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.
753 xfs_buf_t **ialloc_context,
756 struct xfs_mount *mp = tp->t_mountp;
765 * Call the space management code to pick
766 * the on-disk inode to be allocated.
768 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
769 ialloc_context, &ino);
772 if (*ialloc_context || ino == NULLFSINO) {
776 ASSERT(*ialloc_context == NULL);
779 * Get the in-core inode with the lock held exclusively.
780 * This is because we're setting fields here we need
781 * to prevent others from looking at until we're done.
783 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
784 XFS_ILOCK_EXCL, &ip);
791 * We always convert v1 inodes to v2 now - we only support filesystems
792 * with >= v2 inode capability, so there is no reason for ever leaving
793 * an inode in v1 format.
795 if (ip->i_d.di_version == 1)
796 ip->i_d.di_version = 2;
798 inode->i_mode = mode;
799 set_nlink(inode, nlink);
800 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
801 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
802 inode->i_rdev = rdev;
803 xfs_set_projid(ip, prid);
805 if (pip && XFS_INHERIT_GID(pip)) {
806 ip->i_d.di_gid = pip->i_d.di_gid;
807 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
808 inode->i_mode |= S_ISGID;
812 * If the group ID of the new file does not match the effective group
813 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
814 * (and only if the irix_sgid_inherit compatibility variable is set).
816 if ((irix_sgid_inherit) &&
817 (inode->i_mode & S_ISGID) &&
818 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
819 inode->i_mode &= ~S_ISGID;
822 ip->i_d.di_nextents = 0;
823 ASSERT(ip->i_d.di_nblocks == 0);
825 tv = current_time(inode);
830 ip->i_d.di_extsize = 0;
831 ip->i_d.di_dmevmask = 0;
832 ip->i_d.di_dmstate = 0;
833 ip->i_d.di_flags = 0;
835 if (ip->i_d.di_version == 3) {
836 inode->i_version = 1;
837 ip->i_d.di_flags2 = 0;
838 ip->i_d.di_cowextsize = 0;
839 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
840 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
844 flags = XFS_ILOG_CORE;
845 switch (mode & S_IFMT) {
850 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
851 ip->i_df.if_flags = 0;
852 flags |= XFS_ILOG_DEV;
856 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
860 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
861 di_flags |= XFS_DIFLAG_RTINHERIT;
862 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
863 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
864 ip->i_d.di_extsize = pip->i_d.di_extsize;
866 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
867 di_flags |= XFS_DIFLAG_PROJINHERIT;
868 } else if (S_ISREG(mode)) {
869 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
870 di_flags |= XFS_DIFLAG_REALTIME;
871 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
872 di_flags |= XFS_DIFLAG_EXTSIZE;
873 ip->i_d.di_extsize = pip->i_d.di_extsize;
876 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
878 di_flags |= XFS_DIFLAG_NOATIME;
879 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
881 di_flags |= XFS_DIFLAG_NODUMP;
882 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
884 di_flags |= XFS_DIFLAG_SYNC;
885 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
886 xfs_inherit_nosymlinks)
887 di_flags |= XFS_DIFLAG_NOSYMLINKS;
888 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
889 xfs_inherit_nodefrag)
890 di_flags |= XFS_DIFLAG_NODEFRAG;
891 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
892 di_flags |= XFS_DIFLAG_FILESTREAM;
894 ip->i_d.di_flags |= di_flags;
897 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
898 pip->i_d.di_version == 3 &&
899 ip->i_d.di_version == 3) {
900 uint64_t di_flags2 = 0;
902 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
903 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
904 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
906 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
907 di_flags2 |= XFS_DIFLAG2_DAX;
909 ip->i_d.di_flags2 |= di_flags2;
913 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
914 ip->i_df.if_flags = XFS_IFEXTENTS;
915 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
916 ip->i_df.if_u1.if_root = NULL;
922 * Attribute fork settings for new inode.
924 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
925 ip->i_d.di_anextents = 0;
928 * Log the new values stuffed into the inode.
930 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
931 xfs_trans_log_inode(tp, ip, flags);
933 /* now that we have an i_mode we can setup the inode structure */
941 * Allocates a new inode from disk and return a pointer to the
942 * incore copy. This routine will internally commit the current
943 * transaction and allocate a new one if the Space Manager needed
944 * to do an allocation to replenish the inode free-list.
946 * This routine is designed to be called from xfs_create and
952 xfs_trans_t **tpp, /* input: current transaction;
953 output: may be a new transaction. */
954 xfs_inode_t *dp, /* directory within whose allocate
959 prid_t prid, /* project id */
960 int okalloc, /* ok to allocate new space */
961 xfs_inode_t **ipp, /* pointer to inode; it will be
968 xfs_buf_t *ialloc_context = NULL;
974 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
977 * xfs_ialloc will return a pointer to an incore inode if
978 * the Space Manager has an available inode on the free
979 * list. Otherwise, it will do an allocation and replenish
980 * the freelist. Since we can only do one allocation per
981 * transaction without deadlocks, we will need to commit the
982 * current transaction and start a new one. We will then
983 * need to call xfs_ialloc again to get the inode.
985 * If xfs_ialloc did an allocation to replenish the freelist,
986 * it returns the bp containing the head of the freelist as
987 * ialloc_context. We will hold a lock on it across the
988 * transaction commit so that no other process can steal
989 * the inode(s) that we've just allocated.
991 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
992 &ialloc_context, &ip);
995 * Return an error if we were unable to allocate a new inode.
996 * This should only happen if we run out of space on disk or
997 * encounter a disk error.
1003 if (!ialloc_context && !ip) {
1009 * If the AGI buffer is non-NULL, then we were unable to get an
1010 * inode in one operation. We need to commit the current
1011 * transaction and call xfs_ialloc() again. It is guaranteed
1012 * to succeed the second time.
1014 if (ialloc_context) {
1016 * Normally, xfs_trans_commit releases all the locks.
1017 * We call bhold to hang on to the ialloc_context across
1018 * the commit. Holding this buffer prevents any other
1019 * processes from doing any allocations in this
1022 xfs_trans_bhold(tp, ialloc_context);
1025 * We want the quota changes to be associated with the next
1026 * transaction, NOT this one. So, detach the dqinfo from this
1027 * and attach it to the next transaction.
1032 dqinfo = (void *)tp->t_dqinfo;
1033 tp->t_dqinfo = NULL;
1034 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1035 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1038 code = xfs_trans_roll(&tp);
1039 if (committed != NULL)
1043 * Re-attach the quota info that we detached from prev trx.
1046 tp->t_dqinfo = dqinfo;
1047 tp->t_flags |= tflags;
1051 xfs_buf_relse(ialloc_context);
1056 xfs_trans_bjoin(tp, ialloc_context);
1059 * Call ialloc again. Since we've locked out all
1060 * other allocations in this allocation group,
1061 * this call should always succeed.
1063 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1064 okalloc, &ialloc_context, &ip);
1067 * If we get an error at this point, return to the caller
1068 * so that the current transaction can be aborted.
1075 ASSERT(!ialloc_context && ip);
1078 if (committed != NULL)
1089 * Decrement the link count on an inode & log the change. If this causes the
1090 * link count to go to zero, move the inode to AGI unlinked list so that it can
1091 * be freed when the last active reference goes away via xfs_inactive().
1093 static int /* error */
1098 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1100 drop_nlink(VFS_I(ip));
1101 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1103 if (VFS_I(ip)->i_nlink)
1106 return xfs_iunlink(tp, ip);
1110 * Increment the link count on an inode & log the change.
1117 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1119 ASSERT(ip->i_d.di_version > 1);
1120 inc_nlink(VFS_I(ip));
1121 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1128 struct xfs_name *name,
1133 int is_dir = S_ISDIR(mode);
1134 struct xfs_mount *mp = dp->i_mount;
1135 struct xfs_inode *ip = NULL;
1136 struct xfs_trans *tp = NULL;
1138 struct xfs_defer_ops dfops;
1139 xfs_fsblock_t first_block;
1140 bool unlock_dp_on_error = false;
1142 struct xfs_dquot *udqp = NULL;
1143 struct xfs_dquot *gdqp = NULL;
1144 struct xfs_dquot *pdqp = NULL;
1145 struct xfs_trans_res *tres;
1148 trace_xfs_create(dp, name);
1150 if (XFS_FORCED_SHUTDOWN(mp))
1153 prid = xfs_get_initial_prid(dp);
1156 * Make sure that we have allocated dquot(s) on disk.
1158 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1159 xfs_kgid_to_gid(current_fsgid()), prid,
1160 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1161 &udqp, &gdqp, &pdqp);
1166 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1167 tres = &M_RES(mp)->tr_mkdir;
1169 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1170 tres = &M_RES(mp)->tr_create;
1174 * Initially assume that the file does not exist and
1175 * reserve the resources for that case. If that is not
1176 * the case we'll drop the one we have and get a more
1177 * appropriate transaction later.
1179 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1180 if (error == -ENOSPC) {
1181 /* flush outstanding delalloc blocks and retry */
1182 xfs_flush_inodes(mp);
1183 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1185 if (error == -ENOSPC) {
1186 /* No space at all so try a "no-allocation" reservation */
1188 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1191 goto out_release_inode;
1193 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1194 unlock_dp_on_error = true;
1196 xfs_defer_init(&dfops, &first_block);
1199 * Reserve disk quota and the inode.
1201 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1202 pdqp, resblks, 1, 0);
1204 goto out_trans_cancel;
1207 error = xfs_dir_canenter(tp, dp, name);
1209 goto out_trans_cancel;
1213 * A newly created regular or special file just has one directory
1214 * entry pointing to them, but a directory also the "." entry
1215 * pointing to itself.
1217 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1218 prid, resblks > 0, &ip, NULL);
1220 goto out_trans_cancel;
1223 * Now we join the directory inode to the transaction. We do not do it
1224 * earlier because xfs_dir_ialloc might commit the previous transaction
1225 * (and release all the locks). An error from here on will result in
1226 * the transaction cancel unlocking dp so don't do it explicitly in the
1229 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1230 unlock_dp_on_error = false;
1232 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1233 &first_block, &dfops, resblks ?
1234 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1236 ASSERT(error != -ENOSPC);
1237 goto out_trans_cancel;
1239 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1240 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1243 error = xfs_dir_init(tp, ip, dp);
1245 goto out_bmap_cancel;
1247 error = xfs_bumplink(tp, dp);
1249 goto out_bmap_cancel;
1253 * If this is a synchronous mount, make sure that the
1254 * create transaction goes to disk before returning to
1257 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1258 xfs_trans_set_sync(tp);
1261 * Attach the dquot(s) to the inodes and modify them incore.
1262 * These ids of the inode couldn't have changed since the new
1263 * inode has been locked ever since it was created.
1265 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1267 error = xfs_defer_finish(&tp, &dfops);
1269 goto out_bmap_cancel;
1271 error = xfs_trans_commit(tp);
1273 goto out_release_inode;
1275 xfs_qm_dqrele(udqp);
1276 xfs_qm_dqrele(gdqp);
1277 xfs_qm_dqrele(pdqp);
1283 xfs_defer_cancel(&dfops);
1285 xfs_trans_cancel(tp);
1288 * Wait until after the current transaction is aborted to finish the
1289 * setup of the inode and release the inode. This prevents recursive
1290 * transactions and deadlocks from xfs_inactive.
1293 xfs_finish_inode_setup(ip);
1297 xfs_qm_dqrele(udqp);
1298 xfs_qm_dqrele(gdqp);
1299 xfs_qm_dqrele(pdqp);
1301 if (unlock_dp_on_error)
1302 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1308 struct xfs_inode *dp,
1309 struct dentry *dentry,
1311 struct xfs_inode **ipp)
1313 struct xfs_mount *mp = dp->i_mount;
1314 struct xfs_inode *ip = NULL;
1315 struct xfs_trans *tp = NULL;
1318 struct xfs_dquot *udqp = NULL;
1319 struct xfs_dquot *gdqp = NULL;
1320 struct xfs_dquot *pdqp = NULL;
1321 struct xfs_trans_res *tres;
1324 if (XFS_FORCED_SHUTDOWN(mp))
1327 prid = xfs_get_initial_prid(dp);
1330 * Make sure that we have allocated dquot(s) on disk.
1332 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1333 xfs_kgid_to_gid(current_fsgid()), prid,
1334 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1335 &udqp, &gdqp, &pdqp);
1339 resblks = XFS_IALLOC_SPACE_RES(mp);
1340 tres = &M_RES(mp)->tr_create_tmpfile;
1342 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1343 if (error == -ENOSPC) {
1344 /* No space at all so try a "no-allocation" reservation */
1346 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1349 goto out_release_inode;
1351 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1352 pdqp, resblks, 1, 0);
1354 goto out_trans_cancel;
1356 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1357 prid, resblks > 0, &ip, NULL);
1359 goto out_trans_cancel;
1361 if (mp->m_flags & XFS_MOUNT_WSYNC)
1362 xfs_trans_set_sync(tp);
1365 * Attach the dquot(s) to the inodes and modify them incore.
1366 * These ids of the inode couldn't have changed since the new
1367 * inode has been locked ever since it was created.
1369 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1371 error = xfs_iunlink(tp, ip);
1373 goto out_trans_cancel;
1375 error = xfs_trans_commit(tp);
1377 goto out_release_inode;
1379 xfs_qm_dqrele(udqp);
1380 xfs_qm_dqrele(gdqp);
1381 xfs_qm_dqrele(pdqp);
1387 xfs_trans_cancel(tp);
1390 * Wait until after the current transaction is aborted to finish the
1391 * setup of the inode and release the inode. This prevents recursive
1392 * transactions and deadlocks from xfs_inactive.
1395 xfs_finish_inode_setup(ip);
1399 xfs_qm_dqrele(udqp);
1400 xfs_qm_dqrele(gdqp);
1401 xfs_qm_dqrele(pdqp);
1410 struct xfs_name *target_name)
1412 xfs_mount_t *mp = tdp->i_mount;
1415 struct xfs_defer_ops dfops;
1416 xfs_fsblock_t first_block;
1419 trace_xfs_link(tdp, target_name);
1421 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1423 if (XFS_FORCED_SHUTDOWN(mp))
1426 error = xfs_qm_dqattach(sip, 0);
1430 error = xfs_qm_dqattach(tdp, 0);
1434 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1435 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1436 if (error == -ENOSPC) {
1438 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1443 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1445 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1446 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1449 * If we are using project inheritance, we only allow hard link
1450 * creation in our tree when the project IDs are the same; else
1451 * the tree quota mechanism could be circumvented.
1453 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1454 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1460 error = xfs_dir_canenter(tp, tdp, target_name);
1465 xfs_defer_init(&dfops, &first_block);
1468 * Handle initial link state of O_TMPFILE inode
1470 if (VFS_I(sip)->i_nlink == 0) {
1471 error = xfs_iunlink_remove(tp, sip);
1476 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1477 &first_block, &dfops, resblks);
1480 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1481 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1483 error = xfs_bumplink(tp, sip);
1488 * If this is a synchronous mount, make sure that the
1489 * link transaction goes to disk before returning to
1492 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1493 xfs_trans_set_sync(tp);
1495 error = xfs_defer_finish(&tp, &dfops);
1497 xfs_defer_cancel(&dfops);
1501 return xfs_trans_commit(tp);
1504 xfs_trans_cancel(tp);
1510 * Free up the underlying blocks past new_size. The new size must be smaller
1511 * than the current size. This routine can be used both for the attribute and
1512 * data fork, and does not modify the inode size, which is left to the caller.
1514 * The transaction passed to this routine must have made a permanent log
1515 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1516 * given transaction and start new ones, so make sure everything involved in
1517 * the transaction is tidy before calling here. Some transaction will be
1518 * returned to the caller to be committed. The incoming transaction must
1519 * already include the inode, and both inode locks must be held exclusively.
1520 * The inode must also be "held" within the transaction. On return the inode
1521 * will be "held" within the returned transaction. This routine does NOT
1522 * require any disk space to be reserved for it within the transaction.
1524 * If we get an error, we must return with the inode locked and linked into the
1525 * current transaction. This keeps things simple for the higher level code,
1526 * because it always knows that the inode is locked and held in the transaction
1527 * that returns to it whether errors occur or not. We don't mark the inode
1528 * dirty on error so that transactions can be easily aborted if possible.
1531 xfs_itruncate_extents(
1532 struct xfs_trans **tpp,
1533 struct xfs_inode *ip,
1535 xfs_fsize_t new_size)
1537 struct xfs_mount *mp = ip->i_mount;
1538 struct xfs_trans *tp = *tpp;
1539 struct xfs_defer_ops dfops;
1540 xfs_fsblock_t first_block;
1541 xfs_fileoff_t first_unmap_block;
1542 xfs_fileoff_t last_block;
1543 xfs_filblks_t unmap_len;
1547 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1548 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1549 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1550 ASSERT(new_size <= XFS_ISIZE(ip));
1551 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1552 ASSERT(ip->i_itemp != NULL);
1553 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1554 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1556 trace_xfs_itruncate_extents_start(ip, new_size);
1559 * Since it is possible for space to become allocated beyond
1560 * the end of the file (in a crash where the space is allocated
1561 * but the inode size is not yet updated), simply remove any
1562 * blocks which show up between the new EOF and the maximum
1563 * possible file size. If the first block to be removed is
1564 * beyond the maximum file size (ie it is the same as last_block),
1565 * then there is nothing to do.
1567 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1568 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1569 if (first_unmap_block == last_block)
1572 ASSERT(first_unmap_block < last_block);
1573 unmap_len = last_block - first_unmap_block + 1;
1575 xfs_defer_init(&dfops, &first_block);
1576 error = xfs_bunmapi(tp, ip,
1577 first_unmap_block, unmap_len,
1578 xfs_bmapi_aflag(whichfork),
1579 XFS_ITRUNC_MAX_EXTENTS,
1580 &first_block, &dfops,
1583 goto out_bmap_cancel;
1586 * Duplicate the transaction that has the permanent
1587 * reservation and commit the old transaction.
1589 xfs_defer_ijoin(&dfops, ip);
1590 error = xfs_defer_finish(&tp, &dfops);
1592 goto out_bmap_cancel;
1594 error = xfs_trans_roll_inode(&tp, ip);
1599 /* Remove all pending CoW reservations. */
1600 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1606 * Clear the reflink flag if there are no data fork blocks and
1607 * there are no extents staged in the cow fork.
1609 if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) {
1610 if (ip->i_d.di_nblocks == 0)
1611 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1612 xfs_inode_clear_cowblocks_tag(ip);
1616 * Always re-log the inode so that our permanent transaction can keep
1617 * on rolling it forward in the log.
1619 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1621 trace_xfs_itruncate_extents_end(ip, new_size);
1628 * If the bunmapi call encounters an error, return to the caller where
1629 * the transaction can be properly aborted. We just need to make sure
1630 * we're not holding any resources that we were not when we came in.
1632 xfs_defer_cancel(&dfops);
1640 xfs_mount_t *mp = ip->i_mount;
1643 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1646 /* If this is a read-only mount, don't do this (would generate I/O) */
1647 if (mp->m_flags & XFS_MOUNT_RDONLY)
1650 if (!XFS_FORCED_SHUTDOWN(mp)) {
1654 * If we previously truncated this file and removed old data
1655 * in the process, we want to initiate "early" writeout on
1656 * the last close. This is an attempt to combat the notorious
1657 * NULL files problem which is particularly noticeable from a
1658 * truncate down, buffered (re-)write (delalloc), followed by
1659 * a crash. What we are effectively doing here is
1660 * significantly reducing the time window where we'd otherwise
1661 * be exposed to that problem.
1663 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1665 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1666 if (ip->i_delayed_blks > 0) {
1667 error = filemap_flush(VFS_I(ip)->i_mapping);
1674 if (VFS_I(ip)->i_nlink == 0)
1677 if (xfs_can_free_eofblocks(ip, false)) {
1680 * Check if the inode is being opened, written and closed
1681 * frequently and we have delayed allocation blocks outstanding
1682 * (e.g. streaming writes from the NFS server), truncating the
1683 * blocks past EOF will cause fragmentation to occur.
1685 * In this case don't do the truncation, but we have to be
1686 * careful how we detect this case. Blocks beyond EOF show up as
1687 * i_delayed_blks even when the inode is clean, so we need to
1688 * truncate them away first before checking for a dirty release.
1689 * Hence on the first dirty close we will still remove the
1690 * speculative allocation, but after that we will leave it in
1693 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1696 * If we can't get the iolock just skip truncating the blocks
1697 * past EOF because we could deadlock with the mmap_sem
1698 * otherwise. We'll get another chance to drop them once the
1699 * last reference to the inode is dropped, so we'll never leak
1700 * blocks permanently.
1702 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1703 error = xfs_free_eofblocks(ip);
1704 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1709 /* delalloc blocks after truncation means it really is dirty */
1710 if (ip->i_delayed_blks)
1711 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1717 * xfs_inactive_truncate
1719 * Called to perform a truncate when an inode becomes unlinked.
1722 xfs_inactive_truncate(
1723 struct xfs_inode *ip)
1725 struct xfs_mount *mp = ip->i_mount;
1726 struct xfs_trans *tp;
1729 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1731 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1735 xfs_ilock(ip, XFS_ILOCK_EXCL);
1736 xfs_trans_ijoin(tp, ip, 0);
1739 * Log the inode size first to prevent stale data exposure in the event
1740 * of a system crash before the truncate completes. See the related
1741 * comment in xfs_vn_setattr_size() for details.
1743 ip->i_d.di_size = 0;
1744 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1746 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1748 goto error_trans_cancel;
1750 ASSERT(ip->i_d.di_nextents == 0);
1752 error = xfs_trans_commit(tp);
1756 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1760 xfs_trans_cancel(tp);
1762 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1767 * xfs_inactive_ifree()
1769 * Perform the inode free when an inode is unlinked.
1773 struct xfs_inode *ip)
1775 struct xfs_defer_ops dfops;
1776 xfs_fsblock_t first_block;
1777 struct xfs_mount *mp = ip->i_mount;
1778 struct xfs_trans *tp;
1782 * We try to use a per-AG reservation for any block needed by the finobt
1783 * tree, but as the finobt feature predates the per-AG reservation
1784 * support a degraded file system might not have enough space for the
1785 * reservation at mount time. In that case try to dip into the reserved
1788 * Send a warning if the reservation does happen to fail, as the inode
1789 * now remains allocated and sits on the unlinked list until the fs is
1792 if (unlikely(mp->m_inotbt_nores)) {
1793 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1794 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1797 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1800 if (error == -ENOSPC) {
1801 xfs_warn_ratelimited(mp,
1802 "Failed to remove inode(s) from unlinked list. "
1803 "Please free space, unmount and run xfs_repair.");
1805 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1810 xfs_ilock(ip, XFS_ILOCK_EXCL);
1811 xfs_trans_ijoin(tp, ip, 0);
1813 xfs_defer_init(&dfops, &first_block);
1814 error = xfs_ifree(tp, ip, &dfops);
1817 * If we fail to free the inode, shut down. The cancel
1818 * might do that, we need to make sure. Otherwise the
1819 * inode might be lost for a long time or forever.
1821 if (!XFS_FORCED_SHUTDOWN(mp)) {
1822 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1824 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1826 xfs_trans_cancel(tp);
1827 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1832 * Credit the quota account(s). The inode is gone.
1834 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1837 * Just ignore errors at this point. There is nothing we can do except
1838 * to try to keep going. Make sure it's not a silent error.
1840 error = xfs_defer_finish(&tp, &dfops);
1842 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1844 xfs_defer_cancel(&dfops);
1846 error = xfs_trans_commit(tp);
1848 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1851 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1858 * This is called when the vnode reference count for the vnode
1859 * goes to zero. If the file has been unlinked, then it must
1860 * now be truncated. Also, we clear all of the read-ahead state
1861 * kept for the inode here since the file is now closed.
1867 struct xfs_mount *mp;
1872 * If the inode is already free, then there can be nothing
1875 if (VFS_I(ip)->i_mode == 0) {
1876 ASSERT(ip->i_df.if_real_bytes == 0);
1877 ASSERT(ip->i_df.if_broot_bytes == 0);
1882 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1884 /* If this is a read-only mount, don't do this (would generate I/O) */
1885 if (mp->m_flags & XFS_MOUNT_RDONLY)
1888 if (VFS_I(ip)->i_nlink != 0) {
1890 * force is true because we are evicting an inode from the
1891 * cache. Post-eof blocks must be freed, lest we end up with
1892 * broken free space accounting.
1894 * Note: don't bother with iolock here since lockdep complains
1895 * about acquiring it in reclaim context. We have the only
1896 * reference to the inode at this point anyways.
1898 if (xfs_can_free_eofblocks(ip, true))
1899 xfs_free_eofblocks(ip);
1904 if (S_ISREG(VFS_I(ip)->i_mode) &&
1905 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1906 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1909 error = xfs_qm_dqattach(ip, 0);
1913 if (S_ISLNK(VFS_I(ip)->i_mode))
1914 error = xfs_inactive_symlink(ip);
1916 error = xfs_inactive_truncate(ip);
1921 * If there are attributes associated with the file then blow them away
1922 * now. The code calls a routine that recursively deconstructs the
1923 * attribute fork. If also blows away the in-core attribute fork.
1925 if (XFS_IFORK_Q(ip)) {
1926 error = xfs_attr_inactive(ip);
1932 ASSERT(ip->i_d.di_anextents == 0);
1933 ASSERT(ip->i_d.di_forkoff == 0);
1938 error = xfs_inactive_ifree(ip);
1943 * Release the dquots held by inode, if any.
1945 xfs_qm_dqdetach(ip);
1949 * This is called when the inode's link count goes to 0 or we are creating a
1950 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1951 * set to true as the link count is dropped to zero by the VFS after we've
1952 * created the file successfully, so we have to add it to the unlinked list
1953 * while the link count is non-zero.
1955 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1956 * list when the inode is freed.
1960 struct xfs_trans *tp,
1961 struct xfs_inode *ip)
1963 xfs_mount_t *mp = tp->t_mountp;
1973 ASSERT(VFS_I(ip)->i_mode != 0);
1976 * Get the agi buffer first. It ensures lock ordering
1979 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1982 agi = XFS_BUF_TO_AGI(agibp);
1985 * Get the index into the agi hash table for the
1986 * list this inode will go on.
1988 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1990 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1991 ASSERT(agi->agi_unlinked[bucket_index]);
1992 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1994 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1996 * There is already another inode in the bucket we need
1997 * to add ourselves to. Add us at the front of the list.
1998 * Here we put the head pointer into our next pointer,
1999 * and then we fall through to point the head at us.
2001 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2006 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2007 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2008 offset = ip->i_imap.im_boffset +
2009 offsetof(xfs_dinode_t, di_next_unlinked);
2011 /* need to recalc the inode CRC if appropriate */
2012 xfs_dinode_calc_crc(mp, dip);
2014 xfs_trans_inode_buf(tp, ibp);
2015 xfs_trans_log_buf(tp, ibp, offset,
2016 (offset + sizeof(xfs_agino_t) - 1));
2017 xfs_inobp_check(mp, ibp);
2021 * Point the bucket head pointer at the inode being inserted.
2024 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2025 offset = offsetof(xfs_agi_t, agi_unlinked) +
2026 (sizeof(xfs_agino_t) * bucket_index);
2027 xfs_trans_log_buf(tp, agibp, offset,
2028 (offset + sizeof(xfs_agino_t) - 1));
2033 * Pull the on-disk inode from the AGI unlinked list.
2046 xfs_agnumber_t agno;
2048 xfs_agino_t next_agino;
2049 xfs_buf_t *last_ibp;
2050 xfs_dinode_t *last_dip = NULL;
2052 int offset, last_offset = 0;
2056 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2059 * Get the agi buffer first. It ensures lock ordering
2062 error = xfs_read_agi(mp, tp, agno, &agibp);
2066 agi = XFS_BUF_TO_AGI(agibp);
2069 * Get the index into the agi hash table for the
2070 * list this inode will go on.
2072 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2074 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2075 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2076 ASSERT(agi->agi_unlinked[bucket_index]);
2078 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2080 * We're at the head of the list. Get the inode's on-disk
2081 * buffer to see if there is anyone after us on the list.
2082 * Only modify our next pointer if it is not already NULLAGINO.
2083 * This saves us the overhead of dealing with the buffer when
2084 * there is no need to change it.
2086 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2089 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2093 next_agino = be32_to_cpu(dip->di_next_unlinked);
2094 ASSERT(next_agino != 0);
2095 if (next_agino != NULLAGINO) {
2096 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2097 offset = ip->i_imap.im_boffset +
2098 offsetof(xfs_dinode_t, di_next_unlinked);
2100 /* need to recalc the inode CRC if appropriate */
2101 xfs_dinode_calc_crc(mp, dip);
2103 xfs_trans_inode_buf(tp, ibp);
2104 xfs_trans_log_buf(tp, ibp, offset,
2105 (offset + sizeof(xfs_agino_t) - 1));
2106 xfs_inobp_check(mp, ibp);
2108 xfs_trans_brelse(tp, ibp);
2111 * Point the bucket head pointer at the next inode.
2113 ASSERT(next_agino != 0);
2114 ASSERT(next_agino != agino);
2115 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2116 offset = offsetof(xfs_agi_t, agi_unlinked) +
2117 (sizeof(xfs_agino_t) * bucket_index);
2118 xfs_trans_log_buf(tp, agibp, offset,
2119 (offset + sizeof(xfs_agino_t) - 1));
2122 * We need to search the list for the inode being freed.
2124 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2126 while (next_agino != agino) {
2127 struct xfs_imap imap;
2130 xfs_trans_brelse(tp, last_ibp);
2133 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2135 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2138 "%s: xfs_imap returned error %d.",
2143 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2147 "%s: xfs_imap_to_bp returned error %d.",
2152 last_offset = imap.im_boffset;
2153 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2154 ASSERT(next_agino != NULLAGINO);
2155 ASSERT(next_agino != 0);
2159 * Now last_ibp points to the buffer previous to us on the
2160 * unlinked list. Pull us from the list.
2162 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2165 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2169 next_agino = be32_to_cpu(dip->di_next_unlinked);
2170 ASSERT(next_agino != 0);
2171 ASSERT(next_agino != agino);
2172 if (next_agino != NULLAGINO) {
2173 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2174 offset = ip->i_imap.im_boffset +
2175 offsetof(xfs_dinode_t, di_next_unlinked);
2177 /* need to recalc the inode CRC if appropriate */
2178 xfs_dinode_calc_crc(mp, dip);
2180 xfs_trans_inode_buf(tp, ibp);
2181 xfs_trans_log_buf(tp, ibp, offset,
2182 (offset + sizeof(xfs_agino_t) - 1));
2183 xfs_inobp_check(mp, ibp);
2185 xfs_trans_brelse(tp, ibp);
2188 * Point the previous inode on the list to the next inode.
2190 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2191 ASSERT(next_agino != 0);
2192 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2194 /* need to recalc the inode CRC if appropriate */
2195 xfs_dinode_calc_crc(mp, last_dip);
2197 xfs_trans_inode_buf(tp, last_ibp);
2198 xfs_trans_log_buf(tp, last_ibp, offset,
2199 (offset + sizeof(xfs_agino_t) - 1));
2200 xfs_inobp_check(mp, last_ibp);
2206 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2207 * inodes that are in memory - they all must be marked stale and attached to
2208 * the cluster buffer.
2212 xfs_inode_t *free_ip,
2214 struct xfs_icluster *xic)
2216 xfs_mount_t *mp = free_ip->i_mount;
2217 int blks_per_cluster;
2218 int inodes_per_cluster;
2225 xfs_inode_log_item_t *iip;
2226 xfs_log_item_t *lip;
2227 struct xfs_perag *pag;
2230 inum = xic->first_ino;
2231 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2232 blks_per_cluster = xfs_icluster_size_fsb(mp);
2233 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2234 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2236 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2238 * The allocation bitmap tells us which inodes of the chunk were
2239 * physically allocated. Skip the cluster if an inode falls into
2242 ioffset = inum - xic->first_ino;
2243 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2244 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2248 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2249 XFS_INO_TO_AGBNO(mp, inum));
2252 * We obtain and lock the backing buffer first in the process
2253 * here, as we have to ensure that any dirty inode that we
2254 * can't get the flush lock on is attached to the buffer.
2255 * If we scan the in-memory inodes first, then buffer IO can
2256 * complete before we get a lock on it, and hence we may fail
2257 * to mark all the active inodes on the buffer stale.
2259 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2260 mp->m_bsize * blks_per_cluster,
2267 * This buffer may not have been correctly initialised as we
2268 * didn't read it from disk. That's not important because we are
2269 * only using to mark the buffer as stale in the log, and to
2270 * attach stale cached inodes on it. That means it will never be
2271 * dispatched for IO. If it is, we want to know about it, and we
2272 * want it to fail. We can acheive this by adding a write
2273 * verifier to the buffer.
2275 bp->b_ops = &xfs_inode_buf_ops;
2278 * Walk the inodes already attached to the buffer and mark them
2279 * stale. These will all have the flush locks held, so an
2280 * in-memory inode walk can't lock them. By marking them all
2281 * stale first, we will not attempt to lock them in the loop
2282 * below as the XFS_ISTALE flag will be set.
2286 if (lip->li_type == XFS_LI_INODE) {
2287 iip = (xfs_inode_log_item_t *)lip;
2288 ASSERT(iip->ili_logged == 1);
2289 lip->li_cb = xfs_istale_done;
2290 xfs_trans_ail_copy_lsn(mp->m_ail,
2291 &iip->ili_flush_lsn,
2292 &iip->ili_item.li_lsn);
2293 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2295 lip = lip->li_bio_list;
2300 * For each inode in memory attempt to add it to the inode
2301 * buffer and set it up for being staled on buffer IO
2302 * completion. This is safe as we've locked out tail pushing
2303 * and flushing by locking the buffer.
2305 * We have already marked every inode that was part of a
2306 * transaction stale above, which means there is no point in
2307 * even trying to lock them.
2309 for (i = 0; i < inodes_per_cluster; i++) {
2312 ip = radix_tree_lookup(&pag->pag_ici_root,
2313 XFS_INO_TO_AGINO(mp, (inum + i)));
2315 /* Inode not in memory, nothing to do */
2322 * because this is an RCU protected lookup, we could
2323 * find a recently freed or even reallocated inode
2324 * during the lookup. We need to check under the
2325 * i_flags_lock for a valid inode here. Skip it if it
2326 * is not valid, the wrong inode or stale.
2328 spin_lock(&ip->i_flags_lock);
2329 if (ip->i_ino != inum + i ||
2330 __xfs_iflags_test(ip, XFS_ISTALE)) {
2331 spin_unlock(&ip->i_flags_lock);
2335 spin_unlock(&ip->i_flags_lock);
2338 * Don't try to lock/unlock the current inode, but we
2339 * _cannot_ skip the other inodes that we did not find
2340 * in the list attached to the buffer and are not
2341 * already marked stale. If we can't lock it, back off
2344 if (ip != free_ip) {
2345 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2352 * Check the inode number again in case we're
2353 * racing with freeing in xfs_reclaim_inode().
2354 * See the comments in that function for more
2355 * information as to why the initial check is
2358 if (ip->i_ino != inum + i) {
2359 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2367 xfs_iflags_set(ip, XFS_ISTALE);
2370 * we don't need to attach clean inodes or those only
2371 * with unlogged changes (which we throw away, anyway).
2374 if (!iip || xfs_inode_clean(ip)) {
2375 ASSERT(ip != free_ip);
2377 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2381 iip->ili_last_fields = iip->ili_fields;
2382 iip->ili_fields = 0;
2383 iip->ili_fsync_fields = 0;
2384 iip->ili_logged = 1;
2385 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2386 &iip->ili_item.li_lsn);
2388 xfs_buf_attach_iodone(bp, xfs_istale_done,
2392 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2395 xfs_trans_stale_inode_buf(tp, bp);
2396 xfs_trans_binval(tp, bp);
2404 * This is called to return an inode to the inode free list.
2405 * The inode should already be truncated to 0 length and have
2406 * no pages associated with it. This routine also assumes that
2407 * the inode is already a part of the transaction.
2409 * The on-disk copy of the inode will have been added to the list
2410 * of unlinked inodes in the AGI. We need to remove the inode from
2411 * that list atomically with respect to freeing it here.
2417 struct xfs_defer_ops *dfops)
2420 struct xfs_icluster xic = { 0 };
2422 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2423 ASSERT(VFS_I(ip)->i_nlink == 0);
2424 ASSERT(ip->i_d.di_nextents == 0);
2425 ASSERT(ip->i_d.di_anextents == 0);
2426 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2427 ASSERT(ip->i_d.di_nblocks == 0);
2430 * Pull the on-disk inode from the AGI unlinked list.
2432 error = xfs_iunlink_remove(tp, ip);
2436 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2440 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2441 ip->i_d.di_flags = 0;
2442 ip->i_d.di_dmevmask = 0;
2443 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2444 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2445 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2447 * Bump the generation count so no one will be confused
2448 * by reincarnations of this inode.
2450 VFS_I(ip)->i_generation++;
2451 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2454 error = xfs_ifree_cluster(ip, tp, &xic);
2460 * This is called to unpin an inode. The caller must have the inode locked
2461 * in at least shared mode so that the buffer cannot be subsequently pinned
2462 * once someone is waiting for it to be unpinned.
2466 struct xfs_inode *ip)
2468 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2470 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2472 /* Give the log a push to start the unpinning I/O */
2473 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2479 struct xfs_inode *ip)
2481 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2482 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2487 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2488 if (xfs_ipincount(ip))
2490 } while (xfs_ipincount(ip));
2491 finish_wait(wq, &wait.wq_entry);
2496 struct xfs_inode *ip)
2498 if (xfs_ipincount(ip))
2499 __xfs_iunpin_wait(ip);
2503 * Removing an inode from the namespace involves removing the directory entry
2504 * and dropping the link count on the inode. Removing the directory entry can
2505 * result in locking an AGF (directory blocks were freed) and removing a link
2506 * count can result in placing the inode on an unlinked list which results in
2509 * The big problem here is that we have an ordering constraint on AGF and AGI
2510 * locking - inode allocation locks the AGI, then can allocate a new extent for
2511 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2512 * removes the inode from the unlinked list, requiring that we lock the AGI
2513 * first, and then freeing the inode can result in an inode chunk being freed
2514 * and hence freeing disk space requiring that we lock an AGF.
2516 * Hence the ordering that is imposed by other parts of the code is AGI before
2517 * AGF. This means we cannot remove the directory entry before we drop the inode
2518 * reference count and put it on the unlinked list as this results in a lock
2519 * order of AGF then AGI, and this can deadlock against inode allocation and
2520 * freeing. Therefore we must drop the link counts before we remove the
2523 * This is still safe from a transactional point of view - it is not until we
2524 * get to xfs_defer_finish() that we have the possibility of multiple
2525 * transactions in this operation. Hence as long as we remove the directory
2526 * entry and drop the link count in the first transaction of the remove
2527 * operation, there are no transactional constraints on the ordering here.
2532 struct xfs_name *name,
2535 xfs_mount_t *mp = dp->i_mount;
2536 xfs_trans_t *tp = NULL;
2537 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2539 struct xfs_defer_ops dfops;
2540 xfs_fsblock_t first_block;
2543 trace_xfs_remove(dp, name);
2545 if (XFS_FORCED_SHUTDOWN(mp))
2548 error = xfs_qm_dqattach(dp, 0);
2552 error = xfs_qm_dqattach(ip, 0);
2557 * We try to get the real space reservation first,
2558 * allowing for directory btree deletion(s) implying
2559 * possible bmap insert(s). If we can't get the space
2560 * reservation then we use 0 instead, and avoid the bmap
2561 * btree insert(s) in the directory code by, if the bmap
2562 * insert tries to happen, instead trimming the LAST
2563 * block from the directory.
2565 resblks = XFS_REMOVE_SPACE_RES(mp);
2566 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2567 if (error == -ENOSPC) {
2569 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2573 ASSERT(error != -ENOSPC);
2577 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2579 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2580 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2583 * If we're removing a directory perform some additional validation.
2586 ASSERT(VFS_I(ip)->i_nlink >= 2);
2587 if (VFS_I(ip)->i_nlink != 2) {
2589 goto out_trans_cancel;
2591 if (!xfs_dir_isempty(ip)) {
2593 goto out_trans_cancel;
2596 /* Drop the link from ip's "..". */
2597 error = xfs_droplink(tp, dp);
2599 goto out_trans_cancel;
2601 /* Drop the "." link from ip to self. */
2602 error = xfs_droplink(tp, ip);
2604 goto out_trans_cancel;
2607 * When removing a non-directory we need to log the parent
2608 * inode here. For a directory this is done implicitly
2609 * by the xfs_droplink call for the ".." entry.
2611 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2613 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2615 /* Drop the link from dp to ip. */
2616 error = xfs_droplink(tp, ip);
2618 goto out_trans_cancel;
2620 xfs_defer_init(&dfops, &first_block);
2621 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2622 &first_block, &dfops, resblks);
2624 ASSERT(error != -ENOENT);
2625 goto out_bmap_cancel;
2629 * If this is a synchronous mount, make sure that the
2630 * remove transaction goes to disk before returning to
2633 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2634 xfs_trans_set_sync(tp);
2636 error = xfs_defer_finish(&tp, &dfops);
2638 goto out_bmap_cancel;
2640 error = xfs_trans_commit(tp);
2644 if (is_dir && xfs_inode_is_filestream(ip))
2645 xfs_filestream_deassociate(ip);
2650 xfs_defer_cancel(&dfops);
2652 xfs_trans_cancel(tp);
2658 * Enter all inodes for a rename transaction into a sorted array.
2660 #define __XFS_SORT_INODES 5
2662 xfs_sort_for_rename(
2663 struct xfs_inode *dp1, /* in: old (source) directory inode */
2664 struct xfs_inode *dp2, /* in: new (target) directory inode */
2665 struct xfs_inode *ip1, /* in: inode of old entry */
2666 struct xfs_inode *ip2, /* in: inode of new entry */
2667 struct xfs_inode *wip, /* in: whiteout inode */
2668 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2669 int *num_inodes) /* in/out: inodes in array */
2673 ASSERT(*num_inodes == __XFS_SORT_INODES);
2674 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2677 * i_tab contains a list of pointers to inodes. We initialize
2678 * the table here & we'll sort it. We will then use it to
2679 * order the acquisition of the inode locks.
2681 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2694 * Sort the elements via bubble sort. (Remember, there are at
2695 * most 5 elements to sort, so this is adequate.)
2697 for (i = 0; i < *num_inodes; i++) {
2698 for (j = 1; j < *num_inodes; j++) {
2699 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2700 struct xfs_inode *temp = i_tab[j];
2701 i_tab[j] = i_tab[j-1];
2710 struct xfs_trans *tp,
2711 struct xfs_defer_ops *dfops)
2716 * If this is a synchronous mount, make sure that the rename transaction
2717 * goes to disk before returning to the user.
2719 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2720 xfs_trans_set_sync(tp);
2722 error = xfs_defer_finish(&tp, dfops);
2724 xfs_defer_cancel(dfops);
2725 xfs_trans_cancel(tp);
2729 return xfs_trans_commit(tp);
2733 * xfs_cross_rename()
2735 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2739 struct xfs_trans *tp,
2740 struct xfs_inode *dp1,
2741 struct xfs_name *name1,
2742 struct xfs_inode *ip1,
2743 struct xfs_inode *dp2,
2744 struct xfs_name *name2,
2745 struct xfs_inode *ip2,
2746 struct xfs_defer_ops *dfops,
2747 xfs_fsblock_t *first_block,
2755 /* Swap inode number for dirent in first parent */
2756 error = xfs_dir_replace(tp, dp1, name1,
2758 first_block, dfops, spaceres);
2760 goto out_trans_abort;
2762 /* Swap inode number for dirent in second parent */
2763 error = xfs_dir_replace(tp, dp2, name2,
2765 first_block, dfops, spaceres);
2767 goto out_trans_abort;
2770 * If we're renaming one or more directories across different parents,
2771 * update the respective ".." entries (and link counts) to match the new
2775 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2777 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2778 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2779 dp1->i_ino, first_block,
2782 goto out_trans_abort;
2784 /* transfer ip2 ".." reference to dp1 */
2785 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2786 error = xfs_droplink(tp, dp2);
2788 goto out_trans_abort;
2789 error = xfs_bumplink(tp, dp1);
2791 goto out_trans_abort;
2795 * Although ip1 isn't changed here, userspace needs
2796 * to be warned about the change, so that applications
2797 * relying on it (like backup ones), will properly
2800 ip1_flags |= XFS_ICHGTIME_CHG;
2801 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2804 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2805 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2806 dp2->i_ino, first_block,
2809 goto out_trans_abort;
2811 /* transfer ip1 ".." reference to dp2 */
2812 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2813 error = xfs_droplink(tp, dp1);
2815 goto out_trans_abort;
2816 error = xfs_bumplink(tp, dp2);
2818 goto out_trans_abort;
2822 * Although ip2 isn't changed here, userspace needs
2823 * to be warned about the change, so that applications
2824 * relying on it (like backup ones), will properly
2827 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2828 ip2_flags |= XFS_ICHGTIME_CHG;
2833 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2834 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2837 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2838 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2841 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2842 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2844 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2845 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2846 return xfs_finish_rename(tp, dfops);
2849 xfs_defer_cancel(dfops);
2850 xfs_trans_cancel(tp);
2855 * xfs_rename_alloc_whiteout()
2857 * Return a referenced, unlinked, unlocked inode that that can be used as a
2858 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2859 * crash between allocating the inode and linking it into the rename transaction
2860 * recovery will free the inode and we won't leak it.
2863 xfs_rename_alloc_whiteout(
2864 struct xfs_inode *dp,
2865 struct xfs_inode **wip)
2867 struct xfs_inode *tmpfile;
2870 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2875 * Prepare the tmpfile inode as if it were created through the VFS.
2876 * Otherwise, the link increment paths will complain about nlink 0->1.
2877 * Drop the link count as done by d_tmpfile(), complete the inode setup
2878 * and flag it as linkable.
2880 drop_nlink(VFS_I(tmpfile));
2881 xfs_setup_iops(tmpfile);
2882 xfs_finish_inode_setup(tmpfile);
2883 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2894 struct xfs_inode *src_dp,
2895 struct xfs_name *src_name,
2896 struct xfs_inode *src_ip,
2897 struct xfs_inode *target_dp,
2898 struct xfs_name *target_name,
2899 struct xfs_inode *target_ip,
2902 struct xfs_mount *mp = src_dp->i_mount;
2903 struct xfs_trans *tp;
2904 struct xfs_defer_ops dfops;
2905 xfs_fsblock_t first_block;
2906 struct xfs_inode *wip = NULL; /* whiteout inode */
2907 struct xfs_inode *inodes[__XFS_SORT_INODES];
2908 int num_inodes = __XFS_SORT_INODES;
2909 bool new_parent = (src_dp != target_dp);
2910 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2914 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2916 if ((flags & RENAME_EXCHANGE) && !target_ip)
2920 * If we are doing a whiteout operation, allocate the whiteout inode
2921 * we will be placing at the target and ensure the type is set
2924 if (flags & RENAME_WHITEOUT) {
2925 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2926 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2930 /* setup target dirent info as whiteout */
2931 src_name->type = XFS_DIR3_FT_CHRDEV;
2934 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2935 inodes, &num_inodes);
2937 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2938 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2939 if (error == -ENOSPC) {
2941 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2945 goto out_release_wip;
2948 * Attach the dquots to the inodes
2950 error = xfs_qm_vop_rename_dqattach(inodes);
2952 goto out_trans_cancel;
2955 * Lock all the participating inodes. Depending upon whether
2956 * the target_name exists in the target directory, and
2957 * whether the target directory is the same as the source
2958 * directory, we can lock from 2 to 4 inodes.
2960 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2963 * Join all the inodes to the transaction. From this point on,
2964 * we can rely on either trans_commit or trans_cancel to unlock
2967 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2969 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2970 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2972 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2974 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2977 * If we are using project inheritance, we only allow renames
2978 * into our tree when the project IDs are the same; else the
2979 * tree quota mechanism would be circumvented.
2981 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2982 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2984 goto out_trans_cancel;
2987 xfs_defer_init(&dfops, &first_block);
2989 /* RENAME_EXCHANGE is unique from here on. */
2990 if (flags & RENAME_EXCHANGE)
2991 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2992 target_dp, target_name, target_ip,
2993 &dfops, &first_block, spaceres);
2996 * Set up the target.
2998 if (target_ip == NULL) {
3000 * If there's no space reservation, check the entry will
3001 * fit before actually inserting it.
3004 error = xfs_dir_canenter(tp, target_dp, target_name);
3006 goto out_trans_cancel;
3009 * If target does not exist and the rename crosses
3010 * directories, adjust the target directory link count
3011 * to account for the ".." reference from the new entry.
3013 error = xfs_dir_createname(tp, target_dp, target_name,
3014 src_ip->i_ino, &first_block,
3017 goto out_bmap_cancel;
3019 xfs_trans_ichgtime(tp, target_dp,
3020 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3022 if (new_parent && src_is_directory) {
3023 error = xfs_bumplink(tp, target_dp);
3025 goto out_bmap_cancel;
3027 } else { /* target_ip != NULL */
3029 * If target exists and it's a directory, check that both
3030 * target and source are directories and that target can be
3031 * destroyed, or that neither is a directory.
3033 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3035 * Make sure target dir is empty.
3037 if (!(xfs_dir_isempty(target_ip)) ||
3038 (VFS_I(target_ip)->i_nlink > 2)) {
3040 goto out_trans_cancel;
3045 * Link the source inode under the target name.
3046 * If the source inode is a directory and we are moving
3047 * it across directories, its ".." entry will be
3048 * inconsistent until we replace that down below.
3050 * In case there is already an entry with the same
3051 * name at the destination directory, remove it first.
3053 error = xfs_dir_replace(tp, target_dp, target_name,
3055 &first_block, &dfops, spaceres);
3057 goto out_bmap_cancel;
3059 xfs_trans_ichgtime(tp, target_dp,
3060 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3063 * Decrement the link count on the target since the target
3064 * dir no longer points to it.
3066 error = xfs_droplink(tp, target_ip);
3068 goto out_bmap_cancel;
3070 if (src_is_directory) {
3072 * Drop the link from the old "." entry.
3074 error = xfs_droplink(tp, target_ip);
3076 goto out_bmap_cancel;
3078 } /* target_ip != NULL */
3081 * Remove the source.
3083 if (new_parent && src_is_directory) {
3085 * Rewrite the ".." entry to point to the new
3088 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3090 &first_block, &dfops, spaceres);
3091 ASSERT(error != -EEXIST);
3093 goto out_bmap_cancel;
3097 * We always want to hit the ctime on the source inode.
3099 * This isn't strictly required by the standards since the source
3100 * inode isn't really being changed, but old unix file systems did
3101 * it and some incremental backup programs won't work without it.
3103 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3104 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3107 * Adjust the link count on src_dp. This is necessary when
3108 * renaming a directory, either within one parent when
3109 * the target existed, or across two parent directories.
3111 if (src_is_directory && (new_parent || target_ip != NULL)) {
3114 * Decrement link count on src_directory since the
3115 * entry that's moved no longer points to it.
3117 error = xfs_droplink(tp, src_dp);
3119 goto out_bmap_cancel;
3123 * For whiteouts, we only need to update the source dirent with the
3124 * inode number of the whiteout inode rather than removing it
3128 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3129 &first_block, &dfops, spaceres);
3131 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3132 &first_block, &dfops, spaceres);
3134 goto out_bmap_cancel;
3137 * For whiteouts, we need to bump the link count on the whiteout inode.
3138 * This means that failures all the way up to this point leave the inode
3139 * on the unlinked list and so cleanup is a simple matter of dropping
3140 * the remaining reference to it. If we fail here after bumping the link
3141 * count, we're shutting down the filesystem so we'll never see the
3142 * intermediate state on disk.
3145 ASSERT(VFS_I(wip)->i_nlink == 0);
3146 error = xfs_bumplink(tp, wip);
3148 goto out_bmap_cancel;
3149 error = xfs_iunlink_remove(tp, wip);
3151 goto out_bmap_cancel;
3152 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3155 * Now we have a real link, clear the "I'm a tmpfile" state
3156 * flag from the inode so it doesn't accidentally get misused in
3159 VFS_I(wip)->i_state &= ~I_LINKABLE;
3162 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3163 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3165 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3167 error = xfs_finish_rename(tp, &dfops);
3173 xfs_defer_cancel(&dfops);
3175 xfs_trans_cancel(tp);
3184 struct xfs_inode *ip,
3187 struct xfs_mount *mp = ip->i_mount;
3188 struct xfs_perag *pag;
3189 unsigned long first_index, mask;
3190 unsigned long inodes_per_cluster;
3192 struct xfs_inode **cilist;
3193 struct xfs_inode *cip;
3199 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3201 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3202 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3203 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3207 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3208 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3210 /* really need a gang lookup range call here */
3211 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3212 first_index, inodes_per_cluster);
3216 for (i = 0; i < nr_found; i++) {
3222 * because this is an RCU protected lookup, we could find a
3223 * recently freed or even reallocated inode during the lookup.
3224 * We need to check under the i_flags_lock for a valid inode
3225 * here. Skip it if it is not valid or the wrong inode.
3227 spin_lock(&cip->i_flags_lock);
3229 __xfs_iflags_test(cip, XFS_ISTALE)) {
3230 spin_unlock(&cip->i_flags_lock);
3235 * Once we fall off the end of the cluster, no point checking
3236 * any more inodes in the list because they will also all be
3237 * outside the cluster.
3239 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3240 spin_unlock(&cip->i_flags_lock);
3243 spin_unlock(&cip->i_flags_lock);
3246 * Do an un-protected check to see if the inode is dirty and
3247 * is a candidate for flushing. These checks will be repeated
3248 * later after the appropriate locks are acquired.
3250 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3254 * Try to get locks. If any are unavailable or it is pinned,
3255 * then this inode cannot be flushed and is skipped.
3258 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3260 if (!xfs_iflock_nowait(cip)) {
3261 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3264 if (xfs_ipincount(cip)) {
3266 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3272 * Check the inode number again, just to be certain we are not
3273 * racing with freeing in xfs_reclaim_inode(). See the comments
3274 * in that function for more information as to why the initial
3275 * check is not sufficient.
3279 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3284 * arriving here means that this inode can be flushed. First
3285 * re-check that it's dirty before flushing.
3287 if (!xfs_inode_clean(cip)) {
3289 error = xfs_iflush_int(cip, bp);
3291 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3292 goto cluster_corrupt_out;
3298 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3302 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3303 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3314 cluster_corrupt_out:
3316 * Corruption detected in the clustering loop. Invalidate the
3317 * inode buffer and shut down the filesystem.
3321 * Clean up the buffer. If it was delwri, just release it --
3322 * brelse can handle it with no problems. If not, shut down the
3323 * filesystem before releasing the buffer.
3325 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3329 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3331 if (!bufwasdelwri) {
3333 * Just like incore_relse: if we have b_iodone functions,
3334 * mark the buffer as an error and call them. Otherwise
3335 * mark it as stale and brelse.
3338 bp->b_flags &= ~XBF_DONE;
3340 xfs_buf_ioerror(bp, -EIO);
3349 * Unlocks the flush lock
3351 xfs_iflush_abort(cip, false);
3354 return -EFSCORRUPTED;
3358 * Flush dirty inode metadata into the backing buffer.
3360 * The caller must have the inode lock and the inode flush lock held. The
3361 * inode lock will still be held upon return to the caller, and the inode
3362 * flush lock will be released after the inode has reached the disk.
3364 * The caller must write out the buffer returned in *bpp and release it.
3368 struct xfs_inode *ip,
3369 struct xfs_buf **bpp)
3371 struct xfs_mount *mp = ip->i_mount;
3372 struct xfs_buf *bp = NULL;
3373 struct xfs_dinode *dip;
3376 XFS_STATS_INC(mp, xs_iflush_count);
3378 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3379 ASSERT(xfs_isiflocked(ip));
3380 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3381 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3385 xfs_iunpin_wait(ip);
3388 * For stale inodes we cannot rely on the backing buffer remaining
3389 * stale in cache for the remaining life of the stale inode and so
3390 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3391 * inodes below. We have to check this after ensuring the inode is
3392 * unpinned so that it is safe to reclaim the stale inode after the
3395 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3401 * This may have been unpinned because the filesystem is shutting
3402 * down forcibly. If that's the case we must not write this inode
3403 * to disk, because the log record didn't make it to disk.
3405 * We also have to remove the log item from the AIL in this case,
3406 * as we wait for an empty AIL as part of the unmount process.
3408 if (XFS_FORCED_SHUTDOWN(mp)) {
3414 * Get the buffer containing the on-disk inode. We are doing a try-lock
3415 * operation here, so we may get an EAGAIN error. In that case, we
3416 * simply want to return with the inode still dirty.
3418 * If we get any other error, we effectively have a corruption situation
3419 * and we cannot flush the inode, so we treat it the same as failing
3422 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3424 if (error == -EAGAIN) {
3432 * First flush out the inode that xfs_iflush was called with.
3434 error = xfs_iflush_int(ip, bp);
3439 * If the buffer is pinned then push on the log now so we won't
3440 * get stuck waiting in the write for too long.
3442 if (xfs_buf_ispinned(bp))
3443 xfs_log_force(mp, 0);
3447 * see if other inodes can be gathered into this write
3449 error = xfs_iflush_cluster(ip, bp);
3451 goto cluster_corrupt_out;
3459 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3460 cluster_corrupt_out:
3461 error = -EFSCORRUPTED;
3464 * Unlocks the flush lock
3466 xfs_iflush_abort(ip, false);
3472 struct xfs_inode *ip,
3475 struct xfs_inode_log_item *iip = ip->i_itemp;
3476 struct xfs_dinode *dip;
3477 struct xfs_mount *mp = ip->i_mount;
3479 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3480 ASSERT(xfs_isiflocked(ip));
3481 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3482 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3483 ASSERT(iip != NULL && iip->ili_fields != 0);
3484 ASSERT(ip->i_d.di_version > 1);
3486 /* set *dip = inode's place in the buffer */
3487 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3489 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3490 mp, XFS_ERRTAG_IFLUSH_1)) {
3491 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3492 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3493 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3496 if (S_ISREG(VFS_I(ip)->i_mode)) {
3498 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3499 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3500 mp, XFS_ERRTAG_IFLUSH_3)) {
3501 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3502 "%s: Bad regular inode %Lu, ptr 0x%p",
3503 __func__, ip->i_ino, ip);
3506 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3508 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3509 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3510 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3511 mp, XFS_ERRTAG_IFLUSH_4)) {
3512 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3513 "%s: Bad directory inode %Lu, ptr 0x%p",
3514 __func__, ip->i_ino, ip);
3518 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3519 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3520 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3521 "%s: detected corrupt incore inode %Lu, "
3522 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3523 __func__, ip->i_ino,
3524 ip->i_d.di_nextents + ip->i_d.di_anextents,
3525 ip->i_d.di_nblocks, ip);
3528 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3529 mp, XFS_ERRTAG_IFLUSH_6)) {
3530 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3531 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3532 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3537 * Inode item log recovery for v2 inodes are dependent on the
3538 * di_flushiter count for correct sequencing. We bump the flush
3539 * iteration count so we can detect flushes which postdate a log record
3540 * during recovery. This is redundant as we now log every change and
3541 * hence this can't happen but we need to still do it to ensure
3542 * backwards compatibility with old kernels that predate logging all
3545 if (ip->i_d.di_version < 3)
3546 ip->i_d.di_flushiter++;
3548 /* Check the inline directory data. */
3549 if (S_ISDIR(VFS_I(ip)->i_mode) &&
3550 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3551 xfs_dir2_sf_verify(ip))
3555 * Copy the dirty parts of the inode into the on-disk inode. We always
3556 * copy out the core of the inode, because if the inode is dirty at all
3559 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3561 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3562 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3563 ip->i_d.di_flushiter = 0;
3565 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3566 if (XFS_IFORK_Q(ip))
3567 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3568 xfs_inobp_check(mp, bp);
3571 * We've recorded everything logged in the inode, so we'd like to clear
3572 * the ili_fields bits so we don't log and flush things unnecessarily.
3573 * However, we can't stop logging all this information until the data
3574 * we've copied into the disk buffer is written to disk. If we did we
3575 * might overwrite the copy of the inode in the log with all the data
3576 * after re-logging only part of it, and in the face of a crash we
3577 * wouldn't have all the data we need to recover.
3579 * What we do is move the bits to the ili_last_fields field. When
3580 * logging the inode, these bits are moved back to the ili_fields field.
3581 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3582 * know that the information those bits represent is permanently on
3583 * disk. As long as the flush completes before the inode is logged
3584 * again, then both ili_fields and ili_last_fields will be cleared.
3586 * We can play with the ili_fields bits here, because the inode lock
3587 * must be held exclusively in order to set bits there and the flush
3588 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3589 * done routine can tell whether or not to look in the AIL. Also, store
3590 * the current LSN of the inode so that we can tell whether the item has
3591 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3592 * need the AIL lock, because it is a 64 bit value that cannot be read
3595 iip->ili_last_fields = iip->ili_fields;
3596 iip->ili_fields = 0;
3597 iip->ili_fsync_fields = 0;
3598 iip->ili_logged = 1;
3600 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3601 &iip->ili_item.li_lsn);
3604 * Attach the function xfs_iflush_done to the inode's
3605 * buffer. This will remove the inode from the AIL
3606 * and unlock the inode's flush lock when the inode is
3607 * completely written to disk.
3609 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3611 /* generate the checksum. */
3612 xfs_dinode_calc_crc(mp, dip);
3614 ASSERT(bp->b_fspriv != NULL);
3615 ASSERT(bp->b_iodone != NULL);
3619 return -EFSCORRUPTED;