 inode->i_rdev = rdev;
804 xfs_set_projid(ip, prid);
806 if (pip && XFS_INHERIT_GID(pip)) {
807 ip->i_d.di_gid = pip->i_d.di_gid;
808 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
809 inode->i_mode |= S_ISGID;
813 * If the group ID of the new file does not match the effective group
814 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
815 * (and only if the irix_sgid_inherit compatibility variable is set).
817 if ((irix_sgid_inherit) &&
818 (inode->i_mode & S_ISGID) &&
819 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
820 inode->i_mode &= ~S_ISGID;
823 ip->i_d.di_nextents = 0;
824 ASSERT(ip->i_d.di_nblocks == 0);
826 tv = current_time(inode);
831 ip->i_d.di_extsize = 0;
832 ip->i_d.di_dmevmask = 0;
833 ip->i_d.di_dmstate = 0;
834 ip->i_d.di_flags = 0;
836 if (ip->i_d.di_version == 3) {
837 inode_set_iversion(inode, 1);
838 ip->i_d.di_flags2 = 0;
839 ip->i_d.di_cowextsize = 0;
840 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
841 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
845 flags = XFS_ILOG_CORE;
846 switch (mode & S_IFMT) {
851 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
852 ip->i_df.if_flags = 0;
853 flags |= XFS_ILOG_DEV;
857 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
861 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
862 di_flags |= XFS_DIFLAG_RTINHERIT;
863 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
864 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
865 ip->i_d.di_extsize = pip->i_d.di_extsize;
867 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
868 di_flags |= XFS_DIFLAG_PROJINHERIT;
869 } else if (S_ISREG(mode)) {
870 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
871 di_flags |= XFS_DIFLAG_REALTIME;
872 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
873 di_flags |= XFS_DIFLAG_EXTSIZE;
874 ip->i_d.di_extsize = pip->i_d.di_extsize;
877 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
879 di_flags |= XFS_DIFLAG_NOATIME;
880 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
882 di_flags |= XFS_DIFLAG_NODUMP;
883 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
885 di_flags |= XFS_DIFLAG_SYNC;
886 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
887 xfs_inherit_nosymlinks)
888 di_flags |= XFS_DIFLAG_NOSYMLINKS;
889 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
890 xfs_inherit_nodefrag)
891 di_flags |= XFS_DIFLAG_NODEFRAG;
892 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
893 di_flags |= XFS_DIFLAG_FILESTREAM;
895 ip->i_d.di_flags |= di_flags;
898 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
899 pip->i_d.di_version == 3 &&
900 ip->i_d.di_version == 3) {
901 uint64_t di_flags2 = 0;
903 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
904 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
905 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
907 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
908 di_flags2 |= XFS_DIFLAG2_DAX;
910 ip->i_d.di_flags2 |= di_flags2;
914 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
915 ip->i_df.if_flags = XFS_IFEXTENTS;
916 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
917 ip->i_df.if_u1.if_root = NULL;
923 * Attribute fork settings for new inode.
925 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
926 ip->i_d.di_anextents = 0;
929 * Log the new values stuffed into the inode.
931 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
932 xfs_trans_log_inode(tp, ip, flags);
934 /* now that we have an i_mode we can setup the inode structure */
942 * Allocates a new inode from disk and return a pointer to the
943 * incore copy. This routine will internally commit the current
944 * transaction and allocate a new one if the Space Manager needed
945 * to do an allocation to replenish the inode free-list.
947 * This routine is designed to be called from xfs_create and
953 xfs_trans_t **tpp, /* input: current transaction;
954 output: may be a new transaction. */
955 xfs_inode_t *dp, /* directory within whose allocate
960 prid_t prid, /* project id */
961 int okalloc, /* ok to allocate new space */
962 xfs_inode_t **ipp, /* pointer to inode; it will be
969 xfs_buf_t *ialloc_context = NULL;
975 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
978 * xfs_ialloc will return a pointer to an incore inode if
979 * the Space Manager has an available inode on the free
980 * list. Otherwise, it will do an allocation and replenish
981 * the freelist. Since we can only do one allocation per
982 * transaction without deadlocks, we will need to commit the
983 * current transaction and start a new one. We will then
984 * need to call xfs_ialloc again to get the inode.
986 * If xfs_ialloc did an allocation to replenish the freelist,
987 * it returns the bp containing the head of the freelist as
988 * ialloc_context. We will hold a lock on it across the
989 * transaction commit so that no other process can steal
990 * the inode(s) that we've just allocated.
992 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
993 &ialloc_context, &ip);
996 * Return an error if we were unable to allocate a new inode.
997 * This should only happen if we run out of space on disk or
998 * encounter a disk error.
1004 if (!ialloc_context && !ip) {
1010 * If the AGI buffer is non-NULL, then we were unable to get an
1011 * inode in one operation. We need to commit the current
1012 * transaction and call xfs_ialloc() again. It is guaranteed
1013 * to succeed the second time.
1015 if (ialloc_context) {
1017 * Normally, xfs_trans_commit releases all the locks.
1018 * We call bhold to hang on to the ialloc_context across
1019 * the commit. Holding this buffer prevents any other
1020 * processes from doing any allocations in this
1023 xfs_trans_bhold(tp, ialloc_context);
1026 * We want the quota changes to be associated with the next
1027 * transaction, NOT this one. So, detach the dqinfo from this
1028 * and attach it to the next transaction.
1033 dqinfo = (void *)tp->t_dqinfo;
1034 tp->t_dqinfo = NULL;
1035 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1036 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1039 code = xfs_trans_roll(&tp);
1040 if (committed != NULL)
1044 * Re-attach the quota info that we detached from prev trx.
1047 tp->t_dqinfo = dqinfo;
1048 tp->t_flags |= tflags;
1052 xfs_buf_relse(ialloc_context);
1057 xfs_trans_bjoin(tp, ialloc_context);
1060 * Call ialloc again. Since we've locked out all
1061 * other allocations in this allocation group,
1062 * this call should always succeed.
1064 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1065 okalloc, &ialloc_context, &ip);
1068 * If we get an error at this point, return to the caller
1069 * so that the current transaction can be aborted.
1076 ASSERT(!ialloc_context && ip);
1079 if (committed != NULL)
1090 * Decrement the link count on an inode & log the change. If this causes the
1091 * link count to go to zero, move the inode to AGI unlinked list so that it can
1092 * be freed when the last active reference goes away via xfs_inactive().
1094 static int /* error */
1099 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1101 drop_nlink(VFS_I(ip));
1102 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1104 if (VFS_I(ip)->i_nlink)
1107 return xfs_iunlink(tp, ip);
1111 * Increment the link count on an inode & log the change.
1118 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1120 ASSERT(ip->i_d.di_version > 1);
1121 inc_nlink(VFS_I(ip));
1122 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1129 struct xfs_name *name,
1134 int is_dir = S_ISDIR(mode);
1135 struct xfs_mount *mp = dp->i_mount;
1136 struct xfs_inode *ip = NULL;
1137 struct xfs_trans *tp = NULL;
1139 struct xfs_defer_ops dfops;
1140 xfs_fsblock_t first_block;
1141 bool unlock_dp_on_error = false;
1143 struct xfs_dquot *udqp = NULL;
1144 struct xfs_dquot *gdqp = NULL;
1145 struct xfs_dquot *pdqp = NULL;
1146 struct xfs_trans_res *tres;
1149 trace_xfs_create(dp, name);
1151 if (XFS_FORCED_SHUTDOWN(mp))
1154 prid = xfs_get_initial_prid(dp);
1157 * Make sure that we have allocated dquot(s) on disk.
1159 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1160 xfs_kgid_to_gid(current_fsgid()), prid,
1161 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1162 &udqp, &gdqp, &pdqp);
1167 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1168 tres = &M_RES(mp)->tr_mkdir;
1170 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1171 tres = &M_RES(mp)->tr_create;
1175 * Initially assume that the file does not exist and
1176 * reserve the resources for that case. If that is not
1177 * the case we'll drop the one we have and get a more
1178 * appropriate transaction later.
1180 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1181 if (error == -ENOSPC) {
1182 /* flush outstanding delalloc blocks and retry */
1183 xfs_flush_inodes(mp);
1184 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1186 if (error == -ENOSPC) {
1187 /* No space at all so try a "no-allocation" reservation */
1189 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1192 goto out_release_inode;
1194 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1195 unlock_dp_on_error = true;
1197 xfs_defer_init(&dfops, &first_block);
1200 * Reserve disk quota and the inode.
1202 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1203 pdqp, resblks, 1, 0);
1205 goto out_trans_cancel;
1208 error = xfs_dir_canenter(tp, dp, name);
1210 goto out_trans_cancel;
1214 * A newly created regular or special file just has one directory
1215 * entry pointing to them, but a directory also the "." entry
1216 * pointing to itself.
1218 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1219 prid, resblks > 0, &ip, NULL);
1221 goto out_trans_cancel;
1224 * Now we join the directory inode to the transaction. We do not do it
1225 * earlier because xfs_dir_ialloc might commit the previous transaction
1226 * (and release all the locks). An error from here on will result in
1227 * the transaction cancel unlocking dp so don't do it explicitly in the
1230 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1231 unlock_dp_on_error = false;
1233 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1234 &first_block, &dfops, resblks ?
1235 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1237 ASSERT(error != -ENOSPC);
1238 goto out_trans_cancel;
1240 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1241 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1244 error = xfs_dir_init(tp, ip, dp);
1246 goto out_bmap_cancel;
1248 error = xfs_bumplink(tp, dp);
1250 goto out_bmap_cancel;
1254 * If this is a synchronous mount, make sure that the
1255 * create transaction goes to disk before returning to
1258 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1259 xfs_trans_set_sync(tp);
1262 * Attach the dquot(s) to the inodes and modify them incore.
1263 * These ids of the inode couldn't have changed since the new
1264 * inode has been locked ever since it was created.
1266 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1268 error = xfs_defer_finish(&tp, &dfops);
1270 goto out_bmap_cancel;
1272 error = xfs_trans_commit(tp);
1274 goto out_release_inode;
1276 xfs_qm_dqrele(udqp);
1277 xfs_qm_dqrele(gdqp);
1278 xfs_qm_dqrele(pdqp);
1284 xfs_defer_cancel(&dfops);
1286 xfs_trans_cancel(tp);
1289 * Wait until after the current transaction is aborted to finish the
1290 * setup of the inode and release the inode. This prevents recursive
1291 * transactions and deadlocks from xfs_inactive.
1294 xfs_finish_inode_setup(ip);
1298 xfs_qm_dqrele(udqp);
1299 xfs_qm_dqrele(gdqp);
1300 xfs_qm_dqrele(pdqp);
1302 if (unlock_dp_on_error)
1303 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1309 struct xfs_inode *dp,
1310 struct dentry *dentry,
1312 struct xfs_inode **ipp)
1314 struct xfs_mount *mp = dp->i_mount;
1315 struct xfs_inode *ip = NULL;
1316 struct xfs_trans *tp = NULL;
1319 struct xfs_dquot *udqp = NULL;
1320 struct xfs_dquot *gdqp = NULL;
1321 struct xfs_dquot *pdqp = NULL;
1322 struct xfs_trans_res *tres;
1325 if (XFS_FORCED_SHUTDOWN(mp))
1328 prid = xfs_get_initial_prid(dp);
1331 * Make sure that we have allocated dquot(s) on disk.
1333 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1334 xfs_kgid_to_gid(current_fsgid()), prid,
1335 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1336 &udqp, &gdqp, &pdqp);
1340 resblks = XFS_IALLOC_SPACE_RES(mp);
1341 tres = &M_RES(mp)->tr_create_tmpfile;
1343 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1344 if (error == -ENOSPC) {
1345 /* No space at all so try a "no-allocation" reservation */
1347 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1350 goto out_release_inode;
1352 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1353 pdqp, resblks, 1, 0);
1355 goto out_trans_cancel;
1357 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1358 prid, resblks > 0, &ip, NULL);
1360 goto out_trans_cancel;
1362 if (mp->m_flags & XFS_MOUNT_WSYNC)
1363 xfs_trans_set_sync(tp);
1366 * Attach the dquot(s) to the inodes and modify them incore.
1367 * These ids of the inode couldn't have changed since the new
1368 * inode has been locked ever since it was created.
1370 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1372 error = xfs_iunlink(tp, ip);
1374 goto out_trans_cancel;
1376 error = xfs_trans_commit(tp);
1378 goto out_release_inode;
1380 xfs_qm_dqrele(udqp);
1381 xfs_qm_dqrele(gdqp);
1382 xfs_qm_dqrele(pdqp);
1388 xfs_trans_cancel(tp);
1391 * Wait until after the current transaction is aborted to finish the
1392 * setup of the inode and release the inode. This prevents recursive
1393 * transactions and deadlocks from xfs_inactive.
1396 xfs_finish_inode_setup(ip);
1400 xfs_qm_dqrele(udqp);
1401 xfs_qm_dqrele(gdqp);
1402 xfs_qm_dqrele(pdqp);
1411 struct xfs_name *target_name)
1413 xfs_mount_t *mp = tdp->i_mount;
1416 struct xfs_defer_ops dfops;
1417 xfs_fsblock_t first_block;
1420 trace_xfs_link(tdp, target_name);
1422 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1424 if (XFS_FORCED_SHUTDOWN(mp))
1427 error = xfs_qm_dqattach(sip, 0);
1431 error = xfs_qm_dqattach(tdp, 0);
1435 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1436 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1437 if (error == -ENOSPC) {
1439 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1444 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1446 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1447 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1450 * If we are using project inheritance, we only allow hard link
1451 * creation in our tree when the project IDs are the same; else
1452 * the tree quota mechanism could be circumvented.
1454 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1455 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1461 error = xfs_dir_canenter(tp, tdp, target_name);
1466 xfs_defer_init(&dfops, &first_block);
1469 * Handle initial link state of O_TMPFILE inode
1471 if (VFS_I(sip)->i_nlink == 0) {
1472 error = xfs_iunlink_remove(tp, sip);
1477 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1478 &first_block, &dfops, resblks);
1481 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1482 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1484 error = xfs_bumplink(tp, sip);
1489 * If this is a synchronous mount, make sure that the
1490 * link transaction goes to disk before returning to
1493 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1494 xfs_trans_set_sync(tp);
1496 error = xfs_defer_finish(&tp, &dfops);
1498 xfs_defer_cancel(&dfops);
1502 return xfs_trans_commit(tp);
1505 xfs_trans_cancel(tp);
1511 * Free up the underlying blocks past new_size. The new size must be smaller
1512 * than the current size. This routine can be used both for the attribute and
1513 * data fork, and does not modify the inode size, which is left to the caller.
1515 * The transaction passed to this routine must have made a permanent log
1516 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1517 * given transaction and start new ones, so make sure everything involved in
1518 * the transaction is tidy before calling here. Some transaction will be
1519 * returned to the caller to be committed. The incoming transaction must
1520 * already include the inode, and both inode locks must be held exclusively.
1521 * The inode must also be "held" within the transaction. On return the inode
1522 * will be "held" within the returned transaction. This routine does NOT
1523 * require any disk space to be reserved for it within the transaction.
1525 * If we get an error, we must return with the inode locked and linked into the
1526 * current transaction. This keeps things simple for the higher level code,
1527 * because it always knows that the inode is locked and held in the transaction
1528 * that returns to it whether errors occur or not. We don't mark the inode
1529 * dirty on error so that transactions can be easily aborted if possible.
1532 xfs_itruncate_extents(
1533 struct xfs_trans **tpp,
1534 struct xfs_inode *ip,
1536 xfs_fsize_t new_size)
1538 struct xfs_mount *mp = ip->i_mount;
1539 struct xfs_trans *tp = *tpp;
1540 struct xfs_defer_ops dfops;
1541 xfs_fsblock_t first_block;
1542 xfs_fileoff_t first_unmap_block;
1543 xfs_fileoff_t last_block;
1544 xfs_filblks_t unmap_len;
1548 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1549 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1550 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1551 ASSERT(new_size <= XFS_ISIZE(ip));
1552 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1553 ASSERT(ip->i_itemp != NULL);
1554 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1555 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1557 trace_xfs_itruncate_extents_start(ip, new_size);
1560 * Since it is possible for space to become allocated beyond
1561 * the end of the file (in a crash where the space is allocated
1562 * but the inode size is not yet updated), simply remove any
1563 * blocks which show up between the new EOF and the maximum
1564 * possible file size. If the first block to be removed is
1565 * beyond the maximum file size (ie it is the same as last_block),
1566 * then there is nothing to do.
1568 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1569 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1570 if (first_unmap_block == last_block)
1573 ASSERT(first_unmap_block < last_block);
1574 unmap_len = last_block - first_unmap_block + 1;
1576 xfs_defer_init(&dfops, &first_block);
1577 error = xfs_bunmapi(tp, ip,
1578 first_unmap_block, unmap_len,
1579 xfs_bmapi_aflag(whichfork),
1580 XFS_ITRUNC_MAX_EXTENTS,
1581 &first_block, &dfops,
1584 goto out_bmap_cancel;
1587 * Duplicate the transaction that has the permanent
1588 * reservation and commit the old transaction.
1590 xfs_defer_ijoin(&dfops, ip);
1591 error = xfs_defer_finish(&tp, &dfops);
1593 goto out_bmap_cancel;
1595 error = xfs_trans_roll_inode(&tp, ip);
1600 /* Remove all pending CoW reservations. */
1601 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1607 * Clear the reflink flag if there are no data fork blocks and
1608 * there are no extents staged in the cow fork.
1610 if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) {
1611 if (ip->i_d.di_nblocks == 0)
1612 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1613 xfs_inode_clear_cowblocks_tag(ip);
1617 * Always re-log the inode so that our permanent transaction can keep
1618 * on rolling it forward in the log.
1620 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1622 trace_xfs_itruncate_extents_end(ip, new_size);
1629 * If the bunmapi call encounters an error, return to the caller where
1630 * the transaction can be properly aborted. We just need to make sure
1631 * we're not holding any resources that we were not when we came in.
1633 xfs_defer_cancel(&dfops);
1641 xfs_mount_t *mp = ip->i_mount;
1644 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1647 /* If this is a read-only mount, don't do this (would generate I/O) */
1648 if (mp->m_flags & XFS_MOUNT_RDONLY)
1651 if (!XFS_FORCED_SHUTDOWN(mp)) {
1655 * If we previously truncated this file and removed old data
1656 * in the process, we want to initiate "early" writeout on
1657 * the last close. This is an attempt to combat the notorious
1658 * NULL files problem which is particularly noticeable from a
1659 * truncate down, buffered (re-)write (delalloc), followed by
1660 * a crash. What we are effectively doing here is
1661 * significantly reducing the time window where we'd otherwise
1662 * be exposed to that problem.
1664 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1666 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1667 if (ip->i_delayed_blks > 0) {
1668 error = filemap_flush(VFS_I(ip)->i_mapping);
1675 if (VFS_I(ip)->i_nlink == 0)
1678 if (xfs_can_free_eofblocks(ip, false)) {
1681 * Check if the inode is being opened, written and closed
1682 * frequently and we have delayed allocation blocks outstanding
1683 * (e.g. streaming writes from the NFS server), truncating the
1684 * blocks past EOF will cause fragmentation to occur.
1686 * In this case don't do the truncation, but we have to be
1687 * careful how we detect this case. Blocks beyond EOF show up as
1688 * i_delayed_blks even when the inode is clean, so we need to
1689 * truncate them away first before checking for a dirty release.
1690 * Hence on the first dirty close we will still remove the
1691 * speculative allocation, but after that we will leave it in
1694 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1697 * If we can't get the iolock just skip truncating the blocks
1698 * past EOF because we could deadlock with the mmap_sem
1699 * otherwise. We'll get another chance to drop them once the
1700 * last reference to the inode is dropped, so we'll never leak
1701 * blocks permanently.
1703 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1704 error = xfs_free_eofblocks(ip);
1705 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1710 /* delalloc blocks after truncation means it really is dirty */
1711 if (ip->i_delayed_blks)
1712 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1718 * xfs_inactive_truncate
1720 * Called to perform a truncate when an inode becomes unlinked.
1723 xfs_inactive_truncate(
1724 struct xfs_inode *ip)
1726 struct xfs_mount *mp = ip->i_mount;
1727 struct xfs_trans *tp;
1730 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1732 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1736 xfs_ilock(ip, XFS_ILOCK_EXCL);
1737 xfs_trans_ijoin(tp, ip, 0);
1740 * Log the inode size first to prevent stale data exposure in the event
1741 * of a system crash before the truncate completes. See the related
1742 * comment in xfs_vn_setattr_size() for details.
1744 ip->i_d.di_size = 0;
1745 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1747 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1749 goto error_trans_cancel;
1751 ASSERT(ip->i_d.di_nextents == 0);
1753 error = xfs_trans_commit(tp);
1757 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1761 xfs_trans_cancel(tp);
1763 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1768 * xfs_inactive_ifree()
1770 * Perform the inode free when an inode is unlinked.
1774 struct xfs_inode *ip)
1776 struct xfs_defer_ops dfops;
1777 xfs_fsblock_t first_block;
1778 struct xfs_mount *mp = ip->i_mount;
1779 struct xfs_trans *tp;
1783 * We try to use a per-AG reservation for any block needed by the finobt
1784 * tree, but as the finobt feature predates the per-AG reservation
1785 * support a degraded file system might not have enough space for the
1786 * reservation at mount time. In that case try to dip into the reserved
1789 * Send a warning if the reservation does happen to fail, as the inode
1790 * now remains allocated and sits on the unlinked list until the fs is
1793 if (unlikely(mp->m_inotbt_nores)) {
1794 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1795 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1798 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1801 if (error == -ENOSPC) {
1802 xfs_warn_ratelimited(mp,
1803 "Failed to remove inode(s) from unlinked list. "
1804 "Please free space, unmount and run xfs_repair.");
1806 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1811 xfs_ilock(ip, XFS_ILOCK_EXCL);
1812 xfs_trans_ijoin(tp, ip, 0);
1814 xfs_defer_init(&dfops, &first_block);
1815 error = xfs_ifree(tp, ip, &dfops);
1818 * If we fail to free the inode, shut down. The cancel
1819 * might do that, we need to make sure. Otherwise the
1820 * inode might be lost for a long time or forever.
1822 if (!XFS_FORCED_SHUTDOWN(mp)) {
1823 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1825 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1827 xfs_trans_cancel(tp);
1828 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1833 * Credit the quota account(s). The inode is gone.
1835 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1838 * Just ignore errors at this point. There is nothing we can do except
1839 * to try to keep going. Make sure it's not a silent error.
1841 error = xfs_defer_finish(&tp, &dfops);
1843 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1845 xfs_defer_cancel(&dfops);
1847 error = xfs_trans_commit(tp);
1849 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1852 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1859 * This is called when the vnode reference count for the vnode
1860 * goes to zero. If the file has been unlinked, then it must
1861 * now be truncated. Also, we clear all of the read-ahead state
1862 * kept for the inode here since the file is now closed.
1868 struct xfs_mount *mp;
1873 * If the inode is already free, then there can be nothing
1876 if (VFS_I(ip)->i_mode == 0) {
1877 ASSERT(ip->i_df.if_real_bytes == 0);
1878 ASSERT(ip->i_df.if_broot_bytes == 0);
1883 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1885 /* If this is a read-only mount, don't do this (would generate I/O) */
1886 if (mp->m_flags & XFS_MOUNT_RDONLY)
1889 if (VFS_I(ip)->i_nlink != 0) {
1891 * force is true because we are evicting an inode from the
1892 * cache. Post-eof blocks must be freed, lest we end up with
1893 * broken free space accounting.
1895 * Note: don't bother with iolock here since lockdep complains
1896 * about acquiring it in reclaim context. We have the only
1897 * reference to the inode at this point anyways.
1899 if (xfs_can_free_eofblocks(ip, true))
1900 xfs_free_eofblocks(ip);
1905 if (S_ISREG(VFS_I(ip)->i_mode) &&
1906 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1907 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1910 error = xfs_qm_dqattach(ip, 0);
1914 if (S_ISLNK(VFS_I(ip)->i_mode))
1915 error = xfs_inactive_symlink(ip);
1917 error = xfs_inactive_truncate(ip);
1922 * If there are attributes associated with the file then blow them away
1923 * now. The code calls a routine that recursively deconstructs the
1924 * attribute fork. If also blows away the in-core attribute fork.
1926 if (XFS_IFORK_Q(ip)) {
1927 error = xfs_attr_inactive(ip);
1933 ASSERT(ip->i_d.di_anextents == 0);
1934 ASSERT(ip->i_d.di_forkoff == 0);
1939 error = xfs_inactive_ifree(ip);
1944 * Release the dquots held by inode, if any.
1946 xfs_qm_dqdetach(ip);
1950 * This is called when the inode's link count goes to 0 or we are creating a
1951 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1952 * set to true as the link count is dropped to zero by the VFS after we've
1953 * created the file successfully, so we have to add it to the unlinked list
1954 * while the link count is non-zero.
1956 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1957 * list when the inode is freed.
1961 struct xfs_trans *tp,
1962 struct xfs_inode *ip)
1964 xfs_mount_t *mp = tp->t_mountp;
1974 ASSERT(VFS_I(ip)->i_mode != 0);
1977 * Get the agi buffer first. It ensures lock ordering
1980 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1983 agi = XFS_BUF_TO_AGI(agibp);
1986 * Get the index into the agi hash table for the
1987 * list this inode will go on.
1989 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1991 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1992 ASSERT(agi->agi_unlinked[bucket_index]);
1993 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1995 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1997 * There is already another inode in the bucket we need
1998 * to add ourselves to. Add us at the front of the list.
1999 * Here we put the head pointer into our next pointer,
2000 * and then we fall through to point the head at us.
2002 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2007 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2008 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2009 offset = ip->i_imap.im_boffset +
2010 offsetof(xfs_dinode_t, di_next_unlinked);
2012 /* need to recalc the inode CRC if appropriate */
2013 xfs_dinode_calc_crc(mp, dip);
2015 xfs_trans_inode_buf(tp, ibp);
2016 xfs_trans_log_buf(tp, ibp, offset,
2017 (offset + sizeof(xfs_agino_t) - 1));
2018 xfs_inobp_check(mp, ibp);
2022 * Point the bucket head pointer at the inode being inserted.
2025 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2026 offset = offsetof(xfs_agi_t, agi_unlinked) +
2027 (sizeof(xfs_agino_t) * bucket_index);
2028 xfs_trans_log_buf(tp, agibp, offset,
2029 (offset + sizeof(xfs_agino_t) - 1));
2034 * Pull the on-disk inode from the AGI unlinked list.
2047 xfs_agnumber_t agno;
2049 xfs_agino_t next_agino;
2050 xfs_buf_t *last_ibp;
2051 xfs_dinode_t *last_dip = NULL;
2053 int offset, last_offset = 0;
2057 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2060 * Get the agi buffer first. It ensures lock ordering
2063 error = xfs_read_agi(mp, tp, agno, &agibp);
2067 agi = XFS_BUF_TO_AGI(agibp);
2070 * Get the index into the agi hash table for the
2071 * list this inode will go on.
2073 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2075 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2076 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2077 ASSERT(agi->agi_unlinked[bucket_index]);
2079 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2081 * We're at the head of the list. Get the inode's on-disk
2082 * buffer to see if there is anyone after us on the list.
2083 * Only modify our next pointer if it is not already NULLAGINO.
2084 * This saves us the overhead of dealing with the buffer when
2085 * there is no need to change it.
2087 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2090 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2094 next_agino = be32_to_cpu(dip->di_next_unlinked);
2095 ASSERT(next_agino != 0);
2096 if (next_agino != NULLAGINO) {
2097 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2098 offset = ip->i_imap.im_boffset +
2099 offsetof(xfs_dinode_t, di_next_unlinked);
2101 /* need to recalc the inode CRC if appropriate */
2102 xfs_dinode_calc_crc(mp, dip);
2104 xfs_trans_inode_buf(tp, ibp);
2105 xfs_trans_log_buf(tp, ibp, offset,
2106 (offset + sizeof(xfs_agino_t) - 1));
2107 xfs_inobp_check(mp, ibp);
2109 xfs_trans_brelse(tp, ibp);
2112 * Point the bucket head pointer at the next inode.
2114 ASSERT(next_agino != 0);
2115 ASSERT(next_agino != agino);
2116 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2117 offset = offsetof(xfs_agi_t, agi_unlinked) +
2118 (sizeof(xfs_agino_t) * bucket_index);
2119 xfs_trans_log_buf(tp, agibp, offset,
2120 (offset + sizeof(xfs_agino_t) - 1));
2123 * We need to search the list for the inode being freed.
2125 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2127 while (next_agino != agino) {
2128 struct xfs_imap imap;
2131 xfs_trans_brelse(tp, last_ibp);
2134 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2136 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2139 "%s: xfs_imap returned error %d.",
2144 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2148 "%s: xfs_imap_to_bp returned error %d.",
2153 last_offset = imap.im_boffset;
2154 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2155 ASSERT(next_agino != NULLAGINO);
2156 ASSERT(next_agino != 0);
2160 * Now last_ibp points to the buffer previous to us on the
2161 * unlinked list. Pull us from the list.
2163 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2166 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2170 next_agino = be32_to_cpu(dip->di_next_unlinked);
2171 ASSERT(next_agino != 0);
2172 ASSERT(next_agino != agino);
2173 if (next_agino != NULLAGINO) {
2174 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2175 offset = ip->i_imap.im_boffset +
2176 offsetof(xfs_dinode_t, di_next_unlinked);
2178 /* need to recalc the inode CRC if appropriate */
2179 xfs_dinode_calc_crc(mp, dip);
2181 xfs_trans_inode_buf(tp, ibp);
2182 xfs_trans_log_buf(tp, ibp, offset,
2183 (offset + sizeof(xfs_agino_t) - 1));
2184 xfs_inobp_check(mp, ibp);
2186 xfs_trans_brelse(tp, ibp);
2189 * Point the previous inode on the list to the next inode.
2191 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2192 ASSERT(next_agino != 0);
2193 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2195 /* need to recalc the inode CRC if appropriate */
2196 xfs_dinode_calc_crc(mp, last_dip);
2198 xfs_trans_inode_buf(tp, last_ibp);
2199 xfs_trans_log_buf(tp, last_ibp, offset,
2200 (offset + sizeof(xfs_agino_t) - 1));
2201 xfs_inobp_check(mp, last_ibp);
2207 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2208 * inodes that are in memory - they all must be marked stale and attached to
2209 * the cluster buffer.
2213 xfs_inode_t *free_ip,
2215 struct xfs_icluster *xic)
2217 xfs_mount_t *mp = free_ip->i_mount;
2218 int blks_per_cluster;
2219 int inodes_per_cluster;
2226 xfs_inode_log_item_t *iip;
2227 xfs_log_item_t *lip;
2228 struct xfs_perag *pag;
2231 inum = xic->first_ino;
2232 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2233 blks_per_cluster = xfs_icluster_size_fsb(mp);
2234 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2235 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2237 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2239 * The allocation bitmap tells us which inodes of the chunk were
2240 * physically allocated. Skip the cluster if an inode falls into
2243 ioffset = inum - xic->first_ino;
2244 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2245 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2249 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2250 XFS_INO_TO_AGBNO(mp, inum));
2253 * We obtain and lock the backing buffer first in the process
2254 * here, as we have to ensure that any dirty inode that we
2255 * can't get the flush lock on is attached to the buffer.
2256 * If we scan the in-memory inodes first, then buffer IO can
2257 * complete before we get a lock on it, and hence we may fail
2258 * to mark all the active inodes on the buffer stale.
2260 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2261 mp->m_bsize * blks_per_cluster,
2268 * This buffer may not have been correctly initialised as we
2269 * didn't read it from disk. That's not important because we are
2270 * only using to mark the buffer as stale in the log, and to
2271 * attach stale cached inodes on it. That means it will never be
2272 * dispatched for IO. If it is, we want to know about it, and we
2273 * want it to fail. We can acheive this by adding a write
2274 * verifier to the buffer.
2276 bp->b_ops = &xfs_inode_buf_ops;
2279 * Walk the inodes already attached to the buffer and mark them
2280 * stale. These will all have the flush locks held, so an
2281 * in-memory inode walk can't lock them. By marking them all
2282 * stale first, we will not attempt to lock them in the loop
2283 * below as the XFS_ISTALE flag will be set.
2287 if (lip->li_type == XFS_LI_INODE) {
2288 iip = (xfs_inode_log_item_t *)lip;
2289 ASSERT(iip->ili_logged == 1);
2290 lip->li_cb = xfs_istale_done;
2291 xfs_trans_ail_copy_lsn(mp->m_ail,
2292 &iip->ili_flush_lsn,
2293 &iip->ili_item.li_lsn);
2294 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2296 lip = lip->li_bio_list;
2301 * For each inode in memory attempt to add it to the inode
2302 * buffer and set it up for being staled on buffer IO
2303 * completion. This is safe as we've locked out tail pushing
2304 * and flushing by locking the buffer.
2306 * We have already marked every inode that was part of a
2307 * transaction stale above, which means there is no point in
2308 * even trying to lock them.
2310 for (i = 0; i < inodes_per_cluster; i++) {
2313 ip = radix_tree_lookup(&pag->pag_ici_root,
2314 XFS_INO_TO_AGINO(mp, (inum + i)));
2316 /* Inode not in memory, nothing to do */
2323 * because this is an RCU protected lookup, we could
2324 * find a recently freed or even reallocated inode
2325 * during the lookup. We need to check under the
2326 * i_flags_lock for a valid inode here. Skip it if it
2327 * is not valid, the wrong inode or stale.
2329 spin_lock(&ip->i_flags_lock);
2330 if (ip->i_ino != inum + i ||
2331 __xfs_iflags_test(ip, XFS_ISTALE)) {
2332 spin_unlock(&ip->i_flags_lock);
2336 spin_unlock(&ip->i_flags_lock);
2339 * Don't try to lock/unlock the current inode, but we
2340 * _cannot_ skip the other inodes that we did not find
2341 * in the list attached to the buffer and are not
2342 * already marked stale. If we can't lock it, back off
2345 if (ip != free_ip) {
2346 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2353 * Check the inode number again in case we're
2354 * racing with freeing in xfs_reclaim_inode().
2355 * See the comments in that function for more
2356 * information as to why the initial check is
2359 if (ip->i_ino != inum + i) {
2360 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2368 xfs_iflags_set(ip, XFS_ISTALE);
2371 * we don't need to attach clean inodes or those only
2372 * with unlogged changes (which we throw away, anyway).
2375 if (!iip || xfs_inode_clean(ip)) {
2376 ASSERT(ip != free_ip);
2378 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2382 iip->ili_last_fields = iip->ili_fields;
2383 iip->ili_fields = 0;
2384 iip->ili_fsync_fields = 0;
2385 iip->ili_logged = 1;
2386 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2387 &iip->ili_item.li_lsn);
2389 xfs_buf_attach_iodone(bp, xfs_istale_done,
2393 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2396 xfs_trans_stale_inode_buf(tp, bp);
2397 xfs_trans_binval(tp, bp);
2405 * Free any local-format buffers sitting around before we reset to
2409 xfs_ifree_local_data(
2410 struct xfs_inode *ip,
2413 struct xfs_ifork *ifp;
2415 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2418 ifp = XFS_IFORK_PTR(ip, whichfork);
2419 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2423 * This is called to return an inode to the inode free list.
2424 * The inode should already be truncated to 0 length and have
2425 * no pages associated with it. This routine also assumes that
2426 * the inode is already a part of the transaction.
2428 * The on-disk copy of the inode will have been added to the list
2429 * of unlinked inodes in the AGI. We need to remove the inode from
2430 * that list atomically with respect to freeing it here.
2436 struct xfs_defer_ops *dfops)
2439 struct xfs_icluster xic = { 0 };
2441 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2442 ASSERT(VFS_I(ip)->i_nlink == 0);
2443 ASSERT(ip->i_d.di_nextents == 0);
2444 ASSERT(ip->i_d.di_anextents == 0);
2445 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2446 ASSERT(ip->i_d.di_nblocks == 0);
2449 * Pull the on-disk inode from the AGI unlinked list.
2451 error = xfs_iunlink_remove(tp, ip);
2455 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2459 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2460 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2462 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2463 ip->i_d.di_flags = 0;
2464 ip->i_d.di_dmevmask = 0;
2465 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2466 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2467 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2469 * Bump the generation count so no one will be confused
2470 * by reincarnations of this inode.
2472 VFS_I(ip)->i_generation++;
2473 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2476 error = xfs_ifree_cluster(ip, tp, &xic);
2482 * This is called to unpin an inode. The caller must have the inode locked
2483 * in at least shared mode so that the buffer cannot be subsequently pinned
2484 * once someone is waiting for it to be unpinned.
2488 struct xfs_inode *ip)
2490 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2492 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2494 /* Give the log a push to start the unpinning I/O */
2495 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2501 struct xfs_inode *ip)
2503 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2504 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2509 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2510 if (xfs_ipincount(ip))
2512 } while (xfs_ipincount(ip));
2513 finish_wait(wq, &wait.wq_entry);
2518 struct xfs_inode *ip)
2520 if (xfs_ipincount(ip))
2521 __xfs_iunpin_wait(ip);
2525 * Removing an inode from the namespace involves removing the directory entry
2526 * and dropping the link count on the inode. Removing the directory entry can
2527 * result in locking an AGF (directory blocks were freed) and removing a link
2528 * count can result in placing the inode on an unlinked list which results in
2531 * The big problem here is that we have an ordering constraint on AGF and AGI
2532 * locking - inode allocation locks the AGI, then can allocate a new extent for
2533 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2534 * removes the inode from the unlinked list, requiring that we lock the AGI
2535 * first, and then freeing the inode can result in an inode chunk being freed
2536 * and hence freeing disk space requiring that we lock an AGF.
2538 * Hence the ordering that is imposed by other parts of the code is AGI before
2539 * AGF. This means we cannot remove the directory entry before we drop the inode
2540 * reference count and put it on the unlinked list as this results in a lock
2541 * order of AGF then AGI, and this can deadlock against inode allocation and
2542 * freeing. Therefore we must drop the link counts before we remove the
2545 * This is still safe from a transactional point of view - it is not until we
2546 * get to xfs_defer_finish() that we have the possibility of multiple
2547 * transactions in this operation. Hence as long as we remove the directory
2548 * entry and drop the link count in the first transaction of the remove
2549 * operation, there are no transactional constraints on the ordering here.
2554 struct xfs_name *name,
2557 xfs_mount_t *mp = dp->i_mount;
2558 xfs_trans_t *tp = NULL;
2559 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2561 struct xfs_defer_ops dfops;
2562 xfs_fsblock_t first_block;
2565 trace_xfs_remove(dp, name);
2567 if (XFS_FORCED_SHUTDOWN(mp))
2570 error = xfs_qm_dqattach(dp, 0);
2574 error = xfs_qm_dqattach(ip, 0);
2579 * We try to get the real space reservation first,
2580 * allowing for directory btree deletion(s) implying
2581 * possible bmap insert(s). If we can't get the space
2582 * reservation then we use 0 instead, and avoid the bmap
2583 * btree insert(s) in the directory code by, if the bmap
2584 * insert tries to happen, instead trimming the LAST
2585 * block from the directory.
2587 resblks = XFS_REMOVE_SPACE_RES(mp);
2588 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2589 if (error == -ENOSPC) {
2591 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2595 ASSERT(error != -ENOSPC);
2599 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2601 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2602 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2605 * If we're removing a directory perform some additional validation.
2608 ASSERT(VFS_I(ip)->i_nlink >= 2);
2609 if (VFS_I(ip)->i_nlink != 2) {
2611 goto out_trans_cancel;
2613 if (!xfs_dir_isempty(ip)) {
2615 goto out_trans_cancel;
2618 /* Drop the link from ip's "..". */
2619 error = xfs_droplink(tp, dp);
2621 goto out_trans_cancel;
2623 /* Drop the "." link from ip to self. */
2624 error = xfs_droplink(tp, ip);
2626 goto out_trans_cancel;
2629 * When removing a non-directory we need to log the parent
2630 * inode here. For a directory this is done implicitly
2631 * by the xfs_droplink call for the ".." entry.
2633 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2635 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2637 /* Drop the link from dp to ip. */
2638 error = xfs_droplink(tp, ip);
2640 goto out_trans_cancel;
2642 xfs_defer_init(&dfops, &first_block);
2643 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2644 &first_block, &dfops, resblks);
2646 ASSERT(error != -ENOENT);
2647 goto out_bmap_cancel;
2651 * If this is a synchronous mount, make sure that the
2652 * remove transaction goes to disk before returning to
2655 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2656 xfs_trans_set_sync(tp);
2658 error = xfs_defer_finish(&tp, &dfops);
2660 goto out_bmap_cancel;
2662 error = xfs_trans_commit(tp);
2666 if (is_dir && xfs_inode_is_filestream(ip))
2667 xfs_filestream_deassociate(ip);
2672 xfs_defer_cancel(&dfops);
2674 xfs_trans_cancel(tp);
2680 * Enter all inodes for a rename transaction into a sorted array.
2682 #define __XFS_SORT_INODES 5
2684 xfs_sort_for_rename(
2685 struct xfs_inode *dp1, /* in: old (source) directory inode */
2686 struct xfs_inode *dp2, /* in: new (target) directory inode */
2687 struct xfs_inode *ip1, /* in: inode of old entry */
2688 struct xfs_inode *ip2, /* in: inode of new entry */
2689 struct xfs_inode *wip, /* in: whiteout inode */
2690 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2691 int *num_inodes) /* in/out: inodes in array */
2695 ASSERT(*num_inodes == __XFS_SORT_INODES);
2696 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2699 * i_tab contains a list of pointers to inodes. We initialize
2700 * the table here & we'll sort it. We will then use it to
2701 * order the acquisition of the inode locks.
2703 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2716 * Sort the elements via bubble sort. (Remember, there are at
2717 * most 5 elements to sort, so this is adequate.)
2719 for (i = 0; i < *num_inodes; i++) {
2720 for (j = 1; j < *num_inodes; j++) {
2721 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2722 struct xfs_inode *temp = i_tab[j];
2723 i_tab[j] = i_tab[j-1];
2732 struct xfs_trans *tp,
2733 struct xfs_defer_ops *dfops)
2738 * If this is a synchronous mount, make sure that the rename transaction
2739 * goes to disk before returning to the user.
2741 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2742 xfs_trans_set_sync(tp);
2744 error = xfs_defer_finish(&tp, dfops);
2746 xfs_defer_cancel(dfops);
2747 xfs_trans_cancel(tp);
2751 return xfs_trans_commit(tp);
2755 * xfs_cross_rename()
2757 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2761 struct xfs_trans *tp,
2762 struct xfs_inode *dp1,
2763 struct xfs_name *name1,
2764 struct xfs_inode *ip1,
2765 struct xfs_inode *dp2,
2766 struct xfs_name *name2,
2767 struct xfs_inode *ip2,
2768 struct xfs_defer_ops *dfops,
2769 xfs_fsblock_t *first_block,
2777 /* Swap inode number for dirent in first parent */
2778 error = xfs_dir_replace(tp, dp1, name1,
2780 first_block, dfops, spaceres);
2782 goto out_trans_abort;
2784 /* Swap inode number for dirent in second parent */
2785 error = xfs_dir_replace(tp, dp2, name2,
2787 first_block, dfops, spaceres);
2789 goto out_trans_abort;
2792 * If we're renaming one or more directories across different parents,
2793 * update the respective ".." entries (and link counts) to match the new
2797 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2799 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2800 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2801 dp1->i_ino, first_block,
2804 goto out_trans_abort;
2806 /* transfer ip2 ".." reference to dp1 */
2807 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2808 error = xfs_droplink(tp, dp2);
2810 goto out_trans_abort;
2811 error = xfs_bumplink(tp, dp1);
2813 goto out_trans_abort;
2817 * Although ip1 isn't changed here, userspace needs
2818 * to be warned about the change, so that applications
2819 * relying on it (like backup ones), will properly
2822 ip1_flags |= XFS_ICHGTIME_CHG;
2823 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2826 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2827 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2828 dp2->i_ino, first_block,
2831 goto out_trans_abort;
2833 /* transfer ip1 ".." reference to dp2 */
2834 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2835 error = xfs_droplink(tp, dp1);
2837 goto out_trans_abort;
2838 error = xfs_bumplink(tp, dp2);
2840 goto out_trans_abort;
2844 * Although ip2 isn't changed here, userspace needs
2845 * to be warned about the change, so that applications
2846 * relying on it (like backup ones), will properly
2849 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2850 ip2_flags |= XFS_ICHGTIME_CHG;
2855 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2856 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2859 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2860 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2863 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2864 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2866 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2867 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2868 return xfs_finish_rename(tp, dfops);
2871 xfs_defer_cancel(dfops);
2872 xfs_trans_cancel(tp);
2877 * xfs_rename_alloc_whiteout()
2879 * Return a referenced, unlinked, unlocked inode that that can be used as a
2880 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2881 * crash between allocating the inode and linking it into the rename transaction
2882 * recovery will free the inode and we won't leak it.
2885 xfs_rename_alloc_whiteout(
2886 struct xfs_inode *dp,
2887 struct xfs_inode **wip)
2889 struct xfs_inode *tmpfile;
2892 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2897 * Prepare the tmpfile inode as if it were created through the VFS.
2898 * Otherwise, the link increment paths will complain about nlink 0->1.
2899 * Drop the link count as done by d_tmpfile(), complete the inode setup
2900 * and flag it as linkable.
2902 drop_nlink(VFS_I(tmpfile));
2903 xfs_setup_iops(tmpfile);
2904 xfs_finish_inode_setup(tmpfile);
2905 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2916 struct xfs_inode *src_dp,
2917 struct xfs_name *src_name,
2918 struct xfs_inode *src_ip,
2919 struct xfs_inode *target_dp,
2920 struct xfs_name *target_name,
2921 struct xfs_inode *target_ip,
2924 struct xfs_mount *mp = src_dp->i_mount;
2925 struct xfs_trans *tp;
2926 struct xfs_defer_ops dfops;
2927 xfs_fsblock_t first_block;
2928 struct xfs_inode *wip = NULL; /* whiteout inode */
2929 struct xfs_inode *inodes[__XFS_SORT_INODES];
2930 int num_inodes = __XFS_SORT_INODES;
2931 bool new_parent = (src_dp != target_dp);
2932 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2936 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2938 if ((flags & RENAME_EXCHANGE) && !target_ip)
2942 * If we are doing a whiteout operation, allocate the whiteout inode
2943 * we will be placing at the target and ensure the type is set
2946 if (flags & RENAME_WHITEOUT) {
2947 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2948 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2952 /* setup target dirent info as whiteout */
2953 src_name->type = XFS_DIR3_FT_CHRDEV;
2956 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2957 inodes, &num_inodes);
2959 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2960 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2961 if (error == -ENOSPC) {
2963 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2967 goto out_release_wip;
2970 * Attach the dquots to the inodes
2972 error = xfs_qm_vop_rename_dqattach(inodes);
2974 goto out_trans_cancel;
2977 * Lock all the participating inodes. Depending upon whether
2978 * the target_name exists in the target directory, and
2979 * whether the target directory is the same as the source
2980 * directory, we can lock from 2 to 4 inodes.
2982 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2985 * Join all the inodes to the transaction. From this point on,
2986 * we can rely on either trans_commit or trans_cancel to unlock
2989 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2991 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2992 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2994 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2996 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2999 * If we are using project inheritance, we only allow renames
3000 * into our tree when the project IDs are the same; else the
3001 * tree quota mechanism would be circumvented.
3003 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3004 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3006 goto out_trans_cancel;
3009 xfs_defer_init(&dfops, &first_block);
3011 /* RENAME_EXCHANGE is unique from here on. */
3012 if (flags & RENAME_EXCHANGE)
3013 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3014 target_dp, target_name, target_ip,
3015 &dfops, &first_block, spaceres);
3018 * Set up the target.
3020 if (target_ip == NULL) {
3022 * If there's no space reservation, check the entry will
3023 * fit before actually inserting it.
3026 error = xfs_dir_canenter(tp, target_dp, target_name);
3028 goto out_trans_cancel;
3031 * If target does not exist and the rename crosses
3032 * directories, adjust the target directory link count
3033 * to account for the ".." reference from the new entry.
3035 error = xfs_dir_createname(tp, target_dp, target_name,
3036 src_ip->i_ino, &first_block,
3039 goto out_bmap_cancel;
3041 xfs_trans_ichgtime(tp, target_dp,
3042 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3044 if (new_parent && src_is_directory) {
3045 error = xfs_bumplink(tp, target_dp);
3047 goto out_bmap_cancel;
3049 } else { /* target_ip != NULL */
3051 * If target exists and it's a directory, check that both
3052 * target and source are directories and that target can be
3053 * destroyed, or that neither is a directory.
3055 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3057 * Make sure target dir is empty.
3059 if (!(xfs_dir_isempty(target_ip)) ||
3060 (VFS_I(target_ip)->i_nlink > 2)) {
3062 goto out_trans_cancel;
3067 * Link the source inode under the target name.
3068 * If the source inode is a directory and we are moving
3069 * it across directories, its ".." entry will be
3070 * inconsistent until we replace that down below.
3072 * In case there is already an entry with the same
3073 * name at the destination directory, remove it first.
3075 error = xfs_dir_replace(tp, target_dp, target_name,
3077 &first_block, &dfops, spaceres);
3079 goto out_bmap_cancel;
3081 xfs_trans_ichgtime(tp, target_dp,
3082 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3085 * Decrement the link count on the target since the target
3086 * dir no longer points to it.
3088 error = xfs_droplink(tp, target_ip);
3090 goto out_bmap_cancel;
3092 if (src_is_directory) {
3094 * Drop the link from the old "." entry.
3096 error = xfs_droplink(tp, target_ip);
3098 goto out_bmap_cancel;
3100 } /* target_ip != NULL */
3103 * Remove the source.
3105 if (new_parent && src_is_directory) {
3107 * Rewrite the ".." entry to point to the new
3110 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3112 &first_block, &dfops, spaceres);
3113 ASSERT(error != -EEXIST);
3115 goto out_bmap_cancel;
3119 * We always want to hit the ctime on the source inode.
3121 * This isn't strictly required by the standards since the source
3122 * inode isn't really being changed, but old unix file systems did
3123 * it and some incremental backup programs won't work without it.
3125 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3126 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3129 * Adjust the link count on src_dp. This is necessary when
3130 * renaming a directory, either within one parent when
3131 * the target existed, or across two parent directories.
3133 if (src_is_directory && (new_parent || target_ip != NULL)) {
3136 * Decrement link count on src_directory since the
3137 * entry that's moved no longer points to it.
3139 error = xfs_droplink(tp, src_dp);
3141 goto out_bmap_cancel;
3145 * For whiteouts, we only need to update the source dirent with the
3146 * inode number of the whiteout inode rather than removing it
3150 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3151 &first_block, &dfops, spaceres);
3153 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3154 &first_block, &dfops, spaceres);
3156 goto out_bmap_cancel;
3159 * For whiteouts, we need to bump the link count on the whiteout inode.
3160 * This means that failures all the way up to this point leave the inode
3161 * on the unlinked list and so cleanup is a simple matter of dropping
3162 * the remaining reference to it. If we fail here after bumping the link
3163 * count, we're shutting down the filesystem so we'll never see the
3164 * intermediate state on disk.
3167 ASSERT(VFS_I(wip)->i_nlink == 0);
3168 error = xfs_bumplink(tp, wip);
3170 goto out_bmap_cancel;
3171 error = xfs_iunlink_remove(tp, wip);
3173 goto out_bmap_cancel;
3174 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3177 * Now we have a real link, clear the "I'm a tmpfile" state
3178 * flag from the inode so it doesn't accidentally get misused in
3181 VFS_I(wip)->i_state &= ~I_LINKABLE;
3184 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3185 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3187 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3189 error = xfs_finish_rename(tp, &dfops);
3195 xfs_defer_cancel(&dfops);
3197 xfs_trans_cancel(tp);
3206 struct xfs_inode *ip,
3209 struct xfs_mount *mp = ip->i_mount;
3210 struct xfs_perag *pag;
3211 unsigned long first_index, mask;
3212 unsigned long inodes_per_cluster;
3214 struct xfs_inode **cilist;
3215 struct xfs_inode *cip;
3221 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3223 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3224 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3225 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3229 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3230 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3232 /* really need a gang lookup range call here */
3233 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3234 first_index, inodes_per_cluster);
3238 for (i = 0; i < nr_found; i++) {
3244 * because this is an RCU protected lookup, we could find a
3245 * recently freed or even reallocated inode during the lookup.
3246 * We need to check under the i_flags_lock for a valid inode
3247 * here. Skip it if it is not valid or the wrong inode.
3249 spin_lock(&cip->i_flags_lock);
3251 __xfs_iflags_test(cip, XFS_ISTALE)) {
3252 spin_unlock(&cip->i_flags_lock);
3257 * Once we fall off the end of the cluster, no point checking
3258 * any more inodes in the list because they will also all be
3259 * outside the cluster.
3261 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3262 spin_unlock(&cip->i_flags_lock);
3265 spin_unlock(&cip->i_flags_lock);
3268 * Do an un-protected check to see if the inode is dirty and
3269 * is a candidate for flushing. These checks will be repeated
3270 * later after the appropriate locks are acquired.
3272 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3276 * Try to get locks. If any are unavailable or it is pinned,
3277 * then this inode cannot be flushed and is skipped.
3280 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3282 if (!xfs_iflock_nowait(cip)) {
3283 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3286 if (xfs_ipincount(cip)) {
3288 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3294 * Check the inode number again, just to be certain we are not
3295 * racing with freeing in xfs_reclaim_inode(). See the comments
3296 * in that function for more information as to why the initial
3297 * check is not sufficient.
3301 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3306 * arriving here means that this inode can be flushed. First
3307 * re-check that it's dirty before flushing.
3309 if (!xfs_inode_clean(cip)) {
3311 error = xfs_iflush_int(cip, bp);
3313 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3314 goto cluster_corrupt_out;
3320 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3324 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3325 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3336 cluster_corrupt_out:
3338 * Corruption detected in the clustering loop. Invalidate the
3339 * inode buffer and shut down the filesystem.
3343 * Clean up the buffer. If it was delwri, just release it --
3344 * brelse can handle it with no problems. If not, shut down the
3345 * filesystem before releasing the buffer.
3347 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3351 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3353 if (!bufwasdelwri) {
3355 * Just like incore_relse: if we have b_iodone functions,
3356 * mark the buffer as an error and call them. Otherwise
3357 * mark it as stale and brelse.
3360 bp->b_flags &= ~XBF_DONE;
3362 xfs_buf_ioerror(bp, -EIO);
3371 * Unlocks the flush lock
3373 xfs_iflush_abort(cip, false);
3376 return -EFSCORRUPTED;
3380 * Flush dirty inode metadata into the backing buffer.
3382 * The caller must have the inode lock and the inode flush lock held. The
3383 * inode lock will still be held upon return to the caller, and the inode
3384 * flush lock will be released after the inode has reached the disk.
3386 * The caller must write out the buffer returned in *bpp and release it.
3390 struct xfs_inode *ip,
3391 struct xfs_buf **bpp)
3393 struct xfs_mount *mp = ip->i_mount;
3394 struct xfs_buf *bp = NULL;
3395 struct xfs_dinode *dip;
3398 XFS_STATS_INC(mp, xs_iflush_count);
3400 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3401 ASSERT(xfs_isiflocked(ip));
3402 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3403 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3407 xfs_iunpin_wait(ip);
3410 * For stale inodes we cannot rely on the backing buffer remaining
3411 * stale in cache for the remaining life of the stale inode and so
3412 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3413 * inodes below. We have to check this after ensuring the inode is
3414 * unpinned so that it is safe to reclaim the stale inode after the
3417 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3423 * This may have been unpinned because the filesystem is shutting
3424 * down forcibly. If that's the case we must not write this inode
3425 * to disk, because the log record didn't make it to disk.
3427 * We also have to remove the log item from the AIL in this case,
3428 * as we wait for an empty AIL as part of the unmount process.
3430 if (XFS_FORCED_SHUTDOWN(mp)) {
3436 * Get the buffer containing the on-disk inode. We are doing a try-lock
3437 * operation here, so we may get an EAGAIN error. In that case, we
3438 * simply want to return with the inode still dirty.
3440 * If we get any other error, we effectively have a corruption situation
3441 * and we cannot flush the inode, so we treat it the same as failing
3444 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3446 if (error == -EAGAIN) {
3454 * First flush out the inode that xfs_iflush was called with.
3456 error = xfs_iflush_int(ip, bp);
3461 * If the buffer is pinned then push on the log now so we won't
3462 * get stuck waiting in the write for too long.
3464 if (xfs_buf_ispinned(bp))
3465 xfs_log_force(mp, 0);
3469 * see if other inodes can be gathered into this write
3471 error = xfs_iflush_cluster(ip, bp);
3473 goto cluster_corrupt_out;
3481 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3482 cluster_corrupt_out:
3483 error = -EFSCORRUPTED;
3486 * Unlocks the flush lock
3488 xfs_iflush_abort(ip, false);
3494 struct xfs_inode *ip,
3497 struct xfs_inode_log_item *iip = ip->i_itemp;
3498 struct xfs_dinode *dip;
3499 struct xfs_mount *mp = ip->i_mount;
3501 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3502 ASSERT(xfs_isiflocked(ip));
3503 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3504 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3505 ASSERT(iip != NULL && iip->ili_fields != 0);
3506 ASSERT(ip->i_d.di_version > 1);
3508 /* set *dip = inode's place in the buffer */
3509 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3511 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3512 mp, XFS_ERRTAG_IFLUSH_1)) {
3513 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3514 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3515 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3518 if (S_ISREG(VFS_I(ip)->i_mode)) {
3520 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3521 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3522 mp, XFS_ERRTAG_IFLUSH_3)) {
3523 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3524 "%s: Bad regular inode %Lu, ptr 0x%p",
3525 __func__, ip->i_ino, ip);
3528 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3530 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3531 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3532 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3533 mp, XFS_ERRTAG_IFLUSH_4)) {
3534 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3535 "%s: Bad directory inode %Lu, ptr 0x%p",
3536 __func__, ip->i_ino, ip);
3540 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3541 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3542 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3543 "%s: detected corrupt incore inode %Lu, "
3544 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3545 __func__, ip->i_ino,
3546 ip->i_d.di_nextents + ip->i_d.di_anextents,
3547 ip->i_d.di_nblocks, ip);
3550 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3551 mp, XFS_ERRTAG_IFLUSH_6)) {
3552 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3553 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3554 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3559 * Inode item log recovery for v2 inodes are dependent on the
3560 * di_flushiter count for correct sequencing. We bump the flush
3561 * iteration count so we can detect flushes which postdate a log record
3562 * during recovery. This is redundant as we now log every change and
3563 * hence this can't happen but we need to still do it to ensure
3564 * backwards compatibility with old kernels that predate logging all
3567 if (ip->i_d.di_version < 3)
3568 ip->i_d.di_flushiter++;
3570 /* Check the inline directory data. */
3571 if (S_ISDIR(VFS_I(ip)->i_mode) &&
3572 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3573 xfs_dir2_sf_verify(ip))
3577 * Copy the dirty parts of the inode into the on-disk inode. We always
3578 * copy out the core of the inode, because if the inode is dirty at all
3581 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3583 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3584 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3585 ip->i_d.di_flushiter = 0;
3587 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3588 if (XFS_IFORK_Q(ip))
3589 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3590 xfs_inobp_check(mp, bp);
3593 * We've recorded everything logged in the inode, so we'd like to clear
3594 * the ili_fields bits so we don't log and flush things unnecessarily.
3595 * However, we can't stop logging all this information until the data
3596 * we've copied into the disk buffer is written to disk. If we did we
3597 * might overwrite the copy of the inode in the log with all the data
3598 * after re-logging only part of it, and in the face of a crash we
3599 * wouldn't have all the data we need to recover.
3601 * What we do is move the bits to the ili_last_fields field. When
3602 * logging the inode, these bits are moved back to the ili_fields field.
3603 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3604 * know that the information those bits represent is permanently on
3605 * disk. As long as the flush completes before the inode is logged
3606 * again, then both ili_fields and ili_last_fields will be cleared.
3608 * We can play with the ili_fields bits here, because the inode lock
3609 * must be held exclusively in order to set bits there and the flush
3610 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3611 * done routine can tell whether or not to look in the AIL. Also, store
3612 * the current LSN of the inode so that we can tell whether the item has
3613 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3614 * need the AIL lock, because it is a 64 bit value that cannot be read
3617 iip->ili_last_fields = iip->ili_fields;
3618 iip->ili_fields = 0;
3619 iip->ili_fsync_fields = 0;
3620 iip->ili_logged = 1;
3622 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3623 &iip->ili_item.li_lsn);
3626 * Attach the function xfs_iflush_done to the inode's
3627 * buffer. This will remove the inode from the AIL
3628 * and unlock the inode's flush lock when the inode is
3629 * completely written to disk.
3631 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3633 /* generate the checksum. */
3634 xfs_dinode_calc_crc(mp, dip);
3636 ASSERT(bp->b_fspriv != NULL);
3637 ASSERT(bp->b_iodone != NULL);
3641 return -EFSCORRUPTED;