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_inode.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
32 #include "xfs_attr_sf.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_ialloc.h"
40 #include "xfs_bmap_util.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_filestream.h"
44 #include "xfs_cksum.h"
45 #include "xfs_trace.h"
46 #include "xfs_icache.h"
47 #include "xfs_symlink.h"
48 #include "xfs_trans_priv.h"
50 #include "xfs_bmap_btree.h"
52 kmem_zone_t *xfs_inode_zone;
55 * Used in xfs_itruncate_extents(). This is the maximum number of extents
56 * freed from a file in a single transaction.
58 #define XFS_ITRUNC_MAX_EXTENTS 2
60 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
61 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
62 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
65 * helper function to extract extent size hint from inode
71 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
72 return ip->i_d.di_extsize;
73 if (XFS_IS_REALTIME_INODE(ip))
74 return ip->i_mount->m_sb.sb_rextsize;
79 * These two are wrapper routines around the xfs_ilock() routine used to
80 * centralize some grungy code. They are used in places that wish to lock the
81 * inode solely for reading the extents. The reason these places can't just
82 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
83 * bringing in of the extents from disk for a file in b-tree format. If the
84 * inode is in b-tree format, then we need to lock the inode exclusively until
85 * the extents are read in. Locking it exclusively all the time would limit
86 * our parallelism unnecessarily, though. What we do instead is check to see
87 * if the extents have been read in yet, and only lock the inode exclusively
90 * The functions return a value which should be given to the corresponding
94 xfs_ilock_data_map_shared(
97 uint lock_mode = XFS_ILOCK_SHARED;
99 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
100 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
101 lock_mode = XFS_ILOCK_EXCL;
102 xfs_ilock(ip, lock_mode);
107 xfs_ilock_attr_map_shared(
108 struct xfs_inode *ip)
110 uint lock_mode = XFS_ILOCK_SHARED;
112 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
113 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
114 lock_mode = XFS_ILOCK_EXCL;
115 xfs_ilock(ip, lock_mode);
120 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
121 * the i_lock. This routine allows various combinations of the locks to be
124 * The 3 locks should always be ordered so that the IO lock is obtained first,
125 * the mmap lock second and the ilock last in order to prevent deadlock.
127 * Basic locking order:
129 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
131 * mmap_sem locking order:
133 * i_iolock -> page lock -> mmap_sem
134 * mmap_sem -> i_mmap_lock -> page_lock
136 * The difference in mmap_sem locking order mean that we cannot hold the
137 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
138 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
139 * in get_user_pages() to map the user pages into the kernel address space for
140 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
141 * page faults already hold the mmap_sem.
143 * Hence to serialise fully against both syscall and mmap based IO, we need to
144 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
145 * taken in places where we need to invalidate the page cache in a race
146 * free manner (e.g. truncate, hole punch and other extent manipulation
154 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
157 * You can't set both SHARED and EXCL for the same lock,
158 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
159 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
161 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
162 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
163 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
164 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
165 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
166 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
167 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
169 if (lock_flags & XFS_IOLOCK_EXCL)
170 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
171 else if (lock_flags & XFS_IOLOCK_SHARED)
172 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
174 if (lock_flags & XFS_MMAPLOCK_EXCL)
175 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
176 else if (lock_flags & XFS_MMAPLOCK_SHARED)
177 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
179 if (lock_flags & XFS_ILOCK_EXCL)
180 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
181 else if (lock_flags & XFS_ILOCK_SHARED)
182 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
186 * This is just like xfs_ilock(), except that the caller
187 * is guaranteed not to sleep. It returns 1 if it gets
188 * the requested locks and 0 otherwise. If the IO lock is
189 * obtained but the inode lock cannot be, then the IO lock
190 * is dropped before returning.
192 * ip -- the inode being locked
193 * lock_flags -- this parameter indicates the inode's locks to be
194 * to be locked. See the comment for xfs_ilock() for a list
202 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
205 * You can't set both SHARED and EXCL for the same lock,
206 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
207 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
209 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
210 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
211 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
212 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
213 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
214 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
215 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
217 if (lock_flags & XFS_IOLOCK_EXCL) {
218 if (!mrtryupdate(&ip->i_iolock))
220 } else if (lock_flags & XFS_IOLOCK_SHARED) {
221 if (!mrtryaccess(&ip->i_iolock))
225 if (lock_flags & XFS_MMAPLOCK_EXCL) {
226 if (!mrtryupdate(&ip->i_mmaplock))
227 goto out_undo_iolock;
228 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
229 if (!mrtryaccess(&ip->i_mmaplock))
230 goto out_undo_iolock;
233 if (lock_flags & XFS_ILOCK_EXCL) {
234 if (!mrtryupdate(&ip->i_lock))
235 goto out_undo_mmaplock;
236 } else if (lock_flags & XFS_ILOCK_SHARED) {
237 if (!mrtryaccess(&ip->i_lock))
238 goto out_undo_mmaplock;
243 if (lock_flags & XFS_MMAPLOCK_EXCL)
244 mrunlock_excl(&ip->i_mmaplock);
245 else if (lock_flags & XFS_MMAPLOCK_SHARED)
246 mrunlock_shared(&ip->i_mmaplock);
248 if (lock_flags & XFS_IOLOCK_EXCL)
249 mrunlock_excl(&ip->i_iolock);
250 else if (lock_flags & XFS_IOLOCK_SHARED)
251 mrunlock_shared(&ip->i_iolock);
257 * xfs_iunlock() is used to drop the inode locks acquired with
258 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
259 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
260 * that we know which locks to drop.
262 * ip -- the inode being unlocked
263 * lock_flags -- this parameter indicates the inode's locks to be
264 * to be unlocked. See the comment for xfs_ilock() for a list
265 * of valid values for this parameter.
274 * You can't set both SHARED and EXCL for the same lock,
275 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
276 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
278 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
279 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
280 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
281 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
282 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
283 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
284 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
285 ASSERT(lock_flags != 0);
287 if (lock_flags & XFS_IOLOCK_EXCL)
288 mrunlock_excl(&ip->i_iolock);
289 else if (lock_flags & XFS_IOLOCK_SHARED)
290 mrunlock_shared(&ip->i_iolock);
292 if (lock_flags & XFS_MMAPLOCK_EXCL)
293 mrunlock_excl(&ip->i_mmaplock);
294 else if (lock_flags & XFS_MMAPLOCK_SHARED)
295 mrunlock_shared(&ip->i_mmaplock);
297 if (lock_flags & XFS_ILOCK_EXCL)
298 mrunlock_excl(&ip->i_lock);
299 else if (lock_flags & XFS_ILOCK_SHARED)
300 mrunlock_shared(&ip->i_lock);
302 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
306 * give up write locks. the i/o lock cannot be held nested
307 * if it is being demoted.
314 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
316 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
318 if (lock_flags & XFS_ILOCK_EXCL)
319 mrdemote(&ip->i_lock);
320 if (lock_flags & XFS_MMAPLOCK_EXCL)
321 mrdemote(&ip->i_mmaplock);
322 if (lock_flags & XFS_IOLOCK_EXCL)
323 mrdemote(&ip->i_iolock);
325 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
328 #if defined(DEBUG) || defined(XFS_WARN)
334 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
335 if (!(lock_flags & XFS_ILOCK_SHARED))
336 return !!ip->i_lock.mr_writer;
337 return rwsem_is_locked(&ip->i_lock.mr_lock);
340 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
341 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
342 return !!ip->i_mmaplock.mr_writer;
343 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
346 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
347 if (!(lock_flags & XFS_IOLOCK_SHARED))
348 return !!ip->i_iolock.mr_writer;
349 return rwsem_is_locked(&ip->i_iolock.mr_lock);
359 int xfs_small_retries;
360 int xfs_middle_retries;
361 int xfs_lots_retries;
366 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
367 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
368 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
369 * errors and warnings.
371 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
373 xfs_lockdep_subclass_ok(
376 return subclass < MAX_LOCKDEP_SUBCLASSES;
379 #define xfs_lockdep_subclass_ok(subclass) (true)
383 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
384 * value. This can be called for any type of inode lock combination, including
385 * parent locking. Care must be taken to ensure we don't overrun the subclass
386 * storage fields in the class mask we build.
389 xfs_lock_inumorder(int lock_mode, int subclass)
393 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
395 ASSERT(xfs_lockdep_subclass_ok(subclass));
397 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
398 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
399 ASSERT(xfs_lockdep_subclass_ok(subclass +
400 XFS_IOLOCK_PARENT_VAL));
401 class += subclass << XFS_IOLOCK_SHIFT;
402 if (lock_mode & XFS_IOLOCK_PARENT)
403 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
406 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
407 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
408 class += subclass << XFS_MMAPLOCK_SHIFT;
411 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
412 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
413 class += subclass << XFS_ILOCK_SHIFT;
416 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
420 * The following routine will lock n inodes in exclusive mode. We assume the
421 * caller calls us with the inodes in i_ino order.
423 * We need to detect deadlock where an inode that we lock is in the AIL and we
424 * start waiting for another inode that is locked by a thread in a long running
425 * transaction (such as truncate). This can result in deadlock since the long
426 * running trans might need to wait for the inode we just locked in order to
427 * push the tail and free space in the log.
429 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
430 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
431 * lock more than one at a time, lockdep will report false positives saying we
432 * have violated locking orders.
440 int attempts = 0, i, j, try_lock;
444 * Currently supports between 2 and 5 inodes with exclusive locking. We
445 * support an arbitrary depth of locking here, but absolute limits on
446 * inodes depend on the the type of locking and the limits placed by
447 * lockdep annotations in xfs_lock_inumorder. These are all checked by
450 ASSERT(ips && inodes >= 2 && inodes <= 5);
451 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
453 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
455 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
456 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
457 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
458 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
459 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
460 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
462 if (lock_mode & XFS_IOLOCK_EXCL) {
463 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
464 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
470 for (; i < inodes; i++) {
473 if (i && (ips[i] == ips[i - 1])) /* Already locked */
477 * If try_lock is not set yet, make sure all locked inodes are
478 * not in the AIL. If any are, set try_lock to be used later.
481 for (j = (i - 1); j >= 0 && !try_lock; j--) {
482 lp = (xfs_log_item_t *)ips[j]->i_itemp;
483 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
489 * If any of the previous locks we have locked is in the AIL,
490 * we must TRY to get the second and subsequent locks. If
491 * we can't get any, we must release all we have
495 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
499 /* try_lock means we have an inode locked that is in the AIL. */
501 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
505 * Unlock all previous guys and try again. xfs_iunlock will try
506 * to push the tail if the inode is in the AIL.
509 for (j = i - 1; j >= 0; j--) {
511 * Check to see if we've already unlocked this one. Not
512 * the first one going back, and the inode ptr is the
515 if (j != (i - 1) && ips[j] == ips[j + 1])
518 xfs_iunlock(ips[j], lock_mode);
521 if ((attempts % 5) == 0) {
522 delay(1); /* Don't just spin the CPU */
534 if (attempts < 5) xfs_small_retries++;
535 else if (attempts < 100) xfs_middle_retries++;
536 else xfs_lots_retries++;
544 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
545 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
546 * lock more than one at a time, lockdep will report false positives saying we
547 * have violated locking orders.
559 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
560 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
561 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
563 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
565 ASSERT(ip0->i_ino != ip1->i_ino);
567 if (ip0->i_ino > ip1->i_ino) {
574 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
577 * If the first lock we have locked is in the AIL, we must TRY to get
578 * the second lock. If we can't get it, we must release the first one
581 lp = (xfs_log_item_t *)ip0->i_itemp;
582 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
583 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
584 xfs_iunlock(ip0, lock_mode);
585 if ((++attempts % 5) == 0)
586 delay(1); /* Don't just spin the CPU */
590 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
597 struct xfs_inode *ip)
599 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
600 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
603 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
604 if (xfs_isiflocked(ip))
606 } while (!xfs_iflock_nowait(ip));
608 finish_wait(wq, &wait.wait);
619 if (di_flags & XFS_DIFLAG_ANY) {
620 if (di_flags & XFS_DIFLAG_REALTIME)
621 flags |= FS_XFLAG_REALTIME;
622 if (di_flags & XFS_DIFLAG_PREALLOC)
623 flags |= FS_XFLAG_PREALLOC;
624 if (di_flags & XFS_DIFLAG_IMMUTABLE)
625 flags |= FS_XFLAG_IMMUTABLE;
626 if (di_flags & XFS_DIFLAG_APPEND)
627 flags |= FS_XFLAG_APPEND;
628 if (di_flags & XFS_DIFLAG_SYNC)
629 flags |= FS_XFLAG_SYNC;
630 if (di_flags & XFS_DIFLAG_NOATIME)
631 flags |= FS_XFLAG_NOATIME;
632 if (di_flags & XFS_DIFLAG_NODUMP)
633 flags |= FS_XFLAG_NODUMP;
634 if (di_flags & XFS_DIFLAG_RTINHERIT)
635 flags |= FS_XFLAG_RTINHERIT;
636 if (di_flags & XFS_DIFLAG_PROJINHERIT)
637 flags |= FS_XFLAG_PROJINHERIT;
638 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
639 flags |= FS_XFLAG_NOSYMLINKS;
640 if (di_flags & XFS_DIFLAG_EXTSIZE)
641 flags |= FS_XFLAG_EXTSIZE;
642 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
643 flags |= FS_XFLAG_EXTSZINHERIT;
644 if (di_flags & XFS_DIFLAG_NODEFRAG)
645 flags |= FS_XFLAG_NODEFRAG;
646 if (di_flags & XFS_DIFLAG_FILESTREAM)
647 flags |= FS_XFLAG_FILESTREAM;
650 if (di_flags2 & XFS_DIFLAG2_ANY) {
651 if (di_flags2 & XFS_DIFLAG2_DAX)
652 flags |= FS_XFLAG_DAX;
656 flags |= FS_XFLAG_HASATTR;
663 struct xfs_inode *ip)
665 struct xfs_icdinode *dic = &ip->i_d;
667 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
671 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
672 * is allowed, otherwise it has to be an exact match. If a CI match is found,
673 * ci_name->name will point to a the actual name (caller must free) or
674 * will be set to NULL if an exact match is found.
679 struct xfs_name *name,
681 struct xfs_name *ci_name)
686 trace_xfs_lookup(dp, name);
688 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
691 xfs_ilock(dp, XFS_IOLOCK_SHARED);
692 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
696 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
700 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
705 kmem_free(ci_name->name);
707 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
713 * Allocate an inode on disk and return a copy of its in-core version.
714 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
715 * appropriately within the inode. The uid and gid for the inode are
716 * set according to the contents of the given cred structure.
718 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
719 * has a free inode available, call xfs_iget() to obtain the in-core
720 * version of the allocated inode. Finally, fill in the inode and
721 * log its initial contents. In this case, ialloc_context would be
724 * If xfs_dialloc() does not have an available inode, it will replenish
725 * its supply by doing an allocation. Since we can only do one
726 * allocation within a transaction without deadlocks, we must commit
727 * the current transaction before returning the inode itself.
728 * In this case, therefore, we will set ialloc_context and return.
729 * The caller should then commit the current transaction, start a new
730 * transaction, and call xfs_ialloc() again to actually get the inode.
732 * To ensure that some other process does not grab the inode that
733 * was allocated during the first call to xfs_ialloc(), this routine
734 * also returns the [locked] bp pointing to the head of the freelist
735 * as ialloc_context. The caller should hold this buffer across
736 * the commit and pass it back into this routine on the second call.
738 * If we are allocating quota inodes, we do not have a parent inode
739 * to attach to or associate with (i.e. pip == NULL) because they
740 * are not linked into the directory structure - they are attached
741 * directly to the superblock - and so have no parent.
752 xfs_buf_t **ialloc_context,
755 struct xfs_mount *mp = tp->t_mountp;
764 * Call the space management code to pick
765 * the on-disk inode to be allocated.
767 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
768 ialloc_context, &ino);
771 if (*ialloc_context || ino == NULLFSINO) {
775 ASSERT(*ialloc_context == NULL);
778 * Get the in-core inode with the lock held exclusively.
779 * This is because we're setting fields here we need
780 * to prevent others from looking at until we're done.
782 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
783 XFS_ILOCK_EXCL, &ip);
790 * We always convert v1 inodes to v2 now - we only support filesystems
791 * with >= v2 inode capability, so there is no reason for ever leaving
792 * an inode in v1 format.
794 if (ip->i_d.di_version == 1)
795 ip->i_d.di_version = 2;
797 inode->i_mode = mode;
798 set_nlink(inode, nlink);
799 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
800 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
801 xfs_set_projid(ip, prid);
803 if (pip && XFS_INHERIT_GID(pip)) {
804 ip->i_d.di_gid = pip->i_d.di_gid;
805 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
806 inode->i_mode |= S_ISGID;
810 * If the group ID of the new file does not match the effective group
811 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
812 * (and only if the irix_sgid_inherit compatibility variable is set).
814 if ((irix_sgid_inherit) &&
815 (inode->i_mode & S_ISGID) &&
816 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
817 inode->i_mode &= ~S_ISGID;
820 ip->i_d.di_nextents = 0;
821 ASSERT(ip->i_d.di_nblocks == 0);
823 tv = current_fs_time(mp->m_super);
828 ip->i_d.di_extsize = 0;
829 ip->i_d.di_dmevmask = 0;
830 ip->i_d.di_dmstate = 0;
831 ip->i_d.di_flags = 0;
833 if (ip->i_d.di_version == 3) {
834 inode->i_version = 1;
835 ip->i_d.di_flags2 = 0;
836 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
837 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
841 flags = XFS_ILOG_CORE;
842 switch (mode & S_IFMT) {
847 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
848 ip->i_df.if_u2.if_rdev = rdev;
849 ip->i_df.if_flags = 0;
850 flags |= XFS_ILOG_DEV;
854 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
855 uint64_t di_flags2 = 0;
859 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
860 di_flags |= XFS_DIFLAG_RTINHERIT;
861 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
862 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
863 ip->i_d.di_extsize = pip->i_d.di_extsize;
865 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
866 di_flags |= XFS_DIFLAG_PROJINHERIT;
867 } else if (S_ISREG(mode)) {
868 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
869 di_flags |= XFS_DIFLAG_REALTIME;
870 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
871 di_flags |= XFS_DIFLAG_EXTSIZE;
872 ip->i_d.di_extsize = pip->i_d.di_extsize;
875 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
877 di_flags |= XFS_DIFLAG_NOATIME;
878 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
880 di_flags |= XFS_DIFLAG_NODUMP;
881 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
883 di_flags |= XFS_DIFLAG_SYNC;
884 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
885 xfs_inherit_nosymlinks)
886 di_flags |= XFS_DIFLAG_NOSYMLINKS;
887 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
888 xfs_inherit_nodefrag)
889 di_flags |= XFS_DIFLAG_NODEFRAG;
890 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
891 di_flags |= XFS_DIFLAG_FILESTREAM;
892 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
893 di_flags2 |= XFS_DIFLAG2_DAX;
895 ip->i_d.di_flags |= di_flags;
896 ip->i_d.di_flags2 |= di_flags2;
900 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
901 ip->i_df.if_flags = XFS_IFEXTENTS;
902 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
903 ip->i_df.if_u1.if_extents = NULL;
909 * Attribute fork settings for new inode.
911 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
912 ip->i_d.di_anextents = 0;
915 * Log the new values stuffed into the inode.
917 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
918 xfs_trans_log_inode(tp, ip, flags);
920 /* now that we have an i_mode we can setup the inode structure */
928 * Allocates a new inode from disk and return a pointer to the
929 * incore copy. This routine will internally commit the current
930 * transaction and allocate a new one if the Space Manager needed
931 * to do an allocation to replenish the inode free-list.
933 * This routine is designed to be called from xfs_create and
939 xfs_trans_t **tpp, /* input: current transaction;
940 output: may be a new transaction. */
941 xfs_inode_t *dp, /* directory within whose allocate
946 prid_t prid, /* project id */
947 int okalloc, /* ok to allocate new space */
948 xfs_inode_t **ipp, /* pointer to inode; it will be
955 xfs_buf_t *ialloc_context = NULL;
961 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
964 * xfs_ialloc will return a pointer to an incore inode if
965 * the Space Manager has an available inode on the free
966 * list. Otherwise, it will do an allocation and replenish
967 * the freelist. Since we can only do one allocation per
968 * transaction without deadlocks, we will need to commit the
969 * current transaction and start a new one. We will then
970 * need to call xfs_ialloc again to get the inode.
972 * If xfs_ialloc did an allocation to replenish the freelist,
973 * it returns the bp containing the head of the freelist as
974 * ialloc_context. We will hold a lock on it across the
975 * transaction commit so that no other process can steal
976 * the inode(s) that we've just allocated.
978 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
979 &ialloc_context, &ip);
982 * Return an error if we were unable to allocate a new inode.
983 * This should only happen if we run out of space on disk or
984 * encounter a disk error.
990 if (!ialloc_context && !ip) {
996 * If the AGI buffer is non-NULL, then we were unable to get an
997 * inode in one operation. We need to commit the current
998 * transaction and call xfs_ialloc() again. It is guaranteed
999 * to succeed the second time.
1001 if (ialloc_context) {
1003 * Normally, xfs_trans_commit releases all the locks.
1004 * We call bhold to hang on to the ialloc_context across
1005 * the commit. Holding this buffer prevents any other
1006 * processes from doing any allocations in this
1009 xfs_trans_bhold(tp, ialloc_context);
1012 * We want the quota changes to be associated with the next
1013 * transaction, NOT this one. So, detach the dqinfo from this
1014 * and attach it to the next transaction.
1019 dqinfo = (void *)tp->t_dqinfo;
1020 tp->t_dqinfo = NULL;
1021 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1022 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1025 code = xfs_trans_roll(&tp, NULL);
1026 if (committed != NULL)
1030 * Re-attach the quota info that we detached from prev trx.
1033 tp->t_dqinfo = dqinfo;
1034 tp->t_flags |= tflags;
1038 xfs_buf_relse(ialloc_context);
1043 xfs_trans_bjoin(tp, ialloc_context);
1046 * Call ialloc again. Since we've locked out all
1047 * other allocations in this allocation group,
1048 * this call should always succeed.
1050 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1051 okalloc, &ialloc_context, &ip);
1054 * If we get an error at this point, return to the caller
1055 * so that the current transaction can be aborted.
1062 ASSERT(!ialloc_context && ip);
1065 if (committed != NULL)
1076 * Decrement the link count on an inode & log the change. If this causes the
1077 * link count to go to zero, move the inode to AGI unlinked list so that it can
1078 * be freed when the last active reference goes away via xfs_inactive().
1080 static int /* error */
1085 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1087 drop_nlink(VFS_I(ip));
1088 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1090 if (VFS_I(ip)->i_nlink)
1093 return xfs_iunlink(tp, ip);
1097 * Increment the link count on an inode & log the change.
1104 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1106 ASSERT(ip->i_d.di_version > 1);
1107 inc_nlink(VFS_I(ip));
1108 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1115 struct xfs_name *name,
1120 int is_dir = S_ISDIR(mode);
1121 struct xfs_mount *mp = dp->i_mount;
1122 struct xfs_inode *ip = NULL;
1123 struct xfs_trans *tp = NULL;
1125 xfs_bmap_free_t free_list;
1126 xfs_fsblock_t first_block;
1127 bool unlock_dp_on_error = false;
1129 struct xfs_dquot *udqp = NULL;
1130 struct xfs_dquot *gdqp = NULL;
1131 struct xfs_dquot *pdqp = NULL;
1132 struct xfs_trans_res *tres;
1135 trace_xfs_create(dp, name);
1137 if (XFS_FORCED_SHUTDOWN(mp))
1140 prid = xfs_get_initial_prid(dp);
1143 * Make sure that we have allocated dquot(s) on disk.
1145 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1146 xfs_kgid_to_gid(current_fsgid()), prid,
1147 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1148 &udqp, &gdqp, &pdqp);
1154 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1155 tres = &M_RES(mp)->tr_mkdir;
1157 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1158 tres = &M_RES(mp)->tr_create;
1162 * Initially assume that the file does not exist and
1163 * reserve the resources for that case. If that is not
1164 * the case we'll drop the one we have and get a more
1165 * appropriate transaction later.
1167 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1168 if (error == -ENOSPC) {
1169 /* flush outstanding delalloc blocks and retry */
1170 xfs_flush_inodes(mp);
1171 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1173 if (error == -ENOSPC) {
1174 /* No space at all so try a "no-allocation" reservation */
1176 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1179 goto out_release_inode;
1181 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1182 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
1183 unlock_dp_on_error = true;
1185 xfs_bmap_init(&free_list, &first_block);
1188 * Reserve disk quota and the inode.
1190 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1191 pdqp, resblks, 1, 0);
1193 goto out_trans_cancel;
1196 error = xfs_dir_canenter(tp, dp, name);
1198 goto out_trans_cancel;
1202 * A newly created regular or special file just has one directory
1203 * entry pointing to them, but a directory also the "." entry
1204 * pointing to itself.
1206 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1207 prid, resblks > 0, &ip, NULL);
1209 goto out_trans_cancel;
1212 * Now we join the directory inode to the transaction. We do not do it
1213 * earlier because xfs_dir_ialloc might commit the previous transaction
1214 * (and release all the locks). An error from here on will result in
1215 * the transaction cancel unlocking dp so don't do it explicitly in the
1218 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1219 unlock_dp_on_error = false;
1221 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1222 &first_block, &free_list, resblks ?
1223 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1225 ASSERT(error != -ENOSPC);
1226 goto out_trans_cancel;
1228 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1229 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1232 error = xfs_dir_init(tp, ip, dp);
1234 goto out_bmap_cancel;
1236 error = xfs_bumplink(tp, dp);
1238 goto out_bmap_cancel;
1242 * If this is a synchronous mount, make sure that the
1243 * create transaction goes to disk before returning to
1246 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1247 xfs_trans_set_sync(tp);
1250 * Attach the dquot(s) to the inodes and modify them incore.
1251 * These ids of the inode couldn't have changed since the new
1252 * inode has been locked ever since it was created.
1254 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1256 error = xfs_bmap_finish(&tp, &free_list, NULL);
1258 goto out_bmap_cancel;
1260 error = xfs_trans_commit(tp);
1262 goto out_release_inode;
1264 xfs_qm_dqrele(udqp);
1265 xfs_qm_dqrele(gdqp);
1266 xfs_qm_dqrele(pdqp);
1272 xfs_bmap_cancel(&free_list);
1274 xfs_trans_cancel(tp);
1277 * Wait until after the current transaction is aborted to finish the
1278 * setup of the inode and release the inode. This prevents recursive
1279 * transactions and deadlocks from xfs_inactive.
1282 xfs_finish_inode_setup(ip);
1286 xfs_qm_dqrele(udqp);
1287 xfs_qm_dqrele(gdqp);
1288 xfs_qm_dqrele(pdqp);
1290 if (unlock_dp_on_error)
1291 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1297 struct xfs_inode *dp,
1298 struct dentry *dentry,
1300 struct xfs_inode **ipp)
1302 struct xfs_mount *mp = dp->i_mount;
1303 struct xfs_inode *ip = NULL;
1304 struct xfs_trans *tp = NULL;
1307 struct xfs_dquot *udqp = NULL;
1308 struct xfs_dquot *gdqp = NULL;
1309 struct xfs_dquot *pdqp = NULL;
1310 struct xfs_trans_res *tres;
1313 if (XFS_FORCED_SHUTDOWN(mp))
1316 prid = xfs_get_initial_prid(dp);
1319 * Make sure that we have allocated dquot(s) on disk.
1321 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1322 xfs_kgid_to_gid(current_fsgid()), prid,
1323 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1324 &udqp, &gdqp, &pdqp);
1328 resblks = XFS_IALLOC_SPACE_RES(mp);
1329 tres = &M_RES(mp)->tr_create_tmpfile;
1331 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1332 if (error == -ENOSPC) {
1333 /* No space at all so try a "no-allocation" reservation */
1335 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1338 goto out_release_inode;
1340 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1341 pdqp, resblks, 1, 0);
1343 goto out_trans_cancel;
1345 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1346 prid, resblks > 0, &ip, NULL);
1348 goto out_trans_cancel;
1350 if (mp->m_flags & XFS_MOUNT_WSYNC)
1351 xfs_trans_set_sync(tp);
1354 * Attach the dquot(s) to the inodes and modify them incore.
1355 * These ids of the inode couldn't have changed since the new
1356 * inode has been locked ever since it was created.
1358 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1360 error = xfs_iunlink(tp, ip);
1362 goto out_trans_cancel;
1364 error = xfs_trans_commit(tp);
1366 goto out_release_inode;
1368 xfs_qm_dqrele(udqp);
1369 xfs_qm_dqrele(gdqp);
1370 xfs_qm_dqrele(pdqp);
1376 xfs_trans_cancel(tp);
1379 * Wait until after the current transaction is aborted to finish the
1380 * setup of the inode and release the inode. This prevents recursive
1381 * transactions and deadlocks from xfs_inactive.
1384 xfs_finish_inode_setup(ip);
1388 xfs_qm_dqrele(udqp);
1389 xfs_qm_dqrele(gdqp);
1390 xfs_qm_dqrele(pdqp);
1399 struct xfs_name *target_name)
1401 xfs_mount_t *mp = tdp->i_mount;
1404 xfs_bmap_free_t free_list;
1405 xfs_fsblock_t first_block;
1408 trace_xfs_link(tdp, target_name);
1410 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1412 if (XFS_FORCED_SHUTDOWN(mp))
1415 error = xfs_qm_dqattach(sip, 0);
1419 error = xfs_qm_dqattach(tdp, 0);
1423 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1424 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1425 if (error == -ENOSPC) {
1427 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1432 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
1433 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1435 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1436 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1439 * If we are using project inheritance, we only allow hard link
1440 * creation in our tree when the project IDs are the same; else
1441 * the tree quota mechanism could be circumvented.
1443 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1444 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1450 error = xfs_dir_canenter(tp, tdp, target_name);
1455 xfs_bmap_init(&free_list, &first_block);
1458 * Handle initial link state of O_TMPFILE inode
1460 if (VFS_I(sip)->i_nlink == 0) {
1461 error = xfs_iunlink_remove(tp, sip);
1466 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1467 &first_block, &free_list, resblks);
1470 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1471 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1473 error = xfs_bumplink(tp, sip);
1478 * If this is a synchronous mount, make sure that the
1479 * link transaction goes to disk before returning to
1482 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1483 xfs_trans_set_sync(tp);
1485 error = xfs_bmap_finish(&tp, &free_list, NULL);
1487 xfs_bmap_cancel(&free_list);
1491 return xfs_trans_commit(tp);
1494 xfs_trans_cancel(tp);
1500 * Free up the underlying blocks past new_size. The new size must be smaller
1501 * than the current size. This routine can be used both for the attribute and
1502 * data fork, and does not modify the inode size, which is left to the caller.
1504 * The transaction passed to this routine must have made a permanent log
1505 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1506 * given transaction and start new ones, so make sure everything involved in
1507 * the transaction is tidy before calling here. Some transaction will be
1508 * returned to the caller to be committed. The incoming transaction must
1509 * already include the inode, and both inode locks must be held exclusively.
1510 * The inode must also be "held" within the transaction. On return the inode
1511 * will be "held" within the returned transaction. This routine does NOT
1512 * require any disk space to be reserved for it within the transaction.
1514 * If we get an error, we must return with the inode locked and linked into the
1515 * current transaction. This keeps things simple for the higher level code,
1516 * because it always knows that the inode is locked and held in the transaction
1517 * that returns to it whether errors occur or not. We don't mark the inode
1518 * dirty on error so that transactions can be easily aborted if possible.
1521 xfs_itruncate_extents(
1522 struct xfs_trans **tpp,
1523 struct xfs_inode *ip,
1525 xfs_fsize_t new_size)
1527 struct xfs_mount *mp = ip->i_mount;
1528 struct xfs_trans *tp = *tpp;
1529 xfs_bmap_free_t free_list;
1530 xfs_fsblock_t first_block;
1531 xfs_fileoff_t first_unmap_block;
1532 xfs_fileoff_t last_block;
1533 xfs_filblks_t unmap_len;
1537 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1538 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1539 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1540 ASSERT(new_size <= XFS_ISIZE(ip));
1541 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1542 ASSERT(ip->i_itemp != NULL);
1543 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1544 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1546 trace_xfs_itruncate_extents_start(ip, new_size);
1549 * Since it is possible for space to become allocated beyond
1550 * the end of the file (in a crash where the space is allocated
1551 * but the inode size is not yet updated), simply remove any
1552 * blocks which show up between the new EOF and the maximum
1553 * possible file size. If the first block to be removed is
1554 * beyond the maximum file size (ie it is the same as last_block),
1555 * then there is nothing to do.
1557 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1558 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1559 if (first_unmap_block == last_block)
1562 ASSERT(first_unmap_block < last_block);
1563 unmap_len = last_block - first_unmap_block + 1;
1565 xfs_bmap_init(&free_list, &first_block);
1566 error = xfs_bunmapi(tp, ip,
1567 first_unmap_block, unmap_len,
1568 xfs_bmapi_aflag(whichfork),
1569 XFS_ITRUNC_MAX_EXTENTS,
1570 &first_block, &free_list,
1573 goto out_bmap_cancel;
1576 * Duplicate the transaction that has the permanent
1577 * reservation and commit the old transaction.
1579 error = xfs_bmap_finish(&tp, &free_list, ip);
1581 goto out_bmap_cancel;
1583 error = xfs_trans_roll(&tp, ip);
1589 * Always re-log the inode so that our permanent transaction can keep
1590 * on rolling it forward in the log.
1592 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1594 trace_xfs_itruncate_extents_end(ip, new_size);
1601 * If the bunmapi call encounters an error, return to the caller where
1602 * the transaction can be properly aborted. We just need to make sure
1603 * we're not holding any resources that we were not when we came in.
1605 xfs_bmap_cancel(&free_list);
1613 xfs_mount_t *mp = ip->i_mount;
1616 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1619 /* If this is a read-only mount, don't do this (would generate I/O) */
1620 if (mp->m_flags & XFS_MOUNT_RDONLY)
1623 if (!XFS_FORCED_SHUTDOWN(mp)) {
1627 * If we previously truncated this file and removed old data
1628 * in the process, we want to initiate "early" writeout on
1629 * the last close. This is an attempt to combat the notorious
1630 * NULL files problem which is particularly noticeable from a
1631 * truncate down, buffered (re-)write (delalloc), followed by
1632 * a crash. What we are effectively doing here is
1633 * significantly reducing the time window where we'd otherwise
1634 * be exposed to that problem.
1636 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1638 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1639 if (ip->i_delayed_blks > 0) {
1640 error = filemap_flush(VFS_I(ip)->i_mapping);
1647 if (VFS_I(ip)->i_nlink == 0)
1650 if (xfs_can_free_eofblocks(ip, false)) {
1653 * If we can't get the iolock just skip truncating the blocks
1654 * past EOF because we could deadlock with the mmap_sem
1655 * otherwise. We'll get another chance to drop them once the
1656 * last reference to the inode is dropped, so we'll never leak
1657 * blocks permanently.
1659 * Further, check if the inode is being opened, written and
1660 * closed frequently and we have delayed allocation blocks
1661 * outstanding (e.g. streaming writes from the NFS server),
1662 * truncating the blocks past EOF will cause fragmentation to
1665 * In this case don't do the truncation, either, but we have to
1666 * be careful how we detect this case. Blocks beyond EOF show
1667 * up as i_delayed_blks even when the inode is clean, so we
1668 * need to truncate them away first before checking for a dirty
1669 * release. Hence on the first dirty close we will still remove
1670 * the speculative allocation, but after that we will leave it
1673 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1676 error = xfs_free_eofblocks(mp, ip, true);
1677 if (error && error != -EAGAIN)
1680 /* delalloc blocks after truncation means it really is dirty */
1681 if (ip->i_delayed_blks)
1682 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1688 * xfs_inactive_truncate
1690 * Called to perform a truncate when an inode becomes unlinked.
1693 xfs_inactive_truncate(
1694 struct xfs_inode *ip)
1696 struct xfs_mount *mp = ip->i_mount;
1697 struct xfs_trans *tp;
1700 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1702 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1706 xfs_ilock(ip, XFS_ILOCK_EXCL);
1707 xfs_trans_ijoin(tp, ip, 0);
1710 * Log the inode size first to prevent stale data exposure in the event
1711 * of a system crash before the truncate completes. See the related
1712 * comment in xfs_setattr_size() for details.
1714 ip->i_d.di_size = 0;
1715 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1717 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1719 goto error_trans_cancel;
1721 ASSERT(ip->i_d.di_nextents == 0);
1723 error = xfs_trans_commit(tp);
1727 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1731 xfs_trans_cancel(tp);
1733 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1738 * xfs_inactive_ifree()
1740 * Perform the inode free when an inode is unlinked.
1744 struct xfs_inode *ip)
1746 xfs_bmap_free_t free_list;
1747 xfs_fsblock_t first_block;
1748 struct xfs_mount *mp = ip->i_mount;
1749 struct xfs_trans *tp;
1753 * The ifree transaction might need to allocate blocks for record
1754 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1755 * allow ifree to dip into the reserved block pool if necessary.
1757 * Freeing large sets of inodes generally means freeing inode chunks,
1758 * directory and file data blocks, so this should be relatively safe.
1759 * Only under severe circumstances should it be possible to free enough
1760 * inodes to exhaust the reserve block pool via finobt expansion while
1761 * at the same time not creating free space in the filesystem.
1763 * Send a warning if the reservation does happen to fail, as the inode
1764 * now remains allocated and sits on the unlinked list until the fs is
1767 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1768 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
1770 if (error == -ENOSPC) {
1771 xfs_warn_ratelimited(mp,
1772 "Failed to remove inode(s) from unlinked list. "
1773 "Please free space, unmount and run xfs_repair.");
1775 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1780 xfs_ilock(ip, XFS_ILOCK_EXCL);
1781 xfs_trans_ijoin(tp, ip, 0);
1783 xfs_bmap_init(&free_list, &first_block);
1784 error = xfs_ifree(tp, ip, &free_list);
1787 * If we fail to free the inode, shut down. The cancel
1788 * might do that, we need to make sure. Otherwise the
1789 * inode might be lost for a long time or forever.
1791 if (!XFS_FORCED_SHUTDOWN(mp)) {
1792 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1794 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1796 xfs_trans_cancel(tp);
1797 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1802 * Credit the quota account(s). The inode is gone.
1804 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1807 * Just ignore errors at this point. There is nothing we can do except
1808 * to try to keep going. Make sure it's not a silent error.
1810 error = xfs_bmap_finish(&tp, &free_list, NULL);
1812 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1814 xfs_bmap_cancel(&free_list);
1816 error = xfs_trans_commit(tp);
1818 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1821 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1828 * This is called when the vnode reference count for the vnode
1829 * goes to zero. If the file has been unlinked, then it must
1830 * now be truncated. Also, we clear all of the read-ahead state
1831 * kept for the inode here since the file is now closed.
1837 struct xfs_mount *mp;
1842 * If the inode is already free, then there can be nothing
1845 if (VFS_I(ip)->i_mode == 0) {
1846 ASSERT(ip->i_df.if_real_bytes == 0);
1847 ASSERT(ip->i_df.if_broot_bytes == 0);
1853 /* If this is a read-only mount, don't do this (would generate I/O) */
1854 if (mp->m_flags & XFS_MOUNT_RDONLY)
1857 if (VFS_I(ip)->i_nlink != 0) {
1859 * force is true because we are evicting an inode from the
1860 * cache. Post-eof blocks must be freed, lest we end up with
1861 * broken free space accounting.
1863 if (xfs_can_free_eofblocks(ip, true))
1864 xfs_free_eofblocks(mp, ip, false);
1869 if (S_ISREG(VFS_I(ip)->i_mode) &&
1870 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1871 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1874 error = xfs_qm_dqattach(ip, 0);
1878 if (S_ISLNK(VFS_I(ip)->i_mode))
1879 error = xfs_inactive_symlink(ip);
1881 error = xfs_inactive_truncate(ip);
1886 * If there are attributes associated with the file then blow them away
1887 * now. The code calls a routine that recursively deconstructs the
1888 * attribute fork. If also blows away the in-core attribute fork.
1890 if (XFS_IFORK_Q(ip)) {
1891 error = xfs_attr_inactive(ip);
1897 ASSERT(ip->i_d.di_anextents == 0);
1898 ASSERT(ip->i_d.di_forkoff == 0);
1903 error = xfs_inactive_ifree(ip);
1908 * Release the dquots held by inode, if any.
1910 xfs_qm_dqdetach(ip);
1914 * This is called when the inode's link count goes to 0 or we are creating a
1915 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1916 * set to true as the link count is dropped to zero by the VFS after we've
1917 * created the file successfully, so we have to add it to the unlinked list
1918 * while the link count is non-zero.
1920 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1921 * list when the inode is freed.
1925 struct xfs_trans *tp,
1926 struct xfs_inode *ip)
1928 xfs_mount_t *mp = tp->t_mountp;
1938 ASSERT(VFS_I(ip)->i_mode != 0);
1941 * Get the agi buffer first. It ensures lock ordering
1944 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1947 agi = XFS_BUF_TO_AGI(agibp);
1950 * Get the index into the agi hash table for the
1951 * list this inode will go on.
1953 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1955 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1956 ASSERT(agi->agi_unlinked[bucket_index]);
1957 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1959 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1961 * There is already another inode in the bucket we need
1962 * to add ourselves to. Add us at the front of the list.
1963 * Here we put the head pointer into our next pointer,
1964 * and then we fall through to point the head at us.
1966 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1971 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1972 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1973 offset = ip->i_imap.im_boffset +
1974 offsetof(xfs_dinode_t, di_next_unlinked);
1976 /* need to recalc the inode CRC if appropriate */
1977 xfs_dinode_calc_crc(mp, dip);
1979 xfs_trans_inode_buf(tp, ibp);
1980 xfs_trans_log_buf(tp, ibp, offset,
1981 (offset + sizeof(xfs_agino_t) - 1));
1982 xfs_inobp_check(mp, ibp);
1986 * Point the bucket head pointer at the inode being inserted.
1989 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1990 offset = offsetof(xfs_agi_t, agi_unlinked) +
1991 (sizeof(xfs_agino_t) * bucket_index);
1992 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
1993 xfs_trans_log_buf(tp, agibp, offset,
1994 (offset + sizeof(xfs_agino_t) - 1));
1999 * Pull the on-disk inode from the AGI unlinked list.
2012 xfs_agnumber_t agno;
2014 xfs_agino_t next_agino;
2015 xfs_buf_t *last_ibp;
2016 xfs_dinode_t *last_dip = NULL;
2018 int offset, last_offset = 0;
2022 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2025 * Get the agi buffer first. It ensures lock ordering
2028 error = xfs_read_agi(mp, tp, agno, &agibp);
2032 agi = XFS_BUF_TO_AGI(agibp);
2035 * Get the index into the agi hash table for the
2036 * list this inode will go on.
2038 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2040 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2041 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2042 ASSERT(agi->agi_unlinked[bucket_index]);
2044 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2046 * We're at the head of the list. Get the inode's on-disk
2047 * buffer to see if there is anyone after us on the list.
2048 * Only modify our next pointer if it is not already NULLAGINO.
2049 * This saves us the overhead of dealing with the buffer when
2050 * there is no need to change it.
2052 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2055 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2059 next_agino = be32_to_cpu(dip->di_next_unlinked);
2060 ASSERT(next_agino != 0);
2061 if (next_agino != NULLAGINO) {
2062 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2063 offset = ip->i_imap.im_boffset +
2064 offsetof(xfs_dinode_t, di_next_unlinked);
2066 /* need to recalc the inode CRC if appropriate */
2067 xfs_dinode_calc_crc(mp, dip);
2069 xfs_trans_inode_buf(tp, ibp);
2070 xfs_trans_log_buf(tp, ibp, offset,
2071 (offset + sizeof(xfs_agino_t) - 1));
2072 xfs_inobp_check(mp, ibp);
2074 xfs_trans_brelse(tp, ibp);
2077 * Point the bucket head pointer at the next inode.
2079 ASSERT(next_agino != 0);
2080 ASSERT(next_agino != agino);
2081 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2082 offset = offsetof(xfs_agi_t, agi_unlinked) +
2083 (sizeof(xfs_agino_t) * bucket_index);
2084 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2085 xfs_trans_log_buf(tp, agibp, offset,
2086 (offset + sizeof(xfs_agino_t) - 1));
2089 * We need to search the list for the inode being freed.
2091 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2093 while (next_agino != agino) {
2094 struct xfs_imap imap;
2097 xfs_trans_brelse(tp, last_ibp);
2100 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2102 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2105 "%s: xfs_imap returned error %d.",
2110 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2114 "%s: xfs_imap_to_bp returned error %d.",
2119 last_offset = imap.im_boffset;
2120 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2121 ASSERT(next_agino != NULLAGINO);
2122 ASSERT(next_agino != 0);
2126 * Now last_ibp points to the buffer previous to us on the
2127 * unlinked list. Pull us from the list.
2129 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2132 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2136 next_agino = be32_to_cpu(dip->di_next_unlinked);
2137 ASSERT(next_agino != 0);
2138 ASSERT(next_agino != agino);
2139 if (next_agino != NULLAGINO) {
2140 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2141 offset = ip->i_imap.im_boffset +
2142 offsetof(xfs_dinode_t, di_next_unlinked);
2144 /* need to recalc the inode CRC if appropriate */
2145 xfs_dinode_calc_crc(mp, dip);
2147 xfs_trans_inode_buf(tp, ibp);
2148 xfs_trans_log_buf(tp, ibp, offset,
2149 (offset + sizeof(xfs_agino_t) - 1));
2150 xfs_inobp_check(mp, ibp);
2152 xfs_trans_brelse(tp, ibp);
2155 * Point the previous inode on the list to the next inode.
2157 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2158 ASSERT(next_agino != 0);
2159 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2161 /* need to recalc the inode CRC if appropriate */
2162 xfs_dinode_calc_crc(mp, last_dip);
2164 xfs_trans_inode_buf(tp, last_ibp);
2165 xfs_trans_log_buf(tp, last_ibp, offset,
2166 (offset + sizeof(xfs_agino_t) - 1));
2167 xfs_inobp_check(mp, last_ibp);
2173 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2174 * inodes that are in memory - they all must be marked stale and attached to
2175 * the cluster buffer.
2179 xfs_inode_t *free_ip,
2181 struct xfs_icluster *xic)
2183 xfs_mount_t *mp = free_ip->i_mount;
2184 int blks_per_cluster;
2185 int inodes_per_cluster;
2192 xfs_inode_log_item_t *iip;
2193 xfs_log_item_t *lip;
2194 struct xfs_perag *pag;
2197 inum = xic->first_ino;
2198 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2199 blks_per_cluster = xfs_icluster_size_fsb(mp);
2200 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2201 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2203 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2205 * The allocation bitmap tells us which inodes of the chunk were
2206 * physically allocated. Skip the cluster if an inode falls into
2209 ioffset = inum - xic->first_ino;
2210 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2211 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2215 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2216 XFS_INO_TO_AGBNO(mp, inum));
2219 * We obtain and lock the backing buffer first in the process
2220 * here, as we have to ensure that any dirty inode that we
2221 * can't get the flush lock on is attached to the buffer.
2222 * If we scan the in-memory inodes first, then buffer IO can
2223 * complete before we get a lock on it, and hence we may fail
2224 * to mark all the active inodes on the buffer stale.
2226 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2227 mp->m_bsize * blks_per_cluster,
2234 * This buffer may not have been correctly initialised as we
2235 * didn't read it from disk. That's not important because we are
2236 * only using to mark the buffer as stale in the log, and to
2237 * attach stale cached inodes on it. That means it will never be
2238 * dispatched for IO. If it is, we want to know about it, and we
2239 * want it to fail. We can acheive this by adding a write
2240 * verifier to the buffer.
2242 bp->b_ops = &xfs_inode_buf_ops;
2245 * Walk the inodes already attached to the buffer and mark them
2246 * stale. These will all have the flush locks held, so an
2247 * in-memory inode walk can't lock them. By marking them all
2248 * stale first, we will not attempt to lock them in the loop
2249 * below as the XFS_ISTALE flag will be set.
2253 if (lip->li_type == XFS_LI_INODE) {
2254 iip = (xfs_inode_log_item_t *)lip;
2255 ASSERT(iip->ili_logged == 1);
2256 lip->li_cb = xfs_istale_done;
2257 xfs_trans_ail_copy_lsn(mp->m_ail,
2258 &iip->ili_flush_lsn,
2259 &iip->ili_item.li_lsn);
2260 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2262 lip = lip->li_bio_list;
2267 * For each inode in memory attempt to add it to the inode
2268 * buffer and set it up for being staled on buffer IO
2269 * completion. This is safe as we've locked out tail pushing
2270 * and flushing by locking the buffer.
2272 * We have already marked every inode that was part of a
2273 * transaction stale above, which means there is no point in
2274 * even trying to lock them.
2276 for (i = 0; i < inodes_per_cluster; i++) {
2279 ip = radix_tree_lookup(&pag->pag_ici_root,
2280 XFS_INO_TO_AGINO(mp, (inum + i)));
2282 /* Inode not in memory, nothing to do */
2289 * because this is an RCU protected lookup, we could
2290 * find a recently freed or even reallocated inode
2291 * during the lookup. We need to check under the
2292 * i_flags_lock for a valid inode here. Skip it if it
2293 * is not valid, the wrong inode or stale.
2295 spin_lock(&ip->i_flags_lock);
2296 if (ip->i_ino != inum + i ||
2297 __xfs_iflags_test(ip, XFS_ISTALE)) {
2298 spin_unlock(&ip->i_flags_lock);
2302 spin_unlock(&ip->i_flags_lock);
2305 * Don't try to lock/unlock the current inode, but we
2306 * _cannot_ skip the other inodes that we did not find
2307 * in the list attached to the buffer and are not
2308 * already marked stale. If we can't lock it, back off
2311 if (ip != free_ip &&
2312 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2320 xfs_iflags_set(ip, XFS_ISTALE);
2323 * we don't need to attach clean inodes or those only
2324 * with unlogged changes (which we throw away, anyway).
2327 if (!iip || xfs_inode_clean(ip)) {
2328 ASSERT(ip != free_ip);
2330 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2334 iip->ili_last_fields = iip->ili_fields;
2335 iip->ili_fields = 0;
2336 iip->ili_fsync_fields = 0;
2337 iip->ili_logged = 1;
2338 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2339 &iip->ili_item.li_lsn);
2341 xfs_buf_attach_iodone(bp, xfs_istale_done,
2345 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2348 xfs_trans_stale_inode_buf(tp, bp);
2349 xfs_trans_binval(tp, bp);
2357 * This is called to return an inode to the inode free list.
2358 * The inode should already be truncated to 0 length and have
2359 * no pages associated with it. This routine also assumes that
2360 * the inode is already a part of the transaction.
2362 * The on-disk copy of the inode will have been added to the list
2363 * of unlinked inodes in the AGI. We need to remove the inode from
2364 * that list atomically with respect to freeing it here.
2370 xfs_bmap_free_t *flist)
2373 struct xfs_icluster xic = { 0 };
2375 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2376 ASSERT(VFS_I(ip)->i_nlink == 0);
2377 ASSERT(ip->i_d.di_nextents == 0);
2378 ASSERT(ip->i_d.di_anextents == 0);
2379 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2380 ASSERT(ip->i_d.di_nblocks == 0);
2383 * Pull the on-disk inode from the AGI unlinked list.
2385 error = xfs_iunlink_remove(tp, ip);
2389 error = xfs_difree(tp, ip->i_ino, flist, &xic);
2393 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2394 ip->i_d.di_flags = 0;
2395 ip->i_d.di_dmevmask = 0;
2396 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2397 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2398 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2400 * Bump the generation count so no one will be confused
2401 * by reincarnations of this inode.
2403 VFS_I(ip)->i_generation++;
2404 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2407 error = xfs_ifree_cluster(ip, tp, &xic);
2413 * This is called to unpin an inode. The caller must have the inode locked
2414 * in at least shared mode so that the buffer cannot be subsequently pinned
2415 * once someone is waiting for it to be unpinned.
2419 struct xfs_inode *ip)
2421 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2423 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2425 /* Give the log a push to start the unpinning I/O */
2426 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2432 struct xfs_inode *ip)
2434 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2435 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2440 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2441 if (xfs_ipincount(ip))
2443 } while (xfs_ipincount(ip));
2444 finish_wait(wq, &wait.wait);
2449 struct xfs_inode *ip)
2451 if (xfs_ipincount(ip))
2452 __xfs_iunpin_wait(ip);
2456 * Removing an inode from the namespace involves removing the directory entry
2457 * and dropping the link count on the inode. Removing the directory entry can
2458 * result in locking an AGF (directory blocks were freed) and removing a link
2459 * count can result in placing the inode on an unlinked list which results in
2462 * The big problem here is that we have an ordering constraint on AGF and AGI
2463 * locking - inode allocation locks the AGI, then can allocate a new extent for
2464 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2465 * removes the inode from the unlinked list, requiring that we lock the AGI
2466 * first, and then freeing the inode can result in an inode chunk being freed
2467 * and hence freeing disk space requiring that we lock an AGF.
2469 * Hence the ordering that is imposed by other parts of the code is AGI before
2470 * AGF. This means we cannot remove the directory entry before we drop the inode
2471 * reference count and put it on the unlinked list as this results in a lock
2472 * order of AGF then AGI, and this can deadlock against inode allocation and
2473 * freeing. Therefore we must drop the link counts before we remove the
2476 * This is still safe from a transactional point of view - it is not until we
2477 * get to xfs_bmap_finish() that we have the possibility of multiple
2478 * transactions in this operation. Hence as long as we remove the directory
2479 * entry and drop the link count in the first transaction of the remove
2480 * operation, there are no transactional constraints on the ordering here.
2485 struct xfs_name *name,
2488 xfs_mount_t *mp = dp->i_mount;
2489 xfs_trans_t *tp = NULL;
2490 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2492 xfs_bmap_free_t free_list;
2493 xfs_fsblock_t first_block;
2496 trace_xfs_remove(dp, name);
2498 if (XFS_FORCED_SHUTDOWN(mp))
2501 error = xfs_qm_dqattach(dp, 0);
2505 error = xfs_qm_dqattach(ip, 0);
2510 * We try to get the real space reservation first,
2511 * allowing for directory btree deletion(s) implying
2512 * possible bmap insert(s). If we can't get the space
2513 * reservation then we use 0 instead, and avoid the bmap
2514 * btree insert(s) in the directory code by, if the bmap
2515 * insert tries to happen, instead trimming the LAST
2516 * block from the directory.
2518 resblks = XFS_REMOVE_SPACE_RES(mp);
2519 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2520 if (error == -ENOSPC) {
2522 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2526 ASSERT(error != -ENOSPC);
2530 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2531 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2533 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2534 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2537 * If we're removing a directory perform some additional validation.
2540 ASSERT(VFS_I(ip)->i_nlink >= 2);
2541 if (VFS_I(ip)->i_nlink != 2) {
2543 goto out_trans_cancel;
2545 if (!xfs_dir_isempty(ip)) {
2547 goto out_trans_cancel;
2550 /* Drop the link from ip's "..". */
2551 error = xfs_droplink(tp, dp);
2553 goto out_trans_cancel;
2555 /* Drop the "." link from ip to self. */
2556 error = xfs_droplink(tp, ip);
2558 goto out_trans_cancel;
2561 * When removing a non-directory we need to log the parent
2562 * inode here. For a directory this is done implicitly
2563 * by the xfs_droplink call for the ".." entry.
2565 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2567 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2569 /* Drop the link from dp to ip. */
2570 error = xfs_droplink(tp, ip);
2572 goto out_trans_cancel;
2574 xfs_bmap_init(&free_list, &first_block);
2575 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2576 &first_block, &free_list, resblks);
2578 ASSERT(error != -ENOENT);
2579 goto out_bmap_cancel;
2583 * If this is a synchronous mount, make sure that the
2584 * remove transaction goes to disk before returning to
2587 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2588 xfs_trans_set_sync(tp);
2590 error = xfs_bmap_finish(&tp, &free_list, NULL);
2592 goto out_bmap_cancel;
2594 error = xfs_trans_commit(tp);
2598 if (is_dir && xfs_inode_is_filestream(ip))
2599 xfs_filestream_deassociate(ip);
2604 xfs_bmap_cancel(&free_list);
2606 xfs_trans_cancel(tp);
2612 * Enter all inodes for a rename transaction into a sorted array.
2614 #define __XFS_SORT_INODES 5
2616 xfs_sort_for_rename(
2617 struct xfs_inode *dp1, /* in: old (source) directory inode */
2618 struct xfs_inode *dp2, /* in: new (target) directory inode */
2619 struct xfs_inode *ip1, /* in: inode of old entry */
2620 struct xfs_inode *ip2, /* in: inode of new entry */
2621 struct xfs_inode *wip, /* in: whiteout inode */
2622 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2623 int *num_inodes) /* in/out: inodes in array */
2627 ASSERT(*num_inodes == __XFS_SORT_INODES);
2628 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2631 * i_tab contains a list of pointers to inodes. We initialize
2632 * the table here & we'll sort it. We will then use it to
2633 * order the acquisition of the inode locks.
2635 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2648 * Sort the elements via bubble sort. (Remember, there are at
2649 * most 5 elements to sort, so this is adequate.)
2651 for (i = 0; i < *num_inodes; i++) {
2652 for (j = 1; j < *num_inodes; j++) {
2653 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2654 struct xfs_inode *temp = i_tab[j];
2655 i_tab[j] = i_tab[j-1];
2664 struct xfs_trans *tp,
2665 struct xfs_bmap_free *free_list)
2670 * If this is a synchronous mount, make sure that the rename transaction
2671 * goes to disk before returning to the user.
2673 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2674 xfs_trans_set_sync(tp);
2676 error = xfs_bmap_finish(&tp, free_list, NULL);
2678 xfs_bmap_cancel(free_list);
2679 xfs_trans_cancel(tp);
2683 return xfs_trans_commit(tp);
2687 * xfs_cross_rename()
2689 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2693 struct xfs_trans *tp,
2694 struct xfs_inode *dp1,
2695 struct xfs_name *name1,
2696 struct xfs_inode *ip1,
2697 struct xfs_inode *dp2,
2698 struct xfs_name *name2,
2699 struct xfs_inode *ip2,
2700 struct xfs_bmap_free *free_list,
2701 xfs_fsblock_t *first_block,
2709 /* Swap inode number for dirent in first parent */
2710 error = xfs_dir_replace(tp, dp1, name1,
2712 first_block, free_list, spaceres);
2714 goto out_trans_abort;
2716 /* Swap inode number for dirent in second parent */
2717 error = xfs_dir_replace(tp, dp2, name2,
2719 first_block, free_list, spaceres);
2721 goto out_trans_abort;
2724 * If we're renaming one or more directories across different parents,
2725 * update the respective ".." entries (and link counts) to match the new
2729 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2731 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2732 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2733 dp1->i_ino, first_block,
2734 free_list, spaceres);
2736 goto out_trans_abort;
2738 /* transfer ip2 ".." reference to dp1 */
2739 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2740 error = xfs_droplink(tp, dp2);
2742 goto out_trans_abort;
2743 error = xfs_bumplink(tp, dp1);
2745 goto out_trans_abort;
2749 * Although ip1 isn't changed here, userspace needs
2750 * to be warned about the change, so that applications
2751 * relying on it (like backup ones), will properly
2754 ip1_flags |= XFS_ICHGTIME_CHG;
2755 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2758 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2759 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2760 dp2->i_ino, first_block,
2761 free_list, spaceres);
2763 goto out_trans_abort;
2765 /* transfer ip1 ".." reference to dp2 */
2766 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2767 error = xfs_droplink(tp, dp1);
2769 goto out_trans_abort;
2770 error = xfs_bumplink(tp, dp2);
2772 goto out_trans_abort;
2776 * Although ip2 isn't changed here, userspace needs
2777 * to be warned about the change, so that applications
2778 * relying on it (like backup ones), will properly
2781 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2782 ip2_flags |= XFS_ICHGTIME_CHG;
2787 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2788 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2791 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2792 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2795 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2796 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2798 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2799 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2800 return xfs_finish_rename(tp, free_list);
2803 xfs_bmap_cancel(free_list);
2804 xfs_trans_cancel(tp);
2809 * xfs_rename_alloc_whiteout()
2811 * Return a referenced, unlinked, unlocked inode that that can be used as a
2812 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2813 * crash between allocating the inode and linking it into the rename transaction
2814 * recovery will free the inode and we won't leak it.
2817 xfs_rename_alloc_whiteout(
2818 struct xfs_inode *dp,
2819 struct xfs_inode **wip)
2821 struct xfs_inode *tmpfile;
2824 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2829 * Prepare the tmpfile inode as if it were created through the VFS.
2830 * Otherwise, the link increment paths will complain about nlink 0->1.
2831 * Drop the link count as done by d_tmpfile(), complete the inode setup
2832 * and flag it as linkable.
2834 drop_nlink(VFS_I(tmpfile));
2835 xfs_setup_iops(tmpfile);
2836 xfs_finish_inode_setup(tmpfile);
2837 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2848 struct xfs_inode *src_dp,
2849 struct xfs_name *src_name,
2850 struct xfs_inode *src_ip,
2851 struct xfs_inode *target_dp,
2852 struct xfs_name *target_name,
2853 struct xfs_inode *target_ip,
2856 struct xfs_mount *mp = src_dp->i_mount;
2857 struct xfs_trans *tp;
2858 struct xfs_bmap_free free_list;
2859 xfs_fsblock_t first_block;
2860 struct xfs_inode *wip = NULL; /* whiteout inode */
2861 struct xfs_inode *inodes[__XFS_SORT_INODES];
2862 int num_inodes = __XFS_SORT_INODES;
2863 bool new_parent = (src_dp != target_dp);
2864 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2868 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2870 if ((flags & RENAME_EXCHANGE) && !target_ip)
2874 * If we are doing a whiteout operation, allocate the whiteout inode
2875 * we will be placing at the target and ensure the type is set
2878 if (flags & RENAME_WHITEOUT) {
2879 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2880 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2884 /* setup target dirent info as whiteout */
2885 src_name->type = XFS_DIR3_FT_CHRDEV;
2888 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2889 inodes, &num_inodes);
2891 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2892 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2893 if (error == -ENOSPC) {
2895 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2899 goto out_release_wip;
2902 * Attach the dquots to the inodes
2904 error = xfs_qm_vop_rename_dqattach(inodes);
2906 goto out_trans_cancel;
2909 * Lock all the participating inodes. Depending upon whether
2910 * the target_name exists in the target directory, and
2911 * whether the target directory is the same as the source
2912 * directory, we can lock from 2 to 4 inodes.
2915 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2917 xfs_lock_two_inodes(src_dp, target_dp,
2918 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2920 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2923 * Join all the inodes to the transaction. From this point on,
2924 * we can rely on either trans_commit or trans_cancel to unlock
2927 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2929 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2930 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2932 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2934 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2937 * If we are using project inheritance, we only allow renames
2938 * into our tree when the project IDs are the same; else the
2939 * tree quota mechanism would be circumvented.
2941 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2942 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2944 goto out_trans_cancel;
2947 xfs_bmap_init(&free_list, &first_block);
2949 /* RENAME_EXCHANGE is unique from here on. */
2950 if (flags & RENAME_EXCHANGE)
2951 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2952 target_dp, target_name, target_ip,
2953 &free_list, &first_block, spaceres);
2956 * Set up the target.
2958 if (target_ip == NULL) {
2960 * If there's no space reservation, check the entry will
2961 * fit before actually inserting it.
2964 error = xfs_dir_canenter(tp, target_dp, target_name);
2966 goto out_trans_cancel;
2969 * If target does not exist and the rename crosses
2970 * directories, adjust the target directory link count
2971 * to account for the ".." reference from the new entry.
2973 error = xfs_dir_createname(tp, target_dp, target_name,
2974 src_ip->i_ino, &first_block,
2975 &free_list, spaceres);
2977 goto out_bmap_cancel;
2979 xfs_trans_ichgtime(tp, target_dp,
2980 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2982 if (new_parent && src_is_directory) {
2983 error = xfs_bumplink(tp, target_dp);
2985 goto out_bmap_cancel;
2987 } else { /* target_ip != NULL */
2989 * If target exists and it's a directory, check that both
2990 * target and source are directories and that target can be
2991 * destroyed, or that neither is a directory.
2993 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
2995 * Make sure target dir is empty.
2997 if (!(xfs_dir_isempty(target_ip)) ||
2998 (VFS_I(target_ip)->i_nlink > 2)) {
3000 goto out_trans_cancel;
3005 * Link the source inode under the target name.
3006 * If the source inode is a directory and we are moving
3007 * it across directories, its ".." entry will be
3008 * inconsistent until we replace that down below.
3010 * In case there is already an entry with the same
3011 * name at the destination directory, remove it first.
3013 error = xfs_dir_replace(tp, target_dp, target_name,
3015 &first_block, &free_list, spaceres);
3017 goto out_bmap_cancel;
3019 xfs_trans_ichgtime(tp, target_dp,
3020 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3023 * Decrement the link count on the target since the target
3024 * dir no longer points to it.
3026 error = xfs_droplink(tp, target_ip);
3028 goto out_bmap_cancel;
3030 if (src_is_directory) {
3032 * Drop the link from the old "." entry.
3034 error = xfs_droplink(tp, target_ip);
3036 goto out_bmap_cancel;
3038 } /* target_ip != NULL */
3041 * Remove the source.
3043 if (new_parent && src_is_directory) {
3045 * Rewrite the ".." entry to point to the new
3048 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3050 &first_block, &free_list, spaceres);
3051 ASSERT(error != -EEXIST);
3053 goto out_bmap_cancel;
3057 * We always want to hit the ctime on the source inode.
3059 * This isn't strictly required by the standards since the source
3060 * inode isn't really being changed, but old unix file systems did
3061 * it and some incremental backup programs won't work without it.
3063 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3064 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3067 * Adjust the link count on src_dp. This is necessary when
3068 * renaming a directory, either within one parent when
3069 * the target existed, or across two parent directories.
3071 if (src_is_directory && (new_parent || target_ip != NULL)) {
3074 * Decrement link count on src_directory since the
3075 * entry that's moved no longer points to it.
3077 error = xfs_droplink(tp, src_dp);
3079 goto out_bmap_cancel;
3083 * For whiteouts, we only need to update the source dirent with the
3084 * inode number of the whiteout inode rather than removing it
3088 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3089 &first_block, &free_list, spaceres);
3091 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3092 &first_block, &free_list, spaceres);
3094 goto out_bmap_cancel;
3097 * For whiteouts, we need to bump the link count on the whiteout inode.
3098 * This means that failures all the way up to this point leave the inode
3099 * on the unlinked list and so cleanup is a simple matter of dropping
3100 * the remaining reference to it. If we fail here after bumping the link
3101 * count, we're shutting down the filesystem so we'll never see the
3102 * intermediate state on disk.
3105 ASSERT(VFS_I(wip)->i_nlink == 0);
3106 error = xfs_bumplink(tp, wip);
3108 goto out_bmap_cancel;
3109 error = xfs_iunlink_remove(tp, wip);
3111 goto out_bmap_cancel;
3112 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3115 * Now we have a real link, clear the "I'm a tmpfile" state
3116 * flag from the inode so it doesn't accidentally get misused in
3119 VFS_I(wip)->i_state &= ~I_LINKABLE;
3122 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3123 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3125 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3127 error = xfs_finish_rename(tp, &free_list);
3133 xfs_bmap_cancel(&free_list);
3135 xfs_trans_cancel(tp);
3144 struct xfs_inode *ip,
3147 struct xfs_mount *mp = ip->i_mount;
3148 struct xfs_perag *pag;
3149 unsigned long first_index, mask;
3150 unsigned long inodes_per_cluster;
3152 struct xfs_inode **cilist;
3153 struct xfs_inode *cip;
3159 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3161 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3162 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3163 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3167 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3168 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3170 /* really need a gang lookup range call here */
3171 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3172 first_index, inodes_per_cluster);
3176 for (i = 0; i < nr_found; i++) {
3182 * because this is an RCU protected lookup, we could find a
3183 * recently freed or even reallocated inode during the lookup.
3184 * We need to check under the i_flags_lock for a valid inode
3185 * here. Skip it if it is not valid or the wrong inode.
3187 spin_lock(&cip->i_flags_lock);
3189 __xfs_iflags_test(cip, XFS_ISTALE)) {
3190 spin_unlock(&cip->i_flags_lock);
3195 * Once we fall off the end of the cluster, no point checking
3196 * any more inodes in the list because they will also all be
3197 * outside the cluster.
3199 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3200 spin_unlock(&cip->i_flags_lock);
3203 spin_unlock(&cip->i_flags_lock);
3206 * Do an un-protected check to see if the inode is dirty and
3207 * is a candidate for flushing. These checks will be repeated
3208 * later after the appropriate locks are acquired.
3210 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3214 * Try to get locks. If any are unavailable or it is pinned,
3215 * then this inode cannot be flushed and is skipped.
3218 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3220 if (!xfs_iflock_nowait(cip)) {
3221 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3224 if (xfs_ipincount(cip)) {
3226 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3232 * Check the inode number again, just to be certain we are not
3233 * racing with freeing in xfs_reclaim_inode(). See the comments
3234 * in that function for more information as to why the initial
3235 * check is not sufficient.
3239 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3244 * arriving here means that this inode can be flushed. First
3245 * re-check that it's dirty before flushing.
3247 if (!xfs_inode_clean(cip)) {
3249 error = xfs_iflush_int(cip, bp);
3251 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3252 goto cluster_corrupt_out;
3258 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3262 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3263 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3274 cluster_corrupt_out:
3276 * Corruption detected in the clustering loop. Invalidate the
3277 * inode buffer and shut down the filesystem.
3281 * Clean up the buffer. If it was delwri, just release it --
3282 * brelse can handle it with no problems. If not, shut down the
3283 * filesystem before releasing the buffer.
3285 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3289 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3291 if (!bufwasdelwri) {
3293 * Just like incore_relse: if we have b_iodone functions,
3294 * mark the buffer as an error and call them. Otherwise
3295 * mark it as stale and brelse.
3298 bp->b_flags &= ~XBF_DONE;
3300 xfs_buf_ioerror(bp, -EIO);
3309 * Unlocks the flush lock
3311 xfs_iflush_abort(cip, false);
3314 return -EFSCORRUPTED;
3318 * Flush dirty inode metadata into the backing buffer.
3320 * The caller must have the inode lock and the inode flush lock held. The
3321 * inode lock will still be held upon return to the caller, and the inode
3322 * flush lock will be released after the inode has reached the disk.
3324 * The caller must write out the buffer returned in *bpp and release it.
3328 struct xfs_inode *ip,
3329 struct xfs_buf **bpp)
3331 struct xfs_mount *mp = ip->i_mount;
3332 struct xfs_buf *bp = NULL;
3333 struct xfs_dinode *dip;
3336 XFS_STATS_INC(mp, xs_iflush_count);
3338 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3339 ASSERT(xfs_isiflocked(ip));
3340 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3341 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3345 xfs_iunpin_wait(ip);
3348 * For stale inodes we cannot rely on the backing buffer remaining
3349 * stale in cache for the remaining life of the stale inode and so
3350 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3351 * inodes below. We have to check this after ensuring the inode is
3352 * unpinned so that it is safe to reclaim the stale inode after the
3355 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3361 * This may have been unpinned because the filesystem is shutting
3362 * down forcibly. If that's the case we must not write this inode
3363 * to disk, because the log record didn't make it to disk.
3365 * We also have to remove the log item from the AIL in this case,
3366 * as we wait for an empty AIL as part of the unmount process.
3368 if (XFS_FORCED_SHUTDOWN(mp)) {
3374 * Get the buffer containing the on-disk inode. We are doing a try-lock
3375 * operation here, so we may get an EAGAIN error. In that case, we
3376 * simply want to return with the inode still dirty.
3378 * If we get any other error, we effectively have a corruption situation
3379 * and we cannot flush the inode, so we treat it the same as failing
3382 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3384 if (error == -EAGAIN) {
3392 * First flush out the inode that xfs_iflush was called with.
3394 error = xfs_iflush_int(ip, bp);
3399 * If the buffer is pinned then push on the log now so we won't
3400 * get stuck waiting in the write for too long.
3402 if (xfs_buf_ispinned(bp))
3403 xfs_log_force(mp, 0);
3407 * see if other inodes can be gathered into this write
3409 error = xfs_iflush_cluster(ip, bp);
3411 goto cluster_corrupt_out;
3419 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3420 cluster_corrupt_out:
3421 error = -EFSCORRUPTED;
3424 * Unlocks the flush lock
3426 xfs_iflush_abort(ip, false);
3432 struct xfs_inode *ip,
3435 struct xfs_inode_log_item *iip = ip->i_itemp;
3436 struct xfs_dinode *dip;
3437 struct xfs_mount *mp = ip->i_mount;
3439 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3440 ASSERT(xfs_isiflocked(ip));
3441 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3442 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3443 ASSERT(iip != NULL && iip->ili_fields != 0);
3444 ASSERT(ip->i_d.di_version > 1);
3446 /* set *dip = inode's place in the buffer */
3447 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3449 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3450 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3451 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3452 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3453 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3456 if (S_ISREG(VFS_I(ip)->i_mode)) {
3458 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3459 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3460 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3461 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3462 "%s: Bad regular inode %Lu, ptr 0x%p",
3463 __func__, ip->i_ino, ip);
3466 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3468 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3469 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3470 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3471 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3472 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3473 "%s: Bad directory inode %Lu, ptr 0x%p",
3474 __func__, ip->i_ino, ip);
3478 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3479 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3480 XFS_RANDOM_IFLUSH_5)) {
3481 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3482 "%s: detected corrupt incore inode %Lu, "
3483 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3484 __func__, ip->i_ino,
3485 ip->i_d.di_nextents + ip->i_d.di_anextents,
3486 ip->i_d.di_nblocks, ip);
3489 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3490 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3491 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3492 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3493 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3498 * Inode item log recovery for v2 inodes are dependent on the
3499 * di_flushiter count for correct sequencing. We bump the flush
3500 * iteration count so we can detect flushes which postdate a log record
3501 * during recovery. This is redundant as we now log every change and
3502 * hence this can't happen but we need to still do it to ensure
3503 * backwards compatibility with old kernels that predate logging all
3506 if (ip->i_d.di_version < 3)
3507 ip->i_d.di_flushiter++;
3510 * Copy the dirty parts of the inode into the on-disk inode. We always
3511 * copy out the core of the inode, because if the inode is dirty at all
3514 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3516 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3517 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3518 ip->i_d.di_flushiter = 0;
3520 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3521 if (XFS_IFORK_Q(ip))
3522 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3523 xfs_inobp_check(mp, bp);
3526 * We've recorded everything logged in the inode, so we'd like to clear
3527 * the ili_fields bits so we don't log and flush things unnecessarily.
3528 * However, we can't stop logging all this information until the data
3529 * we've copied into the disk buffer is written to disk. If we did we
3530 * might overwrite the copy of the inode in the log with all the data
3531 * after re-logging only part of it, and in the face of a crash we
3532 * wouldn't have all the data we need to recover.
3534 * What we do is move the bits to the ili_last_fields field. When
3535 * logging the inode, these bits are moved back to the ili_fields field.
3536 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3537 * know that the information those bits represent is permanently on
3538 * disk. As long as the flush completes before the inode is logged
3539 * again, then both ili_fields and ili_last_fields will be cleared.
3541 * We can play with the ili_fields bits here, because the inode lock
3542 * must be held exclusively in order to set bits there and the flush
3543 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3544 * done routine can tell whether or not to look in the AIL. Also, store
3545 * the current LSN of the inode so that we can tell whether the item has
3546 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3547 * need the AIL lock, because it is a 64 bit value that cannot be read
3550 iip->ili_last_fields = iip->ili_fields;
3551 iip->ili_fields = 0;
3552 iip->ili_fsync_fields = 0;
3553 iip->ili_logged = 1;
3555 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3556 &iip->ili_item.li_lsn);
3559 * Attach the function xfs_iflush_done to the inode's
3560 * buffer. This will remove the inode from the AIL
3561 * and unlock the inode's flush lock when the inode is
3562 * completely written to disk.
3564 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3566 /* generate the checksum. */
3567 xfs_dinode_calc_crc(mp, dip);
3569 ASSERT(bp->b_fspriv != NULL);
3570 ASSERT(bp->b_iodone != NULL);
3574 return -EFSCORRUPTED;