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(xfs_inode_t *, xfs_buf_t *);
62 STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
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);
617 if (di_flags & XFS_DIFLAG_ANY) {
618 if (di_flags & XFS_DIFLAG_REALTIME)
619 flags |= XFS_XFLAG_REALTIME;
620 if (di_flags & XFS_DIFLAG_PREALLOC)
621 flags |= XFS_XFLAG_PREALLOC;
622 if (di_flags & XFS_DIFLAG_IMMUTABLE)
623 flags |= XFS_XFLAG_IMMUTABLE;
624 if (di_flags & XFS_DIFLAG_APPEND)
625 flags |= XFS_XFLAG_APPEND;
626 if (di_flags & XFS_DIFLAG_SYNC)
627 flags |= XFS_XFLAG_SYNC;
628 if (di_flags & XFS_DIFLAG_NOATIME)
629 flags |= XFS_XFLAG_NOATIME;
630 if (di_flags & XFS_DIFLAG_NODUMP)
631 flags |= XFS_XFLAG_NODUMP;
632 if (di_flags & XFS_DIFLAG_RTINHERIT)
633 flags |= XFS_XFLAG_RTINHERIT;
634 if (di_flags & XFS_DIFLAG_PROJINHERIT)
635 flags |= XFS_XFLAG_PROJINHERIT;
636 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
637 flags |= XFS_XFLAG_NOSYMLINKS;
638 if (di_flags & XFS_DIFLAG_EXTSIZE)
639 flags |= XFS_XFLAG_EXTSIZE;
640 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
641 flags |= XFS_XFLAG_EXTSZINHERIT;
642 if (di_flags & XFS_DIFLAG_NODEFRAG)
643 flags |= XFS_XFLAG_NODEFRAG;
644 if (di_flags & XFS_DIFLAG_FILESTREAM)
645 flags |= XFS_XFLAG_FILESTREAM;
655 xfs_icdinode_t *dic = &ip->i_d;
657 return _xfs_dic2xflags(dic->di_flags) |
658 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
665 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
666 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
670 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
671 * is allowed, otherwise it has to be an exact match. If a CI match is found,
672 * ci_name->name will point to a the actual name (caller must free) or
673 * will be set to NULL if an exact match is found.
678 struct xfs_name *name,
680 struct xfs_name *ci_name)
685 trace_xfs_lookup(dp, name);
687 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
690 xfs_ilock(dp, XFS_IOLOCK_SHARED);
691 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
695 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
699 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
704 kmem_free(ci_name->name);
706 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
712 * Allocate an inode on disk and return a copy of its in-core version.
713 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
714 * appropriately within the inode. The uid and gid for the inode are
715 * set according to the contents of the given cred structure.
717 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
718 * has a free inode available, call xfs_iget() to obtain the in-core
719 * version of the allocated inode. Finally, fill in the inode and
720 * log its initial contents. In this case, ialloc_context would be
723 * If xfs_dialloc() does not have an available inode, it will replenish
724 * its supply by doing an allocation. Since we can only do one
725 * allocation within a transaction without deadlocks, we must commit
726 * the current transaction before returning the inode itself.
727 * In this case, therefore, we will set ialloc_context and return.
728 * The caller should then commit the current transaction, start a new
729 * transaction, and call xfs_ialloc() again to actually get the inode.
731 * To ensure that some other process does not grab the inode that
732 * was allocated during the first call to xfs_ialloc(), this routine
733 * also returns the [locked] bp pointing to the head of the freelist
734 * as ialloc_context. The caller should hold this buffer across
735 * the commit and pass it back into this routine on the second call.
737 * If we are allocating quota inodes, we do not have a parent inode
738 * to attach to or associate with (i.e. pip == NULL) because they
739 * are not linked into the directory structure - they are attached
740 * directly to the superblock - and so have no parent.
751 xfs_buf_t **ialloc_context,
754 struct xfs_mount *mp = tp->t_mountp;
762 * Call the space management code to pick
763 * the on-disk inode to be allocated.
765 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
766 ialloc_context, &ino);
769 if (*ialloc_context || ino == NULLFSINO) {
773 ASSERT(*ialloc_context == NULL);
776 * Get the in-core inode with the lock held exclusively.
777 * This is because we're setting fields here we need
778 * to prevent others from looking at until we're done.
780 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
781 XFS_ILOCK_EXCL, &ip);
787 * We always convert v1 inodes to v2 now - we only support filesystems
788 * with >= v2 inode capability, so there is no reason for ever leaving
789 * an inode in v1 format.
791 if (ip->i_d.di_version == 1)
792 ip->i_d.di_version = 2;
794 ip->i_d.di_mode = mode;
795 ip->i_d.di_onlink = 0;
796 ip->i_d.di_nlink = nlink;
797 ASSERT(ip->i_d.di_nlink == nlink);
798 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
799 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
800 xfs_set_projid(ip, prid);
801 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
803 if (pip && XFS_INHERIT_GID(pip)) {
804 ip->i_d.di_gid = pip->i_d.di_gid;
805 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
806 ip->i_d.di_mode |= S_ISGID;
811 * If the group ID of the new file does not match the effective group
812 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
813 * (and only if the irix_sgid_inherit compatibility variable is set).
815 if ((irix_sgid_inherit) &&
816 (ip->i_d.di_mode & S_ISGID) &&
817 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
818 ip->i_d.di_mode &= ~S_ISGID;
822 ip->i_d.di_nextents = 0;
823 ASSERT(ip->i_d.di_nblocks == 0);
825 tv = current_fs_time(mp->m_super);
826 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
827 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
828 ip->i_d.di_atime = ip->i_d.di_mtime;
829 ip->i_d.di_ctime = ip->i_d.di_mtime;
832 * di_gen will have been taken care of in xfs_iread.
834 ip->i_d.di_extsize = 0;
835 ip->i_d.di_dmevmask = 0;
836 ip->i_d.di_dmstate = 0;
837 ip->i_d.di_flags = 0;
839 if (ip->i_d.di_version == 3) {
840 ASSERT(ip->i_d.di_ino == ino);
841 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_meta_uuid));
843 ip->i_d.di_changecount = 1;
845 ip->i_d.di_flags2 = 0;
846 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
847 ip->i_d.di_crtime = ip->i_d.di_mtime;
851 flags = XFS_ILOG_CORE;
852 switch (mode & S_IFMT) {
857 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
858 ip->i_df.if_u2.if_rdev = rdev;
859 ip->i_df.if_flags = 0;
860 flags |= XFS_ILOG_DEV;
864 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
868 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
869 di_flags |= XFS_DIFLAG_RTINHERIT;
870 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
871 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
872 ip->i_d.di_extsize = pip->i_d.di_extsize;
874 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
875 di_flags |= XFS_DIFLAG_PROJINHERIT;
876 } else if (S_ISREG(mode)) {
877 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
878 di_flags |= XFS_DIFLAG_REALTIME;
879 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
880 di_flags |= XFS_DIFLAG_EXTSIZE;
881 ip->i_d.di_extsize = pip->i_d.di_extsize;
884 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
886 di_flags |= XFS_DIFLAG_NOATIME;
887 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
889 di_flags |= XFS_DIFLAG_NODUMP;
890 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
892 di_flags |= XFS_DIFLAG_SYNC;
893 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
894 xfs_inherit_nosymlinks)
895 di_flags |= XFS_DIFLAG_NOSYMLINKS;
896 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
897 xfs_inherit_nodefrag)
898 di_flags |= XFS_DIFLAG_NODEFRAG;
899 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
900 di_flags |= XFS_DIFLAG_FILESTREAM;
901 ip->i_d.di_flags |= di_flags;
905 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
906 ip->i_df.if_flags = XFS_IFEXTENTS;
907 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
908 ip->i_df.if_u1.if_extents = NULL;
914 * Attribute fork settings for new inode.
916 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
917 ip->i_d.di_anextents = 0;
920 * Log the new values stuffed into the inode.
922 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
923 xfs_trans_log_inode(tp, ip, flags);
925 /* now that we have an i_mode we can setup the inode structure */
933 * Allocates a new inode from disk and return a pointer to the
934 * incore copy. This routine will internally commit the current
935 * transaction and allocate a new one if the Space Manager needed
936 * to do an allocation to replenish the inode free-list.
938 * This routine is designed to be called from xfs_create and
944 xfs_trans_t **tpp, /* input: current transaction;
945 output: may be a new transaction. */
946 xfs_inode_t *dp, /* directory within whose allocate
951 prid_t prid, /* project id */
952 int okalloc, /* ok to allocate new space */
953 xfs_inode_t **ipp, /* pointer to inode; it will be
960 xfs_buf_t *ialloc_context = NULL;
966 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
969 * xfs_ialloc will return a pointer to an incore inode if
970 * the Space Manager has an available inode on the free
971 * list. Otherwise, it will do an allocation and replenish
972 * the freelist. Since we can only do one allocation per
973 * transaction without deadlocks, we will need to commit the
974 * current transaction and start a new one. We will then
975 * need to call xfs_ialloc again to get the inode.
977 * If xfs_ialloc did an allocation to replenish the freelist,
978 * it returns the bp containing the head of the freelist as
979 * ialloc_context. We will hold a lock on it across the
980 * transaction commit so that no other process can steal
981 * the inode(s) that we've just allocated.
983 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
984 &ialloc_context, &ip);
987 * Return an error if we were unable to allocate a new inode.
988 * This should only happen if we run out of space on disk or
989 * encounter a disk error.
995 if (!ialloc_context && !ip) {
1001 * If the AGI buffer is non-NULL, then we were unable to get an
1002 * inode in one operation. We need to commit the current
1003 * transaction and call xfs_ialloc() again. It is guaranteed
1004 * to succeed the second time.
1006 if (ialloc_context) {
1008 * Normally, xfs_trans_commit releases all the locks.
1009 * We call bhold to hang on to the ialloc_context across
1010 * the commit. Holding this buffer prevents any other
1011 * processes from doing any allocations in this
1014 xfs_trans_bhold(tp, ialloc_context);
1017 * We want the quota changes to be associated with the next
1018 * transaction, NOT this one. So, detach the dqinfo from this
1019 * and attach it to the next transaction.
1024 dqinfo = (void *)tp->t_dqinfo;
1025 tp->t_dqinfo = NULL;
1026 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1027 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1030 code = xfs_trans_roll(&tp, 0);
1031 if (committed != NULL)
1035 * Re-attach the quota info that we detached from prev trx.
1038 tp->t_dqinfo = dqinfo;
1039 tp->t_flags |= tflags;
1043 xfs_buf_relse(ialloc_context);
1048 xfs_trans_bjoin(tp, ialloc_context);
1051 * Call ialloc again. Since we've locked out all
1052 * other allocations in this allocation group,
1053 * this call should always succeed.
1055 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1056 okalloc, &ialloc_context, &ip);
1059 * If we get an error at this point, return to the caller
1060 * so that the current transaction can be aborted.
1067 ASSERT(!ialloc_context && ip);
1070 if (committed != NULL)
1081 * Decrement the link count on an inode & log the change.
1082 * If this causes the link count to go to zero, initiate the
1083 * logging activity required to truncate a file.
1092 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1094 ASSERT (ip->i_d.di_nlink > 0);
1096 drop_nlink(VFS_I(ip));
1097 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1100 if (ip->i_d.di_nlink == 0) {
1102 * We're dropping the last link to this file.
1103 * Move the on-disk inode to the AGI unlinked list.
1104 * From xfs_inactive() we will pull the inode from
1105 * the list and free it.
1107 error = xfs_iunlink(tp, ip);
1113 * Increment the link count on an inode & log the change.
1120 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1122 ASSERT(ip->i_d.di_version > 1);
1123 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1125 inc_nlink(VFS_I(ip));
1126 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1133 struct xfs_name *name,
1138 int is_dir = S_ISDIR(mode);
1139 struct xfs_mount *mp = dp->i_mount;
1140 struct xfs_inode *ip = NULL;
1141 struct xfs_trans *tp = NULL;
1143 xfs_bmap_free_t free_list;
1144 xfs_fsblock_t first_block;
1145 bool unlock_dp_on_error = false;
1147 struct xfs_dquot *udqp = NULL;
1148 struct xfs_dquot *gdqp = NULL;
1149 struct xfs_dquot *pdqp = NULL;
1150 struct xfs_trans_res *tres;
1153 trace_xfs_create(dp, name);
1155 if (XFS_FORCED_SHUTDOWN(mp))
1158 prid = xfs_get_initial_prid(dp);
1161 * Make sure that we have allocated dquot(s) on disk.
1163 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1164 xfs_kgid_to_gid(current_fsgid()), prid,
1165 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1166 &udqp, &gdqp, &pdqp);
1172 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1173 tres = &M_RES(mp)->tr_mkdir;
1174 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1176 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1177 tres = &M_RES(mp)->tr_create;
1178 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1182 * Initially assume that the file does not exist and
1183 * reserve the resources for that case. If that is not
1184 * the case we'll drop the one we have and get a more
1185 * appropriate transaction later.
1187 error = xfs_trans_reserve(tp, tres, resblks, 0);
1188 if (error == -ENOSPC) {
1189 /* flush outstanding delalloc blocks and retry */
1190 xfs_flush_inodes(mp);
1191 error = xfs_trans_reserve(tp, tres, resblks, 0);
1193 if (error == -ENOSPC) {
1194 /* No space at all so try a "no-allocation" reservation */
1196 error = xfs_trans_reserve(tp, tres, 0, 0);
1199 goto out_trans_cancel;
1202 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1203 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
1204 unlock_dp_on_error = true;
1206 xfs_bmap_init(&free_list, &first_block);
1209 * Reserve disk quota and the inode.
1211 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1212 pdqp, resblks, 1, 0);
1214 goto out_trans_cancel;
1217 error = xfs_dir_canenter(tp, dp, name);
1219 goto out_trans_cancel;
1223 * A newly created regular or special file just has one directory
1224 * entry pointing to them, but a directory also the "." entry
1225 * pointing to itself.
1227 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1228 prid, resblks > 0, &ip, NULL);
1230 goto out_trans_cancel;
1233 * Now we join the directory inode to the transaction. We do not do it
1234 * earlier because xfs_dir_ialloc might commit the previous transaction
1235 * (and release all the locks). An error from here on will result in
1236 * the transaction cancel unlocking dp so don't do it explicitly in the
1239 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1240 unlock_dp_on_error = false;
1242 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1243 &first_block, &free_list, resblks ?
1244 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1246 ASSERT(error != -ENOSPC);
1247 goto out_trans_cancel;
1249 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1250 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1253 error = xfs_dir_init(tp, ip, dp);
1255 goto out_bmap_cancel;
1257 error = xfs_bumplink(tp, dp);
1259 goto out_bmap_cancel;
1263 * If this is a synchronous mount, make sure that the
1264 * create transaction goes to disk before returning to
1267 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1268 xfs_trans_set_sync(tp);
1271 * Attach the dquot(s) to the inodes and modify them incore.
1272 * These ids of the inode couldn't have changed since the new
1273 * inode has been locked ever since it was created.
1275 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1277 error = xfs_bmap_finish(&tp, &free_list, NULL);
1279 goto out_bmap_cancel;
1281 error = xfs_trans_commit(tp);
1283 goto out_release_inode;
1285 xfs_qm_dqrele(udqp);
1286 xfs_qm_dqrele(gdqp);
1287 xfs_qm_dqrele(pdqp);
1293 xfs_bmap_cancel(&free_list);
1295 xfs_trans_cancel(tp);
1298 * Wait until after the current transaction is aborted to finish the
1299 * setup of the inode and release the inode. This prevents recursive
1300 * transactions and deadlocks from xfs_inactive.
1303 xfs_finish_inode_setup(ip);
1307 xfs_qm_dqrele(udqp);
1308 xfs_qm_dqrele(gdqp);
1309 xfs_qm_dqrele(pdqp);
1311 if (unlock_dp_on_error)
1312 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1318 struct xfs_inode *dp,
1319 struct dentry *dentry,
1321 struct xfs_inode **ipp)
1323 struct xfs_mount *mp = dp->i_mount;
1324 struct xfs_inode *ip = NULL;
1325 struct xfs_trans *tp = NULL;
1328 struct xfs_dquot *udqp = NULL;
1329 struct xfs_dquot *gdqp = NULL;
1330 struct xfs_dquot *pdqp = NULL;
1331 struct xfs_trans_res *tres;
1334 if (XFS_FORCED_SHUTDOWN(mp))
1337 prid = xfs_get_initial_prid(dp);
1340 * Make sure that we have allocated dquot(s) on disk.
1342 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1343 xfs_kgid_to_gid(current_fsgid()), prid,
1344 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1345 &udqp, &gdqp, &pdqp);
1349 resblks = XFS_IALLOC_SPACE_RES(mp);
1350 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1352 tres = &M_RES(mp)->tr_create_tmpfile;
1353 error = xfs_trans_reserve(tp, tres, resblks, 0);
1354 if (error == -ENOSPC) {
1355 /* No space at all so try a "no-allocation" reservation */
1357 error = xfs_trans_reserve(tp, tres, 0, 0);
1360 goto out_trans_cancel;
1362 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1363 pdqp, resblks, 1, 0);
1365 goto out_trans_cancel;
1367 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1368 prid, resblks > 0, &ip, NULL);
1370 goto out_trans_cancel;
1372 if (mp->m_flags & XFS_MOUNT_WSYNC)
1373 xfs_trans_set_sync(tp);
1376 * Attach the dquot(s) to the inodes and modify them incore.
1377 * These ids of the inode couldn't have changed since the new
1378 * inode has been locked ever since it was created.
1380 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1383 error = xfs_iunlink(tp, ip);
1385 goto out_trans_cancel;
1387 error = xfs_trans_commit(tp);
1389 goto out_release_inode;
1391 xfs_qm_dqrele(udqp);
1392 xfs_qm_dqrele(gdqp);
1393 xfs_qm_dqrele(pdqp);
1399 xfs_trans_cancel(tp);
1402 * Wait until after the current transaction is aborted to finish the
1403 * setup of the inode and release the inode. This prevents recursive
1404 * transactions and deadlocks from xfs_inactive.
1407 xfs_finish_inode_setup(ip);
1411 xfs_qm_dqrele(udqp);
1412 xfs_qm_dqrele(gdqp);
1413 xfs_qm_dqrele(pdqp);
1422 struct xfs_name *target_name)
1424 xfs_mount_t *mp = tdp->i_mount;
1427 xfs_bmap_free_t free_list;
1428 xfs_fsblock_t first_block;
1431 trace_xfs_link(tdp, target_name);
1433 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1435 if (XFS_FORCED_SHUTDOWN(mp))
1438 error = xfs_qm_dqattach(sip, 0);
1442 error = xfs_qm_dqattach(tdp, 0);
1446 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1447 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1448 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1449 if (error == -ENOSPC) {
1451 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1456 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
1457 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1459 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1460 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1463 * If we are using project inheritance, we only allow hard link
1464 * creation in our tree when the project IDs are the same; else
1465 * the tree quota mechanism could be circumvented.
1467 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1468 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1474 error = xfs_dir_canenter(tp, tdp, target_name);
1479 xfs_bmap_init(&free_list, &first_block);
1481 if (sip->i_d.di_nlink == 0) {
1482 error = xfs_iunlink_remove(tp, sip);
1487 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1488 &first_block, &free_list, resblks);
1491 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1492 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1494 error = xfs_bumplink(tp, sip);
1499 * If this is a synchronous mount, make sure that the
1500 * link transaction goes to disk before returning to
1503 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1504 xfs_trans_set_sync(tp);
1506 error = xfs_bmap_finish(&tp, &free_list, NULL);
1508 xfs_bmap_cancel(&free_list);
1512 return xfs_trans_commit(tp);
1515 xfs_trans_cancel(tp);
1521 * Free up the underlying blocks past new_size. The new size must be smaller
1522 * than the current size. This routine can be used both for the attribute and
1523 * data fork, and does not modify the inode size, which is left to the caller.
1525 * The transaction passed to this routine must have made a permanent log
1526 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1527 * given transaction and start new ones, so make sure everything involved in
1528 * the transaction is tidy before calling here. Some transaction will be
1529 * returned to the caller to be committed. The incoming transaction must
1530 * already include the inode, and both inode locks must be held exclusively.
1531 * The inode must also be "held" within the transaction. On return the inode
1532 * will be "held" within the returned transaction. This routine does NOT
1533 * require any disk space to be reserved for it within the transaction.
1535 * If we get an error, we must return with the inode locked and linked into the
1536 * current transaction. This keeps things simple for the higher level code,
1537 * because it always knows that the inode is locked and held in the transaction
1538 * that returns to it whether errors occur or not. We don't mark the inode
1539 * dirty on error so that transactions can be easily aborted if possible.
1542 xfs_itruncate_extents(
1543 struct xfs_trans **tpp,
1544 struct xfs_inode *ip,
1546 xfs_fsize_t new_size)
1548 struct xfs_mount *mp = ip->i_mount;
1549 struct xfs_trans *tp = *tpp;
1550 xfs_bmap_free_t free_list;
1551 xfs_fsblock_t first_block;
1552 xfs_fileoff_t first_unmap_block;
1553 xfs_fileoff_t last_block;
1554 xfs_filblks_t unmap_len;
1558 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1559 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1560 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1561 ASSERT(new_size <= XFS_ISIZE(ip));
1562 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1563 ASSERT(ip->i_itemp != NULL);
1564 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1565 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1567 trace_xfs_itruncate_extents_start(ip, new_size);
1570 * Since it is possible for space to become allocated beyond
1571 * the end of the file (in a crash where the space is allocated
1572 * but the inode size is not yet updated), simply remove any
1573 * blocks which show up between the new EOF and the maximum
1574 * possible file size. If the first block to be removed is
1575 * beyond the maximum file size (ie it is the same as last_block),
1576 * then there is nothing to do.
1578 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1579 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1580 if (first_unmap_block == last_block)
1583 ASSERT(first_unmap_block < last_block);
1584 unmap_len = last_block - first_unmap_block + 1;
1586 xfs_bmap_init(&free_list, &first_block);
1587 error = xfs_bunmapi(tp, ip,
1588 first_unmap_block, unmap_len,
1589 xfs_bmapi_aflag(whichfork),
1590 XFS_ITRUNC_MAX_EXTENTS,
1591 &first_block, &free_list,
1594 goto out_bmap_cancel;
1597 * Duplicate the transaction that has the permanent
1598 * reservation and commit the old transaction.
1600 error = xfs_bmap_finish(&tp, &free_list, ip);
1602 goto out_bmap_cancel;
1604 error = xfs_trans_roll(&tp, ip);
1610 * Always re-log the inode so that our permanent transaction can keep
1611 * on rolling it forward in the log.
1613 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1615 trace_xfs_itruncate_extents_end(ip, new_size);
1622 * If the bunmapi call encounters an error, return to the caller where
1623 * the transaction can be properly aborted. We just need to make sure
1624 * we're not holding any resources that we were not when we came in.
1626 xfs_bmap_cancel(&free_list);
1634 xfs_mount_t *mp = ip->i_mount;
1637 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1640 /* If this is a read-only mount, don't do this (would generate I/O) */
1641 if (mp->m_flags & XFS_MOUNT_RDONLY)
1644 if (!XFS_FORCED_SHUTDOWN(mp)) {
1648 * If we previously truncated this file and removed old data
1649 * in the process, we want to initiate "early" writeout on
1650 * the last close. This is an attempt to combat the notorious
1651 * NULL files problem which is particularly noticeable from a
1652 * truncate down, buffered (re-)write (delalloc), followed by
1653 * a crash. What we are effectively doing here is
1654 * significantly reducing the time window where we'd otherwise
1655 * be exposed to that problem.
1657 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1659 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1660 if (ip->i_delayed_blks > 0) {
1661 error = filemap_flush(VFS_I(ip)->i_mapping);
1668 if (ip->i_d.di_nlink == 0)
1671 if (xfs_can_free_eofblocks(ip, false)) {
1674 * If we can't get the iolock just skip truncating the blocks
1675 * past EOF because we could deadlock with the mmap_sem
1676 * otherwise. We'll get another chance to drop them once the
1677 * last reference to the inode is dropped, so we'll never leak
1678 * blocks permanently.
1680 * Further, check if the inode is being opened, written and
1681 * closed frequently and we have delayed allocation blocks
1682 * outstanding (e.g. streaming writes from the NFS server),
1683 * truncating the blocks past EOF will cause fragmentation to
1686 * In this case don't do the truncation, either, but we have to
1687 * be careful how we detect this case. Blocks beyond EOF show
1688 * up as i_delayed_blks even when the inode is clean, so we
1689 * need to truncate them away first before checking for a dirty
1690 * release. Hence on the first dirty close we will still remove
1691 * the speculative allocation, but after that we will leave it
1694 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1697 error = xfs_free_eofblocks(mp, ip, true);
1698 if (error && error != -EAGAIN)
1701 /* delalloc blocks after truncation means it really is dirty */
1702 if (ip->i_delayed_blks)
1703 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1709 * xfs_inactive_truncate
1711 * Called to perform a truncate when an inode becomes unlinked.
1714 xfs_inactive_truncate(
1715 struct xfs_inode *ip)
1717 struct xfs_mount *mp = ip->i_mount;
1718 struct xfs_trans *tp;
1721 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1722 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1724 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1725 xfs_trans_cancel(tp);
1729 xfs_ilock(ip, XFS_ILOCK_EXCL);
1730 xfs_trans_ijoin(tp, ip, 0);
1733 * Log the inode size first to prevent stale data exposure in the event
1734 * of a system crash before the truncate completes. See the related
1735 * comment in xfs_setattr_size() for details.
1737 ip->i_d.di_size = 0;
1738 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1740 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1742 goto error_trans_cancel;
1744 ASSERT(ip->i_d.di_nextents == 0);
1746 error = xfs_trans_commit(tp);
1750 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1754 xfs_trans_cancel(tp);
1756 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1761 * xfs_inactive_ifree()
1763 * Perform the inode free when an inode is unlinked.
1767 struct xfs_inode *ip)
1769 xfs_bmap_free_t free_list;
1770 xfs_fsblock_t first_block;
1771 struct xfs_mount *mp = ip->i_mount;
1772 struct xfs_trans *tp;
1775 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1778 * The ifree transaction might need to allocate blocks for record
1779 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1780 * allow ifree to dip into the reserved block pool if necessary.
1782 * Freeing large sets of inodes generally means freeing inode chunks,
1783 * directory and file data blocks, so this should be relatively safe.
1784 * Only under severe circumstances should it be possible to free enough
1785 * inodes to exhaust the reserve block pool via finobt expansion while
1786 * at the same time not creating free space in the filesystem.
1788 * Send a warning if the reservation does happen to fail, as the inode
1789 * now remains allocated and sits on the unlinked list until the fs is
1792 tp->t_flags |= XFS_TRANS_RESERVE;
1793 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1794 XFS_IFREE_SPACE_RES(mp), 0);
1796 if (error == -ENOSPC) {
1797 xfs_warn_ratelimited(mp,
1798 "Failed to remove inode(s) from unlinked list. "
1799 "Please free space, unmount and run xfs_repair.");
1801 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1803 xfs_trans_cancel(tp);
1807 xfs_ilock(ip, XFS_ILOCK_EXCL);
1808 xfs_trans_ijoin(tp, ip, 0);
1810 xfs_bmap_init(&free_list, &first_block);
1811 error = xfs_ifree(tp, ip, &free_list);
1814 * If we fail to free the inode, shut down. The cancel
1815 * might do that, we need to make sure. Otherwise the
1816 * inode might be lost for a long time or forever.
1818 if (!XFS_FORCED_SHUTDOWN(mp)) {
1819 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1821 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1823 xfs_trans_cancel(tp);
1824 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1829 * Credit the quota account(s). The inode is gone.
1831 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1834 * Just ignore errors at this point. There is nothing we can do except
1835 * to try to keep going. Make sure it's not a silent error.
1837 error = xfs_bmap_finish(&tp, &free_list, NULL);
1839 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1841 xfs_bmap_cancel(&free_list);
1843 error = xfs_trans_commit(tp);
1845 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1848 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1855 * This is called when the vnode reference count for the vnode
1856 * goes to zero. If the file has been unlinked, then it must
1857 * now be truncated. Also, we clear all of the read-ahead state
1858 * kept for the inode here since the file is now closed.
1864 struct xfs_mount *mp;
1869 * If the inode is already free, then there can be nothing
1872 if (ip->i_d.di_mode == 0) {
1873 ASSERT(ip->i_df.if_real_bytes == 0);
1874 ASSERT(ip->i_df.if_broot_bytes == 0);
1880 /* If this is a read-only mount, don't do this (would generate I/O) */
1881 if (mp->m_flags & XFS_MOUNT_RDONLY)
1884 if (ip->i_d.di_nlink != 0) {
1886 * force is true because we are evicting an inode from the
1887 * cache. Post-eof blocks must be freed, lest we end up with
1888 * broken free space accounting.
1890 if (xfs_can_free_eofblocks(ip, true))
1891 xfs_free_eofblocks(mp, ip, false);
1896 if (S_ISREG(ip->i_d.di_mode) &&
1897 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1898 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1901 error = xfs_qm_dqattach(ip, 0);
1905 if (S_ISLNK(ip->i_d.di_mode))
1906 error = xfs_inactive_symlink(ip);
1908 error = xfs_inactive_truncate(ip);
1913 * If there are attributes associated with the file then blow them away
1914 * now. The code calls a routine that recursively deconstructs the
1915 * attribute fork. If also blows away the in-core attribute fork.
1917 if (XFS_IFORK_Q(ip)) {
1918 error = xfs_attr_inactive(ip);
1924 ASSERT(ip->i_d.di_anextents == 0);
1925 ASSERT(ip->i_d.di_forkoff == 0);
1930 error = xfs_inactive_ifree(ip);
1935 * Release the dquots held by inode, if any.
1937 xfs_qm_dqdetach(ip);
1941 * This is called when the inode's link count goes to 0.
1942 * We place the on-disk inode on a list in the AGI. It
1943 * will be pulled from this list when the inode is freed.
1960 ASSERT(ip->i_d.di_nlink == 0);
1961 ASSERT(ip->i_d.di_mode != 0);
1966 * Get the agi buffer first. It ensures lock ordering
1969 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1972 agi = XFS_BUF_TO_AGI(agibp);
1975 * Get the index into the agi hash table for the
1976 * list this inode will go on.
1978 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1980 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1981 ASSERT(agi->agi_unlinked[bucket_index]);
1982 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1984 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1986 * There is already another inode in the bucket we need
1987 * to add ourselves to. Add us at the front of the list.
1988 * Here we put the head pointer into our next pointer,
1989 * and then we fall through to point the head at us.
1991 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1996 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1997 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1998 offset = ip->i_imap.im_boffset +
1999 offsetof(xfs_dinode_t, di_next_unlinked);
2001 /* need to recalc the inode CRC if appropriate */
2002 xfs_dinode_calc_crc(mp, dip);
2004 xfs_trans_inode_buf(tp, ibp);
2005 xfs_trans_log_buf(tp, ibp, offset,
2006 (offset + sizeof(xfs_agino_t) - 1));
2007 xfs_inobp_check(mp, ibp);
2011 * Point the bucket head pointer at the inode being inserted.
2014 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2015 offset = offsetof(xfs_agi_t, agi_unlinked) +
2016 (sizeof(xfs_agino_t) * bucket_index);
2017 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2018 xfs_trans_log_buf(tp, agibp, offset,
2019 (offset + sizeof(xfs_agino_t) - 1));
2024 * Pull the on-disk inode from the AGI unlinked list.
2037 xfs_agnumber_t agno;
2039 xfs_agino_t next_agino;
2040 xfs_buf_t *last_ibp;
2041 xfs_dinode_t *last_dip = NULL;
2043 int offset, last_offset = 0;
2047 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2050 * Get the agi buffer first. It ensures lock ordering
2053 error = xfs_read_agi(mp, tp, agno, &agibp);
2057 agi = XFS_BUF_TO_AGI(agibp);
2060 * Get the index into the agi hash table for the
2061 * list this inode will go on.
2063 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2065 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2066 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2067 ASSERT(agi->agi_unlinked[bucket_index]);
2069 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2071 * We're at the head of the list. Get the inode's on-disk
2072 * buffer to see if there is anyone after us on the list.
2073 * Only modify our next pointer if it is not already NULLAGINO.
2074 * This saves us the overhead of dealing with the buffer when
2075 * there is no need to change it.
2077 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2080 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2084 next_agino = be32_to_cpu(dip->di_next_unlinked);
2085 ASSERT(next_agino != 0);
2086 if (next_agino != NULLAGINO) {
2087 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2088 offset = ip->i_imap.im_boffset +
2089 offsetof(xfs_dinode_t, di_next_unlinked);
2091 /* need to recalc the inode CRC if appropriate */
2092 xfs_dinode_calc_crc(mp, dip);
2094 xfs_trans_inode_buf(tp, ibp);
2095 xfs_trans_log_buf(tp, ibp, offset,
2096 (offset + sizeof(xfs_agino_t) - 1));
2097 xfs_inobp_check(mp, ibp);
2099 xfs_trans_brelse(tp, ibp);
2102 * Point the bucket head pointer at the next inode.
2104 ASSERT(next_agino != 0);
2105 ASSERT(next_agino != agino);
2106 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2107 offset = offsetof(xfs_agi_t, agi_unlinked) +
2108 (sizeof(xfs_agino_t) * bucket_index);
2109 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2110 xfs_trans_log_buf(tp, agibp, offset,
2111 (offset + sizeof(xfs_agino_t) - 1));
2114 * We need to search the list for the inode being freed.
2116 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2118 while (next_agino != agino) {
2119 struct xfs_imap imap;
2122 xfs_trans_brelse(tp, last_ibp);
2125 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2127 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2130 "%s: xfs_imap returned error %d.",
2135 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2139 "%s: xfs_imap_to_bp returned error %d.",
2144 last_offset = imap.im_boffset;
2145 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2146 ASSERT(next_agino != NULLAGINO);
2147 ASSERT(next_agino != 0);
2151 * Now last_ibp points to the buffer previous to us on the
2152 * unlinked list. Pull us from the list.
2154 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2157 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2161 next_agino = be32_to_cpu(dip->di_next_unlinked);
2162 ASSERT(next_agino != 0);
2163 ASSERT(next_agino != agino);
2164 if (next_agino != NULLAGINO) {
2165 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2166 offset = ip->i_imap.im_boffset +
2167 offsetof(xfs_dinode_t, di_next_unlinked);
2169 /* need to recalc the inode CRC if appropriate */
2170 xfs_dinode_calc_crc(mp, dip);
2172 xfs_trans_inode_buf(tp, ibp);
2173 xfs_trans_log_buf(tp, ibp, offset,
2174 (offset + sizeof(xfs_agino_t) - 1));
2175 xfs_inobp_check(mp, ibp);
2177 xfs_trans_brelse(tp, ibp);
2180 * Point the previous inode on the list to the next inode.
2182 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2183 ASSERT(next_agino != 0);
2184 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2186 /* need to recalc the inode CRC if appropriate */
2187 xfs_dinode_calc_crc(mp, last_dip);
2189 xfs_trans_inode_buf(tp, last_ibp);
2190 xfs_trans_log_buf(tp, last_ibp, offset,
2191 (offset + sizeof(xfs_agino_t) - 1));
2192 xfs_inobp_check(mp, last_ibp);
2198 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2199 * inodes that are in memory - they all must be marked stale and attached to
2200 * the cluster buffer.
2204 xfs_inode_t *free_ip,
2206 struct xfs_icluster *xic)
2208 xfs_mount_t *mp = free_ip->i_mount;
2209 int blks_per_cluster;
2210 int inodes_per_cluster;
2217 xfs_inode_log_item_t *iip;
2218 xfs_log_item_t *lip;
2219 struct xfs_perag *pag;
2222 inum = xic->first_ino;
2223 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2224 blks_per_cluster = xfs_icluster_size_fsb(mp);
2225 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2226 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2228 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2230 * The allocation bitmap tells us which inodes of the chunk were
2231 * physically allocated. Skip the cluster if an inode falls into
2234 ioffset = inum - xic->first_ino;
2235 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2236 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2240 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2241 XFS_INO_TO_AGBNO(mp, inum));
2244 * We obtain and lock the backing buffer first in the process
2245 * here, as we have to ensure that any dirty inode that we
2246 * can't get the flush lock on is attached to the buffer.
2247 * If we scan the in-memory inodes first, then buffer IO can
2248 * complete before we get a lock on it, and hence we may fail
2249 * to mark all the active inodes on the buffer stale.
2251 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2252 mp->m_bsize * blks_per_cluster,
2259 * This buffer may not have been correctly initialised as we
2260 * didn't read it from disk. That's not important because we are
2261 * only using to mark the buffer as stale in the log, and to
2262 * attach stale cached inodes on it. That means it will never be
2263 * dispatched for IO. If it is, we want to know about it, and we
2264 * want it to fail. We can acheive this by adding a write
2265 * verifier to the buffer.
2267 bp->b_ops = &xfs_inode_buf_ops;
2270 * Walk the inodes already attached to the buffer and mark them
2271 * stale. These will all have the flush locks held, so an
2272 * in-memory inode walk can't lock them. By marking them all
2273 * stale first, we will not attempt to lock them in the loop
2274 * below as the XFS_ISTALE flag will be set.
2278 if (lip->li_type == XFS_LI_INODE) {
2279 iip = (xfs_inode_log_item_t *)lip;
2280 ASSERT(iip->ili_logged == 1);
2281 lip->li_cb = xfs_istale_done;
2282 xfs_trans_ail_copy_lsn(mp->m_ail,
2283 &iip->ili_flush_lsn,
2284 &iip->ili_item.li_lsn);
2285 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2287 lip = lip->li_bio_list;
2292 * For each inode in memory attempt to add it to the inode
2293 * buffer and set it up for being staled on buffer IO
2294 * completion. This is safe as we've locked out tail pushing
2295 * and flushing by locking the buffer.
2297 * We have already marked every inode that was part of a
2298 * transaction stale above, which means there is no point in
2299 * even trying to lock them.
2301 for (i = 0; i < inodes_per_cluster; i++) {
2304 ip = radix_tree_lookup(&pag->pag_ici_root,
2305 XFS_INO_TO_AGINO(mp, (inum + i)));
2307 /* Inode not in memory, nothing to do */
2314 * because this is an RCU protected lookup, we could
2315 * find a recently freed or even reallocated inode
2316 * during the lookup. We need to check under the
2317 * i_flags_lock for a valid inode here. Skip it if it
2318 * is not valid, the wrong inode or stale.
2320 spin_lock(&ip->i_flags_lock);
2321 if (ip->i_ino != inum + i ||
2322 __xfs_iflags_test(ip, XFS_ISTALE)) {
2323 spin_unlock(&ip->i_flags_lock);
2327 spin_unlock(&ip->i_flags_lock);
2330 * Don't try to lock/unlock the current inode, but we
2331 * _cannot_ skip the other inodes that we did not find
2332 * in the list attached to the buffer and are not
2333 * already marked stale. If we can't lock it, back off
2336 if (ip != free_ip &&
2337 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2345 xfs_iflags_set(ip, XFS_ISTALE);
2348 * we don't need to attach clean inodes or those only
2349 * with unlogged changes (which we throw away, anyway).
2352 if (!iip || xfs_inode_clean(ip)) {
2353 ASSERT(ip != free_ip);
2355 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2359 iip->ili_last_fields = iip->ili_fields;
2360 iip->ili_fields = 0;
2361 iip->ili_fsync_fields = 0;
2362 iip->ili_logged = 1;
2363 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2364 &iip->ili_item.li_lsn);
2366 xfs_buf_attach_iodone(bp, xfs_istale_done,
2370 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2373 xfs_trans_stale_inode_buf(tp, bp);
2374 xfs_trans_binval(tp, bp);
2382 * This is called to return an inode to the inode free list.
2383 * The inode should already be truncated to 0 length and have
2384 * no pages associated with it. This routine also assumes that
2385 * the inode is already a part of the transaction.
2387 * The on-disk copy of the inode will have been added to the list
2388 * of unlinked inodes in the AGI. We need to remove the inode from
2389 * that list atomically with respect to freeing it here.
2395 xfs_bmap_free_t *flist)
2398 struct xfs_icluster xic = { 0 };
2400 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2401 ASSERT(ip->i_d.di_nlink == 0);
2402 ASSERT(ip->i_d.di_nextents == 0);
2403 ASSERT(ip->i_d.di_anextents == 0);
2404 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2405 ASSERT(ip->i_d.di_nblocks == 0);
2408 * Pull the on-disk inode from the AGI unlinked list.
2410 error = xfs_iunlink_remove(tp, ip);
2414 error = xfs_difree(tp, ip->i_ino, flist, &xic);
2418 ip->i_d.di_mode = 0; /* mark incore inode as free */
2419 ip->i_d.di_flags = 0;
2420 ip->i_d.di_dmevmask = 0;
2421 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2422 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2423 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2425 * Bump the generation count so no one will be confused
2426 * by reincarnations of this inode.
2429 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2432 error = xfs_ifree_cluster(ip, tp, &xic);
2438 * This is called to unpin an inode. The caller must have the inode locked
2439 * in at least shared mode so that the buffer cannot be subsequently pinned
2440 * once someone is waiting for it to be unpinned.
2444 struct xfs_inode *ip)
2446 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2448 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2450 /* Give the log a push to start the unpinning I/O */
2451 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2457 struct xfs_inode *ip)
2459 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2460 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2465 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2466 if (xfs_ipincount(ip))
2468 } while (xfs_ipincount(ip));
2469 finish_wait(wq, &wait.wait);
2474 struct xfs_inode *ip)
2476 if (xfs_ipincount(ip))
2477 __xfs_iunpin_wait(ip);
2481 * Removing an inode from the namespace involves removing the directory entry
2482 * and dropping the link count on the inode. Removing the directory entry can
2483 * result in locking an AGF (directory blocks were freed) and removing a link
2484 * count can result in placing the inode on an unlinked list which results in
2487 * The big problem here is that we have an ordering constraint on AGF and AGI
2488 * locking - inode allocation locks the AGI, then can allocate a new extent for
2489 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2490 * removes the inode from the unlinked list, requiring that we lock the AGI
2491 * first, and then freeing the inode can result in an inode chunk being freed
2492 * and hence freeing disk space requiring that we lock an AGF.
2494 * Hence the ordering that is imposed by other parts of the code is AGI before
2495 * AGF. This means we cannot remove the directory entry before we drop the inode
2496 * reference count and put it on the unlinked list as this results in a lock
2497 * order of AGF then AGI, and this can deadlock against inode allocation and
2498 * freeing. Therefore we must drop the link counts before we remove the
2501 * This is still safe from a transactional point of view - it is not until we
2502 * get to xfs_bmap_finish() that we have the possibility of multiple
2503 * transactions in this operation. Hence as long as we remove the directory
2504 * entry and drop the link count in the first transaction of the remove
2505 * operation, there are no transactional constraints on the ordering here.
2510 struct xfs_name *name,
2513 xfs_mount_t *mp = dp->i_mount;
2514 xfs_trans_t *tp = NULL;
2515 int is_dir = S_ISDIR(ip->i_d.di_mode);
2517 xfs_bmap_free_t free_list;
2518 xfs_fsblock_t first_block;
2521 trace_xfs_remove(dp, name);
2523 if (XFS_FORCED_SHUTDOWN(mp))
2526 error = xfs_qm_dqattach(dp, 0);
2530 error = xfs_qm_dqattach(ip, 0);
2535 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2537 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2540 * We try to get the real space reservation first,
2541 * allowing for directory btree deletion(s) implying
2542 * possible bmap insert(s). If we can't get the space
2543 * reservation then we use 0 instead, and avoid the bmap
2544 * btree insert(s) in the directory code by, if the bmap
2545 * insert tries to happen, instead trimming the LAST
2546 * block from the directory.
2548 resblks = XFS_REMOVE_SPACE_RES(mp);
2549 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2550 if (error == -ENOSPC) {
2552 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2555 ASSERT(error != -ENOSPC);
2556 goto out_trans_cancel;
2559 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2560 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2562 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2563 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2566 * If we're removing a directory perform some additional validation.
2569 ASSERT(ip->i_d.di_nlink >= 2);
2570 if (ip->i_d.di_nlink != 2) {
2572 goto out_trans_cancel;
2574 if (!xfs_dir_isempty(ip)) {
2576 goto out_trans_cancel;
2579 /* Drop the link from ip's "..". */
2580 error = xfs_droplink(tp, dp);
2582 goto out_trans_cancel;
2584 /* Drop the "." link from ip to self. */
2585 error = xfs_droplink(tp, ip);
2587 goto out_trans_cancel;
2590 * When removing a non-directory we need to log the parent
2591 * inode here. For a directory this is done implicitly
2592 * by the xfs_droplink call for the ".." entry.
2594 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2596 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2598 /* Drop the link from dp to ip. */
2599 error = xfs_droplink(tp, ip);
2601 goto out_trans_cancel;
2603 xfs_bmap_init(&free_list, &first_block);
2604 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2605 &first_block, &free_list, resblks);
2607 ASSERT(error != -ENOENT);
2608 goto out_bmap_cancel;
2612 * If this is a synchronous mount, make sure that the
2613 * remove transaction goes to disk before returning to
2616 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2617 xfs_trans_set_sync(tp);
2619 error = xfs_bmap_finish(&tp, &free_list, NULL);
2621 goto out_bmap_cancel;
2623 error = xfs_trans_commit(tp);
2627 if (is_dir && xfs_inode_is_filestream(ip))
2628 xfs_filestream_deassociate(ip);
2633 xfs_bmap_cancel(&free_list);
2635 xfs_trans_cancel(tp);
2641 * Enter all inodes for a rename transaction into a sorted array.
2643 #define __XFS_SORT_INODES 5
2645 xfs_sort_for_rename(
2646 struct xfs_inode *dp1, /* in: old (source) directory inode */
2647 struct xfs_inode *dp2, /* in: new (target) directory inode */
2648 struct xfs_inode *ip1, /* in: inode of old entry */
2649 struct xfs_inode *ip2, /* in: inode of new entry */
2650 struct xfs_inode *wip, /* in: whiteout inode */
2651 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2652 int *num_inodes) /* in/out: inodes in array */
2656 ASSERT(*num_inodes == __XFS_SORT_INODES);
2657 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2660 * i_tab contains a list of pointers to inodes. We initialize
2661 * the table here & we'll sort it. We will then use it to
2662 * order the acquisition of the inode locks.
2664 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2677 * Sort the elements via bubble sort. (Remember, there are at
2678 * most 5 elements to sort, so this is adequate.)
2680 for (i = 0; i < *num_inodes; i++) {
2681 for (j = 1; j < *num_inodes; j++) {
2682 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2683 struct xfs_inode *temp = i_tab[j];
2684 i_tab[j] = i_tab[j-1];
2693 struct xfs_trans *tp,
2694 struct xfs_bmap_free *free_list)
2699 * If this is a synchronous mount, make sure that the rename transaction
2700 * goes to disk before returning to the user.
2702 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2703 xfs_trans_set_sync(tp);
2705 error = xfs_bmap_finish(&tp, free_list, NULL);
2707 xfs_bmap_cancel(free_list);
2708 xfs_trans_cancel(tp);
2712 return xfs_trans_commit(tp);
2716 * xfs_cross_rename()
2718 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2722 struct xfs_trans *tp,
2723 struct xfs_inode *dp1,
2724 struct xfs_name *name1,
2725 struct xfs_inode *ip1,
2726 struct xfs_inode *dp2,
2727 struct xfs_name *name2,
2728 struct xfs_inode *ip2,
2729 struct xfs_bmap_free *free_list,
2730 xfs_fsblock_t *first_block,
2738 /* Swap inode number for dirent in first parent */
2739 error = xfs_dir_replace(tp, dp1, name1,
2741 first_block, free_list, spaceres);
2743 goto out_trans_abort;
2745 /* Swap inode number for dirent in second parent */
2746 error = xfs_dir_replace(tp, dp2, name2,
2748 first_block, free_list, spaceres);
2750 goto out_trans_abort;
2753 * If we're renaming one or more directories across different parents,
2754 * update the respective ".." entries (and link counts) to match the new
2758 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2760 if (S_ISDIR(ip2->i_d.di_mode)) {
2761 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2762 dp1->i_ino, first_block,
2763 free_list, spaceres);
2765 goto out_trans_abort;
2767 /* transfer ip2 ".." reference to dp1 */
2768 if (!S_ISDIR(ip1->i_d.di_mode)) {
2769 error = xfs_droplink(tp, dp2);
2771 goto out_trans_abort;
2772 error = xfs_bumplink(tp, dp1);
2774 goto out_trans_abort;
2778 * Although ip1 isn't changed here, userspace needs
2779 * to be warned about the change, so that applications
2780 * relying on it (like backup ones), will properly
2783 ip1_flags |= XFS_ICHGTIME_CHG;
2784 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2787 if (S_ISDIR(ip1->i_d.di_mode)) {
2788 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2789 dp2->i_ino, first_block,
2790 free_list, spaceres);
2792 goto out_trans_abort;
2794 /* transfer ip1 ".." reference to dp2 */
2795 if (!S_ISDIR(ip2->i_d.di_mode)) {
2796 error = xfs_droplink(tp, dp1);
2798 goto out_trans_abort;
2799 error = xfs_bumplink(tp, dp2);
2801 goto out_trans_abort;
2805 * Although ip2 isn't changed here, userspace needs
2806 * to be warned about the change, so that applications
2807 * relying on it (like backup ones), will properly
2810 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2811 ip2_flags |= XFS_ICHGTIME_CHG;
2816 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2817 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2820 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2821 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2824 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2825 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2827 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2828 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2829 return xfs_finish_rename(tp, free_list);
2832 xfs_bmap_cancel(free_list);
2833 xfs_trans_cancel(tp);
2838 * xfs_rename_alloc_whiteout()
2840 * Return a referenced, unlinked, unlocked inode that that can be used as a
2841 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2842 * crash between allocating the inode and linking it into the rename transaction
2843 * recovery will free the inode and we won't leak it.
2846 xfs_rename_alloc_whiteout(
2847 struct xfs_inode *dp,
2848 struct xfs_inode **wip)
2850 struct xfs_inode *tmpfile;
2853 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2858 * Prepare the tmpfile inode as if it were created through the VFS.
2859 * Otherwise, the link increment paths will complain about nlink 0->1.
2860 * Drop the link count as done by d_tmpfile(), complete the inode setup
2861 * and flag it as linkable.
2863 drop_nlink(VFS_I(tmpfile));
2864 xfs_finish_inode_setup(tmpfile);
2865 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2876 struct xfs_inode *src_dp,
2877 struct xfs_name *src_name,
2878 struct xfs_inode *src_ip,
2879 struct xfs_inode *target_dp,
2880 struct xfs_name *target_name,
2881 struct xfs_inode *target_ip,
2884 struct xfs_mount *mp = src_dp->i_mount;
2885 struct xfs_trans *tp;
2886 struct xfs_bmap_free free_list;
2887 xfs_fsblock_t first_block;
2888 struct xfs_inode *wip = NULL; /* whiteout inode */
2889 struct xfs_inode *inodes[__XFS_SORT_INODES];
2890 int num_inodes = __XFS_SORT_INODES;
2891 bool new_parent = (src_dp != target_dp);
2892 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2896 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2898 if ((flags & RENAME_EXCHANGE) && !target_ip)
2902 * If we are doing a whiteout operation, allocate the whiteout inode
2903 * we will be placing at the target and ensure the type is set
2906 if (flags & RENAME_WHITEOUT) {
2907 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2908 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2912 /* setup target dirent info as whiteout */
2913 src_name->type = XFS_DIR3_FT_CHRDEV;
2916 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2917 inodes, &num_inodes);
2919 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2920 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2921 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2922 if (error == -ENOSPC) {
2924 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2927 goto out_trans_cancel;
2930 * Attach the dquots to the inodes
2932 error = xfs_qm_vop_rename_dqattach(inodes);
2934 goto out_trans_cancel;
2937 * Lock all the participating inodes. Depending upon whether
2938 * the target_name exists in the target directory, and
2939 * whether the target directory is the same as the source
2940 * directory, we can lock from 2 to 4 inodes.
2943 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2945 xfs_lock_two_inodes(src_dp, target_dp,
2946 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2948 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2951 * Join all the inodes to the transaction. From this point on,
2952 * we can rely on either trans_commit or trans_cancel to unlock
2955 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2957 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2958 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2960 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2962 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2965 * If we are using project inheritance, we only allow renames
2966 * into our tree when the project IDs are the same; else the
2967 * tree quota mechanism would be circumvented.
2969 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2970 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2972 goto out_trans_cancel;
2975 xfs_bmap_init(&free_list, &first_block);
2977 /* RENAME_EXCHANGE is unique from here on. */
2978 if (flags & RENAME_EXCHANGE)
2979 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2980 target_dp, target_name, target_ip,
2981 &free_list, &first_block, spaceres);
2984 * Set up the target.
2986 if (target_ip == NULL) {
2988 * If there's no space reservation, check the entry will
2989 * fit before actually inserting it.
2992 error = xfs_dir_canenter(tp, target_dp, target_name);
2994 goto out_trans_cancel;
2997 * If target does not exist and the rename crosses
2998 * directories, adjust the target directory link count
2999 * to account for the ".." reference from the new entry.
3001 error = xfs_dir_createname(tp, target_dp, target_name,
3002 src_ip->i_ino, &first_block,
3003 &free_list, spaceres);
3005 goto out_bmap_cancel;
3007 xfs_trans_ichgtime(tp, target_dp,
3008 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3010 if (new_parent && src_is_directory) {
3011 error = xfs_bumplink(tp, target_dp);
3013 goto out_bmap_cancel;
3015 } else { /* target_ip != NULL */
3017 * If target exists and it's a directory, check that both
3018 * target and source are directories and that target can be
3019 * destroyed, or that neither is a directory.
3021 if (S_ISDIR(target_ip->i_d.di_mode)) {
3023 * Make sure target dir is empty.
3025 if (!(xfs_dir_isempty(target_ip)) ||
3026 (target_ip->i_d.di_nlink > 2)) {
3028 goto out_trans_cancel;
3033 * Link the source inode under the target name.
3034 * If the source inode is a directory and we are moving
3035 * it across directories, its ".." entry will be
3036 * inconsistent until we replace that down below.
3038 * In case there is already an entry with the same
3039 * name at the destination directory, remove it first.
3041 error = xfs_dir_replace(tp, target_dp, target_name,
3043 &first_block, &free_list, spaceres);
3045 goto out_bmap_cancel;
3047 xfs_trans_ichgtime(tp, target_dp,
3048 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3051 * Decrement the link count on the target since the target
3052 * dir no longer points to it.
3054 error = xfs_droplink(tp, target_ip);
3056 goto out_bmap_cancel;
3058 if (src_is_directory) {
3060 * Drop the link from the old "." entry.
3062 error = xfs_droplink(tp, target_ip);
3064 goto out_bmap_cancel;
3066 } /* target_ip != NULL */
3069 * Remove the source.
3071 if (new_parent && src_is_directory) {
3073 * Rewrite the ".." entry to point to the new
3076 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3078 &first_block, &free_list, spaceres);
3079 ASSERT(error != -EEXIST);
3081 goto out_bmap_cancel;
3085 * We always want to hit the ctime on the source inode.
3087 * This isn't strictly required by the standards since the source
3088 * inode isn't really being changed, but old unix file systems did
3089 * it and some incremental backup programs won't work without it.
3091 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3092 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3095 * Adjust the link count on src_dp. This is necessary when
3096 * renaming a directory, either within one parent when
3097 * the target existed, or across two parent directories.
3099 if (src_is_directory && (new_parent || target_ip != NULL)) {
3102 * Decrement link count on src_directory since the
3103 * entry that's moved no longer points to it.
3105 error = xfs_droplink(tp, src_dp);
3107 goto out_bmap_cancel;
3111 * For whiteouts, we only need to update the source dirent with the
3112 * inode number of the whiteout inode rather than removing it
3116 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3117 &first_block, &free_list, spaceres);
3119 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3120 &first_block, &free_list, spaceres);
3122 goto out_bmap_cancel;
3125 * For whiteouts, we need to bump the link count on the whiteout inode.
3126 * This means that failures all the way up to this point leave the inode
3127 * on the unlinked list and so cleanup is a simple matter of dropping
3128 * the remaining reference to it. If we fail here after bumping the link
3129 * count, we're shutting down the filesystem so we'll never see the
3130 * intermediate state on disk.
3133 ASSERT(VFS_I(wip)->i_nlink == 0 && wip->i_d.di_nlink == 0);
3134 error = xfs_bumplink(tp, wip);
3136 goto out_bmap_cancel;
3137 error = xfs_iunlink_remove(tp, wip);
3139 goto out_bmap_cancel;
3140 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3143 * Now we have a real link, clear the "I'm a tmpfile" state
3144 * flag from the inode so it doesn't accidentally get misused in
3147 VFS_I(wip)->i_state &= ~I_LINKABLE;
3150 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3151 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3153 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3155 error = xfs_finish_rename(tp, &free_list);
3161 xfs_bmap_cancel(&free_list);
3163 xfs_trans_cancel(tp);
3174 xfs_mount_t *mp = ip->i_mount;
3175 struct xfs_perag *pag;
3176 unsigned long first_index, mask;
3177 unsigned long inodes_per_cluster;
3179 xfs_inode_t **ilist;
3186 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3188 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3189 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3190 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
3194 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3195 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3197 /* really need a gang lookup range call here */
3198 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3199 first_index, inodes_per_cluster);
3203 for (i = 0; i < nr_found; i++) {
3209 * because this is an RCU protected lookup, we could find a
3210 * recently freed or even reallocated inode during the lookup.
3211 * We need to check under the i_flags_lock for a valid inode
3212 * here. Skip it if it is not valid or the wrong inode.
3214 spin_lock(&ip->i_flags_lock);
3216 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3217 spin_unlock(&ip->i_flags_lock);
3220 spin_unlock(&ip->i_flags_lock);
3223 * Do an un-protected check to see if the inode is dirty and
3224 * is a candidate for flushing. These checks will be repeated
3225 * later after the appropriate locks are acquired.
3227 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3231 * Try to get locks. If any are unavailable or it is pinned,
3232 * then this inode cannot be flushed and is skipped.
3235 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3237 if (!xfs_iflock_nowait(iq)) {
3238 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3241 if (xfs_ipincount(iq)) {
3243 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3248 * arriving here means that this inode can be flushed. First
3249 * re-check that it's dirty before flushing.
3251 if (!xfs_inode_clean(iq)) {
3253 error = xfs_iflush_int(iq, bp);
3255 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3256 goto cluster_corrupt_out;
3262 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3266 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3267 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3278 cluster_corrupt_out:
3280 * Corruption detected in the clustering loop. Invalidate the
3281 * inode buffer and shut down the filesystem.
3285 * Clean up the buffer. If it was delwri, just release it --
3286 * brelse can handle it with no problems. If not, shut down the
3287 * filesystem before releasing the buffer.
3289 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3293 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3295 if (!bufwasdelwri) {
3297 * Just like incore_relse: if we have b_iodone functions,
3298 * mark the buffer as an error and call them. Otherwise
3299 * mark it as stale and brelse.
3304 xfs_buf_ioerror(bp, -EIO);
3313 * Unlocks the flush lock
3315 xfs_iflush_abort(iq, false);
3318 return -EFSCORRUPTED;
3322 * Flush dirty inode metadata into the backing buffer.
3324 * The caller must have the inode lock and the inode flush lock held. The
3325 * inode lock will still be held upon return to the caller, and the inode
3326 * flush lock will be released after the inode has reached the disk.
3328 * The caller must write out the buffer returned in *bpp and release it.
3332 struct xfs_inode *ip,
3333 struct xfs_buf **bpp)
3335 struct xfs_mount *mp = ip->i_mount;
3337 struct xfs_dinode *dip;
3340 XFS_STATS_INC(mp, xs_iflush_count);
3342 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3343 ASSERT(xfs_isiflocked(ip));
3344 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3345 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3349 xfs_iunpin_wait(ip);
3352 * For stale inodes we cannot rely on the backing buffer remaining
3353 * stale in cache for the remaining life of the stale inode and so
3354 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3355 * inodes below. We have to check this after ensuring the inode is
3356 * unpinned so that it is safe to reclaim the stale inode after the
3359 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3365 * This may have been unpinned because the filesystem is shutting
3366 * down forcibly. If that's the case we must not write this inode
3367 * to disk, because the log record didn't make it to disk.
3369 * We also have to remove the log item from the AIL in this case,
3370 * as we wait for an empty AIL as part of the unmount process.
3372 if (XFS_FORCED_SHUTDOWN(mp)) {
3378 * Get the buffer containing the on-disk inode.
3380 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3388 * First flush out the inode that xfs_iflush was called with.
3390 error = xfs_iflush_int(ip, bp);
3395 * If the buffer is pinned then push on the log now so we won't
3396 * get stuck waiting in the write for too long.
3398 if (xfs_buf_ispinned(bp))
3399 xfs_log_force(mp, 0);
3403 * see if other inodes can be gathered into this write
3405 error = xfs_iflush_cluster(ip, bp);
3407 goto cluster_corrupt_out;
3414 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3415 cluster_corrupt_out:
3416 error = -EFSCORRUPTED;
3419 * Unlocks the flush lock
3421 xfs_iflush_abort(ip, false);
3427 struct xfs_inode *ip,
3430 struct xfs_inode_log_item *iip = ip->i_itemp;
3431 struct xfs_dinode *dip;
3432 struct xfs_mount *mp = ip->i_mount;
3434 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3435 ASSERT(xfs_isiflocked(ip));
3436 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3437 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3438 ASSERT(iip != NULL && iip->ili_fields != 0);
3439 ASSERT(ip->i_d.di_version > 1);
3441 /* set *dip = inode's place in the buffer */
3442 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3444 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3445 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3446 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3447 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3448 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3451 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3452 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3453 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3454 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3455 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3458 if (S_ISREG(ip->i_d.di_mode)) {
3460 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3461 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3462 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3463 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3464 "%s: Bad regular inode %Lu, ptr 0x%p",
3465 __func__, ip->i_ino, ip);
3468 } else if (S_ISDIR(ip->i_d.di_mode)) {
3470 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3471 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3472 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3473 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3474 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3475 "%s: Bad directory inode %Lu, ptr 0x%p",
3476 __func__, ip->i_ino, ip);
3480 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3481 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3482 XFS_RANDOM_IFLUSH_5)) {
3483 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3484 "%s: detected corrupt incore inode %Lu, "
3485 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3486 __func__, ip->i_ino,
3487 ip->i_d.di_nextents + ip->i_d.di_anextents,
3488 ip->i_d.di_nblocks, ip);
3491 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3492 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3493 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3494 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3495 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3500 * Inode item log recovery for v2 inodes are dependent on the
3501 * di_flushiter count for correct sequencing. We bump the flush
3502 * iteration count so we can detect flushes which postdate a log record
3503 * during recovery. This is redundant as we now log every change and
3504 * hence this can't happen but we need to still do it to ensure
3505 * backwards compatibility with old kernels that predate logging all
3508 if (ip->i_d.di_version < 3)
3509 ip->i_d.di_flushiter++;
3512 * Copy the dirty parts of the inode into the on-disk
3513 * inode. We always copy out the core of the inode,
3514 * because if the inode is dirty at all the core must
3517 xfs_dinode_to_disk(dip, &ip->i_d);
3519 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3520 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3521 ip->i_d.di_flushiter = 0;
3523 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3524 if (XFS_IFORK_Q(ip))
3525 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3526 xfs_inobp_check(mp, bp);
3529 * We've recorded everything logged in the inode, so we'd like to clear
3530 * the ili_fields bits so we don't log and flush things unnecessarily.
3531 * However, we can't stop logging all this information until the data
3532 * we've copied into the disk buffer is written to disk. If we did we
3533 * might overwrite the copy of the inode in the log with all the data
3534 * after re-logging only part of it, and in the face of a crash we
3535 * wouldn't have all the data we need to recover.
3537 * What we do is move the bits to the ili_last_fields field. When
3538 * logging the inode, these bits are moved back to the ili_fields field.
3539 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3540 * know that the information those bits represent is permanently on
3541 * disk. As long as the flush completes before the inode is logged
3542 * again, then both ili_fields and ili_last_fields will be cleared.
3544 * We can play with the ili_fields bits here, because the inode lock
3545 * must be held exclusively in order to set bits there and the flush
3546 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3547 * done routine can tell whether or not to look in the AIL. Also, store
3548 * the current LSN of the inode so that we can tell whether the item has
3549 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3550 * need the AIL lock, because it is a 64 bit value that cannot be read
3553 iip->ili_last_fields = iip->ili_fields;
3554 iip->ili_fields = 0;
3555 iip->ili_fsync_fields = 0;
3556 iip->ili_logged = 1;
3558 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3559 &iip->ili_item.li_lsn);
3562 * Attach the function xfs_iflush_done to the inode's
3563 * buffer. This will remove the inode from the AIL
3564 * and unlock the inode's flush lock when the inode is
3565 * completely written to disk.
3567 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3569 /* update the lsn in the on disk inode if required */
3570 if (ip->i_d.di_version == 3)
3571 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3573 /* generate the checksum. */
3574 xfs_dinode_calc_crc(mp, dip);
3576 ASSERT(bp->b_fspriv != NULL);
3577 ASSERT(bp->b_iodone != NULL);
3581 return -EFSCORRUPTED;