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_format.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_btree.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_error.h"
43 #include "xfs_utils.h"
44 #include "xfs_quota.h"
45 #include "xfs_filestream.h"
46 #include "xfs_vnodeops.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
51 kmem_zone_t *xfs_inode_zone;
54 * Used in xfs_itruncate_extents(). This is the maximum number of extents
55 * freed from a file in a single transaction.
57 #define XFS_ITRUNC_MAX_EXTENTS 2
59 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
62 * helper function to extract extent size hint from inode
68 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
69 return ip->i_d.di_extsize;
70 if (XFS_IS_REALTIME_INODE(ip))
71 return ip->i_mount->m_sb.sb_rextsize;
76 * This is a wrapper routine around the xfs_ilock() routine used to centralize
77 * some grungy code. It is used in places that wish to lock the inode solely
78 * for reading the extents. The reason these places can't just call
79 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
80 * extents from disk for a file in b-tree format. If the inode is in b-tree
81 * format, then we need to lock the inode exclusively until the extents are read
82 * in. Locking it exclusively all the time would limit our parallelism
83 * unnecessarily, though. What we do instead is check to see if the extents
84 * have been read in yet, and only lock the inode exclusively if they have not.
86 * The function returns a value which should be given to the corresponding
87 * xfs_iunlock_map_shared(). This value is the mode in which the lock was
96 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
97 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
98 lock_mode = XFS_ILOCK_EXCL;
100 lock_mode = XFS_ILOCK_SHARED;
103 xfs_ilock(ip, lock_mode);
109 * This is simply the unlock routine to go with xfs_ilock_map_shared().
110 * All it does is call xfs_iunlock() with the given lock_mode.
113 xfs_iunlock_map_shared(
115 unsigned int lock_mode)
117 xfs_iunlock(ip, lock_mode);
121 * The xfs inode contains 2 locks: a multi-reader lock called the
122 * i_iolock and a multi-reader lock called the i_lock. This routine
123 * allows either or both of the locks to be obtained.
125 * The 2 locks should always be ordered so that the IO lock is
126 * obtained first in order to prevent deadlock.
128 * ip -- the inode being locked
129 * lock_flags -- this parameter indicates the inode's locks
130 * to be locked. It can be:
135 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
136 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
137 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
138 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
145 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
148 * You can't set both SHARED and EXCL for the same lock,
149 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
150 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
152 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
153 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
154 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
155 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
156 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
158 if (lock_flags & XFS_IOLOCK_EXCL)
159 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
160 else if (lock_flags & XFS_IOLOCK_SHARED)
161 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
163 if (lock_flags & XFS_ILOCK_EXCL)
164 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
165 else if (lock_flags & XFS_ILOCK_SHARED)
166 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
170 * This is just like xfs_ilock(), except that the caller
171 * is guaranteed not to sleep. It returns 1 if it gets
172 * the requested locks and 0 otherwise. If the IO lock is
173 * obtained but the inode lock cannot be, then the IO lock
174 * is dropped before returning.
176 * ip -- the inode being locked
177 * lock_flags -- this parameter indicates the inode's locks to be
178 * to be locked. See the comment for xfs_ilock() for a list
186 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
189 * You can't set both SHARED and EXCL for the same lock,
190 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
191 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
193 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
194 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
195 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
196 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
197 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
199 if (lock_flags & XFS_IOLOCK_EXCL) {
200 if (!mrtryupdate(&ip->i_iolock))
202 } else if (lock_flags & XFS_IOLOCK_SHARED) {
203 if (!mrtryaccess(&ip->i_iolock))
206 if (lock_flags & XFS_ILOCK_EXCL) {
207 if (!mrtryupdate(&ip->i_lock))
208 goto out_undo_iolock;
209 } else if (lock_flags & XFS_ILOCK_SHARED) {
210 if (!mrtryaccess(&ip->i_lock))
211 goto out_undo_iolock;
216 if (lock_flags & XFS_IOLOCK_EXCL)
217 mrunlock_excl(&ip->i_iolock);
218 else if (lock_flags & XFS_IOLOCK_SHARED)
219 mrunlock_shared(&ip->i_iolock);
225 * xfs_iunlock() is used to drop the inode locks acquired with
226 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
227 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
228 * that we know which locks to drop.
230 * ip -- the inode being unlocked
231 * lock_flags -- this parameter indicates the inode's locks to be
232 * to be unlocked. See the comment for xfs_ilock() for a list
233 * of valid values for this parameter.
242 * You can't set both SHARED and EXCL for the same lock,
243 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
244 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
246 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
247 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
248 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
249 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
250 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
251 ASSERT(lock_flags != 0);
253 if (lock_flags & XFS_IOLOCK_EXCL)
254 mrunlock_excl(&ip->i_iolock);
255 else if (lock_flags & XFS_IOLOCK_SHARED)
256 mrunlock_shared(&ip->i_iolock);
258 if (lock_flags & XFS_ILOCK_EXCL)
259 mrunlock_excl(&ip->i_lock);
260 else if (lock_flags & XFS_ILOCK_SHARED)
261 mrunlock_shared(&ip->i_lock);
263 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
267 * give up write locks. the i/o lock cannot be held nested
268 * if it is being demoted.
275 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
276 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
278 if (lock_flags & XFS_ILOCK_EXCL)
279 mrdemote(&ip->i_lock);
280 if (lock_flags & XFS_IOLOCK_EXCL)
281 mrdemote(&ip->i_iolock);
283 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
286 #if defined(DEBUG) || defined(XFS_WARN)
292 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
293 if (!(lock_flags & XFS_ILOCK_SHARED))
294 return !!ip->i_lock.mr_writer;
295 return rwsem_is_locked(&ip->i_lock.mr_lock);
298 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
299 if (!(lock_flags & XFS_IOLOCK_SHARED))
300 return !!ip->i_iolock.mr_writer;
301 return rwsem_is_locked(&ip->i_iolock.mr_lock);
311 struct xfs_inode *ip)
313 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
314 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
317 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
318 if (xfs_isiflocked(ip))
320 } while (!xfs_iflock_nowait(ip));
322 finish_wait(wq, &wait.wait);
331 if (di_flags & XFS_DIFLAG_ANY) {
332 if (di_flags & XFS_DIFLAG_REALTIME)
333 flags |= XFS_XFLAG_REALTIME;
334 if (di_flags & XFS_DIFLAG_PREALLOC)
335 flags |= XFS_XFLAG_PREALLOC;
336 if (di_flags & XFS_DIFLAG_IMMUTABLE)
337 flags |= XFS_XFLAG_IMMUTABLE;
338 if (di_flags & XFS_DIFLAG_APPEND)
339 flags |= XFS_XFLAG_APPEND;
340 if (di_flags & XFS_DIFLAG_SYNC)
341 flags |= XFS_XFLAG_SYNC;
342 if (di_flags & XFS_DIFLAG_NOATIME)
343 flags |= XFS_XFLAG_NOATIME;
344 if (di_flags & XFS_DIFLAG_NODUMP)
345 flags |= XFS_XFLAG_NODUMP;
346 if (di_flags & XFS_DIFLAG_RTINHERIT)
347 flags |= XFS_XFLAG_RTINHERIT;
348 if (di_flags & XFS_DIFLAG_PROJINHERIT)
349 flags |= XFS_XFLAG_PROJINHERIT;
350 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
351 flags |= XFS_XFLAG_NOSYMLINKS;
352 if (di_flags & XFS_DIFLAG_EXTSIZE)
353 flags |= XFS_XFLAG_EXTSIZE;
354 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
355 flags |= XFS_XFLAG_EXTSZINHERIT;
356 if (di_flags & XFS_DIFLAG_NODEFRAG)
357 flags |= XFS_XFLAG_NODEFRAG;
358 if (di_flags & XFS_DIFLAG_FILESTREAM)
359 flags |= XFS_XFLAG_FILESTREAM;
369 xfs_icdinode_t *dic = &ip->i_d;
371 return _xfs_dic2xflags(dic->di_flags) |
372 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
379 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
380 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
384 * Allocate an inode on disk and return a copy of its in-core version.
385 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
386 * appropriately within the inode. The uid and gid for the inode are
387 * set according to the contents of the given cred structure.
389 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
390 * has a free inode available, call xfs_iget() to obtain the in-core
391 * version of the allocated inode. Finally, fill in the inode and
392 * log its initial contents. In this case, ialloc_context would be
395 * If xfs_dialloc() does not have an available inode, it will replenish
396 * its supply by doing an allocation. Since we can only do one
397 * allocation within a transaction without deadlocks, we must commit
398 * the current transaction before returning the inode itself.
399 * In this case, therefore, we will set ialloc_context and return.
400 * The caller should then commit the current transaction, start a new
401 * transaction, and call xfs_ialloc() again to actually get the inode.
403 * To ensure that some other process does not grab the inode that
404 * was allocated during the first call to xfs_ialloc(), this routine
405 * also returns the [locked] bp pointing to the head of the freelist
406 * as ialloc_context. The caller should hold this buffer across
407 * the commit and pass it back into this routine on the second call.
409 * If we are allocating quota inodes, we do not have a parent inode
410 * to attach to or associate with (i.e. pip == NULL) because they
411 * are not linked into the directory structure - they are attached
412 * directly to the superblock - and so have no parent.
423 xfs_buf_t **ialloc_context,
426 struct xfs_mount *mp = tp->t_mountp;
435 * Call the space management code to pick
436 * the on-disk inode to be allocated.
438 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
439 ialloc_context, &ino);
442 if (*ialloc_context || ino == NULLFSINO) {
446 ASSERT(*ialloc_context == NULL);
449 * Get the in-core inode with the lock held exclusively.
450 * This is because we're setting fields here we need
451 * to prevent others from looking at until we're done.
453 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
454 XFS_ILOCK_EXCL, &ip);
459 ip->i_d.di_mode = mode;
460 ip->i_d.di_onlink = 0;
461 ip->i_d.di_nlink = nlink;
462 ASSERT(ip->i_d.di_nlink == nlink);
463 ip->i_d.di_uid = current_fsuid();
464 ip->i_d.di_gid = current_fsgid();
465 xfs_set_projid(ip, prid);
466 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
469 * If the superblock version is up to where we support new format
470 * inodes and this is currently an old format inode, then change
471 * the inode version number now. This way we only do the conversion
472 * here rather than here and in the flush/logging code.
474 if (xfs_sb_version_hasnlink(&mp->m_sb) &&
475 ip->i_d.di_version == 1) {
476 ip->i_d.di_version = 2;
478 * We've already zeroed the old link count, the projid field,
484 * Project ids won't be stored on disk if we are using a version 1 inode.
486 if ((prid != 0) && (ip->i_d.di_version == 1))
487 xfs_bump_ino_vers2(tp, ip);
489 if (pip && XFS_INHERIT_GID(pip)) {
490 ip->i_d.di_gid = pip->i_d.di_gid;
491 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
492 ip->i_d.di_mode |= S_ISGID;
497 * If the group ID of the new file does not match the effective group
498 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
499 * (and only if the irix_sgid_inherit compatibility variable is set).
501 if ((irix_sgid_inherit) &&
502 (ip->i_d.di_mode & S_ISGID) &&
503 (!in_group_p((gid_t)ip->i_d.di_gid))) {
504 ip->i_d.di_mode &= ~S_ISGID;
508 ip->i_d.di_nextents = 0;
509 ASSERT(ip->i_d.di_nblocks == 0);
512 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
513 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
514 ip->i_d.di_atime = ip->i_d.di_mtime;
515 ip->i_d.di_ctime = ip->i_d.di_mtime;
518 * di_gen will have been taken care of in xfs_iread.
520 ip->i_d.di_extsize = 0;
521 ip->i_d.di_dmevmask = 0;
522 ip->i_d.di_dmstate = 0;
523 ip->i_d.di_flags = 0;
525 if (ip->i_d.di_version == 3) {
526 ASSERT(ip->i_d.di_ino == ino);
527 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
529 ip->i_d.di_changecount = 1;
531 ip->i_d.di_flags2 = 0;
532 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
533 ip->i_d.di_crtime = ip->i_d.di_mtime;
537 flags = XFS_ILOG_CORE;
538 switch (mode & S_IFMT) {
543 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
544 ip->i_df.if_u2.if_rdev = rdev;
545 ip->i_df.if_flags = 0;
546 flags |= XFS_ILOG_DEV;
550 * we can't set up filestreams until after the VFS inode
551 * is set up properly.
553 if (pip && xfs_inode_is_filestream(pip))
557 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
561 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
562 di_flags |= XFS_DIFLAG_RTINHERIT;
563 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
564 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
565 ip->i_d.di_extsize = pip->i_d.di_extsize;
567 } else if (S_ISREG(mode)) {
568 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
569 di_flags |= XFS_DIFLAG_REALTIME;
570 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
571 di_flags |= XFS_DIFLAG_EXTSIZE;
572 ip->i_d.di_extsize = pip->i_d.di_extsize;
575 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
577 di_flags |= XFS_DIFLAG_NOATIME;
578 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
580 di_flags |= XFS_DIFLAG_NODUMP;
581 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
583 di_flags |= XFS_DIFLAG_SYNC;
584 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
585 xfs_inherit_nosymlinks)
586 di_flags |= XFS_DIFLAG_NOSYMLINKS;
587 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
588 di_flags |= XFS_DIFLAG_PROJINHERIT;
589 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
590 xfs_inherit_nodefrag)
591 di_flags |= XFS_DIFLAG_NODEFRAG;
592 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
593 di_flags |= XFS_DIFLAG_FILESTREAM;
594 ip->i_d.di_flags |= di_flags;
598 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
599 ip->i_df.if_flags = XFS_IFEXTENTS;
600 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
601 ip->i_df.if_u1.if_extents = NULL;
607 * Attribute fork settings for new inode.
609 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
610 ip->i_d.di_anextents = 0;
613 * Log the new values stuffed into the inode.
615 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
616 xfs_trans_log_inode(tp, ip, flags);
618 /* now that we have an i_mode we can setup inode ops and unlock */
621 /* now we have set up the vfs inode we can associate the filestream */
623 error = xfs_filestream_associate(pip, ip);
627 xfs_iflags_set(ip, XFS_IFILESTREAM);
635 * Free up the underlying blocks past new_size. The new size must be smaller
636 * than the current size. This routine can be used both for the attribute and
637 * data fork, and does not modify the inode size, which is left to the caller.
639 * The transaction passed to this routine must have made a permanent log
640 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
641 * given transaction and start new ones, so make sure everything involved in
642 * the transaction is tidy before calling here. Some transaction will be
643 * returned to the caller to be committed. The incoming transaction must
644 * already include the inode, and both inode locks must be held exclusively.
645 * The inode must also be "held" within the transaction. On return the inode
646 * will be "held" within the returned transaction. This routine does NOT
647 * require any disk space to be reserved for it within the transaction.
649 * If we get an error, we must return with the inode locked and linked into the
650 * current transaction. This keeps things simple for the higher level code,
651 * because it always knows that the inode is locked and held in the transaction
652 * that returns to it whether errors occur or not. We don't mark the inode
653 * dirty on error so that transactions can be easily aborted if possible.
656 xfs_itruncate_extents(
657 struct xfs_trans **tpp,
658 struct xfs_inode *ip,
660 xfs_fsize_t new_size)
662 struct xfs_mount *mp = ip->i_mount;
663 struct xfs_trans *tp = *tpp;
664 struct xfs_trans *ntp;
665 xfs_bmap_free_t free_list;
666 xfs_fsblock_t first_block;
667 xfs_fileoff_t first_unmap_block;
668 xfs_fileoff_t last_block;
669 xfs_filblks_t unmap_len;
674 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
675 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
676 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
677 ASSERT(new_size <= XFS_ISIZE(ip));
678 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
679 ASSERT(ip->i_itemp != NULL);
680 ASSERT(ip->i_itemp->ili_lock_flags == 0);
681 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
683 trace_xfs_itruncate_extents_start(ip, new_size);
686 * Since it is possible for space to become allocated beyond
687 * the end of the file (in a crash where the space is allocated
688 * but the inode size is not yet updated), simply remove any
689 * blocks which show up between the new EOF and the maximum
690 * possible file size. If the first block to be removed is
691 * beyond the maximum file size (ie it is the same as last_block),
692 * then there is nothing to do.
694 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
695 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
696 if (first_unmap_block == last_block)
699 ASSERT(first_unmap_block < last_block);
700 unmap_len = last_block - first_unmap_block + 1;
702 xfs_bmap_init(&free_list, &first_block);
703 error = xfs_bunmapi(tp, ip,
704 first_unmap_block, unmap_len,
705 xfs_bmapi_aflag(whichfork),
706 XFS_ITRUNC_MAX_EXTENTS,
707 &first_block, &free_list,
710 goto out_bmap_cancel;
713 * Duplicate the transaction that has the permanent
714 * reservation and commit the old transaction.
716 error = xfs_bmap_finish(&tp, &free_list, &committed);
718 xfs_trans_ijoin(tp, ip, 0);
720 goto out_bmap_cancel;
724 * Mark the inode dirty so it will be logged and
725 * moved forward in the log as part of every commit.
727 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
730 ntp = xfs_trans_dup(tp);
731 error = xfs_trans_commit(tp, 0);
734 xfs_trans_ijoin(tp, ip, 0);
740 * Transaction commit worked ok so we can drop the extra ticket
741 * reference that we gained in xfs_trans_dup()
743 xfs_log_ticket_put(tp->t_ticket);
744 error = xfs_trans_reserve(tp, 0,
745 XFS_ITRUNCATE_LOG_RES(mp), 0,
746 XFS_TRANS_PERM_LOG_RES,
747 XFS_ITRUNCATE_LOG_COUNT);
753 * Always re-log the inode so that our permanent transaction can keep
754 * on rolling it forward in the log.
756 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
758 trace_xfs_itruncate_extents_end(ip, new_size);
765 * If the bunmapi call encounters an error, return to the caller where
766 * the transaction can be properly aborted. We just need to make sure
767 * we're not holding any resources that we were not when we came in.
769 xfs_bmap_cancel(&free_list);
774 * This is called when the inode's link count goes to 0.
775 * We place the on-disk inode on a list in the AGI. It
776 * will be pulled from this list when the inode is freed.
793 ASSERT(ip->i_d.di_nlink == 0);
794 ASSERT(ip->i_d.di_mode != 0);
799 * Get the agi buffer first. It ensures lock ordering
802 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
805 agi = XFS_BUF_TO_AGI(agibp);
808 * Get the index into the agi hash table for the
809 * list this inode will go on.
811 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
813 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
814 ASSERT(agi->agi_unlinked[bucket_index]);
815 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
817 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
819 * There is already another inode in the bucket we need
820 * to add ourselves to. Add us at the front of the list.
821 * Here we put the head pointer into our next pointer,
822 * and then we fall through to point the head at us.
824 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
829 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
830 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
831 offset = ip->i_imap.im_boffset +
832 offsetof(xfs_dinode_t, di_next_unlinked);
834 /* need to recalc the inode CRC if appropriate */
835 xfs_dinode_calc_crc(mp, dip);
837 xfs_trans_inode_buf(tp, ibp);
838 xfs_trans_log_buf(tp, ibp, offset,
839 (offset + sizeof(xfs_agino_t) - 1));
840 xfs_inobp_check(mp, ibp);
844 * Point the bucket head pointer at the inode being inserted.
847 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
848 offset = offsetof(xfs_agi_t, agi_unlinked) +
849 (sizeof(xfs_agino_t) * bucket_index);
850 xfs_trans_log_buf(tp, agibp, offset,
851 (offset + sizeof(xfs_agino_t) - 1));
856 * Pull the on-disk inode from the AGI unlinked list.
871 xfs_agino_t next_agino;
873 xfs_dinode_t *last_dip = NULL;
875 int offset, last_offset = 0;
879 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
882 * Get the agi buffer first. It ensures lock ordering
885 error = xfs_read_agi(mp, tp, agno, &agibp);
889 agi = XFS_BUF_TO_AGI(agibp);
892 * Get the index into the agi hash table for the
893 * list this inode will go on.
895 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
897 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
898 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
899 ASSERT(agi->agi_unlinked[bucket_index]);
901 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
903 * We're at the head of the list. Get the inode's on-disk
904 * buffer to see if there is anyone after us on the list.
905 * Only modify our next pointer if it is not already NULLAGINO.
906 * This saves us the overhead of dealing with the buffer when
907 * there is no need to change it.
909 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
912 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
916 next_agino = be32_to_cpu(dip->di_next_unlinked);
917 ASSERT(next_agino != 0);
918 if (next_agino != NULLAGINO) {
919 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
920 offset = ip->i_imap.im_boffset +
921 offsetof(xfs_dinode_t, di_next_unlinked);
923 /* need to recalc the inode CRC if appropriate */
924 xfs_dinode_calc_crc(mp, dip);
926 xfs_trans_inode_buf(tp, ibp);
927 xfs_trans_log_buf(tp, ibp, offset,
928 (offset + sizeof(xfs_agino_t) - 1));
929 xfs_inobp_check(mp, ibp);
931 xfs_trans_brelse(tp, ibp);
934 * Point the bucket head pointer at the next inode.
936 ASSERT(next_agino != 0);
937 ASSERT(next_agino != agino);
938 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
939 offset = offsetof(xfs_agi_t, agi_unlinked) +
940 (sizeof(xfs_agino_t) * bucket_index);
941 xfs_trans_log_buf(tp, agibp, offset,
942 (offset + sizeof(xfs_agino_t) - 1));
945 * We need to search the list for the inode being freed.
947 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
949 while (next_agino != agino) {
950 struct xfs_imap imap;
953 xfs_trans_brelse(tp, last_ibp);
956 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
958 error = xfs_imap(mp, tp, next_ino, &imap, 0);
961 "%s: xfs_imap returned error %d.",
966 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
970 "%s: xfs_imap_to_bp returned error %d.",
975 last_offset = imap.im_boffset;
976 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
977 ASSERT(next_agino != NULLAGINO);
978 ASSERT(next_agino != 0);
982 * Now last_ibp points to the buffer previous to us on the
983 * unlinked list. Pull us from the list.
985 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
988 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
992 next_agino = be32_to_cpu(dip->di_next_unlinked);
993 ASSERT(next_agino != 0);
994 ASSERT(next_agino != agino);
995 if (next_agino != NULLAGINO) {
996 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
997 offset = ip->i_imap.im_boffset +
998 offsetof(xfs_dinode_t, di_next_unlinked);
1000 /* need to recalc the inode CRC if appropriate */
1001 xfs_dinode_calc_crc(mp, dip);
1003 xfs_trans_inode_buf(tp, ibp);
1004 xfs_trans_log_buf(tp, ibp, offset,
1005 (offset + sizeof(xfs_agino_t) - 1));
1006 xfs_inobp_check(mp, ibp);
1008 xfs_trans_brelse(tp, ibp);
1011 * Point the previous inode on the list to the next inode.
1013 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1014 ASSERT(next_agino != 0);
1015 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1017 /* need to recalc the inode CRC if appropriate */
1018 xfs_dinode_calc_crc(mp, last_dip);
1020 xfs_trans_inode_buf(tp, last_ibp);
1021 xfs_trans_log_buf(tp, last_ibp, offset,
1022 (offset + sizeof(xfs_agino_t) - 1));
1023 xfs_inobp_check(mp, last_ibp);
1029 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1030 * inodes that are in memory - they all must be marked stale and attached to
1031 * the cluster buffer.
1035 xfs_inode_t *free_ip,
1039 xfs_mount_t *mp = free_ip->i_mount;
1040 int blks_per_cluster;
1047 xfs_inode_log_item_t *iip;
1048 xfs_log_item_t *lip;
1049 struct xfs_perag *pag;
1051 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1052 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1053 blks_per_cluster = 1;
1054 ninodes = mp->m_sb.sb_inopblock;
1055 nbufs = XFS_IALLOC_BLOCKS(mp);
1057 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1058 mp->m_sb.sb_blocksize;
1059 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1060 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1063 for (j = 0; j < nbufs; j++, inum += ninodes) {
1064 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1065 XFS_INO_TO_AGBNO(mp, inum));
1068 * We obtain and lock the backing buffer first in the process
1069 * here, as we have to ensure that any dirty inode that we
1070 * can't get the flush lock on is attached to the buffer.
1071 * If we scan the in-memory inodes first, then buffer IO can
1072 * complete before we get a lock on it, and hence we may fail
1073 * to mark all the active inodes on the buffer stale.
1075 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1076 mp->m_bsize * blks_per_cluster,
1083 * This buffer may not have been correctly initialised as we
1084 * didn't read it from disk. That's not important because we are
1085 * only using to mark the buffer as stale in the log, and to
1086 * attach stale cached inodes on it. That means it will never be
1087 * dispatched for IO. If it is, we want to know about it, and we
1088 * want it to fail. We can acheive this by adding a write
1089 * verifier to the buffer.
1091 bp->b_ops = &xfs_inode_buf_ops;
1094 * Walk the inodes already attached to the buffer and mark them
1095 * stale. These will all have the flush locks held, so an
1096 * in-memory inode walk can't lock them. By marking them all
1097 * stale first, we will not attempt to lock them in the loop
1098 * below as the XFS_ISTALE flag will be set.
1102 if (lip->li_type == XFS_LI_INODE) {
1103 iip = (xfs_inode_log_item_t *)lip;
1104 ASSERT(iip->ili_logged == 1);
1105 lip->li_cb = xfs_istale_done;
1106 xfs_trans_ail_copy_lsn(mp->m_ail,
1107 &iip->ili_flush_lsn,
1108 &iip->ili_item.li_lsn);
1109 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1111 lip = lip->li_bio_list;
1116 * For each inode in memory attempt to add it to the inode
1117 * buffer and set it up for being staled on buffer IO
1118 * completion. This is safe as we've locked out tail pushing
1119 * and flushing by locking the buffer.
1121 * We have already marked every inode that was part of a
1122 * transaction stale above, which means there is no point in
1123 * even trying to lock them.
1125 for (i = 0; i < ninodes; i++) {
1128 ip = radix_tree_lookup(&pag->pag_ici_root,
1129 XFS_INO_TO_AGINO(mp, (inum + i)));
1131 /* Inode not in memory, nothing to do */
1138 * because this is an RCU protected lookup, we could
1139 * find a recently freed or even reallocated inode
1140 * during the lookup. We need to check under the
1141 * i_flags_lock for a valid inode here. Skip it if it
1142 * is not valid, the wrong inode or stale.
1144 spin_lock(&ip->i_flags_lock);
1145 if (ip->i_ino != inum + i ||
1146 __xfs_iflags_test(ip, XFS_ISTALE)) {
1147 spin_unlock(&ip->i_flags_lock);
1151 spin_unlock(&ip->i_flags_lock);
1154 * Don't try to lock/unlock the current inode, but we
1155 * _cannot_ skip the other inodes that we did not find
1156 * in the list attached to the buffer and are not
1157 * already marked stale. If we can't lock it, back off
1160 if (ip != free_ip &&
1161 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1169 xfs_iflags_set(ip, XFS_ISTALE);
1172 * we don't need to attach clean inodes or those only
1173 * with unlogged changes (which we throw away, anyway).
1176 if (!iip || xfs_inode_clean(ip)) {
1177 ASSERT(ip != free_ip);
1179 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1183 iip->ili_last_fields = iip->ili_fields;
1184 iip->ili_fields = 0;
1185 iip->ili_logged = 1;
1186 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1187 &iip->ili_item.li_lsn);
1189 xfs_buf_attach_iodone(bp, xfs_istale_done,
1193 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1196 xfs_trans_stale_inode_buf(tp, bp);
1197 xfs_trans_binval(tp, bp);
1205 * This is called to return an inode to the inode free list.
1206 * The inode should already be truncated to 0 length and have
1207 * no pages associated with it. This routine also assumes that
1208 * the inode is already a part of the transaction.
1210 * The on-disk copy of the inode will have been added to the list
1211 * of unlinked inodes in the AGI. We need to remove the inode from
1212 * that list atomically with respect to freeing it here.
1218 xfs_bmap_free_t *flist)
1222 xfs_ino_t first_ino;
1224 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1225 ASSERT(ip->i_d.di_nlink == 0);
1226 ASSERT(ip->i_d.di_nextents == 0);
1227 ASSERT(ip->i_d.di_anextents == 0);
1228 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
1229 ASSERT(ip->i_d.di_nblocks == 0);
1232 * Pull the on-disk inode from the AGI unlinked list.
1234 error = xfs_iunlink_remove(tp, ip);
1238 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1242 ip->i_d.di_mode = 0; /* mark incore inode as free */
1243 ip->i_d.di_flags = 0;
1244 ip->i_d.di_dmevmask = 0;
1245 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
1246 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1247 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1249 * Bump the generation count so no one will be confused
1250 * by reincarnations of this inode.
1253 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1256 error = xfs_ifree_cluster(ip, tp, first_ino);
1262 * This is called to unpin an inode. The caller must have the inode locked
1263 * in at least shared mode so that the buffer cannot be subsequently pinned
1264 * once someone is waiting for it to be unpinned.
1268 struct xfs_inode *ip)
1270 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1272 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
1274 /* Give the log a push to start the unpinning I/O */
1275 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
1281 struct xfs_inode *ip)
1283 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
1284 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
1289 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1290 if (xfs_ipincount(ip))
1292 } while (xfs_ipincount(ip));
1293 finish_wait(wq, &wait.wait);
1298 struct xfs_inode *ip)
1300 if (xfs_ipincount(ip))
1301 __xfs_iunpin_wait(ip);
1309 xfs_mount_t *mp = ip->i_mount;
1310 struct xfs_perag *pag;
1311 unsigned long first_index, mask;
1312 unsigned long inodes_per_cluster;
1314 xfs_inode_t **ilist;
1321 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1323 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
1324 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
1325 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
1329 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
1330 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
1332 /* really need a gang lookup range call here */
1333 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
1334 first_index, inodes_per_cluster);
1338 for (i = 0; i < nr_found; i++) {
1344 * because this is an RCU protected lookup, we could find a
1345 * recently freed or even reallocated inode during the lookup.
1346 * We need to check under the i_flags_lock for a valid inode
1347 * here. Skip it if it is not valid or the wrong inode.
1349 spin_lock(&ip->i_flags_lock);
1351 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
1352 spin_unlock(&ip->i_flags_lock);
1355 spin_unlock(&ip->i_flags_lock);
1358 * Do an un-protected check to see if the inode is dirty and
1359 * is a candidate for flushing. These checks will be repeated
1360 * later after the appropriate locks are acquired.
1362 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
1366 * Try to get locks. If any are unavailable or it is pinned,
1367 * then this inode cannot be flushed and is skipped.
1370 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
1372 if (!xfs_iflock_nowait(iq)) {
1373 xfs_iunlock(iq, XFS_ILOCK_SHARED);
1376 if (xfs_ipincount(iq)) {
1378 xfs_iunlock(iq, XFS_ILOCK_SHARED);
1383 * arriving here means that this inode can be flushed. First
1384 * re-check that it's dirty before flushing.
1386 if (!xfs_inode_clean(iq)) {
1388 error = xfs_iflush_int(iq, bp);
1390 xfs_iunlock(iq, XFS_ILOCK_SHARED);
1391 goto cluster_corrupt_out;
1397 xfs_iunlock(iq, XFS_ILOCK_SHARED);
1401 XFS_STATS_INC(xs_icluster_flushcnt);
1402 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
1413 cluster_corrupt_out:
1415 * Corruption detected in the clustering loop. Invalidate the
1416 * inode buffer and shut down the filesystem.
1420 * Clean up the buffer. If it was delwri, just release it --
1421 * brelse can handle it with no problems. If not, shut down the
1422 * filesystem before releasing the buffer.
1424 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
1428 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1430 if (!bufwasdelwri) {
1432 * Just like incore_relse: if we have b_iodone functions,
1433 * mark the buffer as an error and call them. Otherwise
1434 * mark it as stale and brelse.
1439 xfs_buf_ioerror(bp, EIO);
1440 xfs_buf_ioend(bp, 0);
1448 * Unlocks the flush lock
1450 xfs_iflush_abort(iq, false);
1453 return XFS_ERROR(EFSCORRUPTED);
1457 * Flush dirty inode metadata into the backing buffer.
1459 * The caller must have the inode lock and the inode flush lock held. The
1460 * inode lock will still be held upon return to the caller, and the inode
1461 * flush lock will be released after the inode has reached the disk.
1463 * The caller must write out the buffer returned in *bpp and release it.
1467 struct xfs_inode *ip,
1468 struct xfs_buf **bpp)
1470 struct xfs_mount *mp = ip->i_mount;
1472 struct xfs_dinode *dip;
1475 XFS_STATS_INC(xs_iflush_count);
1477 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1478 ASSERT(xfs_isiflocked(ip));
1479 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
1480 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
1484 xfs_iunpin_wait(ip);
1487 * For stale inodes we cannot rely on the backing buffer remaining
1488 * stale in cache for the remaining life of the stale inode and so
1489 * xfs_imap_to_bp() below may give us a buffer that no longer contains
1490 * inodes below. We have to check this after ensuring the inode is
1491 * unpinned so that it is safe to reclaim the stale inode after the
1494 if (xfs_iflags_test(ip, XFS_ISTALE)) {
1500 * This may have been unpinned because the filesystem is shutting
1501 * down forcibly. If that's the case we must not write this inode
1502 * to disk, because the log record didn't make it to disk.
1504 * We also have to remove the log item from the AIL in this case,
1505 * as we wait for an empty AIL as part of the unmount process.
1507 if (XFS_FORCED_SHUTDOWN(mp)) {
1508 error = XFS_ERROR(EIO);
1513 * Get the buffer containing the on-disk inode.
1515 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
1523 * First flush out the inode that xfs_iflush was called with.
1525 error = xfs_iflush_int(ip, bp);
1530 * If the buffer is pinned then push on the log now so we won't
1531 * get stuck waiting in the write for too long.
1533 if (xfs_buf_ispinned(bp))
1534 xfs_log_force(mp, 0);
1538 * see if other inodes can be gathered into this write
1540 error = xfs_iflush_cluster(ip, bp);
1542 goto cluster_corrupt_out;
1549 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1550 cluster_corrupt_out:
1551 error = XFS_ERROR(EFSCORRUPTED);
1554 * Unlocks the flush lock
1556 xfs_iflush_abort(ip, false);
1563 struct xfs_inode *ip,
1566 struct xfs_inode_log_item *iip = ip->i_itemp;
1567 struct xfs_dinode *dip;
1568 struct xfs_mount *mp = ip->i_mount;
1570 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1571 ASSERT(xfs_isiflocked(ip));
1572 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
1573 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
1574 ASSERT(iip != NULL && iip->ili_fields != 0);
1576 /* set *dip = inode's place in the buffer */
1577 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
1579 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
1580 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
1581 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
1582 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
1583 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
1586 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
1587 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
1588 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
1589 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
1590 __func__, ip->i_ino, ip, ip->i_d.di_magic);
1593 if (S_ISREG(ip->i_d.di_mode)) {
1595 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
1596 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
1597 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
1598 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
1599 "%s: Bad regular inode %Lu, ptr 0x%p",
1600 __func__, ip->i_ino, ip);
1603 } else if (S_ISDIR(ip->i_d.di_mode)) {
1605 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
1606 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
1607 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
1608 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
1609 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
1610 "%s: Bad directory inode %Lu, ptr 0x%p",
1611 __func__, ip->i_ino, ip);
1615 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
1616 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
1617 XFS_RANDOM_IFLUSH_5)) {
1618 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
1619 "%s: detected corrupt incore inode %Lu, "
1620 "total extents = %d, nblocks = %Ld, ptr 0x%p",
1621 __func__, ip->i_ino,
1622 ip->i_d.di_nextents + ip->i_d.di_anextents,
1623 ip->i_d.di_nblocks, ip);
1626 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
1627 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
1628 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
1629 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
1630 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
1635 * Inode item log recovery for v1/v2 inodes are dependent on the
1636 * di_flushiter count for correct sequencing. We bump the flush
1637 * iteration count so we can detect flushes which postdate a log record
1638 * during recovery. This is redundant as we now log every change and
1639 * hence this can't happen but we need to still do it to ensure
1640 * backwards compatibility with old kernels that predate logging all
1643 if (ip->i_d.di_version < 3)
1644 ip->i_d.di_flushiter++;
1647 * Copy the dirty parts of the inode into the on-disk
1648 * inode. We always copy out the core of the inode,
1649 * because if the inode is dirty at all the core must
1652 xfs_dinode_to_disk(dip, &ip->i_d);
1654 /* Wrap, we never let the log put out DI_MAX_FLUSH */
1655 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
1656 ip->i_d.di_flushiter = 0;
1659 * If this is really an old format inode and the superblock version
1660 * has not been updated to support only new format inodes, then
1661 * convert back to the old inode format. If the superblock version
1662 * has been updated, then make the conversion permanent.
1664 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
1665 if (ip->i_d.di_version == 1) {
1666 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1670 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
1671 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
1674 * The superblock version has already been bumped,
1675 * so just make the conversion to the new inode
1678 ip->i_d.di_version = 2;
1679 dip->di_version = 2;
1680 ip->i_d.di_onlink = 0;
1682 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1683 memset(&(dip->di_pad[0]), 0,
1684 sizeof(dip->di_pad));
1685 ASSERT(xfs_get_projid(ip) == 0);
1689 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
1690 if (XFS_IFORK_Q(ip))
1691 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
1692 xfs_inobp_check(mp, bp);
1695 * We've recorded everything logged in the inode, so we'd like to clear
1696 * the ili_fields bits so we don't log and flush things unnecessarily.
1697 * However, we can't stop logging all this information until the data
1698 * we've copied into the disk buffer is written to disk. If we did we
1699 * might overwrite the copy of the inode in the log with all the data
1700 * after re-logging only part of it, and in the face of a crash we
1701 * wouldn't have all the data we need to recover.
1703 * What we do is move the bits to the ili_last_fields field. When
1704 * logging the inode, these bits are moved back to the ili_fields field.
1705 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
1706 * know that the information those bits represent is permanently on
1707 * disk. As long as the flush completes before the inode is logged
1708 * again, then both ili_fields and ili_last_fields will be cleared.
1710 * We can play with the ili_fields bits here, because the inode lock
1711 * must be held exclusively in order to set bits there and the flush
1712 * lock protects the ili_last_fields bits. Set ili_logged so the flush
1713 * done routine can tell whether or not to look in the AIL. Also, store
1714 * the current LSN of the inode so that we can tell whether the item has
1715 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
1716 * need the AIL lock, because it is a 64 bit value that cannot be read
1719 iip->ili_last_fields = iip->ili_fields;
1720 iip->ili_fields = 0;
1721 iip->ili_logged = 1;
1723 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1724 &iip->ili_item.li_lsn);
1727 * Attach the function xfs_iflush_done to the inode's
1728 * buffer. This will remove the inode from the AIL
1729 * and unlock the inode's flush lock when the inode is
1730 * completely written to disk.
1732 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
1734 /* update the lsn in the on disk inode if required */
1735 if (ip->i_d.di_version == 3)
1736 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
1738 /* generate the checksum. */
1739 xfs_dinode_calc_crc(mp, dip);
1741 ASSERT(bp->b_fspriv != NULL);
1742 ASSERT(bp->b_iodone != NULL);
1746 return XFS_ERROR(EFSCORRUPTED);
1750 * Test whether it is appropriate to check an inode for and free post EOF
1751 * blocks. The 'force' parameter determines whether we should also consider
1752 * regular files that are marked preallocated or append-only.
1755 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
1757 /* prealloc/delalloc exists only on regular files */
1758 if (!S_ISREG(ip->i_d.di_mode))
1762 * Zero sized files with no cached pages and delalloc blocks will not
1763 * have speculative prealloc/delalloc blocks to remove.
1765 if (VFS_I(ip)->i_size == 0 &&
1766 VN_CACHED(VFS_I(ip)) == 0 &&
1767 ip->i_delayed_blks == 0)
1770 /* If we haven't read in the extent list, then don't do it now. */
1771 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
1775 * Do not free real preallocated or append-only files unless the file
1776 * has delalloc blocks and we are forced to remove them.
1778 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
1779 if (!force || ip->i_delayed_blks == 0)