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_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
53 #include "xfs_dir2_priv.h"
55 kmem_zone_t *xfs_inode_zone;
58 * Used in xfs_itruncate_extents(). This is the maximum number of extents
59 * freed from a file in a single transaction.
61 #define XFS_ITRUNC_MAX_EXTENTS 2
63 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
64 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
65 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
68 * helper function to extract extent size hint from inode
74 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
75 return ip->i_d.di_extsize;
76 if (XFS_IS_REALTIME_INODE(ip))
77 return ip->i_mount->m_sb.sb_rextsize;
82 * Helper function to extract CoW extent size hint from inode.
83 * Between the extent size hint and the CoW extent size hint, we
84 * return the greater of the two. If the value is zero (automatic),
85 * use the default size.
88 xfs_get_cowextsz_hint(
94 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
95 a = ip->i_d.di_cowextsize;
96 b = xfs_get_extsz_hint(ip);
100 return XFS_DEFAULT_COWEXTSZ_HINT;
105 * These two are wrapper routines around the xfs_ilock() routine used to
106 * centralize some grungy code. They are used in places that wish to lock the
107 * inode solely for reading the extents. The reason these places can't just
108 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
109 * bringing in of the extents from disk for a file in b-tree format. If the
110 * inode is in b-tree format, then we need to lock the inode exclusively until
111 * the extents are read in. Locking it exclusively all the time would limit
112 * our parallelism unnecessarily, though. What we do instead is check to see
113 * if the extents have been read in yet, and only lock the inode exclusively
116 * The functions return a value which should be given to the corresponding
117 * xfs_iunlock() call.
120 xfs_ilock_data_map_shared(
121 struct xfs_inode *ip)
123 uint lock_mode = XFS_ILOCK_SHARED;
125 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
126 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
127 lock_mode = XFS_ILOCK_EXCL;
128 xfs_ilock(ip, lock_mode);
133 xfs_ilock_attr_map_shared(
134 struct xfs_inode *ip)
136 uint lock_mode = XFS_ILOCK_SHARED;
138 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
139 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
140 lock_mode = XFS_ILOCK_EXCL;
141 xfs_ilock(ip, lock_mode);
146 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
147 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
148 * various combinations of the locks to be obtained.
150 * The 3 locks should always be ordered so that the IO lock is obtained first,
151 * the mmap lock second and the ilock last in order to prevent deadlock.
153 * Basic locking order:
155 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
157 * mmap_sem locking order:
159 * i_rwsem -> page lock -> mmap_sem
160 * mmap_sem -> i_mmap_lock -> page_lock
162 * The difference in mmap_sem locking order mean that we cannot hold the
163 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
164 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
165 * in get_user_pages() to map the user pages into the kernel address space for
166 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
167 * page faults already hold the mmap_sem.
169 * Hence to serialise fully against both syscall and mmap based IO, we need to
170 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
171 * taken in places where we need to invalidate the page cache in a race
172 * free manner (e.g. truncate, hole punch and other extent manipulation
180 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
183 * You can't set both SHARED and EXCL for the same lock,
184 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
185 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
187 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
188 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
189 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
190 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
191 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
192 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
193 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
195 if (lock_flags & XFS_IOLOCK_EXCL) {
196 down_write_nested(&VFS_I(ip)->i_rwsem,
197 XFS_IOLOCK_DEP(lock_flags));
198 } else if (lock_flags & XFS_IOLOCK_SHARED) {
199 down_read_nested(&VFS_I(ip)->i_rwsem,
200 XFS_IOLOCK_DEP(lock_flags));
203 if (lock_flags & XFS_MMAPLOCK_EXCL)
204 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
205 else if (lock_flags & XFS_MMAPLOCK_SHARED)
206 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
208 if (lock_flags & XFS_ILOCK_EXCL)
209 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
210 else if (lock_flags & XFS_ILOCK_SHARED)
211 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
215 * This is just like xfs_ilock(), except that the caller
216 * is guaranteed not to sleep. It returns 1 if it gets
217 * the requested locks and 0 otherwise. If the IO lock is
218 * obtained but the inode lock cannot be, then the IO lock
219 * is dropped before returning.
221 * ip -- the inode being locked
222 * lock_flags -- this parameter indicates the inode's locks to be
223 * to be locked. See the comment for xfs_ilock() for a list
231 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
234 * You can't set both SHARED and EXCL for the same lock,
235 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
236 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
238 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
239 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
240 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
241 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
242 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
243 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
244 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
246 if (lock_flags & XFS_IOLOCK_EXCL) {
247 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
249 } else if (lock_flags & XFS_IOLOCK_SHARED) {
250 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
254 if (lock_flags & XFS_MMAPLOCK_EXCL) {
255 if (!mrtryupdate(&ip->i_mmaplock))
256 goto out_undo_iolock;
257 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
258 if (!mrtryaccess(&ip->i_mmaplock))
259 goto out_undo_iolock;
262 if (lock_flags & XFS_ILOCK_EXCL) {
263 if (!mrtryupdate(&ip->i_lock))
264 goto out_undo_mmaplock;
265 } else if (lock_flags & XFS_ILOCK_SHARED) {
266 if (!mrtryaccess(&ip->i_lock))
267 goto out_undo_mmaplock;
272 if (lock_flags & XFS_MMAPLOCK_EXCL)
273 mrunlock_excl(&ip->i_mmaplock);
274 else if (lock_flags & XFS_MMAPLOCK_SHARED)
275 mrunlock_shared(&ip->i_mmaplock);
277 if (lock_flags & XFS_IOLOCK_EXCL)
278 up_write(&VFS_I(ip)->i_rwsem);
279 else if (lock_flags & XFS_IOLOCK_SHARED)
280 up_read(&VFS_I(ip)->i_rwsem);
286 * xfs_iunlock() is used to drop the inode locks acquired with
287 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
288 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
289 * that we know which locks to drop.
291 * ip -- the inode being unlocked
292 * lock_flags -- this parameter indicates the inode's locks to be
293 * to be unlocked. See the comment for xfs_ilock() for a list
294 * of valid values for this parameter.
303 * You can't set both SHARED and EXCL for the same lock,
304 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
305 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
307 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
308 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
309 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
310 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
311 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
312 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
313 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
314 ASSERT(lock_flags != 0);
316 if (lock_flags & XFS_IOLOCK_EXCL)
317 up_write(&VFS_I(ip)->i_rwsem);
318 else if (lock_flags & XFS_IOLOCK_SHARED)
319 up_read(&VFS_I(ip)->i_rwsem);
321 if (lock_flags & XFS_MMAPLOCK_EXCL)
322 mrunlock_excl(&ip->i_mmaplock);
323 else if (lock_flags & XFS_MMAPLOCK_SHARED)
324 mrunlock_shared(&ip->i_mmaplock);
326 if (lock_flags & XFS_ILOCK_EXCL)
327 mrunlock_excl(&ip->i_lock);
328 else if (lock_flags & XFS_ILOCK_SHARED)
329 mrunlock_shared(&ip->i_lock);
331 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
335 * give up write locks. the i/o lock cannot be held nested
336 * if it is being demoted.
343 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
345 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
347 if (lock_flags & XFS_ILOCK_EXCL)
348 mrdemote(&ip->i_lock);
349 if (lock_flags & XFS_MMAPLOCK_EXCL)
350 mrdemote(&ip->i_mmaplock);
351 if (lock_flags & XFS_IOLOCK_EXCL)
352 downgrade_write(&VFS_I(ip)->i_rwsem);
354 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
357 #if defined(DEBUG) || defined(XFS_WARN)
363 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
364 if (!(lock_flags & XFS_ILOCK_SHARED))
365 return !!ip->i_lock.mr_writer;
366 return rwsem_is_locked(&ip->i_lock.mr_lock);
369 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
370 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
371 return !!ip->i_mmaplock.mr_writer;
372 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
375 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
376 if (!(lock_flags & XFS_IOLOCK_SHARED))
377 return !debug_locks ||
378 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
379 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
389 int xfs_small_retries;
390 int xfs_middle_retries;
391 int xfs_lots_retries;
396 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
397 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
398 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
399 * errors and warnings.
401 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
403 xfs_lockdep_subclass_ok(
406 return subclass < MAX_LOCKDEP_SUBCLASSES;
409 #define xfs_lockdep_subclass_ok(subclass) (true)
413 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
414 * value. This can be called for any type of inode lock combination, including
415 * parent locking. Care must be taken to ensure we don't overrun the subclass
416 * storage fields in the class mask we build.
419 xfs_lock_inumorder(int lock_mode, int subclass)
423 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
425 ASSERT(xfs_lockdep_subclass_ok(subclass));
427 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
428 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
429 class += subclass << XFS_IOLOCK_SHIFT;
432 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
433 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
434 class += subclass << XFS_MMAPLOCK_SHIFT;
437 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
438 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
439 class += subclass << XFS_ILOCK_SHIFT;
442 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
446 * The following routine will lock n inodes in exclusive mode. We assume the
447 * caller calls us with the inodes in i_ino order.
449 * We need to detect deadlock where an inode that we lock is in the AIL and we
450 * start waiting for another inode that is locked by a thread in a long running
451 * transaction (such as truncate). This can result in deadlock since the long
452 * running trans might need to wait for the inode we just locked in order to
453 * push the tail and free space in the log.
455 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
456 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
457 * lock more than one at a time, lockdep will report false positives saying we
458 * have violated locking orders.
466 int attempts = 0, i, j, try_lock;
470 * Currently supports between 2 and 5 inodes with exclusive locking. We
471 * support an arbitrary depth of locking here, but absolute limits on
472 * inodes depend on the the type of locking and the limits placed by
473 * lockdep annotations in xfs_lock_inumorder. These are all checked by
476 ASSERT(ips && inodes >= 2 && inodes <= 5);
477 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
479 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
481 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
482 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
483 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
484 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
486 if (lock_mode & XFS_IOLOCK_EXCL) {
487 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
488 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
489 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
494 for (; i < inodes; i++) {
497 if (i && (ips[i] == ips[i - 1])) /* Already locked */
501 * If try_lock is not set yet, make sure all locked inodes are
502 * not in the AIL. If any are, set try_lock to be used later.
505 for (j = (i - 1); j >= 0 && !try_lock; j--) {
506 lp = (xfs_log_item_t *)ips[j]->i_itemp;
507 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
513 * If any of the previous locks we have locked is in the AIL,
514 * we must TRY to get the second and subsequent locks. If
515 * we can't get any, we must release all we have
519 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
523 /* try_lock means we have an inode locked that is in the AIL. */
525 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
529 * Unlock all previous guys and try again. xfs_iunlock will try
530 * to push the tail if the inode is in the AIL.
533 for (j = i - 1; j >= 0; j--) {
535 * Check to see if we've already unlocked this one. Not
536 * the first one going back, and the inode ptr is the
539 if (j != (i - 1) && ips[j] == ips[j + 1])
542 xfs_iunlock(ips[j], lock_mode);
545 if ((attempts % 5) == 0) {
546 delay(1); /* Don't just spin the CPU */
558 if (attempts < 5) xfs_small_retries++;
559 else if (attempts < 100) xfs_middle_retries++;
560 else xfs_lots_retries++;
568 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
569 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
570 * lock more than one at a time, lockdep will report false positives saying we
571 * have violated locking orders.
583 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
584 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
585 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
587 ASSERT(ip0->i_ino != ip1->i_ino);
589 if (ip0->i_ino > ip1->i_ino) {
596 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
599 * If the first lock we have locked is in the AIL, we must TRY to get
600 * the second lock. If we can't get it, we must release the first one
603 lp = (xfs_log_item_t *)ip0->i_itemp;
604 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
605 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
606 xfs_iunlock(ip0, lock_mode);
607 if ((++attempts % 5) == 0)
608 delay(1); /* Don't just spin the CPU */
612 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
619 struct xfs_inode *ip)
621 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
622 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
625 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
626 if (xfs_isiflocked(ip))
628 } while (!xfs_iflock_nowait(ip));
630 finish_wait(wq, &wait.wq_entry);
641 if (di_flags & XFS_DIFLAG_ANY) {
642 if (di_flags & XFS_DIFLAG_REALTIME)
643 flags |= FS_XFLAG_REALTIME;
644 if (di_flags & XFS_DIFLAG_PREALLOC)
645 flags |= FS_XFLAG_PREALLOC;
646 if (di_flags & XFS_DIFLAG_IMMUTABLE)
647 flags |= FS_XFLAG_IMMUTABLE;
648 if (di_flags & XFS_DIFLAG_APPEND)
649 flags |= FS_XFLAG_APPEND;
650 if (di_flags & XFS_DIFLAG_SYNC)
651 flags |= FS_XFLAG_SYNC;
652 if (di_flags & XFS_DIFLAG_NOATIME)
653 flags |= FS_XFLAG_NOATIME;
654 if (di_flags & XFS_DIFLAG_NODUMP)
655 flags |= FS_XFLAG_NODUMP;
656 if (di_flags & XFS_DIFLAG_RTINHERIT)
657 flags |= FS_XFLAG_RTINHERIT;
658 if (di_flags & XFS_DIFLAG_PROJINHERIT)
659 flags |= FS_XFLAG_PROJINHERIT;
660 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
661 flags |= FS_XFLAG_NOSYMLINKS;
662 if (di_flags & XFS_DIFLAG_EXTSIZE)
663 flags |= FS_XFLAG_EXTSIZE;
664 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
665 flags |= FS_XFLAG_EXTSZINHERIT;
666 if (di_flags & XFS_DIFLAG_NODEFRAG)
667 flags |= FS_XFLAG_NODEFRAG;
668 if (di_flags & XFS_DIFLAG_FILESTREAM)
669 flags |= FS_XFLAG_FILESTREAM;
672 if (di_flags2 & XFS_DIFLAG2_ANY) {
673 if (di_flags2 & XFS_DIFLAG2_DAX)
674 flags |= FS_XFLAG_DAX;
675 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
676 flags |= FS_XFLAG_COWEXTSIZE;
680 flags |= FS_XFLAG_HASATTR;
687 struct xfs_inode *ip)
689 struct xfs_icdinode *dic = &ip->i_d;
691 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
695 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
696 * is allowed, otherwise it has to be an exact match. If a CI match is found,
697 * ci_name->name will point to a the actual name (caller must free) or
698 * will be set to NULL if an exact match is found.
703 struct xfs_name *name,
705 struct xfs_name *ci_name)
710 trace_xfs_lookup(dp, name);
712 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
715 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
719 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
727 kmem_free(ci_name->name);
734 * Allocate an inode on disk and return a copy of its in-core version.
735 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
736 * appropriately within the inode. The uid and gid for the inode are
737 * set according to the contents of the given cred structure.
739 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
740 * has a free inode available, call xfs_iget() to obtain the in-core
741 * version of the allocated inode. Finally, fill in the inode and
742 * log its initial contents. In this case, ialloc_context would be
745 * If xfs_dialloc() does not have an available inode, it will replenish
746 * its supply by doing an allocation. Since we can only do one
747 * allocation within a transaction without deadlocks, we must commit
748 * the current transaction before returning the inode itself.
749 * In this case, therefore, we will set ialloc_context and return.
750 * The caller should then commit the current transaction, start a new
751 * transaction, and call xfs_ialloc() again to actually get the inode.
753 * To ensure that some other process does not grab the inode that
754 * was allocated during the first call to xfs_ialloc(), this routine
755 * also returns the [locked] bp pointing to the head of the freelist
756 * as ialloc_context. The caller should hold this buffer across
757 * the commit and pass it back into this routine on the second call.
759 * If we are allocating quota inodes, we do not have a parent inode
760 * to attach to or associate with (i.e. pip == NULL) because they
761 * are not linked into the directory structure - they are attached
762 * directly to the superblock - and so have no parent.
773 xfs_buf_t **ialloc_context,
776 struct xfs_mount *mp = tp->t_mountp;
785 * Call the space management code to pick
786 * the on-disk inode to be allocated.
788 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
789 ialloc_context, &ino);
792 if (*ialloc_context || ino == NULLFSINO) {
796 ASSERT(*ialloc_context == NULL);
799 * Get the in-core inode with the lock held exclusively.
800 * This is because we're setting fields here we need
801 * to prevent others from looking at until we're done.
803 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
804 XFS_ILOCK_EXCL, &ip);
811 * We always convert v1 inodes to v2 now - we only support filesystems
812 * with >= v2 inode capability, so there is no reason for ever leaving
813 * an inode in v1 format.
815 if (ip->i_d.di_version == 1)
816 ip->i_d.di_version = 2;
818 inode->i_mode = mode;
819 set_nlink(inode, nlink);
820 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
821 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
822 xfs_set_projid(ip, prid);
824 if (pip && XFS_INHERIT_GID(pip)) {
825 ip->i_d.di_gid = pip->i_d.di_gid;
826 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
827 inode->i_mode |= S_ISGID;
831 * If the group ID of the new file does not match the effective group
832 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
833 * (and only if the irix_sgid_inherit compatibility variable is set).
835 if ((irix_sgid_inherit) &&
836 (inode->i_mode & S_ISGID) &&
837 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
838 inode->i_mode &= ~S_ISGID;
841 ip->i_d.di_nextents = 0;
842 ASSERT(ip->i_d.di_nblocks == 0);
844 tv = current_time(inode);
849 ip->i_d.di_extsize = 0;
850 ip->i_d.di_dmevmask = 0;
851 ip->i_d.di_dmstate = 0;
852 ip->i_d.di_flags = 0;
854 if (ip->i_d.di_version == 3) {
855 inode->i_version = 1;
856 ip->i_d.di_flags2 = 0;
857 ip->i_d.di_cowextsize = 0;
858 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
859 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
863 flags = XFS_ILOG_CORE;
864 switch (mode & S_IFMT) {
869 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
870 ip->i_df.if_u2.if_rdev = rdev;
871 ip->i_df.if_flags = 0;
872 flags |= XFS_ILOG_DEV;
876 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
880 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
881 di_flags |= XFS_DIFLAG_RTINHERIT;
882 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
883 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
884 ip->i_d.di_extsize = pip->i_d.di_extsize;
886 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
887 di_flags |= XFS_DIFLAG_PROJINHERIT;
888 } else if (S_ISREG(mode)) {
889 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
890 di_flags |= XFS_DIFLAG_REALTIME;
891 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
892 di_flags |= XFS_DIFLAG_EXTSIZE;
893 ip->i_d.di_extsize = pip->i_d.di_extsize;
896 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
898 di_flags |= XFS_DIFLAG_NOATIME;
899 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
901 di_flags |= XFS_DIFLAG_NODUMP;
902 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
904 di_flags |= XFS_DIFLAG_SYNC;
905 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
906 xfs_inherit_nosymlinks)
907 di_flags |= XFS_DIFLAG_NOSYMLINKS;
908 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
909 xfs_inherit_nodefrag)
910 di_flags |= XFS_DIFLAG_NODEFRAG;
911 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
912 di_flags |= XFS_DIFLAG_FILESTREAM;
914 ip->i_d.di_flags |= di_flags;
917 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
918 pip->i_d.di_version == 3 &&
919 ip->i_d.di_version == 3) {
920 uint64_t di_flags2 = 0;
922 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
923 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
924 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
926 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
927 di_flags2 |= XFS_DIFLAG2_DAX;
929 ip->i_d.di_flags2 |= di_flags2;
933 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
934 ip->i_df.if_flags = XFS_IFEXTENTS;
935 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
936 ip->i_df.if_u1.if_extents = NULL;
942 * Attribute fork settings for new inode.
944 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
945 ip->i_d.di_anextents = 0;
948 * Log the new values stuffed into the inode.
950 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
951 xfs_trans_log_inode(tp, ip, flags);
953 /* now that we have an i_mode we can setup the inode structure */
961 * Allocates a new inode from disk and return a pointer to the
962 * incore copy. This routine will internally commit the current
963 * transaction and allocate a new one if the Space Manager needed
964 * to do an allocation to replenish the inode free-list.
966 * This routine is designed to be called from xfs_create and
972 xfs_trans_t **tpp, /* input: current transaction;
973 output: may be a new transaction. */
974 xfs_inode_t *dp, /* directory within whose allocate
979 prid_t prid, /* project id */
980 int okalloc, /* ok to allocate new space */
981 xfs_inode_t **ipp, /* pointer to inode; it will be
988 xfs_buf_t *ialloc_context = NULL;
994 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
997 * xfs_ialloc will return a pointer to an incore inode if
998 * the Space Manager has an available inode on the free
999 * list. Otherwise, it will do an allocation and replenish
1000 * the freelist. Since we can only do one allocation per
1001 * transaction without deadlocks, we will need to commit the
1002 * current transaction and start a new one. We will then
1003 * need to call xfs_ialloc again to get the inode.
1005 * If xfs_ialloc did an allocation to replenish the freelist,
1006 * it returns the bp containing the head of the freelist as
1007 * ialloc_context. We will hold a lock on it across the
1008 * transaction commit so that no other process can steal
1009 * the inode(s) that we've just allocated.
1011 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1012 &ialloc_context, &ip);
1015 * Return an error if we were unable to allocate a new inode.
1016 * This should only happen if we run out of space on disk or
1017 * encounter a disk error.
1023 if (!ialloc_context && !ip) {
1029 * If the AGI buffer is non-NULL, then we were unable to get an
1030 * inode in one operation. We need to commit the current
1031 * transaction and call xfs_ialloc() again. It is guaranteed
1032 * to succeed the second time.
1034 if (ialloc_context) {
1036 * Normally, xfs_trans_commit releases all the locks.
1037 * We call bhold to hang on to the ialloc_context across
1038 * the commit. Holding this buffer prevents any other
1039 * processes from doing any allocations in this
1042 xfs_trans_bhold(tp, ialloc_context);
1045 * We want the quota changes to be associated with the next
1046 * transaction, NOT this one. So, detach the dqinfo from this
1047 * and attach it to the next transaction.
1052 dqinfo = (void *)tp->t_dqinfo;
1053 tp->t_dqinfo = NULL;
1054 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1055 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1058 code = xfs_trans_roll(&tp);
1059 if (committed != NULL)
1063 * Re-attach the quota info that we detached from prev trx.
1066 tp->t_dqinfo = dqinfo;
1067 tp->t_flags |= tflags;
1071 xfs_buf_relse(ialloc_context);
1076 xfs_trans_bjoin(tp, ialloc_context);
1079 * Call ialloc again. Since we've locked out all
1080 * other allocations in this allocation group,
1081 * this call should always succeed.
1083 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1084 okalloc, &ialloc_context, &ip);
1087 * If we get an error at this point, return to the caller
1088 * so that the current transaction can be aborted.
1095 ASSERT(!ialloc_context && ip);
1098 if (committed != NULL)
1109 * Decrement the link count on an inode & log the change. If this causes the
1110 * link count to go to zero, move the inode to AGI unlinked list so that it can
1111 * be freed when the last active reference goes away via xfs_inactive().
1113 static int /* error */
1118 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1120 drop_nlink(VFS_I(ip));
1121 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1123 if (VFS_I(ip)->i_nlink)
1126 return xfs_iunlink(tp, ip);
1130 * Increment the link count on an inode & log the change.
1137 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1139 ASSERT(ip->i_d.di_version > 1);
1140 inc_nlink(VFS_I(ip));
1141 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1148 struct xfs_name *name,
1153 int is_dir = S_ISDIR(mode);
1154 struct xfs_mount *mp = dp->i_mount;
1155 struct xfs_inode *ip = NULL;
1156 struct xfs_trans *tp = NULL;
1158 struct xfs_defer_ops dfops;
1159 xfs_fsblock_t first_block;
1160 bool unlock_dp_on_error = false;
1162 struct xfs_dquot *udqp = NULL;
1163 struct xfs_dquot *gdqp = NULL;
1164 struct xfs_dquot *pdqp = NULL;
1165 struct xfs_trans_res *tres;
1168 trace_xfs_create(dp, name);
1170 if (XFS_FORCED_SHUTDOWN(mp))
1173 prid = xfs_get_initial_prid(dp);
1176 * Make sure that we have allocated dquot(s) on disk.
1178 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1179 xfs_kgid_to_gid(current_fsgid()), prid,
1180 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1181 &udqp, &gdqp, &pdqp);
1187 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1188 tres = &M_RES(mp)->tr_mkdir;
1190 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1191 tres = &M_RES(mp)->tr_create;
1195 * Initially assume that the file does not exist and
1196 * reserve the resources for that case. If that is not
1197 * the case we'll drop the one we have and get a more
1198 * appropriate transaction later.
1200 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1201 if (error == -ENOSPC) {
1202 /* flush outstanding delalloc blocks and retry */
1203 xfs_flush_inodes(mp);
1204 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1206 if (error == -ENOSPC) {
1207 /* No space at all so try a "no-allocation" reservation */
1209 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1212 goto out_release_inode;
1214 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1215 unlock_dp_on_error = true;
1217 xfs_defer_init(&dfops, &first_block);
1220 * Reserve disk quota and the inode.
1222 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1223 pdqp, resblks, 1, 0);
1225 goto out_trans_cancel;
1228 error = xfs_dir_canenter(tp, dp, name);
1230 goto out_trans_cancel;
1234 * A newly created regular or special file just has one directory
1235 * entry pointing to them, but a directory also the "." entry
1236 * pointing to itself.
1238 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1239 prid, resblks > 0, &ip, NULL);
1241 goto out_trans_cancel;
1244 * Now we join the directory inode to the transaction. We do not do it
1245 * earlier because xfs_dir_ialloc might commit the previous transaction
1246 * (and release all the locks). An error from here on will result in
1247 * the transaction cancel unlocking dp so don't do it explicitly in the
1250 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1251 unlock_dp_on_error = false;
1253 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1254 &first_block, &dfops, resblks ?
1255 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1257 ASSERT(error != -ENOSPC);
1258 goto out_trans_cancel;
1260 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1261 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1264 error = xfs_dir_init(tp, ip, dp);
1266 goto out_bmap_cancel;
1268 error = xfs_bumplink(tp, dp);
1270 goto out_bmap_cancel;
1274 * If this is a synchronous mount, make sure that the
1275 * create transaction goes to disk before returning to
1278 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1279 xfs_trans_set_sync(tp);
1282 * Attach the dquot(s) to the inodes and modify them incore.
1283 * These ids of the inode couldn't have changed since the new
1284 * inode has been locked ever since it was created.
1286 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1288 error = xfs_defer_finish(&tp, &dfops);
1290 goto out_bmap_cancel;
1292 error = xfs_trans_commit(tp);
1294 goto out_release_inode;
1296 xfs_qm_dqrele(udqp);
1297 xfs_qm_dqrele(gdqp);
1298 xfs_qm_dqrele(pdqp);
1304 xfs_defer_cancel(&dfops);
1306 xfs_trans_cancel(tp);
1309 * Wait until after the current transaction is aborted to finish the
1310 * setup of the inode and release the inode. This prevents recursive
1311 * transactions and deadlocks from xfs_inactive.
1314 xfs_finish_inode_setup(ip);
1318 xfs_qm_dqrele(udqp);
1319 xfs_qm_dqrele(gdqp);
1320 xfs_qm_dqrele(pdqp);
1322 if (unlock_dp_on_error)
1323 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1329 struct xfs_inode *dp,
1330 struct dentry *dentry,
1332 struct xfs_inode **ipp)
1334 struct xfs_mount *mp = dp->i_mount;
1335 struct xfs_inode *ip = NULL;
1336 struct xfs_trans *tp = NULL;
1339 struct xfs_dquot *udqp = NULL;
1340 struct xfs_dquot *gdqp = NULL;
1341 struct xfs_dquot *pdqp = NULL;
1342 struct xfs_trans_res *tres;
1345 if (XFS_FORCED_SHUTDOWN(mp))
1348 prid = xfs_get_initial_prid(dp);
1351 * Make sure that we have allocated dquot(s) on disk.
1353 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1354 xfs_kgid_to_gid(current_fsgid()), prid,
1355 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1356 &udqp, &gdqp, &pdqp);
1360 resblks = XFS_IALLOC_SPACE_RES(mp);
1361 tres = &M_RES(mp)->tr_create_tmpfile;
1363 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1364 if (error == -ENOSPC) {
1365 /* No space at all so try a "no-allocation" reservation */
1367 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1370 goto out_release_inode;
1372 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1373 pdqp, resblks, 1, 0);
1375 goto out_trans_cancel;
1377 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1378 prid, resblks > 0, &ip, NULL);
1380 goto out_trans_cancel;
1382 if (mp->m_flags & XFS_MOUNT_WSYNC)
1383 xfs_trans_set_sync(tp);
1386 * Attach the dquot(s) to the inodes and modify them incore.
1387 * These ids of the inode couldn't have changed since the new
1388 * inode has been locked ever since it was created.
1390 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1392 error = xfs_iunlink(tp, ip);
1394 goto out_trans_cancel;
1396 error = xfs_trans_commit(tp);
1398 goto out_release_inode;
1400 xfs_qm_dqrele(udqp);
1401 xfs_qm_dqrele(gdqp);
1402 xfs_qm_dqrele(pdqp);
1408 xfs_trans_cancel(tp);
1411 * Wait until after the current transaction is aborted to finish the
1412 * setup of the inode and release the inode. This prevents recursive
1413 * transactions and deadlocks from xfs_inactive.
1416 xfs_finish_inode_setup(ip);
1420 xfs_qm_dqrele(udqp);
1421 xfs_qm_dqrele(gdqp);
1422 xfs_qm_dqrele(pdqp);
1431 struct xfs_name *target_name)
1433 xfs_mount_t *mp = tdp->i_mount;
1436 struct xfs_defer_ops dfops;
1437 xfs_fsblock_t first_block;
1440 trace_xfs_link(tdp, target_name);
1442 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1444 if (XFS_FORCED_SHUTDOWN(mp))
1447 error = xfs_qm_dqattach(sip, 0);
1451 error = xfs_qm_dqattach(tdp, 0);
1455 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1456 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1457 if (error == -ENOSPC) {
1459 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1464 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1466 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1467 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1470 * If we are using project inheritance, we only allow hard link
1471 * creation in our tree when the project IDs are the same; else
1472 * the tree quota mechanism could be circumvented.
1474 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1475 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1481 error = xfs_dir_canenter(tp, tdp, target_name);
1486 xfs_defer_init(&dfops, &first_block);
1489 * Handle initial link state of O_TMPFILE inode
1491 if (VFS_I(sip)->i_nlink == 0) {
1492 error = xfs_iunlink_remove(tp, sip);
1497 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1498 &first_block, &dfops, resblks);
1501 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1502 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1504 error = xfs_bumplink(tp, sip);
1509 * If this is a synchronous mount, make sure that the
1510 * link transaction goes to disk before returning to
1513 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1514 xfs_trans_set_sync(tp);
1516 error = xfs_defer_finish(&tp, &dfops);
1518 xfs_defer_cancel(&dfops);
1522 return xfs_trans_commit(tp);
1525 xfs_trans_cancel(tp);
1531 * Free up the underlying blocks past new_size. The new size must be smaller
1532 * than the current size. This routine can be used both for the attribute and
1533 * data fork, and does not modify the inode size, which is left to the caller.
1535 * The transaction passed to this routine must have made a permanent log
1536 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1537 * given transaction and start new ones, so make sure everything involved in
1538 * the transaction is tidy before calling here. Some transaction will be
1539 * returned to the caller to be committed. The incoming transaction must
1540 * already include the inode, and both inode locks must be held exclusively.
1541 * The inode must also be "held" within the transaction. On return the inode
1542 * will be "held" within the returned transaction. This routine does NOT
1543 * require any disk space to be reserved for it within the transaction.
1545 * If we get an error, we must return with the inode locked and linked into the
1546 * current transaction. This keeps things simple for the higher level code,
1547 * because it always knows that the inode is locked and held in the transaction
1548 * that returns to it whether errors occur or not. We don't mark the inode
1549 * dirty on error so that transactions can be easily aborted if possible.
1552 xfs_itruncate_extents(
1553 struct xfs_trans **tpp,
1554 struct xfs_inode *ip,
1556 xfs_fsize_t new_size)
1558 struct xfs_mount *mp = ip->i_mount;
1559 struct xfs_trans *tp = *tpp;
1560 struct xfs_defer_ops dfops;
1561 xfs_fsblock_t first_block;
1562 xfs_fileoff_t first_unmap_block;
1563 xfs_fileoff_t last_block;
1564 xfs_filblks_t unmap_len;
1568 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1569 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1570 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1571 ASSERT(new_size <= XFS_ISIZE(ip));
1572 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1573 ASSERT(ip->i_itemp != NULL);
1574 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1575 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1577 trace_xfs_itruncate_extents_start(ip, new_size);
1580 * Since it is possible for space to become allocated beyond
1581 * the end of the file (in a crash where the space is allocated
1582 * but the inode size is not yet updated), simply remove any
1583 * blocks which show up between the new EOF and the maximum
1584 * possible file size. If the first block to be removed is
1585 * beyond the maximum file size (ie it is the same as last_block),
1586 * then there is nothing to do.
1588 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1589 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1590 if (first_unmap_block == last_block)
1593 ASSERT(first_unmap_block < last_block);
1594 unmap_len = last_block - first_unmap_block + 1;
1596 xfs_defer_init(&dfops, &first_block);
1597 error = xfs_bunmapi(tp, ip,
1598 first_unmap_block, unmap_len,
1599 xfs_bmapi_aflag(whichfork),
1600 XFS_ITRUNC_MAX_EXTENTS,
1601 &first_block, &dfops,
1604 goto out_bmap_cancel;
1607 * Duplicate the transaction that has the permanent
1608 * reservation and commit the old transaction.
1610 xfs_defer_ijoin(&dfops, ip);
1611 error = xfs_defer_finish(&tp, &dfops);
1613 goto out_bmap_cancel;
1615 error = xfs_trans_roll_inode(&tp, ip);
1620 /* Remove all pending CoW reservations. */
1621 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1627 * Clear the reflink flag if there are no data fork blocks and
1628 * there are no extents staged in the cow fork.
1630 if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) {
1631 if (ip->i_d.di_nblocks == 0)
1632 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1633 xfs_inode_clear_cowblocks_tag(ip);
1637 * Always re-log the inode so that our permanent transaction can keep
1638 * on rolling it forward in the log.
1640 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1642 trace_xfs_itruncate_extents_end(ip, new_size);
1649 * If the bunmapi call encounters an error, return to the caller where
1650 * the transaction can be properly aborted. We just need to make sure
1651 * we're not holding any resources that we were not when we came in.
1653 xfs_defer_cancel(&dfops);
1661 xfs_mount_t *mp = ip->i_mount;
1664 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1667 /* If this is a read-only mount, don't do this (would generate I/O) */
1668 if (mp->m_flags & XFS_MOUNT_RDONLY)
1671 if (!XFS_FORCED_SHUTDOWN(mp)) {
1675 * If we previously truncated this file and removed old data
1676 * in the process, we want to initiate "early" writeout on
1677 * the last close. This is an attempt to combat the notorious
1678 * NULL files problem which is particularly noticeable from a
1679 * truncate down, buffered (re-)write (delalloc), followed by
1680 * a crash. What we are effectively doing here is
1681 * significantly reducing the time window where we'd otherwise
1682 * be exposed to that problem.
1684 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1686 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1687 if (ip->i_delayed_blks > 0) {
1688 error = filemap_flush(VFS_I(ip)->i_mapping);
1695 if (VFS_I(ip)->i_nlink == 0)
1698 if (xfs_can_free_eofblocks(ip, false)) {
1701 * Check if the inode is being opened, written and closed
1702 * frequently and we have delayed allocation blocks outstanding
1703 * (e.g. streaming writes from the NFS server), truncating the
1704 * blocks past EOF will cause fragmentation to occur.
1706 * In this case don't do the truncation, but we have to be
1707 * careful how we detect this case. Blocks beyond EOF show up as
1708 * i_delayed_blks even when the inode is clean, so we need to
1709 * truncate them away first before checking for a dirty release.
1710 * Hence on the first dirty close we will still remove the
1711 * speculative allocation, but after that we will leave it in
1714 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1717 * If we can't get the iolock just skip truncating the blocks
1718 * past EOF because we could deadlock with the mmap_sem
1719 * otherwise. We'll get another chance to drop them once the
1720 * last reference to the inode is dropped, so we'll never leak
1721 * blocks permanently.
1723 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1724 error = xfs_free_eofblocks(ip);
1725 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1730 /* delalloc blocks after truncation means it really is dirty */
1731 if (ip->i_delayed_blks)
1732 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1738 * xfs_inactive_truncate
1740 * Called to perform a truncate when an inode becomes unlinked.
1743 xfs_inactive_truncate(
1744 struct xfs_inode *ip)
1746 struct xfs_mount *mp = ip->i_mount;
1747 struct xfs_trans *tp;
1750 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1752 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1756 xfs_ilock(ip, XFS_ILOCK_EXCL);
1757 xfs_trans_ijoin(tp, ip, 0);
1760 * Log the inode size first to prevent stale data exposure in the event
1761 * of a system crash before the truncate completes. See the related
1762 * comment in xfs_vn_setattr_size() for details.
1764 ip->i_d.di_size = 0;
1765 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1767 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1769 goto error_trans_cancel;
1771 ASSERT(ip->i_d.di_nextents == 0);
1773 error = xfs_trans_commit(tp);
1777 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1781 xfs_trans_cancel(tp);
1783 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1788 * xfs_inactive_ifree()
1790 * Perform the inode free when an inode is unlinked.
1794 struct xfs_inode *ip)
1796 struct xfs_defer_ops dfops;
1797 xfs_fsblock_t first_block;
1798 struct xfs_mount *mp = ip->i_mount;
1799 struct xfs_trans *tp;
1803 * We try to use a per-AG reservation for any block needed by the finobt
1804 * tree, but as the finobt feature predates the per-AG reservation
1805 * support a degraded file system might not have enough space for the
1806 * reservation at mount time. In that case try to dip into the reserved
1809 * Send a warning if the reservation does happen to fail, as the inode
1810 * now remains allocated and sits on the unlinked list until the fs is
1813 if (unlikely(mp->m_inotbt_nores)) {
1814 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1815 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1818 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1821 if (error == -ENOSPC) {
1822 xfs_warn_ratelimited(mp,
1823 "Failed to remove inode(s) from unlinked list. "
1824 "Please free space, unmount and run xfs_repair.");
1826 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1831 xfs_ilock(ip, XFS_ILOCK_EXCL);
1832 xfs_trans_ijoin(tp, ip, 0);
1834 xfs_defer_init(&dfops, &first_block);
1835 error = xfs_ifree(tp, ip, &dfops);
1838 * If we fail to free the inode, shut down. The cancel
1839 * might do that, we need to make sure. Otherwise the
1840 * inode might be lost for a long time or forever.
1842 if (!XFS_FORCED_SHUTDOWN(mp)) {
1843 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1845 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1847 xfs_trans_cancel(tp);
1848 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1853 * Credit the quota account(s). The inode is gone.
1855 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1858 * Just ignore errors at this point. There is nothing we can do except
1859 * to try to keep going. Make sure it's not a silent error.
1861 error = xfs_defer_finish(&tp, &dfops);
1863 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1865 xfs_defer_cancel(&dfops);
1867 error = xfs_trans_commit(tp);
1869 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1872 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1879 * This is called when the vnode reference count for the vnode
1880 * goes to zero. If the file has been unlinked, then it must
1881 * now be truncated. Also, we clear all of the read-ahead state
1882 * kept for the inode here since the file is now closed.
1888 struct xfs_mount *mp;
1893 * If the inode is already free, then there can be nothing
1896 if (VFS_I(ip)->i_mode == 0) {
1897 ASSERT(ip->i_df.if_real_bytes == 0);
1898 ASSERT(ip->i_df.if_broot_bytes == 0);
1903 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1905 /* If this is a read-only mount, don't do this (would generate I/O) */
1906 if (mp->m_flags & XFS_MOUNT_RDONLY)
1909 if (VFS_I(ip)->i_nlink != 0) {
1911 * force is true because we are evicting an inode from the
1912 * cache. Post-eof blocks must be freed, lest we end up with
1913 * broken free space accounting.
1915 * Note: don't bother with iolock here since lockdep complains
1916 * about acquiring it in reclaim context. We have the only
1917 * reference to the inode at this point anyways.
1919 if (xfs_can_free_eofblocks(ip, true))
1920 xfs_free_eofblocks(ip);
1925 if (S_ISREG(VFS_I(ip)->i_mode) &&
1926 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1927 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1930 error = xfs_qm_dqattach(ip, 0);
1934 if (S_ISLNK(VFS_I(ip)->i_mode))
1935 error = xfs_inactive_symlink(ip);
1937 error = xfs_inactive_truncate(ip);
1942 * If there are attributes associated with the file then blow them away
1943 * now. The code calls a routine that recursively deconstructs the
1944 * attribute fork. If also blows away the in-core attribute fork.
1946 if (XFS_IFORK_Q(ip)) {
1947 error = xfs_attr_inactive(ip);
1953 ASSERT(ip->i_d.di_anextents == 0);
1954 ASSERT(ip->i_d.di_forkoff == 0);
1959 error = xfs_inactive_ifree(ip);
1964 * Release the dquots held by inode, if any.
1966 xfs_qm_dqdetach(ip);
1970 * This is called when the inode's link count goes to 0 or we are creating a
1971 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1972 * set to true as the link count is dropped to zero by the VFS after we've
1973 * created the file successfully, so we have to add it to the unlinked list
1974 * while the link count is non-zero.
1976 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1977 * list when the inode is freed.
1981 struct xfs_trans *tp,
1982 struct xfs_inode *ip)
1984 xfs_mount_t *mp = tp->t_mountp;
1994 ASSERT(VFS_I(ip)->i_mode != 0);
1997 * Get the agi buffer first. It ensures lock ordering
2000 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
2003 agi = XFS_BUF_TO_AGI(agibp);
2006 * Get the index into the agi hash table for the
2007 * list this inode will go on.
2009 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2011 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2012 ASSERT(agi->agi_unlinked[bucket_index]);
2013 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2015 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2017 * There is already another inode in the bucket we need
2018 * to add ourselves to. Add us at the front of the list.
2019 * Here we put the head pointer into our next pointer,
2020 * and then we fall through to point the head at us.
2022 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2027 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2028 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2029 offset = ip->i_imap.im_boffset +
2030 offsetof(xfs_dinode_t, di_next_unlinked);
2032 /* need to recalc the inode CRC if appropriate */
2033 xfs_dinode_calc_crc(mp, dip);
2035 xfs_trans_inode_buf(tp, ibp);
2036 xfs_trans_log_buf(tp, ibp, offset,
2037 (offset + sizeof(xfs_agino_t) - 1));
2038 xfs_inobp_check(mp, ibp);
2042 * Point the bucket head pointer at the inode being inserted.
2045 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2046 offset = offsetof(xfs_agi_t, agi_unlinked) +
2047 (sizeof(xfs_agino_t) * bucket_index);
2048 xfs_trans_log_buf(tp, agibp, offset,
2049 (offset + sizeof(xfs_agino_t) - 1));
2054 * Pull the on-disk inode from the AGI unlinked list.
2067 xfs_agnumber_t agno;
2069 xfs_agino_t next_agino;
2070 xfs_buf_t *last_ibp;
2071 xfs_dinode_t *last_dip = NULL;
2073 int offset, last_offset = 0;
2077 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2080 * Get the agi buffer first. It ensures lock ordering
2083 error = xfs_read_agi(mp, tp, agno, &agibp);
2087 agi = XFS_BUF_TO_AGI(agibp);
2090 * Get the index into the agi hash table for the
2091 * list this inode will go on.
2093 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2095 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2096 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2097 ASSERT(agi->agi_unlinked[bucket_index]);
2099 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2101 * We're at the head of the list. Get the inode's on-disk
2102 * buffer to see if there is anyone after us on the list.
2103 * Only modify our next pointer if it is not already NULLAGINO.
2104 * This saves us the overhead of dealing with the buffer when
2105 * there is no need to change it.
2107 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2110 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2114 next_agino = be32_to_cpu(dip->di_next_unlinked);
2115 ASSERT(next_agino != 0);
2116 if (next_agino != NULLAGINO) {
2117 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2118 offset = ip->i_imap.im_boffset +
2119 offsetof(xfs_dinode_t, di_next_unlinked);
2121 /* need to recalc the inode CRC if appropriate */
2122 xfs_dinode_calc_crc(mp, dip);
2124 xfs_trans_inode_buf(tp, ibp);
2125 xfs_trans_log_buf(tp, ibp, offset,
2126 (offset + sizeof(xfs_agino_t) - 1));
2127 xfs_inobp_check(mp, ibp);
2129 xfs_trans_brelse(tp, ibp);
2132 * Point the bucket head pointer at the next inode.
2134 ASSERT(next_agino != 0);
2135 ASSERT(next_agino != agino);
2136 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2137 offset = offsetof(xfs_agi_t, agi_unlinked) +
2138 (sizeof(xfs_agino_t) * bucket_index);
2139 xfs_trans_log_buf(tp, agibp, offset,
2140 (offset + sizeof(xfs_agino_t) - 1));
2143 * We need to search the list for the inode being freed.
2145 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2147 while (next_agino != agino) {
2148 struct xfs_imap imap;
2151 xfs_trans_brelse(tp, last_ibp);
2154 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2156 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2159 "%s: xfs_imap returned error %d.",
2164 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2168 "%s: xfs_imap_to_bp returned error %d.",
2173 last_offset = imap.im_boffset;
2174 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2175 ASSERT(next_agino != NULLAGINO);
2176 ASSERT(next_agino != 0);
2180 * Now last_ibp points to the buffer previous to us on the
2181 * unlinked list. Pull us from the list.
2183 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2186 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2190 next_agino = be32_to_cpu(dip->di_next_unlinked);
2191 ASSERT(next_agino != 0);
2192 ASSERT(next_agino != agino);
2193 if (next_agino != NULLAGINO) {
2194 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2195 offset = ip->i_imap.im_boffset +
2196 offsetof(xfs_dinode_t, di_next_unlinked);
2198 /* need to recalc the inode CRC if appropriate */
2199 xfs_dinode_calc_crc(mp, dip);
2201 xfs_trans_inode_buf(tp, ibp);
2202 xfs_trans_log_buf(tp, ibp, offset,
2203 (offset + sizeof(xfs_agino_t) - 1));
2204 xfs_inobp_check(mp, ibp);
2206 xfs_trans_brelse(tp, ibp);
2209 * Point the previous inode on the list to the next inode.
2211 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2212 ASSERT(next_agino != 0);
2213 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2215 /* need to recalc the inode CRC if appropriate */
2216 xfs_dinode_calc_crc(mp, last_dip);
2218 xfs_trans_inode_buf(tp, last_ibp);
2219 xfs_trans_log_buf(tp, last_ibp, offset,
2220 (offset + sizeof(xfs_agino_t) - 1));
2221 xfs_inobp_check(mp, last_ibp);
2227 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2228 * inodes that are in memory - they all must be marked stale and attached to
2229 * the cluster buffer.
2233 xfs_inode_t *free_ip,
2235 struct xfs_icluster *xic)
2237 xfs_mount_t *mp = free_ip->i_mount;
2238 int blks_per_cluster;
2239 int inodes_per_cluster;
2246 xfs_inode_log_item_t *iip;
2247 xfs_log_item_t *lip;
2248 struct xfs_perag *pag;
2251 inum = xic->first_ino;
2252 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2253 blks_per_cluster = xfs_icluster_size_fsb(mp);
2254 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2255 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2257 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2259 * The allocation bitmap tells us which inodes of the chunk were
2260 * physically allocated. Skip the cluster if an inode falls into
2263 ioffset = inum - xic->first_ino;
2264 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2265 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2269 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2270 XFS_INO_TO_AGBNO(mp, inum));
2273 * We obtain and lock the backing buffer first in the process
2274 * here, as we have to ensure that any dirty inode that we
2275 * can't get the flush lock on is attached to the buffer.
2276 * If we scan the in-memory inodes first, then buffer IO can
2277 * complete before we get a lock on it, and hence we may fail
2278 * to mark all the active inodes on the buffer stale.
2280 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2281 mp->m_bsize * blks_per_cluster,
2288 * This buffer may not have been correctly initialised as we
2289 * didn't read it from disk. That's not important because we are
2290 * only using to mark the buffer as stale in the log, and to
2291 * attach stale cached inodes on it. That means it will never be
2292 * dispatched for IO. If it is, we want to know about it, and we
2293 * want it to fail. We can acheive this by adding a write
2294 * verifier to the buffer.
2296 bp->b_ops = &xfs_inode_buf_ops;
2299 * Walk the inodes already attached to the buffer and mark them
2300 * stale. These will all have the flush locks held, so an
2301 * in-memory inode walk can't lock them. By marking them all
2302 * stale first, we will not attempt to lock them in the loop
2303 * below as the XFS_ISTALE flag will be set.
2307 if (lip->li_type == XFS_LI_INODE) {
2308 iip = (xfs_inode_log_item_t *)lip;
2309 ASSERT(iip->ili_logged == 1);
2310 lip->li_cb = xfs_istale_done;
2311 xfs_trans_ail_copy_lsn(mp->m_ail,
2312 &iip->ili_flush_lsn,
2313 &iip->ili_item.li_lsn);
2314 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2316 lip = lip->li_bio_list;
2321 * For each inode in memory attempt to add it to the inode
2322 * buffer and set it up for being staled on buffer IO
2323 * completion. This is safe as we've locked out tail pushing
2324 * and flushing by locking the buffer.
2326 * We have already marked every inode that was part of a
2327 * transaction stale above, which means there is no point in
2328 * even trying to lock them.
2330 for (i = 0; i < inodes_per_cluster; i++) {
2333 ip = radix_tree_lookup(&pag->pag_ici_root,
2334 XFS_INO_TO_AGINO(mp, (inum + i)));
2336 /* Inode not in memory, nothing to do */
2343 * because this is an RCU protected lookup, we could
2344 * find a recently freed or even reallocated inode
2345 * during the lookup. We need to check under the
2346 * i_flags_lock for a valid inode here. Skip it if it
2347 * is not valid, the wrong inode or stale.
2349 spin_lock(&ip->i_flags_lock);
2350 if (ip->i_ino != inum + i ||
2351 __xfs_iflags_test(ip, XFS_ISTALE)) {
2352 spin_unlock(&ip->i_flags_lock);
2356 spin_unlock(&ip->i_flags_lock);
2359 * Don't try to lock/unlock the current inode, but we
2360 * _cannot_ skip the other inodes that we did not find
2361 * in the list attached to the buffer and are not
2362 * already marked stale. If we can't lock it, back off
2365 if (ip != free_ip) {
2366 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2373 * Check the inode number again in case we're
2374 * racing with freeing in xfs_reclaim_inode().
2375 * See the comments in that function for more
2376 * information as to why the initial check is
2379 if (ip->i_ino != inum + i) {
2380 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2388 xfs_iflags_set(ip, XFS_ISTALE);
2391 * we don't need to attach clean inodes or those only
2392 * with unlogged changes (which we throw away, anyway).
2395 if (!iip || xfs_inode_clean(ip)) {
2396 ASSERT(ip != free_ip);
2398 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2402 iip->ili_last_fields = iip->ili_fields;
2403 iip->ili_fields = 0;
2404 iip->ili_fsync_fields = 0;
2405 iip->ili_logged = 1;
2406 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2407 &iip->ili_item.li_lsn);
2409 xfs_buf_attach_iodone(bp, xfs_istale_done,
2413 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2416 xfs_trans_stale_inode_buf(tp, bp);
2417 xfs_trans_binval(tp, bp);
2425 * Free any local-format buffers sitting around before we reset to
2429 xfs_ifree_local_data(
2430 struct xfs_inode *ip,
2433 struct xfs_ifork *ifp;
2435 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2438 ifp = XFS_IFORK_PTR(ip, whichfork);
2439 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2443 * This is called to return an inode to the inode free list.
2444 * The inode should already be truncated to 0 length and have
2445 * no pages associated with it. This routine also assumes that
2446 * the inode is already a part of the transaction.
2448 * The on-disk copy of the inode will have been added to the list
2449 * of unlinked inodes in the AGI. We need to remove the inode from
2450 * that list atomically with respect to freeing it here.
2456 struct xfs_defer_ops *dfops)
2459 struct xfs_icluster xic = { 0 };
2461 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2462 ASSERT(VFS_I(ip)->i_nlink == 0);
2463 ASSERT(ip->i_d.di_nextents == 0);
2464 ASSERT(ip->i_d.di_anextents == 0);
2465 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2466 ASSERT(ip->i_d.di_nblocks == 0);
2469 * Pull the on-disk inode from the AGI unlinked list.
2471 error = xfs_iunlink_remove(tp, ip);
2475 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2479 xfs_ifree_local_data(ip, XFS_DATA_FORK);
2480 xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2482 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2483 ip->i_d.di_flags = 0;
2484 ip->i_d.di_dmevmask = 0;
2485 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2486 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2487 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2489 * Bump the generation count so no one will be confused
2490 * by reincarnations of this inode.
2492 VFS_I(ip)->i_generation++;
2493 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2496 error = xfs_ifree_cluster(ip, tp, &xic);
2502 * This is called to unpin an inode. The caller must have the inode locked
2503 * in at least shared mode so that the buffer cannot be subsequently pinned
2504 * once someone is waiting for it to be unpinned.
2508 struct xfs_inode *ip)
2510 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2512 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2514 /* Give the log a push to start the unpinning I/O */
2515 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2521 struct xfs_inode *ip)
2523 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2524 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2529 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2530 if (xfs_ipincount(ip))
2532 } while (xfs_ipincount(ip));
2533 finish_wait(wq, &wait.wq_entry);
2538 struct xfs_inode *ip)
2540 if (xfs_ipincount(ip))
2541 __xfs_iunpin_wait(ip);
2545 * Removing an inode from the namespace involves removing the directory entry
2546 * and dropping the link count on the inode. Removing the directory entry can
2547 * result in locking an AGF (directory blocks were freed) and removing a link
2548 * count can result in placing the inode on an unlinked list which results in
2551 * The big problem here is that we have an ordering constraint on AGF and AGI
2552 * locking - inode allocation locks the AGI, then can allocate a new extent for
2553 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2554 * removes the inode from the unlinked list, requiring that we lock the AGI
2555 * first, and then freeing the inode can result in an inode chunk being freed
2556 * and hence freeing disk space requiring that we lock an AGF.
2558 * Hence the ordering that is imposed by other parts of the code is AGI before
2559 * AGF. This means we cannot remove the directory entry before we drop the inode
2560 * reference count and put it on the unlinked list as this results in a lock
2561 * order of AGF then AGI, and this can deadlock against inode allocation and
2562 * freeing. Therefore we must drop the link counts before we remove the
2565 * This is still safe from a transactional point of view - it is not until we
2566 * get to xfs_defer_finish() that we have the possibility of multiple
2567 * transactions in this operation. Hence as long as we remove the directory
2568 * entry and drop the link count in the first transaction of the remove
2569 * operation, there are no transactional constraints on the ordering here.
2574 struct xfs_name *name,
2577 xfs_mount_t *mp = dp->i_mount;
2578 xfs_trans_t *tp = NULL;
2579 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2581 struct xfs_defer_ops dfops;
2582 xfs_fsblock_t first_block;
2585 trace_xfs_remove(dp, name);
2587 if (XFS_FORCED_SHUTDOWN(mp))
2590 error = xfs_qm_dqattach(dp, 0);
2594 error = xfs_qm_dqattach(ip, 0);
2599 * We try to get the real space reservation first,
2600 * allowing for directory btree deletion(s) implying
2601 * possible bmap insert(s). If we can't get the space
2602 * reservation then we use 0 instead, and avoid the bmap
2603 * btree insert(s) in the directory code by, if the bmap
2604 * insert tries to happen, instead trimming the LAST
2605 * block from the directory.
2607 resblks = XFS_REMOVE_SPACE_RES(mp);
2608 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2609 if (error == -ENOSPC) {
2611 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2615 ASSERT(error != -ENOSPC);
2619 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2621 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2622 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2625 * If we're removing a directory perform some additional validation.
2628 ASSERT(VFS_I(ip)->i_nlink >= 2);
2629 if (VFS_I(ip)->i_nlink != 2) {
2631 goto out_trans_cancel;
2633 if (!xfs_dir_isempty(ip)) {
2635 goto out_trans_cancel;
2638 /* Drop the link from ip's "..". */
2639 error = xfs_droplink(tp, dp);
2641 goto out_trans_cancel;
2643 /* Drop the "." link from ip to self. */
2644 error = xfs_droplink(tp, ip);
2646 goto out_trans_cancel;
2649 * When removing a non-directory we need to log the parent
2650 * inode here. For a directory this is done implicitly
2651 * by the xfs_droplink call for the ".." entry.
2653 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2655 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2657 /* Drop the link from dp to ip. */
2658 error = xfs_droplink(tp, ip);
2660 goto out_trans_cancel;
2662 xfs_defer_init(&dfops, &first_block);
2663 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2664 &first_block, &dfops, resblks);
2666 ASSERT(error != -ENOENT);
2667 goto out_bmap_cancel;
2671 * If this is a synchronous mount, make sure that the
2672 * remove transaction goes to disk before returning to
2675 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2676 xfs_trans_set_sync(tp);
2678 error = xfs_defer_finish(&tp, &dfops);
2680 goto out_bmap_cancel;
2682 error = xfs_trans_commit(tp);
2686 if (is_dir && xfs_inode_is_filestream(ip))
2687 xfs_filestream_deassociate(ip);
2692 xfs_defer_cancel(&dfops);
2694 xfs_trans_cancel(tp);
2700 * Enter all inodes for a rename transaction into a sorted array.
2702 #define __XFS_SORT_INODES 5
2704 xfs_sort_for_rename(
2705 struct xfs_inode *dp1, /* in: old (source) directory inode */
2706 struct xfs_inode *dp2, /* in: new (target) directory inode */
2707 struct xfs_inode *ip1, /* in: inode of old entry */
2708 struct xfs_inode *ip2, /* in: inode of new entry */
2709 struct xfs_inode *wip, /* in: whiteout inode */
2710 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2711 int *num_inodes) /* in/out: inodes in array */
2715 ASSERT(*num_inodes == __XFS_SORT_INODES);
2716 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2719 * i_tab contains a list of pointers to inodes. We initialize
2720 * the table here & we'll sort it. We will then use it to
2721 * order the acquisition of the inode locks.
2723 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2736 * Sort the elements via bubble sort. (Remember, there are at
2737 * most 5 elements to sort, so this is adequate.)
2739 for (i = 0; i < *num_inodes; i++) {
2740 for (j = 1; j < *num_inodes; j++) {
2741 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2742 struct xfs_inode *temp = i_tab[j];
2743 i_tab[j] = i_tab[j-1];
2752 struct xfs_trans *tp,
2753 struct xfs_defer_ops *dfops)
2758 * If this is a synchronous mount, make sure that the rename transaction
2759 * goes to disk before returning to the user.
2761 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2762 xfs_trans_set_sync(tp);
2764 error = xfs_defer_finish(&tp, dfops);
2766 xfs_defer_cancel(dfops);
2767 xfs_trans_cancel(tp);
2771 return xfs_trans_commit(tp);
2775 * xfs_cross_rename()
2777 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2781 struct xfs_trans *tp,
2782 struct xfs_inode *dp1,
2783 struct xfs_name *name1,
2784 struct xfs_inode *ip1,
2785 struct xfs_inode *dp2,
2786 struct xfs_name *name2,
2787 struct xfs_inode *ip2,
2788 struct xfs_defer_ops *dfops,
2789 xfs_fsblock_t *first_block,
2797 /* Swap inode number for dirent in first parent */
2798 error = xfs_dir_replace(tp, dp1, name1,
2800 first_block, dfops, spaceres);
2802 goto out_trans_abort;
2804 /* Swap inode number for dirent in second parent */
2805 error = xfs_dir_replace(tp, dp2, name2,
2807 first_block, dfops, spaceres);
2809 goto out_trans_abort;
2812 * If we're renaming one or more directories across different parents,
2813 * update the respective ".." entries (and link counts) to match the new
2817 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2819 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2820 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2821 dp1->i_ino, first_block,
2824 goto out_trans_abort;
2826 /* transfer ip2 ".." reference to dp1 */
2827 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2828 error = xfs_droplink(tp, dp2);
2830 goto out_trans_abort;
2831 error = xfs_bumplink(tp, dp1);
2833 goto out_trans_abort;
2837 * Although ip1 isn't changed here, userspace needs
2838 * to be warned about the change, so that applications
2839 * relying on it (like backup ones), will properly
2842 ip1_flags |= XFS_ICHGTIME_CHG;
2843 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2846 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2847 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2848 dp2->i_ino, first_block,
2851 goto out_trans_abort;
2853 /* transfer ip1 ".." reference to dp2 */
2854 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2855 error = xfs_droplink(tp, dp1);
2857 goto out_trans_abort;
2858 error = xfs_bumplink(tp, dp2);
2860 goto out_trans_abort;
2864 * Although ip2 isn't changed here, userspace needs
2865 * to be warned about the change, so that applications
2866 * relying on it (like backup ones), will properly
2869 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2870 ip2_flags |= XFS_ICHGTIME_CHG;
2875 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2876 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2879 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2880 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2883 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2884 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2886 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2887 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2888 return xfs_finish_rename(tp, dfops);
2891 xfs_defer_cancel(dfops);
2892 xfs_trans_cancel(tp);
2897 * xfs_rename_alloc_whiteout()
2899 * Return a referenced, unlinked, unlocked inode that that can be used as a
2900 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2901 * crash between allocating the inode and linking it into the rename transaction
2902 * recovery will free the inode and we won't leak it.
2905 xfs_rename_alloc_whiteout(
2906 struct xfs_inode *dp,
2907 struct xfs_inode **wip)
2909 struct xfs_inode *tmpfile;
2912 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2917 * Prepare the tmpfile inode as if it were created through the VFS.
2918 * Otherwise, the link increment paths will complain about nlink 0->1.
2919 * Drop the link count as done by d_tmpfile(), complete the inode setup
2920 * and flag it as linkable.
2922 drop_nlink(VFS_I(tmpfile));
2923 xfs_setup_iops(tmpfile);
2924 xfs_finish_inode_setup(tmpfile);
2925 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2936 struct xfs_inode *src_dp,
2937 struct xfs_name *src_name,
2938 struct xfs_inode *src_ip,
2939 struct xfs_inode *target_dp,
2940 struct xfs_name *target_name,
2941 struct xfs_inode *target_ip,
2944 struct xfs_mount *mp = src_dp->i_mount;
2945 struct xfs_trans *tp;
2946 struct xfs_defer_ops dfops;
2947 xfs_fsblock_t first_block;
2948 struct xfs_inode *wip = NULL; /* whiteout inode */
2949 struct xfs_inode *inodes[__XFS_SORT_INODES];
2950 int num_inodes = __XFS_SORT_INODES;
2951 bool new_parent = (src_dp != target_dp);
2952 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2956 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2958 if ((flags & RENAME_EXCHANGE) && !target_ip)
2962 * If we are doing a whiteout operation, allocate the whiteout inode
2963 * we will be placing at the target and ensure the type is set
2966 if (flags & RENAME_WHITEOUT) {
2967 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2968 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2972 /* setup target dirent info as whiteout */
2973 src_name->type = XFS_DIR3_FT_CHRDEV;
2976 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2977 inodes, &num_inodes);
2979 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2980 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2981 if (error == -ENOSPC) {
2983 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2987 goto out_release_wip;
2990 * Attach the dquots to the inodes
2992 error = xfs_qm_vop_rename_dqattach(inodes);
2994 goto out_trans_cancel;
2997 * Lock all the participating inodes. Depending upon whether
2998 * the target_name exists in the target directory, and
2999 * whether the target directory is the same as the source
3000 * directory, we can lock from 2 to 4 inodes.
3002 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3005 * Join all the inodes to the transaction. From this point on,
3006 * we can rely on either trans_commit or trans_cancel to unlock
3009 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3011 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3012 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3014 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3016 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3019 * If we are using project inheritance, we only allow renames
3020 * into our tree when the project IDs are the same; else the
3021 * tree quota mechanism would be circumvented.
3023 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3024 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3026 goto out_trans_cancel;
3029 xfs_defer_init(&dfops, &first_block);
3031 /* RENAME_EXCHANGE is unique from here on. */
3032 if (flags & RENAME_EXCHANGE)
3033 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3034 target_dp, target_name, target_ip,
3035 &dfops, &first_block, spaceres);
3038 * Check for expected errors before we dirty the transaction
3039 * so we can return an error without a transaction abort.
3041 if (target_ip == NULL) {
3043 * If there's no space reservation, check the entry will
3044 * fit before actually inserting it.
3047 error = xfs_dir_canenter(tp, target_dp, target_name);
3049 goto out_trans_cancel;
3053 * If target exists and it's a directory, check that whether
3054 * it can be destroyed.
3056 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3057 (!xfs_dir_isempty(target_ip) ||
3058 (VFS_I(target_ip)->i_nlink > 2))) {
3060 goto out_trans_cancel;
3065 * Directory entry creation below may acquire the AGF. Remove
3066 * the whiteout from the unlinked list first to preserve correct
3067 * AGI/AGF locking order. This dirties the transaction so failures
3068 * after this point will abort and log recovery will clean up the
3071 * For whiteouts, we need to bump the link count on the whiteout
3072 * inode. After this point, we have a real link, clear the tmpfile
3073 * state flag from the inode so it doesn't accidentally get misused
3077 ASSERT(VFS_I(wip)->i_nlink == 0);
3078 error = xfs_iunlink_remove(tp, wip);
3080 goto out_trans_cancel;
3082 xfs_bumplink(tp, wip);
3083 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3084 VFS_I(wip)->i_state &= ~I_LINKABLE;
3088 * Set up the target.
3090 if (target_ip == NULL) {
3092 * If target does not exist and the rename crosses
3093 * directories, adjust the target directory link count
3094 * to account for the ".." reference from the new entry.
3096 error = xfs_dir_createname(tp, target_dp, target_name,
3097 src_ip->i_ino, &first_block,
3100 goto out_bmap_cancel;
3102 xfs_trans_ichgtime(tp, target_dp,
3103 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3105 if (new_parent && src_is_directory) {
3106 error = xfs_bumplink(tp, target_dp);
3108 goto out_bmap_cancel;
3110 } else { /* target_ip != NULL */
3112 * Link the source inode under the target name.
3113 * If the source inode is a directory and we are moving
3114 * it across directories, its ".." entry will be
3115 * inconsistent until we replace that down below.
3117 * In case there is already an entry with the same
3118 * name at the destination directory, remove it first.
3120 error = xfs_dir_replace(tp, target_dp, target_name,
3122 &first_block, &dfops, spaceres);
3124 goto out_bmap_cancel;
3126 xfs_trans_ichgtime(tp, target_dp,
3127 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3130 * Decrement the link count on the target since the target
3131 * dir no longer points to it.
3133 error = xfs_droplink(tp, target_ip);
3135 goto out_bmap_cancel;
3137 if (src_is_directory) {
3139 * Drop the link from the old "." entry.
3141 error = xfs_droplink(tp, target_ip);
3143 goto out_bmap_cancel;
3145 } /* target_ip != NULL */
3148 * Remove the source.
3150 if (new_parent && src_is_directory) {
3152 * Rewrite the ".." entry to point to the new
3155 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3157 &first_block, &dfops, spaceres);
3158 ASSERT(error != -EEXIST);
3160 goto out_bmap_cancel;
3164 * We always want to hit the ctime on the source inode.
3166 * This isn't strictly required by the standards since the source
3167 * inode isn't really being changed, but old unix file systems did
3168 * it and some incremental backup programs won't work without it.
3170 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3171 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3174 * Adjust the link count on src_dp. This is necessary when
3175 * renaming a directory, either within one parent when
3176 * the target existed, or across two parent directories.
3178 if (src_is_directory && (new_parent || target_ip != NULL)) {
3181 * Decrement link count on src_directory since the
3182 * entry that's moved no longer points to it.
3184 error = xfs_droplink(tp, src_dp);
3186 goto out_bmap_cancel;
3190 * For whiteouts, we only need to update the source dirent with the
3191 * inode number of the whiteout inode rather than removing it
3195 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3196 &first_block, &dfops, spaceres);
3198 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3199 &first_block, &dfops, spaceres);
3201 goto out_bmap_cancel;
3203 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3204 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3206 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3208 error = xfs_finish_rename(tp, &dfops);
3214 xfs_defer_cancel(&dfops);
3216 xfs_trans_cancel(tp);
3225 struct xfs_inode *ip,
3228 struct xfs_mount *mp = ip->i_mount;
3229 struct xfs_perag *pag;
3230 unsigned long first_index, mask;
3231 unsigned long inodes_per_cluster;
3233 struct xfs_inode **cilist;
3234 struct xfs_inode *cip;
3240 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3242 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3243 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3244 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3248 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3249 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3251 /* really need a gang lookup range call here */
3252 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3253 first_index, inodes_per_cluster);
3257 for (i = 0; i < nr_found; i++) {
3263 * because this is an RCU protected lookup, we could find a
3264 * recently freed or even reallocated inode during the lookup.
3265 * We need to check under the i_flags_lock for a valid inode
3266 * here. Skip it if it is not valid or the wrong inode.
3268 spin_lock(&cip->i_flags_lock);
3270 __xfs_iflags_test(cip, XFS_ISTALE)) {
3271 spin_unlock(&cip->i_flags_lock);
3276 * Once we fall off the end of the cluster, no point checking
3277 * any more inodes in the list because they will also all be
3278 * outside the cluster.
3280 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3281 spin_unlock(&cip->i_flags_lock);
3284 spin_unlock(&cip->i_flags_lock);
3287 * Do an un-protected check to see if the inode is dirty and
3288 * is a candidate for flushing. These checks will be repeated
3289 * later after the appropriate locks are acquired.
3291 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3295 * Try to get locks. If any are unavailable or it is pinned,
3296 * then this inode cannot be flushed and is skipped.
3299 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3301 if (!xfs_iflock_nowait(cip)) {
3302 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3305 if (xfs_ipincount(cip)) {
3307 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3313 * Check the inode number again, just to be certain we are not
3314 * racing with freeing in xfs_reclaim_inode(). See the comments
3315 * in that function for more information as to why the initial
3316 * check is not sufficient.
3320 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3325 * arriving here means that this inode can be flushed. First
3326 * re-check that it's dirty before flushing.
3328 if (!xfs_inode_clean(cip)) {
3330 error = xfs_iflush_int(cip, bp);
3332 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3333 goto cluster_corrupt_out;
3339 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3343 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3344 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3355 cluster_corrupt_out:
3357 * Corruption detected in the clustering loop. Invalidate the
3358 * inode buffer and shut down the filesystem.
3362 * Clean up the buffer. If it was delwri, just release it --
3363 * brelse can handle it with no problems. If not, shut down the
3364 * filesystem before releasing the buffer.
3366 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3370 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3372 if (!bufwasdelwri) {
3374 * Just like incore_relse: if we have b_iodone functions,
3375 * mark the buffer as an error and call them. Otherwise
3376 * mark it as stale and brelse.
3379 bp->b_flags &= ~XBF_DONE;
3381 xfs_buf_ioerror(bp, -EIO);
3390 * Unlocks the flush lock
3392 xfs_iflush_abort(cip, false);
3395 return -EFSCORRUPTED;
3399 * Flush dirty inode metadata into the backing buffer.
3401 * The caller must have the inode lock and the inode flush lock held. The
3402 * inode lock will still be held upon return to the caller, and the inode
3403 * flush lock will be released after the inode has reached the disk.
3405 * The caller must write out the buffer returned in *bpp and release it.
3409 struct xfs_inode *ip,
3410 struct xfs_buf **bpp)
3412 struct xfs_mount *mp = ip->i_mount;
3413 struct xfs_buf *bp = NULL;
3414 struct xfs_dinode *dip;
3417 XFS_STATS_INC(mp, xs_iflush_count);
3419 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3420 ASSERT(xfs_isiflocked(ip));
3421 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3422 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3426 xfs_iunpin_wait(ip);
3429 * For stale inodes we cannot rely on the backing buffer remaining
3430 * stale in cache for the remaining life of the stale inode and so
3431 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3432 * inodes below. We have to check this after ensuring the inode is
3433 * unpinned so that it is safe to reclaim the stale inode after the
3436 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3442 * This may have been unpinned because the filesystem is shutting
3443 * down forcibly. If that's the case we must not write this inode
3444 * to disk, because the log record didn't make it to disk.
3446 * We also have to remove the log item from the AIL in this case,
3447 * as we wait for an empty AIL as part of the unmount process.
3449 if (XFS_FORCED_SHUTDOWN(mp)) {
3455 * Get the buffer containing the on-disk inode. We are doing a try-lock
3456 * operation here, so we may get an EAGAIN error. In that case, we
3457 * simply want to return with the inode still dirty.
3459 * If we get any other error, we effectively have a corruption situation
3460 * and we cannot flush the inode, so we treat it the same as failing
3463 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3465 if (error == -EAGAIN) {
3473 * First flush out the inode that xfs_iflush was called with.
3475 error = xfs_iflush_int(ip, bp);
3480 * If the buffer is pinned then push on the log now so we won't
3481 * get stuck waiting in the write for too long.
3483 if (xfs_buf_ispinned(bp))
3484 xfs_log_force(mp, 0);
3488 * see if other inodes can be gathered into this write
3490 error = xfs_iflush_cluster(ip, bp);
3492 goto cluster_corrupt_out;
3500 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3501 cluster_corrupt_out:
3502 error = -EFSCORRUPTED;
3505 * Unlocks the flush lock
3507 xfs_iflush_abort(ip, false);
3513 struct xfs_inode *ip,
3516 struct xfs_inode_log_item *iip = ip->i_itemp;
3517 struct xfs_dinode *dip;
3518 struct xfs_mount *mp = ip->i_mount;
3520 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3521 ASSERT(xfs_isiflocked(ip));
3522 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3523 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3524 ASSERT(iip != NULL && iip->ili_fields != 0);
3525 ASSERT(ip->i_d.di_version > 1);
3527 /* set *dip = inode's place in the buffer */
3528 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3530 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3531 mp, XFS_ERRTAG_IFLUSH_1)) {
3532 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3533 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3534 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3537 if (S_ISREG(VFS_I(ip)->i_mode)) {
3539 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3540 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3541 mp, XFS_ERRTAG_IFLUSH_3)) {
3542 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3543 "%s: Bad regular inode %Lu, ptr 0x%p",
3544 __func__, ip->i_ino, ip);
3547 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3549 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3550 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3551 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3552 mp, XFS_ERRTAG_IFLUSH_4)) {
3553 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3554 "%s: Bad directory inode %Lu, ptr 0x%p",
3555 __func__, ip->i_ino, ip);
3559 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3560 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3561 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3562 "%s: detected corrupt incore inode %Lu, "
3563 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3564 __func__, ip->i_ino,
3565 ip->i_d.di_nextents + ip->i_d.di_anextents,
3566 ip->i_d.di_nblocks, ip);
3569 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3570 mp, XFS_ERRTAG_IFLUSH_6)) {
3571 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3572 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3573 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3578 * Inode item log recovery for v2 inodes are dependent on the
3579 * di_flushiter count for correct sequencing. We bump the flush
3580 * iteration count so we can detect flushes which postdate a log record
3581 * during recovery. This is redundant as we now log every change and
3582 * hence this can't happen but we need to still do it to ensure
3583 * backwards compatibility with old kernels that predate logging all
3586 if (ip->i_d.di_version < 3)
3587 ip->i_d.di_flushiter++;
3589 /* Check the inline directory data. */
3590 if (S_ISDIR(VFS_I(ip)->i_mode) &&
3591 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3592 xfs_dir2_sf_verify(ip))
3596 * Copy the dirty parts of the inode into the on-disk inode. We always
3597 * copy out the core of the inode, because if the inode is dirty at all
3600 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3602 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3603 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3604 ip->i_d.di_flushiter = 0;
3606 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3607 if (XFS_IFORK_Q(ip))
3608 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3609 xfs_inobp_check(mp, bp);
3612 * We've recorded everything logged in the inode, so we'd like to clear
3613 * the ili_fields bits so we don't log and flush things unnecessarily.
3614 * However, we can't stop logging all this information until the data
3615 * we've copied into the disk buffer is written to disk. If we did we
3616 * might overwrite the copy of the inode in the log with all the data
3617 * after re-logging only part of it, and in the face of a crash we
3618 * wouldn't have all the data we need to recover.
3620 * What we do is move the bits to the ili_last_fields field. When
3621 * logging the inode, these bits are moved back to the ili_fields field.
3622 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3623 * know that the information those bits represent is permanently on
3624 * disk. As long as the flush completes before the inode is logged
3625 * again, then both ili_fields and ili_last_fields will be cleared.
3627 * We can play with the ili_fields bits here, because the inode lock
3628 * must be held exclusively in order to set bits there and the flush
3629 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3630 * done routine can tell whether or not to look in the AIL. Also, store
3631 * the current LSN of the inode so that we can tell whether the item has
3632 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3633 * need the AIL lock, because it is a 64 bit value that cannot be read
3636 iip->ili_last_fields = iip->ili_fields;
3637 iip->ili_fields = 0;
3638 iip->ili_fsync_fields = 0;
3639 iip->ili_logged = 1;
3641 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3642 &iip->ili_item.li_lsn);
3645 * Attach the function xfs_iflush_done to the inode's
3646 * buffer. This will remove the inode from the AIL
3647 * and unlock the inode's flush lock when the inode is
3648 * completely written to disk.
3650 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3652 /* generate the checksum. */
3653 xfs_dinode_calc_crc(mp, dip);
3655 ASSERT(bp->b_fspriv != NULL);
3656 ASSERT(bp->b_iodone != NULL);
3660 return -EFSCORRUPTED;