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>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
33 #include "xfs_attr_sf.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_errortag.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_filestream.h"
46 #include "xfs_cksum.h"
47 #include "xfs_trace.h"
48 #include "xfs_icache.h"
49 #include "xfs_symlink.h"
50 #include "xfs_trans_priv.h"
52 #include "xfs_bmap_btree.h"
53 #include "xfs_reflink.h"
54 #include "xfs_dir2_priv.h"
56 kmem_zone_t *xfs_inode_zone;
59 * Used in xfs_itruncate_extents(). This is the maximum number of extents
60 * freed from a file in a single transaction.
62 #define XFS_ITRUNC_MAX_EXTENTS 2
64 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
65 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
66 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
69 * helper function to extract extent size hint from inode
75 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
76 return ip->i_d.di_extsize;
77 if (XFS_IS_REALTIME_INODE(ip))
78 return ip->i_mount->m_sb.sb_rextsize;
83 * Helper function to extract CoW extent size hint from inode.
84 * Between the extent size hint and the CoW extent size hint, we
85 * return the greater of the two. If the value is zero (automatic),
86 * use the default size.
89 xfs_get_cowextsz_hint(
95 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
96 a = ip->i_d.di_cowextsize;
97 b = xfs_get_extsz_hint(ip);
101 return XFS_DEFAULT_COWEXTSZ_HINT;
106 * These two are wrapper routines around the xfs_ilock() routine used to
107 * centralize some grungy code. They are used in places that wish to lock the
108 * inode solely for reading the extents. The reason these places can't just
109 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
110 * bringing in of the extents from disk for a file in b-tree format. If the
111 * inode is in b-tree format, then we need to lock the inode exclusively until
112 * the extents are read in. Locking it exclusively all the time would limit
113 * our parallelism unnecessarily, though. What we do instead is check to see
114 * if the extents have been read in yet, and only lock the inode exclusively
117 * The functions return a value which should be given to the corresponding
118 * xfs_iunlock() call.
121 xfs_ilock_data_map_shared(
122 struct xfs_inode *ip)
124 uint lock_mode = XFS_ILOCK_SHARED;
126 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
127 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
128 lock_mode = XFS_ILOCK_EXCL;
129 xfs_ilock(ip, lock_mode);
134 xfs_ilock_attr_map_shared(
135 struct xfs_inode *ip)
137 uint lock_mode = XFS_ILOCK_SHARED;
139 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
140 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
141 lock_mode = XFS_ILOCK_EXCL;
142 xfs_ilock(ip, lock_mode);
147 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
148 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
149 * various combinations of the locks to be obtained.
151 * The 3 locks should always be ordered so that the IO lock is obtained first,
152 * the mmap lock second and the ilock last in order to prevent deadlock.
154 * Basic locking order:
156 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
158 * mmap_sem locking order:
160 * i_rwsem -> page lock -> mmap_sem
161 * mmap_sem -> i_mmap_lock -> page_lock
163 * The difference in mmap_sem locking order mean that we cannot hold the
164 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
165 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
166 * in get_user_pages() to map the user pages into the kernel address space for
167 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
168 * page faults already hold the mmap_sem.
170 * Hence to serialise fully against both syscall and mmap based IO, we need to
171 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
172 * taken in places where we need to invalidate the page cache in a race
173 * free manner (e.g. truncate, hole punch and other extent manipulation
181 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
184 * You can't set both SHARED and EXCL for the same lock,
185 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
186 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
188 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
189 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
190 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
191 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
192 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
193 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
194 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
196 if (lock_flags & XFS_IOLOCK_EXCL) {
197 down_write_nested(&VFS_I(ip)->i_rwsem,
198 XFS_IOLOCK_DEP(lock_flags));
199 } else if (lock_flags & XFS_IOLOCK_SHARED) {
200 down_read_nested(&VFS_I(ip)->i_rwsem,
201 XFS_IOLOCK_DEP(lock_flags));
204 if (lock_flags & XFS_MMAPLOCK_EXCL)
205 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
206 else if (lock_flags & XFS_MMAPLOCK_SHARED)
207 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
209 if (lock_flags & XFS_ILOCK_EXCL)
210 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
211 else if (lock_flags & XFS_ILOCK_SHARED)
212 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
216 * This is just like xfs_ilock(), except that the caller
217 * is guaranteed not to sleep. It returns 1 if it gets
218 * the requested locks and 0 otherwise. If the IO lock is
219 * obtained but the inode lock cannot be, then the IO lock
220 * is dropped before returning.
222 * ip -- the inode being locked
223 * lock_flags -- this parameter indicates the inode's locks to be
224 * to be locked. See the comment for xfs_ilock() for a list
232 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
235 * You can't set both SHARED and EXCL for the same lock,
236 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
237 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
239 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
240 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
241 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
242 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
243 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
244 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
245 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
247 if (lock_flags & XFS_IOLOCK_EXCL) {
248 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
250 } else if (lock_flags & XFS_IOLOCK_SHARED) {
251 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
255 if (lock_flags & XFS_MMAPLOCK_EXCL) {
256 if (!mrtryupdate(&ip->i_mmaplock))
257 goto out_undo_iolock;
258 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
259 if (!mrtryaccess(&ip->i_mmaplock))
260 goto out_undo_iolock;
263 if (lock_flags & XFS_ILOCK_EXCL) {
264 if (!mrtryupdate(&ip->i_lock))
265 goto out_undo_mmaplock;
266 } else if (lock_flags & XFS_ILOCK_SHARED) {
267 if (!mrtryaccess(&ip->i_lock))
268 goto out_undo_mmaplock;
273 if (lock_flags & XFS_MMAPLOCK_EXCL)
274 mrunlock_excl(&ip->i_mmaplock);
275 else if (lock_flags & XFS_MMAPLOCK_SHARED)
276 mrunlock_shared(&ip->i_mmaplock);
278 if (lock_flags & XFS_IOLOCK_EXCL)
279 up_write(&VFS_I(ip)->i_rwsem);
280 else if (lock_flags & XFS_IOLOCK_SHARED)
281 up_read(&VFS_I(ip)->i_rwsem);
287 * xfs_iunlock() is used to drop the inode locks acquired with
288 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
289 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
290 * that we know which locks to drop.
292 * ip -- the inode being unlocked
293 * lock_flags -- this parameter indicates the inode's locks to be
294 * to be unlocked. See the comment for xfs_ilock() for a list
295 * of valid values for this parameter.
304 * You can't set both SHARED and EXCL for the same lock,
305 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
306 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
308 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
309 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
310 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
311 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
312 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
313 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
314 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
315 ASSERT(lock_flags != 0);
317 if (lock_flags & XFS_IOLOCK_EXCL)
318 up_write(&VFS_I(ip)->i_rwsem);
319 else if (lock_flags & XFS_IOLOCK_SHARED)
320 up_read(&VFS_I(ip)->i_rwsem);
322 if (lock_flags & XFS_MMAPLOCK_EXCL)
323 mrunlock_excl(&ip->i_mmaplock);
324 else if (lock_flags & XFS_MMAPLOCK_SHARED)
325 mrunlock_shared(&ip->i_mmaplock);
327 if (lock_flags & XFS_ILOCK_EXCL)
328 mrunlock_excl(&ip->i_lock);
329 else if (lock_flags & XFS_ILOCK_SHARED)
330 mrunlock_shared(&ip->i_lock);
332 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
336 * give up write locks. the i/o lock cannot be held nested
337 * if it is being demoted.
344 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
346 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
348 if (lock_flags & XFS_ILOCK_EXCL)
349 mrdemote(&ip->i_lock);
350 if (lock_flags & XFS_MMAPLOCK_EXCL)
351 mrdemote(&ip->i_mmaplock);
352 if (lock_flags & XFS_IOLOCK_EXCL)
353 downgrade_write(&VFS_I(ip)->i_rwsem);
355 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
358 #if defined(DEBUG) || defined(XFS_WARN)
364 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
365 if (!(lock_flags & XFS_ILOCK_SHARED))
366 return !!ip->i_lock.mr_writer;
367 return rwsem_is_locked(&ip->i_lock.mr_lock);
370 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
371 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
372 return !!ip->i_mmaplock.mr_writer;
373 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
376 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
377 if (!(lock_flags & XFS_IOLOCK_SHARED))
378 return !debug_locks ||
379 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
380 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
389 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
390 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
391 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
392 * errors and warnings.
394 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
396 xfs_lockdep_subclass_ok(
399 return subclass < MAX_LOCKDEP_SUBCLASSES;
402 #define xfs_lockdep_subclass_ok(subclass) (true)
406 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
407 * value. This can be called for any type of inode lock combination, including
408 * parent locking. Care must be taken to ensure we don't overrun the subclass
409 * storage fields in the class mask we build.
412 xfs_lock_inumorder(int lock_mode, int subclass)
416 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
418 ASSERT(xfs_lockdep_subclass_ok(subclass));
420 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
421 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
422 class += subclass << XFS_IOLOCK_SHIFT;
425 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
426 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
427 class += subclass << XFS_MMAPLOCK_SHIFT;
430 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
431 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
432 class += subclass << XFS_ILOCK_SHIFT;
435 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
439 * The following routine will lock n inodes in exclusive mode. We assume the
440 * caller calls us with the inodes in i_ino order.
442 * We need to detect deadlock where an inode that we lock is in the AIL and we
443 * start waiting for another inode that is locked by a thread in a long running
444 * transaction (such as truncate). This can result in deadlock since the long
445 * running trans might need to wait for the inode we just locked in order to
446 * push the tail and free space in the log.
448 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
449 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
450 * lock more than one at a time, lockdep will report false positives saying we
451 * have violated locking orders.
459 int attempts = 0, i, j, try_lock;
463 * Currently supports between 2 and 5 inodes with exclusive locking. We
464 * support an arbitrary depth of locking here, but absolute limits on
465 * inodes depend on the the type of locking and the limits placed by
466 * lockdep annotations in xfs_lock_inumorder. These are all checked by
469 ASSERT(ips && inodes >= 2 && inodes <= 5);
470 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
472 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
474 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
475 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
476 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
477 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
479 if (lock_mode & XFS_IOLOCK_EXCL) {
480 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
481 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
482 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
487 for (; i < inodes; i++) {
490 if (i && (ips[i] == ips[i - 1])) /* Already locked */
494 * If try_lock is not set yet, make sure all locked inodes are
495 * not in the AIL. If any are, set try_lock to be used later.
498 for (j = (i - 1); j >= 0 && !try_lock; j--) {
499 lp = (xfs_log_item_t *)ips[j]->i_itemp;
500 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
506 * If any of the previous locks we have locked is in the AIL,
507 * we must TRY to get the second and subsequent locks. If
508 * we can't get any, we must release all we have
512 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
516 /* try_lock means we have an inode locked that is in the AIL. */
518 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
522 * Unlock all previous guys and try again. xfs_iunlock will try
523 * to push the tail if the inode is in the AIL.
526 for (j = i - 1; j >= 0; j--) {
528 * Check to see if we've already unlocked this one. Not
529 * the first one going back, and the inode ptr is the
532 if (j != (i - 1) && ips[j] == ips[j + 1])
535 xfs_iunlock(ips[j], lock_mode);
538 if ((attempts % 5) == 0) {
539 delay(1); /* Don't just spin the CPU */
548 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
549 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
550 * lock more than one at a time, lockdep will report false positives saying we
551 * have violated locking orders.
563 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
564 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
565 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
567 ASSERT(ip0->i_ino != ip1->i_ino);
569 if (ip0->i_ino > ip1->i_ino) {
576 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
579 * If the first lock we have locked is in the AIL, we must TRY to get
580 * the second lock. If we can't get it, we must release the first one
583 lp = (xfs_log_item_t *)ip0->i_itemp;
584 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
585 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
586 xfs_iunlock(ip0, lock_mode);
587 if ((++attempts % 5) == 0)
588 delay(1); /* Don't just spin the CPU */
592 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
599 struct xfs_inode *ip)
601 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
602 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
605 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
606 if (xfs_isiflocked(ip))
608 } while (!xfs_iflock_nowait(ip));
610 finish_wait(wq, &wait.wq_entry);
621 if (di_flags & XFS_DIFLAG_ANY) {
622 if (di_flags & XFS_DIFLAG_REALTIME)
623 flags |= FS_XFLAG_REALTIME;
624 if (di_flags & XFS_DIFLAG_PREALLOC)
625 flags |= FS_XFLAG_PREALLOC;
626 if (di_flags & XFS_DIFLAG_IMMUTABLE)
627 flags |= FS_XFLAG_IMMUTABLE;
628 if (di_flags & XFS_DIFLAG_APPEND)
629 flags |= FS_XFLAG_APPEND;
630 if (di_flags & XFS_DIFLAG_SYNC)
631 flags |= FS_XFLAG_SYNC;
632 if (di_flags & XFS_DIFLAG_NOATIME)
633 flags |= FS_XFLAG_NOATIME;
634 if (di_flags & XFS_DIFLAG_NODUMP)
635 flags |= FS_XFLAG_NODUMP;
636 if (di_flags & XFS_DIFLAG_RTINHERIT)
637 flags |= FS_XFLAG_RTINHERIT;
638 if (di_flags & XFS_DIFLAG_PROJINHERIT)
639 flags |= FS_XFLAG_PROJINHERIT;
640 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
641 flags |= FS_XFLAG_NOSYMLINKS;
642 if (di_flags & XFS_DIFLAG_EXTSIZE)
643 flags |= FS_XFLAG_EXTSIZE;
644 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
645 flags |= FS_XFLAG_EXTSZINHERIT;
646 if (di_flags & XFS_DIFLAG_NODEFRAG)
647 flags |= FS_XFLAG_NODEFRAG;
648 if (di_flags & XFS_DIFLAG_FILESTREAM)
649 flags |= FS_XFLAG_FILESTREAM;
652 if (di_flags2 & XFS_DIFLAG2_ANY) {
653 if (di_flags2 & XFS_DIFLAG2_DAX)
654 flags |= FS_XFLAG_DAX;
655 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
656 flags |= FS_XFLAG_COWEXTSIZE;
660 flags |= FS_XFLAG_HASATTR;
667 struct xfs_inode *ip)
669 struct xfs_icdinode *dic = &ip->i_d;
671 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
675 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
676 * is allowed, otherwise it has to be an exact match. If a CI match is found,
677 * ci_name->name will point to a the actual name (caller must free) or
678 * will be set to NULL if an exact match is found.
683 struct xfs_name *name,
685 struct xfs_name *ci_name)
690 trace_xfs_lookup(dp, name);
692 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
695 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
699 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
707 kmem_free(ci_name->name);
714 * Allocate an inode on disk and return a copy of its in-core version.
715 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
716 * appropriately within the inode. The uid and gid for the inode are
717 * set according to the contents of the given cred structure.
719 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
720 * has a free inode available, call xfs_iget() to obtain the in-core
721 * version of the allocated inode. Finally, fill in the inode and
722 * log its initial contents. In this case, ialloc_context would be
725 * If xfs_dialloc() does not have an available inode, it will replenish
726 * its supply by doing an allocation. Since we can only do one
727 * allocation within a transaction without deadlocks, we must commit
728 * the current transaction before returning the inode itself.
729 * In this case, therefore, we will set ialloc_context and return.
730 * The caller should then commit the current transaction, start a new
731 * transaction, and call xfs_ialloc() again to actually get the inode.
733 * To ensure that some other process does not grab the inode that
734 * was allocated during the first call to xfs_ialloc(), this routine
735 * also returns the [locked] bp pointing to the head of the freelist
736 * as ialloc_context. The caller should hold this buffer across
737 * the commit and pass it back into this routine on the second call.
739 * If we are allocating quota inodes, we do not have a parent inode
740 * to attach to or associate with (i.e. pip == NULL) because they
741 * are not linked into the directory structure - they are attached
742 * directly to the superblock - and so have no parent.
752 xfs_buf_t **ialloc_context,
755 struct xfs_mount *mp = tp->t_mountp;
764 * Call the space management code to pick
765 * the on-disk inode to be allocated.
767 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
768 ialloc_context, &ino);
771 if (*ialloc_context || ino == NULLFSINO) {
775 ASSERT(*ialloc_context == NULL);
778 * Get the in-core inode with the lock held exclusively.
779 * This is because we're setting fields here we need
780 * to prevent others from looking at until we're done.
782 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
783 XFS_ILOCK_EXCL, &ip);
790 * We always convert v1 inodes to v2 now - we only support filesystems
791 * with >= v2 inode capability, so there is no reason for ever leaving
792 * an inode in v1 format.
794 if (ip->i_d.di_version == 1)
795 ip->i_d.di_version = 2;
797 inode->i_mode = mode;
798 set_nlink(inode, nlink);
799 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
800 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
801 inode->i_rdev = rdev;
802 xfs_set_projid(ip, prid);
804 if (pip && XFS_INHERIT_GID(pip)) {
805 ip->i_d.di_gid = pip->i_d.di_gid;
806 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
807 inode->i_mode |= S_ISGID;
811 * If the group ID of the new file does not match the effective group
812 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
813 * (and only if the irix_sgid_inherit compatibility variable is set).
815 if ((irix_sgid_inherit) &&
816 (inode->i_mode & S_ISGID) &&
817 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
818 inode->i_mode &= ~S_ISGID;
821 ip->i_d.di_nextents = 0;
822 ASSERT(ip->i_d.di_nblocks == 0);
824 tv = current_time(inode);
829 ip->i_d.di_extsize = 0;
830 ip->i_d.di_dmevmask = 0;
831 ip->i_d.di_dmstate = 0;
832 ip->i_d.di_flags = 0;
834 if (ip->i_d.di_version == 3) {
835 inode->i_version = 1;
836 ip->i_d.di_flags2 = 0;
837 ip->i_d.di_cowextsize = 0;
838 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
839 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
843 flags = XFS_ILOG_CORE;
844 switch (mode & S_IFMT) {
849 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
850 ip->i_df.if_flags = 0;
851 flags |= XFS_ILOG_DEV;
855 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
859 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
860 di_flags |= XFS_DIFLAG_RTINHERIT;
861 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
862 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
863 ip->i_d.di_extsize = pip->i_d.di_extsize;
865 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
866 di_flags |= XFS_DIFLAG_PROJINHERIT;
867 } else if (S_ISREG(mode)) {
868 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
869 di_flags |= XFS_DIFLAG_REALTIME;
870 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
871 di_flags |= XFS_DIFLAG_EXTSIZE;
872 ip->i_d.di_extsize = pip->i_d.di_extsize;
875 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
877 di_flags |= XFS_DIFLAG_NOATIME;
878 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
880 di_flags |= XFS_DIFLAG_NODUMP;
881 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
883 di_flags |= XFS_DIFLAG_SYNC;
884 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
885 xfs_inherit_nosymlinks)
886 di_flags |= XFS_DIFLAG_NOSYMLINKS;
887 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
888 xfs_inherit_nodefrag)
889 di_flags |= XFS_DIFLAG_NODEFRAG;
890 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
891 di_flags |= XFS_DIFLAG_FILESTREAM;
893 ip->i_d.di_flags |= di_flags;
896 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
897 pip->i_d.di_version == 3 &&
898 ip->i_d.di_version == 3) {
899 uint64_t di_flags2 = 0;
901 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
902 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 di_flags2 |= XFS_DIFLAG2_DAX;
908 ip->i_d.di_flags2 |= di_flags2;
912 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
913 ip->i_df.if_flags = XFS_IFEXTENTS;
914 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
915 ip->i_df.if_u1.if_root = NULL;
921 * Attribute fork settings for new inode.
923 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
924 ip->i_d.di_anextents = 0;
927 * Log the new values stuffed into the inode.
929 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
930 xfs_trans_log_inode(tp, ip, flags);
932 /* now that we have an i_mode we can setup the inode structure */
940 * Allocates a new inode from disk and return a pointer to the
941 * incore copy. This routine will internally commit the current
942 * transaction and allocate a new one if the Space Manager needed
943 * to do an allocation to replenish the inode free-list.
945 * This routine is designed to be called from xfs_create and
951 xfs_trans_t **tpp, /* input: current transaction;
952 output: may be a new transaction. */
953 xfs_inode_t *dp, /* directory within whose allocate
958 prid_t prid, /* project id */
959 xfs_inode_t **ipp, /* pointer to inode; it will be
966 xfs_buf_t *ialloc_context = NULL;
972 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
975 * xfs_ialloc will return a pointer to an incore inode if
976 * the Space Manager has an available inode on the free
977 * list. Otherwise, it will do an allocation and replenish
978 * the freelist. Since we can only do one allocation per
979 * transaction without deadlocks, we will need to commit the
980 * current transaction and start a new one. We will then
981 * need to call xfs_ialloc again to get the inode.
983 * If xfs_ialloc did an allocation to replenish the freelist,
984 * it returns the bp containing the head of the freelist as
985 * ialloc_context. We will hold a lock on it across the
986 * transaction commit so that no other process can steal
987 * the inode(s) that we've just allocated.
989 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
993 * Return an error if we were unable to allocate a new inode.
994 * This should only happen if we run out of space on disk or
995 * encounter a disk error.
1001 if (!ialloc_context && !ip) {
1007 * If the AGI buffer is non-NULL, then we were unable to get an
1008 * inode in one operation. We need to commit the current
1009 * transaction and call xfs_ialloc() again. It is guaranteed
1010 * to succeed the second time.
1012 if (ialloc_context) {
1014 * Normally, xfs_trans_commit releases all the locks.
1015 * We call bhold to hang on to the ialloc_context across
1016 * the commit. Holding this buffer prevents any other
1017 * processes from doing any allocations in this
1020 xfs_trans_bhold(tp, ialloc_context);
1023 * We want the quota changes to be associated with the next
1024 * transaction, NOT this one. So, detach the dqinfo from this
1025 * and attach it to the next transaction.
1030 dqinfo = (void *)tp->t_dqinfo;
1031 tp->t_dqinfo = NULL;
1032 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1033 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1036 code = xfs_trans_roll(&tp);
1037 if (committed != NULL)
1041 * Re-attach the quota info that we detached from prev trx.
1044 tp->t_dqinfo = dqinfo;
1045 tp->t_flags |= tflags;
1049 xfs_buf_relse(ialloc_context);
1054 xfs_trans_bjoin(tp, ialloc_context);
1057 * Call ialloc again. Since we've locked out all
1058 * other allocations in this allocation group,
1059 * this call should always succeed.
1061 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1062 &ialloc_context, &ip);
1065 * If we get an error at this point, return to the caller
1066 * so that the current transaction can be aborted.
1073 ASSERT(!ialloc_context && ip);
1076 if (committed != NULL)
1087 * Decrement the link count on an inode & log the change. If this causes the
1088 * link count to go to zero, move the inode to AGI unlinked list so that it can
1089 * be freed when the last active reference goes away via xfs_inactive().
1091 static int /* error */
1096 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1098 drop_nlink(VFS_I(ip));
1099 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1101 if (VFS_I(ip)->i_nlink)
1104 return xfs_iunlink(tp, ip);
1108 * Increment the link count on an inode & log the change.
1115 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1117 ASSERT(ip->i_d.di_version > 1);
1118 inc_nlink(VFS_I(ip));
1119 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1126 struct xfs_name *name,
1131 int is_dir = S_ISDIR(mode);
1132 struct xfs_mount *mp = dp->i_mount;
1133 struct xfs_inode *ip = NULL;
1134 struct xfs_trans *tp = NULL;
1136 struct xfs_defer_ops dfops;
1137 xfs_fsblock_t first_block;
1138 bool unlock_dp_on_error = false;
1140 struct xfs_dquot *udqp = NULL;
1141 struct xfs_dquot *gdqp = NULL;
1142 struct xfs_dquot *pdqp = NULL;
1143 struct xfs_trans_res *tres;
1146 trace_xfs_create(dp, name);
1148 if (XFS_FORCED_SHUTDOWN(mp))
1151 prid = xfs_get_initial_prid(dp);
1154 * Make sure that we have allocated dquot(s) on disk.
1156 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1157 xfs_kgid_to_gid(current_fsgid()), prid,
1158 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1159 &udqp, &gdqp, &pdqp);
1164 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1165 tres = &M_RES(mp)->tr_mkdir;
1167 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1168 tres = &M_RES(mp)->tr_create;
1172 * Initially assume that the file does not exist and
1173 * reserve the resources for that case. If that is not
1174 * the case we'll drop the one we have and get a more
1175 * appropriate transaction later.
1177 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1178 if (error == -ENOSPC) {
1179 /* flush outstanding delalloc blocks and retry */
1180 xfs_flush_inodes(mp);
1181 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1184 goto out_release_inode;
1186 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1187 unlock_dp_on_error = true;
1189 xfs_defer_init(&dfops, &first_block);
1192 * Reserve disk quota and the inode.
1194 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1195 pdqp, resblks, 1, 0);
1197 goto out_trans_cancel;
1200 * A newly created regular or special file just has one directory
1201 * entry pointing to them, but a directory also the "." entry
1202 * pointing to itself.
1204 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip,
1207 goto out_trans_cancel;
1210 * Now we join the directory inode to the transaction. We do not do it
1211 * earlier because xfs_dir_ialloc might commit the previous transaction
1212 * (and release all the locks). An error from here on will result in
1213 * the transaction cancel unlocking dp so don't do it explicitly in the
1216 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1217 unlock_dp_on_error = false;
1219 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1220 &first_block, &dfops, resblks ?
1221 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1223 ASSERT(error != -ENOSPC);
1224 goto out_trans_cancel;
1226 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1227 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1230 error = xfs_dir_init(tp, ip, dp);
1232 goto out_bmap_cancel;
1234 error = xfs_bumplink(tp, dp);
1236 goto out_bmap_cancel;
1240 * If this is a synchronous mount, make sure that the
1241 * create transaction goes to disk before returning to
1244 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1245 xfs_trans_set_sync(tp);
1248 * Attach the dquot(s) to the inodes and modify them incore.
1249 * These ids of the inode couldn't have changed since the new
1250 * inode has been locked ever since it was created.
1252 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1254 error = xfs_defer_finish(&tp, &dfops);
1256 goto out_bmap_cancel;
1258 error = xfs_trans_commit(tp);
1260 goto out_release_inode;
1262 xfs_qm_dqrele(udqp);
1263 xfs_qm_dqrele(gdqp);
1264 xfs_qm_dqrele(pdqp);
1270 xfs_defer_cancel(&dfops);
1272 xfs_trans_cancel(tp);
1275 * Wait until after the current transaction is aborted to finish the
1276 * setup of the inode and release the inode. This prevents recursive
1277 * transactions and deadlocks from xfs_inactive.
1280 xfs_finish_inode_setup(ip);
1284 xfs_qm_dqrele(udqp);
1285 xfs_qm_dqrele(gdqp);
1286 xfs_qm_dqrele(pdqp);
1288 if (unlock_dp_on_error)
1289 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1295 struct xfs_inode *dp,
1296 struct dentry *dentry,
1298 struct xfs_inode **ipp)
1300 struct xfs_mount *mp = dp->i_mount;
1301 struct xfs_inode *ip = NULL;
1302 struct xfs_trans *tp = NULL;
1305 struct xfs_dquot *udqp = NULL;
1306 struct xfs_dquot *gdqp = NULL;
1307 struct xfs_dquot *pdqp = NULL;
1308 struct xfs_trans_res *tres;
1311 if (XFS_FORCED_SHUTDOWN(mp))
1314 prid = xfs_get_initial_prid(dp);
1317 * Make sure that we have allocated dquot(s) on disk.
1319 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1320 xfs_kgid_to_gid(current_fsgid()), prid,
1321 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1322 &udqp, &gdqp, &pdqp);
1326 resblks = XFS_IALLOC_SPACE_RES(mp);
1327 tres = &M_RES(mp)->tr_create_tmpfile;
1329 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1331 goto out_release_inode;
1333 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1334 pdqp, resblks, 1, 0);
1336 goto out_trans_cancel;
1338 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip, NULL);
1340 goto out_trans_cancel;
1342 if (mp->m_flags & XFS_MOUNT_WSYNC)
1343 xfs_trans_set_sync(tp);
1346 * Attach the dquot(s) to the inodes and modify them incore.
1347 * These ids of the inode couldn't have changed since the new
1348 * inode has been locked ever since it was created.
1350 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1352 error = xfs_iunlink(tp, ip);
1354 goto out_trans_cancel;
1356 error = xfs_trans_commit(tp);
1358 goto out_release_inode;
1360 xfs_qm_dqrele(udqp);
1361 xfs_qm_dqrele(gdqp);
1362 xfs_qm_dqrele(pdqp);
1368 xfs_trans_cancel(tp);
1371 * Wait until after the current transaction is aborted to finish the
1372 * setup of the inode and release the inode. This prevents recursive
1373 * transactions and deadlocks from xfs_inactive.
1376 xfs_finish_inode_setup(ip);
1380 xfs_qm_dqrele(udqp);
1381 xfs_qm_dqrele(gdqp);
1382 xfs_qm_dqrele(pdqp);
1391 struct xfs_name *target_name)
1393 xfs_mount_t *mp = tdp->i_mount;
1396 struct xfs_defer_ops dfops;
1397 xfs_fsblock_t first_block;
1400 trace_xfs_link(tdp, target_name);
1402 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1404 if (XFS_FORCED_SHUTDOWN(mp))
1407 error = xfs_qm_dqattach(sip, 0);
1411 error = xfs_qm_dqattach(tdp, 0);
1415 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1416 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1417 if (error == -ENOSPC) {
1419 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1424 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1426 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1427 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1430 * If we are using project inheritance, we only allow hard link
1431 * creation in our tree when the project IDs are the same; else
1432 * the tree quota mechanism could be circumvented.
1434 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1435 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1441 error = xfs_dir_canenter(tp, tdp, target_name);
1446 xfs_defer_init(&dfops, &first_block);
1449 * Handle initial link state of O_TMPFILE inode
1451 if (VFS_I(sip)->i_nlink == 0) {
1452 error = xfs_iunlink_remove(tp, sip);
1457 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1458 &first_block, &dfops, resblks);
1461 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1462 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1464 error = xfs_bumplink(tp, sip);
1469 * If this is a synchronous mount, make sure that the
1470 * link transaction goes to disk before returning to
1473 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1474 xfs_trans_set_sync(tp);
1476 error = xfs_defer_finish(&tp, &dfops);
1478 xfs_defer_cancel(&dfops);
1482 return xfs_trans_commit(tp);
1485 xfs_trans_cancel(tp);
1490 /* Clear the reflink flag and the cowblocks tag if possible. */
1492 xfs_itruncate_clear_reflink_flags(
1493 struct xfs_inode *ip)
1495 struct xfs_ifork *dfork;
1496 struct xfs_ifork *cfork;
1498 if (!xfs_is_reflink_inode(ip))
1500 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1501 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1502 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1503 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1504 if (cfork->if_bytes == 0)
1505 xfs_inode_clear_cowblocks_tag(ip);
1509 * Free up the underlying blocks past new_size. The new size must be smaller
1510 * than the current size. This routine can be used both for the attribute and
1511 * data fork, and does not modify the inode size, which is left to the caller.
1513 * The transaction passed to this routine must have made a permanent log
1514 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1515 * given transaction and start new ones, so make sure everything involved in
1516 * the transaction is tidy before calling here. Some transaction will be
1517 * returned to the caller to be committed. The incoming transaction must
1518 * already include the inode, and both inode locks must be held exclusively.
1519 * The inode must also be "held" within the transaction. On return the inode
1520 * will be "held" within the returned transaction. This routine does NOT
1521 * require any disk space to be reserved for it within the transaction.
1523 * If we get an error, we must return with the inode locked and linked into the
1524 * current transaction. This keeps things simple for the higher level code,
1525 * because it always knows that the inode is locked and held in the transaction
1526 * that returns to it whether errors occur or not. We don't mark the inode
1527 * dirty on error so that transactions can be easily aborted if possible.
1530 xfs_itruncate_extents(
1531 struct xfs_trans **tpp,
1532 struct xfs_inode *ip,
1534 xfs_fsize_t new_size)
1536 struct xfs_mount *mp = ip->i_mount;
1537 struct xfs_trans *tp = *tpp;
1538 struct xfs_defer_ops dfops;
1539 xfs_fsblock_t first_block;
1540 xfs_fileoff_t first_unmap_block;
1541 xfs_fileoff_t last_block;
1542 xfs_filblks_t unmap_len;
1546 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1547 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1548 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1549 ASSERT(new_size <= XFS_ISIZE(ip));
1550 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1551 ASSERT(ip->i_itemp != NULL);
1552 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1553 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1555 trace_xfs_itruncate_extents_start(ip, new_size);
1558 * Since it is possible for space to become allocated beyond
1559 * the end of the file (in a crash where the space is allocated
1560 * but the inode size is not yet updated), simply remove any
1561 * blocks which show up between the new EOF and the maximum
1562 * possible file size. If the first block to be removed is
1563 * beyond the maximum file size (ie it is the same as last_block),
1564 * then there is nothing to do.
1566 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1567 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1568 if (first_unmap_block == last_block)
1571 ASSERT(first_unmap_block < last_block);
1572 unmap_len = last_block - first_unmap_block + 1;
1574 xfs_defer_init(&dfops, &first_block);
1575 error = xfs_bunmapi(tp, ip,
1576 first_unmap_block, unmap_len,
1577 xfs_bmapi_aflag(whichfork),
1578 XFS_ITRUNC_MAX_EXTENTS,
1579 &first_block, &dfops,
1582 goto out_bmap_cancel;
1585 * Duplicate the transaction that has the permanent
1586 * reservation and commit the old transaction.
1588 xfs_defer_ijoin(&dfops, ip);
1589 error = xfs_defer_finish(&tp, &dfops);
1591 goto out_bmap_cancel;
1593 error = xfs_trans_roll_inode(&tp, ip);
1598 /* Remove all pending CoW reservations. */
1599 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1604 xfs_itruncate_clear_reflink_flags(ip);
1607 * Always re-log the inode so that our permanent transaction can keep
1608 * on rolling it forward in the log.
1610 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1612 trace_xfs_itruncate_extents_end(ip, new_size);
1619 * If the bunmapi call encounters an error, return to the caller where
1620 * the transaction can be properly aborted. We just need to make sure
1621 * we're not holding any resources that we were not when we came in.
1623 xfs_defer_cancel(&dfops);
1631 xfs_mount_t *mp = ip->i_mount;
1634 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1637 /* If this is a read-only mount, don't do this (would generate I/O) */
1638 if (mp->m_flags & XFS_MOUNT_RDONLY)
1641 if (!XFS_FORCED_SHUTDOWN(mp)) {
1645 * If we previously truncated this file and removed old data
1646 * in the process, we want to initiate "early" writeout on
1647 * the last close. This is an attempt to combat the notorious
1648 * NULL files problem which is particularly noticeable from a
1649 * truncate down, buffered (re-)write (delalloc), followed by
1650 * a crash. What we are effectively doing here is
1651 * significantly reducing the time window where we'd otherwise
1652 * be exposed to that problem.
1654 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1656 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1657 if (ip->i_delayed_blks > 0) {
1658 error = filemap_flush(VFS_I(ip)->i_mapping);
1665 if (VFS_I(ip)->i_nlink == 0)
1668 if (xfs_can_free_eofblocks(ip, false)) {
1671 * Check if the inode is being opened, written and closed
1672 * frequently and we have delayed allocation blocks outstanding
1673 * (e.g. streaming writes from the NFS server), truncating the
1674 * blocks past EOF will cause fragmentation to occur.
1676 * In this case don't do the truncation, but we have to be
1677 * careful how we detect this case. Blocks beyond EOF show up as
1678 * i_delayed_blks even when the inode is clean, so we need to
1679 * truncate them away first before checking for a dirty release.
1680 * Hence on the first dirty close we will still remove the
1681 * speculative allocation, but after that we will leave it in
1684 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1687 * If we can't get the iolock just skip truncating the blocks
1688 * past EOF because we could deadlock with the mmap_sem
1689 * otherwise. We'll get another chance to drop them once the
1690 * last reference to the inode is dropped, so we'll never leak
1691 * blocks permanently.
1693 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1694 error = xfs_free_eofblocks(ip);
1695 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1700 /* delalloc blocks after truncation means it really is dirty */
1701 if (ip->i_delayed_blks)
1702 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1708 * xfs_inactive_truncate
1710 * Called to perform a truncate when an inode becomes unlinked.
1713 xfs_inactive_truncate(
1714 struct xfs_inode *ip)
1716 struct xfs_mount *mp = ip->i_mount;
1717 struct xfs_trans *tp;
1720 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1722 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1726 xfs_ilock(ip, XFS_ILOCK_EXCL);
1727 xfs_trans_ijoin(tp, ip, 0);
1730 * Log the inode size first to prevent stale data exposure in the event
1731 * of a system crash before the truncate completes. See the related
1732 * comment in xfs_vn_setattr_size() for details.
1734 ip->i_d.di_size = 0;
1735 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1737 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1739 goto error_trans_cancel;
1741 ASSERT(ip->i_d.di_nextents == 0);
1743 error = xfs_trans_commit(tp);
1747 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1751 xfs_trans_cancel(tp);
1753 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1758 * xfs_inactive_ifree()
1760 * Perform the inode free when an inode is unlinked.
1764 struct xfs_inode *ip)
1766 struct xfs_defer_ops dfops;
1767 xfs_fsblock_t first_block;
1768 struct xfs_mount *mp = ip->i_mount;
1769 struct xfs_trans *tp;
1773 * We try to use a per-AG reservation for any block needed by the finobt
1774 * tree, but as the finobt feature predates the per-AG reservation
1775 * support a degraded file system might not have enough space for the
1776 * reservation at mount time. In that case try to dip into the reserved
1779 * Send a warning if the reservation does happen to fail, as the inode
1780 * now remains allocated and sits on the unlinked list until the fs is
1783 if (unlikely(mp->m_inotbt_nores)) {
1784 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1785 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1788 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1791 if (error == -ENOSPC) {
1792 xfs_warn_ratelimited(mp,
1793 "Failed to remove inode(s) from unlinked list. "
1794 "Please free space, unmount and run xfs_repair.");
1796 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1801 xfs_ilock(ip, XFS_ILOCK_EXCL);
1802 xfs_trans_ijoin(tp, ip, 0);
1804 xfs_defer_init(&dfops, &first_block);
1805 error = xfs_ifree(tp, ip, &dfops);
1808 * If we fail to free the inode, shut down. The cancel
1809 * might do that, we need to make sure. Otherwise the
1810 * inode might be lost for a long time or forever.
1812 if (!XFS_FORCED_SHUTDOWN(mp)) {
1813 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1815 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1817 xfs_trans_cancel(tp);
1818 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1823 * Credit the quota account(s). The inode is gone.
1825 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1828 * Just ignore errors at this point. There is nothing we can do except
1829 * to try to keep going. Make sure it's not a silent error.
1831 error = xfs_defer_finish(&tp, &dfops);
1833 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1835 xfs_defer_cancel(&dfops);
1837 error = xfs_trans_commit(tp);
1839 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1842 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1849 * This is called when the vnode reference count for the vnode
1850 * goes to zero. If the file has been unlinked, then it must
1851 * now be truncated. Also, we clear all of the read-ahead state
1852 * kept for the inode here since the file is now closed.
1858 struct xfs_mount *mp;
1863 * If the inode is already free, then there can be nothing
1866 if (VFS_I(ip)->i_mode == 0) {
1867 ASSERT(ip->i_df.if_real_bytes == 0);
1868 ASSERT(ip->i_df.if_broot_bytes == 0);
1873 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1875 /* If this is a read-only mount, don't do this (would generate I/O) */
1876 if (mp->m_flags & XFS_MOUNT_RDONLY)
1879 if (VFS_I(ip)->i_nlink != 0) {
1881 * force is true because we are evicting an inode from the
1882 * cache. Post-eof blocks must be freed, lest we end up with
1883 * broken free space accounting.
1885 * Note: don't bother with iolock here since lockdep complains
1886 * about acquiring it in reclaim context. We have the only
1887 * reference to the inode at this point anyways.
1889 if (xfs_can_free_eofblocks(ip, true))
1890 xfs_free_eofblocks(ip);
1895 if (S_ISREG(VFS_I(ip)->i_mode) &&
1896 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1897 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1900 error = xfs_qm_dqattach(ip, 0);
1904 if (S_ISLNK(VFS_I(ip)->i_mode))
1905 error = xfs_inactive_symlink(ip);
1907 error = xfs_inactive_truncate(ip);
1912 * If there are attributes associated with the file then blow them away
1913 * now. The code calls a routine that recursively deconstructs the
1914 * attribute fork. If also blows away the in-core attribute fork.
1916 if (XFS_IFORK_Q(ip)) {
1917 error = xfs_attr_inactive(ip);
1923 ASSERT(ip->i_d.di_anextents == 0);
1924 ASSERT(ip->i_d.di_forkoff == 0);
1929 error = xfs_inactive_ifree(ip);
1934 * Release the dquots held by inode, if any.
1936 xfs_qm_dqdetach(ip);
1940 * This is called when the inode's link count goes to 0 or we are creating a
1941 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1942 * set to true as the link count is dropped to zero by the VFS after we've
1943 * created the file successfully, so we have to add it to the unlinked list
1944 * while the link count is non-zero.
1946 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1947 * list when the inode is freed.
1951 struct xfs_trans *tp,
1952 struct xfs_inode *ip)
1954 xfs_mount_t *mp = tp->t_mountp;
1964 ASSERT(VFS_I(ip)->i_mode != 0);
1967 * Get the agi buffer first. It ensures lock ordering
1970 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1973 agi = XFS_BUF_TO_AGI(agibp);
1976 * Get the index into the agi hash table for the
1977 * list this inode will go on.
1979 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1981 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1982 ASSERT(agi->agi_unlinked[bucket_index]);
1983 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1985 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1987 * There is already another inode in the bucket we need
1988 * to add ourselves to. Add us at the front of the list.
1989 * Here we put the head pointer into our next pointer,
1990 * and then we fall through to point the head at us.
1992 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1997 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1998 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1999 offset = ip->i_imap.im_boffset +
2000 offsetof(xfs_dinode_t, di_next_unlinked);
2002 /* need to recalc the inode CRC if appropriate */
2003 xfs_dinode_calc_crc(mp, dip);
2005 xfs_trans_inode_buf(tp, ibp);
2006 xfs_trans_log_buf(tp, ibp, offset,
2007 (offset + sizeof(xfs_agino_t) - 1));
2008 xfs_inobp_check(mp, ibp);
2012 * Point the bucket head pointer at the inode being inserted.
2015 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2016 offset = offsetof(xfs_agi_t, agi_unlinked) +
2017 (sizeof(xfs_agino_t) * bucket_index);
2018 xfs_trans_log_buf(tp, agibp, offset,
2019 (offset + sizeof(xfs_agino_t) - 1));
2024 * Pull the on-disk inode from the AGI unlinked list.
2037 xfs_agnumber_t agno;
2039 xfs_agino_t next_agino;
2040 xfs_buf_t *last_ibp;
2041 xfs_dinode_t *last_dip = NULL;
2043 int offset, last_offset = 0;
2047 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2050 * Get the agi buffer first. It ensures lock ordering
2053 error = xfs_read_agi(mp, tp, agno, &agibp);
2057 agi = XFS_BUF_TO_AGI(agibp);
2060 * Get the index into the agi hash table for the
2061 * list this inode will go on.
2063 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2065 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2066 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2067 ASSERT(agi->agi_unlinked[bucket_index]);
2069 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2071 * We're at the head of the list. Get the inode's on-disk
2072 * buffer to see if there is anyone after us on the list.
2073 * Only modify our next pointer if it is not already NULLAGINO.
2074 * This saves us the overhead of dealing with the buffer when
2075 * there is no need to change it.
2077 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2080 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2084 next_agino = be32_to_cpu(dip->di_next_unlinked);
2085 ASSERT(next_agino != 0);
2086 if (next_agino != NULLAGINO) {
2087 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2088 offset = ip->i_imap.im_boffset +
2089 offsetof(xfs_dinode_t, di_next_unlinked);
2091 /* need to recalc the inode CRC if appropriate */
2092 xfs_dinode_calc_crc(mp, dip);
2094 xfs_trans_inode_buf(tp, ibp);
2095 xfs_trans_log_buf(tp, ibp, offset,
2096 (offset + sizeof(xfs_agino_t) - 1));
2097 xfs_inobp_check(mp, ibp);
2099 xfs_trans_brelse(tp, ibp);
2102 * Point the bucket head pointer at the next inode.
2104 ASSERT(next_agino != 0);
2105 ASSERT(next_agino != agino);
2106 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2107 offset = offsetof(xfs_agi_t, agi_unlinked) +
2108 (sizeof(xfs_agino_t) * bucket_index);
2109 xfs_trans_log_buf(tp, agibp, offset,
2110 (offset + sizeof(xfs_agino_t) - 1));
2113 * We need to search the list for the inode being freed.
2115 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2117 while (next_agino != agino) {
2118 struct xfs_imap imap;
2121 xfs_trans_brelse(tp, last_ibp);
2124 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2126 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2129 "%s: xfs_imap returned error %d.",
2134 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2138 "%s: xfs_imap_to_bp returned error %d.",
2143 last_offset = imap.im_boffset;
2144 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2145 ASSERT(next_agino != NULLAGINO);
2146 ASSERT(next_agino != 0);
2150 * Now last_ibp points to the buffer previous to us on the
2151 * unlinked list. Pull us from the list.
2153 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2156 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2160 next_agino = be32_to_cpu(dip->di_next_unlinked);
2161 ASSERT(next_agino != 0);
2162 ASSERT(next_agino != agino);
2163 if (next_agino != NULLAGINO) {
2164 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2165 offset = ip->i_imap.im_boffset +
2166 offsetof(xfs_dinode_t, di_next_unlinked);
2168 /* need to recalc the inode CRC if appropriate */
2169 xfs_dinode_calc_crc(mp, dip);
2171 xfs_trans_inode_buf(tp, ibp);
2172 xfs_trans_log_buf(tp, ibp, offset,
2173 (offset + sizeof(xfs_agino_t) - 1));
2174 xfs_inobp_check(mp, ibp);
2176 xfs_trans_brelse(tp, ibp);
2179 * Point the previous inode on the list to the next inode.
2181 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2182 ASSERT(next_agino != 0);
2183 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2185 /* need to recalc the inode CRC if appropriate */
2186 xfs_dinode_calc_crc(mp, last_dip);
2188 xfs_trans_inode_buf(tp, last_ibp);
2189 xfs_trans_log_buf(tp, last_ibp, offset,
2190 (offset + sizeof(xfs_agino_t) - 1));
2191 xfs_inobp_check(mp, last_ibp);
2197 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2198 * inodes that are in memory - they all must be marked stale and attached to
2199 * the cluster buffer.
2203 xfs_inode_t *free_ip,
2205 struct xfs_icluster *xic)
2207 xfs_mount_t *mp = free_ip->i_mount;
2208 int blks_per_cluster;
2209 int inodes_per_cluster;
2216 xfs_inode_log_item_t *iip;
2217 xfs_log_item_t *lip;
2218 struct xfs_perag *pag;
2221 inum = xic->first_ino;
2222 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2223 blks_per_cluster = xfs_icluster_size_fsb(mp);
2224 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2225 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2227 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2229 * The allocation bitmap tells us which inodes of the chunk were
2230 * physically allocated. Skip the cluster if an inode falls into
2233 ioffset = inum - xic->first_ino;
2234 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2235 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2239 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2240 XFS_INO_TO_AGBNO(mp, inum));
2243 * We obtain and lock the backing buffer first in the process
2244 * here, as we have to ensure that any dirty inode that we
2245 * can't get the flush lock on is attached to the buffer.
2246 * If we scan the in-memory inodes first, then buffer IO can
2247 * complete before we get a lock on it, and hence we may fail
2248 * to mark all the active inodes on the buffer stale.
2250 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2251 mp->m_bsize * blks_per_cluster,
2258 * This buffer may not have been correctly initialised as we
2259 * didn't read it from disk. That's not important because we are
2260 * only using to mark the buffer as stale in the log, and to
2261 * attach stale cached inodes on it. That means it will never be
2262 * dispatched for IO. If it is, we want to know about it, and we
2263 * want it to fail. We can acheive this by adding a write
2264 * verifier to the buffer.
2266 bp->b_ops = &xfs_inode_buf_ops;
2269 * Walk the inodes already attached to the buffer and mark them
2270 * stale. These will all have the flush locks held, so an
2271 * in-memory inode walk can't lock them. By marking them all
2272 * stale first, we will not attempt to lock them in the loop
2273 * below as the XFS_ISTALE flag will be set.
2277 if (lip->li_type == XFS_LI_INODE) {
2278 iip = (xfs_inode_log_item_t *)lip;
2279 ASSERT(iip->ili_logged == 1);
2280 lip->li_cb = xfs_istale_done;
2281 xfs_trans_ail_copy_lsn(mp->m_ail,
2282 &iip->ili_flush_lsn,
2283 &iip->ili_item.li_lsn);
2284 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2286 lip = lip->li_bio_list;
2291 * For each inode in memory attempt to add it to the inode
2292 * buffer and set it up for being staled on buffer IO
2293 * completion. This is safe as we've locked out tail pushing
2294 * and flushing by locking the buffer.
2296 * We have already marked every inode that was part of a
2297 * transaction stale above, which means there is no point in
2298 * even trying to lock them.
2300 for (i = 0; i < inodes_per_cluster; i++) {
2303 ip = radix_tree_lookup(&pag->pag_ici_root,
2304 XFS_INO_TO_AGINO(mp, (inum + i)));
2306 /* Inode not in memory, nothing to do */
2313 * because this is an RCU protected lookup, we could
2314 * find a recently freed or even reallocated inode
2315 * during the lookup. We need to check under the
2316 * i_flags_lock for a valid inode here. Skip it if it
2317 * is not valid, the wrong inode or stale.
2319 spin_lock(&ip->i_flags_lock);
2320 if (ip->i_ino != inum + i ||
2321 __xfs_iflags_test(ip, XFS_ISTALE)) {
2322 spin_unlock(&ip->i_flags_lock);
2326 spin_unlock(&ip->i_flags_lock);
2329 * Don't try to lock/unlock the current inode, but we
2330 * _cannot_ skip the other inodes that we did not find
2331 * in the list attached to the buffer and are not
2332 * already marked stale. If we can't lock it, back off
2335 if (ip != free_ip) {
2336 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2343 * Check the inode number again in case we're
2344 * racing with freeing in xfs_reclaim_inode().
2345 * See the comments in that function for more
2346 * information as to why the initial check is
2349 if (ip->i_ino != inum + i) {
2350 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2358 xfs_iflags_set(ip, XFS_ISTALE);
2361 * we don't need to attach clean inodes or those only
2362 * with unlogged changes (which we throw away, anyway).
2365 if (!iip || xfs_inode_clean(ip)) {
2366 ASSERT(ip != free_ip);
2368 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2372 iip->ili_last_fields = iip->ili_fields;
2373 iip->ili_fields = 0;
2374 iip->ili_fsync_fields = 0;
2375 iip->ili_logged = 1;
2376 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2377 &iip->ili_item.li_lsn);
2379 xfs_buf_attach_iodone(bp, xfs_istale_done,
2383 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2386 xfs_trans_stale_inode_buf(tp, bp);
2387 xfs_trans_binval(tp, bp);
2395 * Free any local-format buffers sitting around before we reset to
2399 xfs_ifree_local_data(
2400 struct xfs_inode *ip,
2403 struct xfs_ifork *ifp;
2405 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2408 ifp = XFS_IFORK_PTR(ip, whichfork);
2409 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2413 * This is called to return an inode to the inode free list.
2414 * The inode should already be truncated to 0 length and have
2415 * no pages associated with it. This routine also assumes that
2416 * the inode is already a part of the transaction.
2418 * The on-disk copy of the inode will have been added to the list
2419 * of unlinked inodes in the AGI. We need to remove the inode from
2420 * that list atomically with respect to freeing it here.
2426 struct xfs_defer_ops *dfops)
2429 struct xfs_icluster xic = { 0 };
2431 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2432 ASSERT(VFS_I(ip)->i_nlink == 0);
2433 ASSERT(ip->i_d.di_nextents == 0);
2434 ASSERT(ip->i_d.di_anextents == 0);
2435 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2436 ASSERT(ip->i_d.di_nblocks == 0);
2439 * Pull the on-disk inode from the AGI unlinked list.
2441 error = xfs_iunlink_remove(tp, ip);
2445 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2449 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2450 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2452 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2453 ip->i_d.di_flags = 0;
2454 ip->i_d.di_dmevmask = 0;
2455 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2456 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2457 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2459 * Bump the generation count so no one will be confused
2460 * by reincarnations of this inode.
2462 VFS_I(ip)->i_generation++;
2463 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2466 error = xfs_ifree_cluster(ip, tp, &xic);
2472 * This is called to unpin an inode. The caller must have the inode locked
2473 * in at least shared mode so that the buffer cannot be subsequently pinned
2474 * once someone is waiting for it to be unpinned.
2478 struct xfs_inode *ip)
2480 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2482 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2484 /* Give the log a push to start the unpinning I/O */
2485 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2491 struct xfs_inode *ip)
2493 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2494 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2499 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2500 if (xfs_ipincount(ip))
2502 } while (xfs_ipincount(ip));
2503 finish_wait(wq, &wait.wq_entry);
2508 struct xfs_inode *ip)
2510 if (xfs_ipincount(ip))
2511 __xfs_iunpin_wait(ip);
2515 * Removing an inode from the namespace involves removing the directory entry
2516 * and dropping the link count on the inode. Removing the directory entry can
2517 * result in locking an AGF (directory blocks were freed) and removing a link
2518 * count can result in placing the inode on an unlinked list which results in
2521 * The big problem here is that we have an ordering constraint on AGF and AGI
2522 * locking - inode allocation locks the AGI, then can allocate a new extent for
2523 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2524 * removes the inode from the unlinked list, requiring that we lock the AGI
2525 * first, and then freeing the inode can result in an inode chunk being freed
2526 * and hence freeing disk space requiring that we lock an AGF.
2528 * Hence the ordering that is imposed by other parts of the code is AGI before
2529 * AGF. This means we cannot remove the directory entry before we drop the inode
2530 * reference count and put it on the unlinked list as this results in a lock
2531 * order of AGF then AGI, and this can deadlock against inode allocation and
2532 * freeing. Therefore we must drop the link counts before we remove the
2535 * This is still safe from a transactional point of view - it is not until we
2536 * get to xfs_defer_finish() that we have the possibility of multiple
2537 * transactions in this operation. Hence as long as we remove the directory
2538 * entry and drop the link count in the first transaction of the remove
2539 * operation, there are no transactional constraints on the ordering here.
2544 struct xfs_name *name,
2547 xfs_mount_t *mp = dp->i_mount;
2548 xfs_trans_t *tp = NULL;
2549 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2551 struct xfs_defer_ops dfops;
2552 xfs_fsblock_t first_block;
2555 trace_xfs_remove(dp, name);
2557 if (XFS_FORCED_SHUTDOWN(mp))
2560 error = xfs_qm_dqattach(dp, 0);
2564 error = xfs_qm_dqattach(ip, 0);
2569 * We try to get the real space reservation first,
2570 * allowing for directory btree deletion(s) implying
2571 * possible bmap insert(s). If we can't get the space
2572 * reservation then we use 0 instead, and avoid the bmap
2573 * btree insert(s) in the directory code by, if the bmap
2574 * insert tries to happen, instead trimming the LAST
2575 * block from the directory.
2577 resblks = XFS_REMOVE_SPACE_RES(mp);
2578 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2579 if (error == -ENOSPC) {
2581 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2585 ASSERT(error != -ENOSPC);
2589 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2591 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2592 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2595 * If we're removing a directory perform some additional validation.
2598 ASSERT(VFS_I(ip)->i_nlink >= 2);
2599 if (VFS_I(ip)->i_nlink != 2) {
2601 goto out_trans_cancel;
2603 if (!xfs_dir_isempty(ip)) {
2605 goto out_trans_cancel;
2608 /* Drop the link from ip's "..". */
2609 error = xfs_droplink(tp, dp);
2611 goto out_trans_cancel;
2613 /* Drop the "." link from ip to self. */
2614 error = xfs_droplink(tp, ip);
2616 goto out_trans_cancel;
2619 * When removing a non-directory we need to log the parent
2620 * inode here. For a directory this is done implicitly
2621 * by the xfs_droplink call for the ".." entry.
2623 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2625 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2627 /* Drop the link from dp to ip. */
2628 error = xfs_droplink(tp, ip);
2630 goto out_trans_cancel;
2632 xfs_defer_init(&dfops, &first_block);
2633 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2634 &first_block, &dfops, resblks);
2636 ASSERT(error != -ENOENT);
2637 goto out_bmap_cancel;
2641 * If this is a synchronous mount, make sure that the
2642 * remove transaction goes to disk before returning to
2645 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2646 xfs_trans_set_sync(tp);
2648 error = xfs_defer_finish(&tp, &dfops);
2650 goto out_bmap_cancel;
2652 error = xfs_trans_commit(tp);
2656 if (is_dir && xfs_inode_is_filestream(ip))
2657 xfs_filestream_deassociate(ip);
2662 xfs_defer_cancel(&dfops);
2664 xfs_trans_cancel(tp);
2670 * Enter all inodes for a rename transaction into a sorted array.
2672 #define __XFS_SORT_INODES 5
2674 xfs_sort_for_rename(
2675 struct xfs_inode *dp1, /* in: old (source) directory inode */
2676 struct xfs_inode *dp2, /* in: new (target) directory inode */
2677 struct xfs_inode *ip1, /* in: inode of old entry */
2678 struct xfs_inode *ip2, /* in: inode of new entry */
2679 struct xfs_inode *wip, /* in: whiteout inode */
2680 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2681 int *num_inodes) /* in/out: inodes in array */
2685 ASSERT(*num_inodes == __XFS_SORT_INODES);
2686 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2689 * i_tab contains a list of pointers to inodes. We initialize
2690 * the table here & we'll sort it. We will then use it to
2691 * order the acquisition of the inode locks.
2693 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2706 * Sort the elements via bubble sort. (Remember, there are at
2707 * most 5 elements to sort, so this is adequate.)
2709 for (i = 0; i < *num_inodes; i++) {
2710 for (j = 1; j < *num_inodes; j++) {
2711 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2712 struct xfs_inode *temp = i_tab[j];
2713 i_tab[j] = i_tab[j-1];
2722 struct xfs_trans *tp,
2723 struct xfs_defer_ops *dfops)
2728 * If this is a synchronous mount, make sure that the rename transaction
2729 * goes to disk before returning to the user.
2731 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2732 xfs_trans_set_sync(tp);
2734 error = xfs_defer_finish(&tp, dfops);
2736 xfs_defer_cancel(dfops);
2737 xfs_trans_cancel(tp);
2741 return xfs_trans_commit(tp);
2745 * xfs_cross_rename()
2747 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2751 struct xfs_trans *tp,
2752 struct xfs_inode *dp1,
2753 struct xfs_name *name1,
2754 struct xfs_inode *ip1,
2755 struct xfs_inode *dp2,
2756 struct xfs_name *name2,
2757 struct xfs_inode *ip2,
2758 struct xfs_defer_ops *dfops,
2759 xfs_fsblock_t *first_block,
2767 /* Swap inode number for dirent in first parent */
2768 error = xfs_dir_replace(tp, dp1, name1,
2770 first_block, dfops, spaceres);
2772 goto out_trans_abort;
2774 /* Swap inode number for dirent in second parent */
2775 error = xfs_dir_replace(tp, dp2, name2,
2777 first_block, dfops, spaceres);
2779 goto out_trans_abort;
2782 * If we're renaming one or more directories across different parents,
2783 * update the respective ".." entries (and link counts) to match the new
2787 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2789 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2790 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2791 dp1->i_ino, first_block,
2794 goto out_trans_abort;
2796 /* transfer ip2 ".." reference to dp1 */
2797 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2798 error = xfs_droplink(tp, dp2);
2800 goto out_trans_abort;
2801 error = xfs_bumplink(tp, dp1);
2803 goto out_trans_abort;
2807 * Although ip1 isn't changed here, userspace needs
2808 * to be warned about the change, so that applications
2809 * relying on it (like backup ones), will properly
2812 ip1_flags |= XFS_ICHGTIME_CHG;
2813 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2816 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2817 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2818 dp2->i_ino, first_block,
2821 goto out_trans_abort;
2823 /* transfer ip1 ".." reference to dp2 */
2824 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2825 error = xfs_droplink(tp, dp1);
2827 goto out_trans_abort;
2828 error = xfs_bumplink(tp, dp2);
2830 goto out_trans_abort;
2834 * Although ip2 isn't changed here, userspace needs
2835 * to be warned about the change, so that applications
2836 * relying on it (like backup ones), will properly
2839 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2840 ip2_flags |= XFS_ICHGTIME_CHG;
2845 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2846 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2849 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2850 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2853 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2854 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2856 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2857 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2858 return xfs_finish_rename(tp, dfops);
2861 xfs_defer_cancel(dfops);
2862 xfs_trans_cancel(tp);
2867 * xfs_rename_alloc_whiteout()
2869 * Return a referenced, unlinked, unlocked inode that that can be used as a
2870 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2871 * crash between allocating the inode and linking it into the rename transaction
2872 * recovery will free the inode and we won't leak it.
2875 xfs_rename_alloc_whiteout(
2876 struct xfs_inode *dp,
2877 struct xfs_inode **wip)
2879 struct xfs_inode *tmpfile;
2882 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2887 * Prepare the tmpfile inode as if it were created through the VFS.
2888 * Otherwise, the link increment paths will complain about nlink 0->1.
2889 * Drop the link count as done by d_tmpfile(), complete the inode setup
2890 * and flag it as linkable.
2892 drop_nlink(VFS_I(tmpfile));
2893 xfs_setup_iops(tmpfile);
2894 xfs_finish_inode_setup(tmpfile);
2895 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2906 struct xfs_inode *src_dp,
2907 struct xfs_name *src_name,
2908 struct xfs_inode *src_ip,
2909 struct xfs_inode *target_dp,
2910 struct xfs_name *target_name,
2911 struct xfs_inode *target_ip,
2914 struct xfs_mount *mp = src_dp->i_mount;
2915 struct xfs_trans *tp;
2916 struct xfs_defer_ops dfops;
2917 xfs_fsblock_t first_block;
2918 struct xfs_inode *wip = NULL; /* whiteout inode */
2919 struct xfs_inode *inodes[__XFS_SORT_INODES];
2920 int num_inodes = __XFS_SORT_INODES;
2921 bool new_parent = (src_dp != target_dp);
2922 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2926 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2928 if ((flags & RENAME_EXCHANGE) && !target_ip)
2932 * If we are doing a whiteout operation, allocate the whiteout inode
2933 * we will be placing at the target and ensure the type is set
2936 if (flags & RENAME_WHITEOUT) {
2937 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2938 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2942 /* setup target dirent info as whiteout */
2943 src_name->type = XFS_DIR3_FT_CHRDEV;
2946 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2947 inodes, &num_inodes);
2949 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2950 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2951 if (error == -ENOSPC) {
2953 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2957 goto out_release_wip;
2960 * Attach the dquots to the inodes
2962 error = xfs_qm_vop_rename_dqattach(inodes);
2964 goto out_trans_cancel;
2967 * Lock all the participating inodes. Depending upon whether
2968 * the target_name exists in the target directory, and
2969 * whether the target directory is the same as the source
2970 * directory, we can lock from 2 to 4 inodes.
2972 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2975 * Join all the inodes to the transaction. From this point on,
2976 * we can rely on either trans_commit or trans_cancel to unlock
2979 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2981 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2982 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2984 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2986 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2989 * If we are using project inheritance, we only allow renames
2990 * into our tree when the project IDs are the same; else the
2991 * tree quota mechanism would be circumvented.
2993 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2994 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2996 goto out_trans_cancel;
2999 xfs_defer_init(&dfops, &first_block);
3001 /* RENAME_EXCHANGE is unique from here on. */
3002 if (flags & RENAME_EXCHANGE)
3003 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3004 target_dp, target_name, target_ip,
3005 &dfops, &first_block, spaceres);
3008 * Set up the target.
3010 if (target_ip == NULL) {
3012 * If there's no space reservation, check the entry will
3013 * fit before actually inserting it.
3016 error = xfs_dir_canenter(tp, target_dp, target_name);
3018 goto out_trans_cancel;
3021 * If target does not exist and the rename crosses
3022 * directories, adjust the target directory link count
3023 * to account for the ".." reference from the new entry.
3025 error = xfs_dir_createname(tp, target_dp, target_name,
3026 src_ip->i_ino, &first_block,
3029 goto out_bmap_cancel;
3031 xfs_trans_ichgtime(tp, target_dp,
3032 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3034 if (new_parent && src_is_directory) {
3035 error = xfs_bumplink(tp, target_dp);
3037 goto out_bmap_cancel;
3039 } else { /* target_ip != NULL */
3041 * If target exists and it's a directory, check that both
3042 * target and source are directories and that target can be
3043 * destroyed, or that neither is a directory.
3045 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3047 * Make sure target dir is empty.
3049 if (!(xfs_dir_isempty(target_ip)) ||
3050 (VFS_I(target_ip)->i_nlink > 2)) {
3052 goto out_trans_cancel;
3057 * Link the source inode under the target name.
3058 * If the source inode is a directory and we are moving
3059 * it across directories, its ".." entry will be
3060 * inconsistent until we replace that down below.
3062 * In case there is already an entry with the same
3063 * name at the destination directory, remove it first.
3065 error = xfs_dir_replace(tp, target_dp, target_name,
3067 &first_block, &dfops, spaceres);
3069 goto out_bmap_cancel;
3071 xfs_trans_ichgtime(tp, target_dp,
3072 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3075 * Decrement the link count on the target since the target
3076 * dir no longer points to it.
3078 error = xfs_droplink(tp, target_ip);
3080 goto out_bmap_cancel;
3082 if (src_is_directory) {
3084 * Drop the link from the old "." entry.
3086 error = xfs_droplink(tp, target_ip);
3088 goto out_bmap_cancel;
3090 } /* target_ip != NULL */
3093 * Remove the source.
3095 if (new_parent && src_is_directory) {
3097 * Rewrite the ".." entry to point to the new
3100 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3102 &first_block, &dfops, spaceres);
3103 ASSERT(error != -EEXIST);
3105 goto out_bmap_cancel;
3109 * We always want to hit the ctime on the source inode.
3111 * This isn't strictly required by the standards since the source
3112 * inode isn't really being changed, but old unix file systems did
3113 * it and some incremental backup programs won't work without it.
3115 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3116 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3119 * Adjust the link count on src_dp. This is necessary when
3120 * renaming a directory, either within one parent when
3121 * the target existed, or across two parent directories.
3123 if (src_is_directory && (new_parent || target_ip != NULL)) {
3126 * Decrement link count on src_directory since the
3127 * entry that's moved no longer points to it.
3129 error = xfs_droplink(tp, src_dp);
3131 goto out_bmap_cancel;
3135 * For whiteouts, we only need to update the source dirent with the
3136 * inode number of the whiteout inode rather than removing it
3140 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3141 &first_block, &dfops, spaceres);
3143 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3144 &first_block, &dfops, spaceres);
3146 goto out_bmap_cancel;
3149 * For whiteouts, we need to bump the link count on the whiteout inode.
3150 * This means that failures all the way up to this point leave the inode
3151 * on the unlinked list and so cleanup is a simple matter of dropping
3152 * the remaining reference to it. If we fail here after bumping the link
3153 * count, we're shutting down the filesystem so we'll never see the
3154 * intermediate state on disk.
3157 ASSERT(VFS_I(wip)->i_nlink == 0);
3158 error = xfs_bumplink(tp, wip);
3160 goto out_bmap_cancel;
3161 error = xfs_iunlink_remove(tp, wip);
3163 goto out_bmap_cancel;
3164 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3167 * Now we have a real link, clear the "I'm a tmpfile" state
3168 * flag from the inode so it doesn't accidentally get misused in
3171 VFS_I(wip)->i_state &= ~I_LINKABLE;
3174 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3175 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3177 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3179 error = xfs_finish_rename(tp, &dfops);
3185 xfs_defer_cancel(&dfops);
3187 xfs_trans_cancel(tp);
3196 struct xfs_inode *ip,
3199 struct xfs_mount *mp = ip->i_mount;
3200 struct xfs_perag *pag;
3201 unsigned long first_index, mask;
3202 unsigned long inodes_per_cluster;
3204 struct xfs_inode **cilist;
3205 struct xfs_inode *cip;
3211 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3213 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3214 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3215 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3219 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3220 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3222 /* really need a gang lookup range call here */
3223 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3224 first_index, inodes_per_cluster);
3228 for (i = 0; i < nr_found; i++) {
3234 * because this is an RCU protected lookup, we could find a
3235 * recently freed or even reallocated inode during the lookup.
3236 * We need to check under the i_flags_lock for a valid inode
3237 * here. Skip it if it is not valid or the wrong inode.
3239 spin_lock(&cip->i_flags_lock);
3241 __xfs_iflags_test(cip, XFS_ISTALE)) {
3242 spin_unlock(&cip->i_flags_lock);
3247 * Once we fall off the end of the cluster, no point checking
3248 * any more inodes in the list because they will also all be
3249 * outside the cluster.
3251 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3252 spin_unlock(&cip->i_flags_lock);
3255 spin_unlock(&cip->i_flags_lock);
3258 * Do an un-protected check to see if the inode is dirty and
3259 * is a candidate for flushing. These checks will be repeated
3260 * later after the appropriate locks are acquired.
3262 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3266 * Try to get locks. If any are unavailable or it is pinned,
3267 * then this inode cannot be flushed and is skipped.
3270 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3272 if (!xfs_iflock_nowait(cip)) {
3273 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3276 if (xfs_ipincount(cip)) {
3278 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3284 * Check the inode number again, just to be certain we are not
3285 * racing with freeing in xfs_reclaim_inode(). See the comments
3286 * in that function for more information as to why the initial
3287 * check is not sufficient.
3291 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3296 * arriving here means that this inode can be flushed. First
3297 * re-check that it's dirty before flushing.
3299 if (!xfs_inode_clean(cip)) {
3301 error = xfs_iflush_int(cip, bp);
3303 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3304 goto cluster_corrupt_out;
3310 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3314 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3315 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3326 cluster_corrupt_out:
3328 * Corruption detected in the clustering loop. Invalidate the
3329 * inode buffer and shut down the filesystem.
3333 * Clean up the buffer. If it was delwri, just release it --
3334 * brelse can handle it with no problems. If not, shut down the
3335 * filesystem before releasing the buffer.
3337 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3341 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3343 if (!bufwasdelwri) {
3345 * Just like incore_relse: if we have b_iodone functions,
3346 * mark the buffer as an error and call them. Otherwise
3347 * mark it as stale and brelse.
3350 bp->b_flags &= ~XBF_DONE;
3352 xfs_buf_ioerror(bp, -EIO);
3361 * Unlocks the flush lock
3363 xfs_iflush_abort(cip, false);
3366 return -EFSCORRUPTED;
3370 * Flush dirty inode metadata into the backing buffer.
3372 * The caller must have the inode lock and the inode flush lock held. The
3373 * inode lock will still be held upon return to the caller, and the inode
3374 * flush lock will be released after the inode has reached the disk.
3376 * The caller must write out the buffer returned in *bpp and release it.
3380 struct xfs_inode *ip,
3381 struct xfs_buf **bpp)
3383 struct xfs_mount *mp = ip->i_mount;
3384 struct xfs_buf *bp = NULL;
3385 struct xfs_dinode *dip;
3388 XFS_STATS_INC(mp, xs_iflush_count);
3390 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3391 ASSERT(xfs_isiflocked(ip));
3392 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3393 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3397 xfs_iunpin_wait(ip);
3400 * For stale inodes we cannot rely on the backing buffer remaining
3401 * stale in cache for the remaining life of the stale inode and so
3402 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3403 * inodes below. We have to check this after ensuring the inode is
3404 * unpinned so that it is safe to reclaim the stale inode after the
3407 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3413 * This may have been unpinned because the filesystem is shutting
3414 * down forcibly. If that's the case we must not write this inode
3415 * to disk, because the log record didn't make it to disk.
3417 * We also have to remove the log item from the AIL in this case,
3418 * as we wait for an empty AIL as part of the unmount process.
3420 if (XFS_FORCED_SHUTDOWN(mp)) {
3426 * Get the buffer containing the on-disk inode. We are doing a try-lock
3427 * operation here, so we may get an EAGAIN error. In that case, we
3428 * simply want to return with the inode still dirty.
3430 * If we get any other error, we effectively have a corruption situation
3431 * and we cannot flush the inode, so we treat it the same as failing
3434 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3436 if (error == -EAGAIN) {
3444 * First flush out the inode that xfs_iflush was called with.
3446 error = xfs_iflush_int(ip, bp);
3451 * If the buffer is pinned then push on the log now so we won't
3452 * get stuck waiting in the write for too long.
3454 if (xfs_buf_ispinned(bp))
3455 xfs_log_force(mp, 0);
3459 * see if other inodes can be gathered into this write
3461 error = xfs_iflush_cluster(ip, bp);
3463 goto cluster_corrupt_out;
3471 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3472 cluster_corrupt_out:
3473 error = -EFSCORRUPTED;
3476 * Unlocks the flush lock
3478 xfs_iflush_abort(ip, false);
3484 struct xfs_inode *ip,
3487 struct xfs_inode_log_item *iip = ip->i_itemp;
3488 struct xfs_dinode *dip;
3489 struct xfs_mount *mp = ip->i_mount;
3491 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3492 ASSERT(xfs_isiflocked(ip));
3493 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3494 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3495 ASSERT(iip != NULL && iip->ili_fields != 0);
3496 ASSERT(ip->i_d.di_version > 1);
3498 /* set *dip = inode's place in the buffer */
3499 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3501 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3502 mp, XFS_ERRTAG_IFLUSH_1)) {
3503 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3504 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3505 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3508 if (S_ISREG(VFS_I(ip)->i_mode)) {
3510 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3511 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3512 mp, XFS_ERRTAG_IFLUSH_3)) {
3513 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3514 "%s: Bad regular inode %Lu, ptr 0x%p",
3515 __func__, ip->i_ino, ip);
3518 } else if (S_ISDIR(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 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3523 mp, XFS_ERRTAG_IFLUSH_4)) {
3524 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3525 "%s: Bad directory inode %Lu, ptr 0x%p",
3526 __func__, ip->i_ino, ip);
3530 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3531 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3532 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3533 "%s: detected corrupt incore inode %Lu, "
3534 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3535 __func__, ip->i_ino,
3536 ip->i_d.di_nextents + ip->i_d.di_anextents,
3537 ip->i_d.di_nblocks, ip);
3540 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3541 mp, XFS_ERRTAG_IFLUSH_6)) {
3542 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3543 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3544 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3549 * Inode item log recovery for v2 inodes are dependent on the
3550 * di_flushiter count for correct sequencing. We bump the flush
3551 * iteration count so we can detect flushes which postdate a log record
3552 * during recovery. This is redundant as we now log every change and
3553 * hence this can't happen but we need to still do it to ensure
3554 * backwards compatibility with old kernels that predate logging all
3557 if (ip->i_d.di_version < 3)
3558 ip->i_d.di_flushiter++;
3560 /* Check the inline directory data. */
3561 if (S_ISDIR(VFS_I(ip)->i_mode) &&
3562 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3563 xfs_dir2_sf_verify(ip))
3567 * Copy the dirty parts of the inode into the on-disk inode. We always
3568 * copy out the core of the inode, because if the inode is dirty at all
3571 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3573 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3574 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3575 ip->i_d.di_flushiter = 0;
3577 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3578 if (XFS_IFORK_Q(ip))
3579 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3580 xfs_inobp_check(mp, bp);
3583 * We've recorded everything logged in the inode, so we'd like to clear
3584 * the ili_fields bits so we don't log and flush things unnecessarily.
3585 * However, we can't stop logging all this information until the data
3586 * we've copied into the disk buffer is written to disk. If we did we
3587 * might overwrite the copy of the inode in the log with all the data
3588 * after re-logging only part of it, and in the face of a crash we
3589 * wouldn't have all the data we need to recover.
3591 * What we do is move the bits to the ili_last_fields field. When
3592 * logging the inode, these bits are moved back to the ili_fields field.
3593 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3594 * know that the information those bits represent is permanently on
3595 * disk. As long as the flush completes before the inode is logged
3596 * again, then both ili_fields and ili_last_fields will be cleared.
3598 * We can play with the ili_fields bits here, because the inode lock
3599 * must be held exclusively in order to set bits there and the flush
3600 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3601 * done routine can tell whether or not to look in the AIL. Also, store
3602 * the current LSN of the inode so that we can tell whether the item has
3603 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3604 * need the AIL lock, because it is a 64 bit value that cannot be read
3607 iip->ili_last_fields = iip->ili_fields;
3608 iip->ili_fields = 0;
3609 iip->ili_fsync_fields = 0;
3610 iip->ili_logged = 1;
3612 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3613 &iip->ili_item.li_lsn);
3616 * Attach the function xfs_iflush_done to the inode's
3617 * buffer. This will remove the inode from the AIL
3618 * and unlock the inode's flush lock when the inode is
3619 * completely written to disk.
3621 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3623 /* generate the checksum. */
3624 xfs_dinode_calc_crc(mp, dip);
3626 ASSERT(bp->b_fspriv != NULL);
3627 ASSERT(bp->b_iodone != NULL);
3631 return -EFSCORRUPTED;