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 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
367 * value. This can be called for any type of inode lock combination, including
368 * parent locking. Care must be taken to ensure we don't overrun the subclass
369 * storage fields in the class mask we build.
372 xfs_lock_inumorder(int lock_mode, int subclass)
376 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
379 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
380 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
381 ASSERT(subclass + XFS_IOLOCK_PARENT_VAL <
382 MAX_LOCKDEP_SUBCLASSES);
383 class += subclass << XFS_IOLOCK_SHIFT;
384 if (lock_mode & XFS_IOLOCK_PARENT)
385 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
388 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
389 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
390 class += subclass << XFS_MMAPLOCK_SHIFT;
393 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
394 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
395 class += subclass << XFS_ILOCK_SHIFT;
398 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
402 * The following routine will lock n inodes in exclusive mode. We assume the
403 * caller calls us with the inodes in i_ino order.
405 * We need to detect deadlock where an inode that we lock is in the AIL and we
406 * start waiting for another inode that is locked by a thread in a long running
407 * transaction (such as truncate). This can result in deadlock since the long
408 * running trans might need to wait for the inode we just locked in order to
409 * push the tail and free space in the log.
411 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
412 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
413 * lock more than one at a time, lockdep will report false positives saying we
414 * have violated locking orders.
422 int attempts = 0, i, j, try_lock;
426 * Currently supports between 2 and 5 inodes with exclusive locking. We
427 * support an arbitrary depth of locking here, but absolute limits on
428 * inodes depend on the the type of locking and the limits placed by
429 * lockdep annotations in xfs_lock_inumorder. These are all checked by
432 ASSERT(ips && inodes >= 2 && inodes <= 5);
433 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
435 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
437 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
438 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
439 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
440 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
441 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
442 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
444 if (lock_mode & XFS_IOLOCK_EXCL) {
445 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
446 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
447 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
452 for (; i < inodes; i++) {
455 if (i && (ips[i] == ips[i - 1])) /* Already locked */
459 * If try_lock is not set yet, make sure all locked inodes are
460 * not in the AIL. If any are, set try_lock to be used later.
463 for (j = (i - 1); j >= 0 && !try_lock; j--) {
464 lp = (xfs_log_item_t *)ips[j]->i_itemp;
465 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
471 * If any of the previous locks we have locked is in the AIL,
472 * we must TRY to get the second and subsequent locks. If
473 * we can't get any, we must release all we have
477 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
481 /* try_lock means we have an inode locked that is in the AIL. */
483 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
487 * Unlock all previous guys and try again. xfs_iunlock will try
488 * to push the tail if the inode is in the AIL.
491 for (j = i - 1; j >= 0; j--) {
493 * Check to see if we've already unlocked this one. Not
494 * the first one going back, and the inode ptr is the
497 if (j != (i - 1) && ips[j] == ips[j + 1])
500 xfs_iunlock(ips[j], lock_mode);
503 if ((attempts % 5) == 0) {
504 delay(1); /* Don't just spin the CPU */
516 if (attempts < 5) xfs_small_retries++;
517 else if (attempts < 100) xfs_middle_retries++;
518 else xfs_lots_retries++;
526 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
527 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
528 * lock more than one at a time, lockdep will report false positives saying we
529 * have violated locking orders.
541 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
542 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
543 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
544 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
545 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
547 ASSERT(ip0->i_ino != ip1->i_ino);
549 if (ip0->i_ino > ip1->i_ino) {
556 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
559 * If the first lock we have locked is in the AIL, we must TRY to get
560 * the second lock. If we can't get it, we must release the first one
563 lp = (xfs_log_item_t *)ip0->i_itemp;
564 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
565 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
566 xfs_iunlock(ip0, lock_mode);
567 if ((++attempts % 5) == 0)
568 delay(1); /* Don't just spin the CPU */
572 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
579 struct xfs_inode *ip)
581 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
582 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
585 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
586 if (xfs_isiflocked(ip))
588 } while (!xfs_iflock_nowait(ip));
590 finish_wait(wq, &wait.wait);
599 if (di_flags & XFS_DIFLAG_ANY) {
600 if (di_flags & XFS_DIFLAG_REALTIME)
601 flags |= XFS_XFLAG_REALTIME;
602 if (di_flags & XFS_DIFLAG_PREALLOC)
603 flags |= XFS_XFLAG_PREALLOC;
604 if (di_flags & XFS_DIFLAG_IMMUTABLE)
605 flags |= XFS_XFLAG_IMMUTABLE;
606 if (di_flags & XFS_DIFLAG_APPEND)
607 flags |= XFS_XFLAG_APPEND;
608 if (di_flags & XFS_DIFLAG_SYNC)
609 flags |= XFS_XFLAG_SYNC;
610 if (di_flags & XFS_DIFLAG_NOATIME)
611 flags |= XFS_XFLAG_NOATIME;
612 if (di_flags & XFS_DIFLAG_NODUMP)
613 flags |= XFS_XFLAG_NODUMP;
614 if (di_flags & XFS_DIFLAG_RTINHERIT)
615 flags |= XFS_XFLAG_RTINHERIT;
616 if (di_flags & XFS_DIFLAG_PROJINHERIT)
617 flags |= XFS_XFLAG_PROJINHERIT;
618 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
619 flags |= XFS_XFLAG_NOSYMLINKS;
620 if (di_flags & XFS_DIFLAG_EXTSIZE)
621 flags |= XFS_XFLAG_EXTSIZE;
622 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
623 flags |= XFS_XFLAG_EXTSZINHERIT;
624 if (di_flags & XFS_DIFLAG_NODEFRAG)
625 flags |= XFS_XFLAG_NODEFRAG;
626 if (di_flags & XFS_DIFLAG_FILESTREAM)
627 flags |= XFS_XFLAG_FILESTREAM;
637 xfs_icdinode_t *dic = &ip->i_d;
639 return _xfs_dic2xflags(dic->di_flags) |
640 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
647 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
648 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
652 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
653 * is allowed, otherwise it has to be an exact match. If a CI match is found,
654 * ci_name->name will point to a the actual name (caller must free) or
655 * will be set to NULL if an exact match is found.
660 struct xfs_name *name,
662 struct xfs_name *ci_name)
668 trace_xfs_lookup(dp, name);
670 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
673 lock_mode = xfs_ilock_data_map_shared(dp);
674 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
675 xfs_iunlock(dp, lock_mode);
680 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
688 kmem_free(ci_name->name);
695 * Allocate an inode on disk and return a copy of its in-core version.
696 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
697 * appropriately within the inode. The uid and gid for the inode are
698 * set according to the contents of the given cred structure.
700 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
701 * has a free inode available, call xfs_iget() to obtain the in-core
702 * version of the allocated inode. Finally, fill in the inode and
703 * log its initial contents. In this case, ialloc_context would be
706 * If xfs_dialloc() does not have an available inode, it will replenish
707 * its supply by doing an allocation. Since we can only do one
708 * allocation within a transaction without deadlocks, we must commit
709 * the current transaction before returning the inode itself.
710 * In this case, therefore, we will set ialloc_context and return.
711 * The caller should then commit the current transaction, start a new
712 * transaction, and call xfs_ialloc() again to actually get the inode.
714 * To ensure that some other process does not grab the inode that
715 * was allocated during the first call to xfs_ialloc(), this routine
716 * also returns the [locked] bp pointing to the head of the freelist
717 * as ialloc_context. The caller should hold this buffer across
718 * the commit and pass it back into this routine on the second call.
720 * If we are allocating quota inodes, we do not have a parent inode
721 * to attach to or associate with (i.e. pip == NULL) because they
722 * are not linked into the directory structure - they are attached
723 * directly to the superblock - and so have no parent.
734 xfs_buf_t **ialloc_context,
737 struct xfs_mount *mp = tp->t_mountp;
745 * Call the space management code to pick
746 * the on-disk inode to be allocated.
748 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
749 ialloc_context, &ino);
752 if (*ialloc_context || ino == NULLFSINO) {
756 ASSERT(*ialloc_context == NULL);
759 * Get the in-core inode with the lock held exclusively.
760 * This is because we're setting fields here we need
761 * to prevent others from looking at until we're done.
763 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
764 XFS_ILOCK_EXCL, &ip);
770 * We always convert v1 inodes to v2 now - we only support filesystems
771 * with >= v2 inode capability, so there is no reason for ever leaving
772 * an inode in v1 format.
774 if (ip->i_d.di_version == 1)
775 ip->i_d.di_version = 2;
777 ip->i_d.di_mode = mode;
778 ip->i_d.di_onlink = 0;
779 ip->i_d.di_nlink = nlink;
780 ASSERT(ip->i_d.di_nlink == nlink);
781 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
782 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
783 xfs_set_projid(ip, prid);
784 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
786 if (pip && XFS_INHERIT_GID(pip)) {
787 ip->i_d.di_gid = pip->i_d.di_gid;
788 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
789 ip->i_d.di_mode |= S_ISGID;
794 * If the group ID of the new file does not match the effective group
795 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
796 * (and only if the irix_sgid_inherit compatibility variable is set).
798 if ((irix_sgid_inherit) &&
799 (ip->i_d.di_mode & S_ISGID) &&
800 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
801 ip->i_d.di_mode &= ~S_ISGID;
805 ip->i_d.di_nextents = 0;
806 ASSERT(ip->i_d.di_nblocks == 0);
808 tv = current_fs_time(mp->m_super);
809 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
810 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
811 ip->i_d.di_atime = ip->i_d.di_mtime;
812 ip->i_d.di_ctime = ip->i_d.di_mtime;
815 * di_gen will have been taken care of in xfs_iread.
817 ip->i_d.di_extsize = 0;
818 ip->i_d.di_dmevmask = 0;
819 ip->i_d.di_dmstate = 0;
820 ip->i_d.di_flags = 0;
822 if (ip->i_d.di_version == 3) {
823 ASSERT(ip->i_d.di_ino == ino);
824 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_meta_uuid));
826 ip->i_d.di_changecount = 1;
828 ip->i_d.di_flags2 = 0;
829 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
830 ip->i_d.di_crtime = ip->i_d.di_mtime;
834 flags = XFS_ILOG_CORE;
835 switch (mode & S_IFMT) {
840 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
841 ip->i_df.if_u2.if_rdev = rdev;
842 ip->i_df.if_flags = 0;
843 flags |= XFS_ILOG_DEV;
847 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
851 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
852 di_flags |= XFS_DIFLAG_RTINHERIT;
853 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
854 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
855 ip->i_d.di_extsize = pip->i_d.di_extsize;
857 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
858 di_flags |= XFS_DIFLAG_PROJINHERIT;
859 } else if (S_ISREG(mode)) {
860 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
861 di_flags |= XFS_DIFLAG_REALTIME;
862 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
863 di_flags |= XFS_DIFLAG_EXTSIZE;
864 ip->i_d.di_extsize = pip->i_d.di_extsize;
867 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
869 di_flags |= XFS_DIFLAG_NOATIME;
870 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
872 di_flags |= XFS_DIFLAG_NODUMP;
873 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
875 di_flags |= XFS_DIFLAG_SYNC;
876 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
877 xfs_inherit_nosymlinks)
878 di_flags |= XFS_DIFLAG_NOSYMLINKS;
879 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
880 xfs_inherit_nodefrag)
881 di_flags |= XFS_DIFLAG_NODEFRAG;
882 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
883 di_flags |= XFS_DIFLAG_FILESTREAM;
884 ip->i_d.di_flags |= di_flags;
888 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
889 ip->i_df.if_flags = XFS_IFEXTENTS;
890 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
891 ip->i_df.if_u1.if_extents = NULL;
897 * Attribute fork settings for new inode.
899 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
900 ip->i_d.di_anextents = 0;
903 * Log the new values stuffed into the inode.
905 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
906 xfs_trans_log_inode(tp, ip, flags);
908 /* now that we have an i_mode we can setup the inode structure */
916 * Allocates a new inode from disk and return a pointer to the
917 * incore copy. This routine will internally commit the current
918 * transaction and allocate a new one if the Space Manager needed
919 * to do an allocation to replenish the inode free-list.
921 * This routine is designed to be called from xfs_create and
927 xfs_trans_t **tpp, /* input: current transaction;
928 output: may be a new transaction. */
929 xfs_inode_t *dp, /* directory within whose allocate
934 prid_t prid, /* project id */
935 int okalloc, /* ok to allocate new space */
936 xfs_inode_t **ipp, /* pointer to inode; it will be
943 xfs_buf_t *ialloc_context = NULL;
949 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
952 * xfs_ialloc will return a pointer to an incore inode if
953 * the Space Manager has an available inode on the free
954 * list. Otherwise, it will do an allocation and replenish
955 * the freelist. Since we can only do one allocation per
956 * transaction without deadlocks, we will need to commit the
957 * current transaction and start a new one. We will then
958 * need to call xfs_ialloc again to get the inode.
960 * If xfs_ialloc did an allocation to replenish the freelist,
961 * it returns the bp containing the head of the freelist as
962 * ialloc_context. We will hold a lock on it across the
963 * transaction commit so that no other process can steal
964 * the inode(s) that we've just allocated.
966 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
967 &ialloc_context, &ip);
970 * Return an error if we were unable to allocate a new inode.
971 * This should only happen if we run out of space on disk or
972 * encounter a disk error.
978 if (!ialloc_context && !ip) {
984 * If the AGI buffer is non-NULL, then we were unable to get an
985 * inode in one operation. We need to commit the current
986 * transaction and call xfs_ialloc() again. It is guaranteed
987 * to succeed the second time.
989 if (ialloc_context) {
991 * Normally, xfs_trans_commit releases all the locks.
992 * We call bhold to hang on to the ialloc_context across
993 * the commit. Holding this buffer prevents any other
994 * processes from doing any allocations in this
997 xfs_trans_bhold(tp, ialloc_context);
1000 * We want the quota changes to be associated with the next
1001 * transaction, NOT this one. So, detach the dqinfo from this
1002 * and attach it to the next transaction.
1007 dqinfo = (void *)tp->t_dqinfo;
1008 tp->t_dqinfo = NULL;
1009 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1010 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1013 code = xfs_trans_roll(&tp, 0);
1014 if (committed != NULL)
1018 * Re-attach the quota info that we detached from prev trx.
1021 tp->t_dqinfo = dqinfo;
1022 tp->t_flags |= tflags;
1026 xfs_buf_relse(ialloc_context);
1031 xfs_trans_bjoin(tp, ialloc_context);
1034 * Call ialloc again. Since we've locked out all
1035 * other allocations in this allocation group,
1036 * this call should always succeed.
1038 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1039 okalloc, &ialloc_context, &ip);
1042 * If we get an error at this point, return to the caller
1043 * so that the current transaction can be aborted.
1050 ASSERT(!ialloc_context && ip);
1053 if (committed != NULL)
1064 * Decrement the link count on an inode & log the change.
1065 * If this causes the link count to go to zero, initiate the
1066 * logging activity required to truncate a file.
1075 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1077 ASSERT (ip->i_d.di_nlink > 0);
1079 drop_nlink(VFS_I(ip));
1080 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1083 if (ip->i_d.di_nlink == 0) {
1085 * We're dropping the last link to this file.
1086 * Move the on-disk inode to the AGI unlinked list.
1087 * From xfs_inactive() we will pull the inode from
1088 * the list and free it.
1090 error = xfs_iunlink(tp, ip);
1096 * Increment the link count on an inode & log the change.
1103 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1105 ASSERT(ip->i_d.di_version > 1);
1106 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1108 inc_nlink(VFS_I(ip));
1109 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1116 struct xfs_name *name,
1121 int is_dir = S_ISDIR(mode);
1122 struct xfs_mount *mp = dp->i_mount;
1123 struct xfs_inode *ip = NULL;
1124 struct xfs_trans *tp = NULL;
1126 xfs_bmap_free_t free_list;
1127 xfs_fsblock_t first_block;
1128 bool unlock_dp_on_error = false;
1131 struct xfs_dquot *udqp = NULL;
1132 struct xfs_dquot *gdqp = NULL;
1133 struct xfs_dquot *pdqp = NULL;
1134 struct xfs_trans_res *tres;
1137 trace_xfs_create(dp, name);
1139 if (XFS_FORCED_SHUTDOWN(mp))
1142 prid = xfs_get_initial_prid(dp);
1145 * Make sure that we have allocated dquot(s) on disk.
1147 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1148 xfs_kgid_to_gid(current_fsgid()), prid,
1149 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1150 &udqp, &gdqp, &pdqp);
1156 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1157 tres = &M_RES(mp)->tr_mkdir;
1158 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1160 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1161 tres = &M_RES(mp)->tr_create;
1162 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1166 * Initially assume that the file does not exist and
1167 * reserve the resources for that case. If that is not
1168 * the case we'll drop the one we have and get a more
1169 * appropriate transaction later.
1171 error = xfs_trans_reserve(tp, tres, resblks, 0);
1172 if (error == -ENOSPC) {
1173 /* flush outstanding delalloc blocks and retry */
1174 xfs_flush_inodes(mp);
1175 error = xfs_trans_reserve(tp, tres, resblks, 0);
1177 if (error == -ENOSPC) {
1178 /* No space at all so try a "no-allocation" reservation */
1180 error = xfs_trans_reserve(tp, tres, 0, 0);
1183 goto out_trans_cancel;
1186 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1187 unlock_dp_on_error = true;
1189 xfs_bmap_init(&free_list, &first_block);
1192 * Reserve disk quota and the inode.
1194 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1195 pdqp, resblks, 1, 0);
1197 goto out_trans_cancel;
1200 error = xfs_dir_canenter(tp, dp, name);
1202 goto out_trans_cancel;
1206 * A newly created regular or special file just has one directory
1207 * entry pointing to them, but a directory also the "." entry
1208 * pointing to itself.
1210 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1211 prid, resblks > 0, &ip, &committed);
1213 if (error == -ENOSPC)
1214 goto out_trans_cancel;
1215 goto out_trans_cancel;
1219 * Now we join the directory inode to the transaction. We do not do it
1220 * earlier because xfs_dir_ialloc might commit the previous transaction
1221 * (and release all the locks). An error from here on will result in
1222 * the transaction cancel unlocking dp so don't do it explicitly in the
1225 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1226 unlock_dp_on_error = false;
1228 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1229 &first_block, &free_list, resblks ?
1230 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1232 ASSERT(error != -ENOSPC);
1233 goto out_trans_cancel;
1235 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1236 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1239 error = xfs_dir_init(tp, ip, dp);
1241 goto out_bmap_cancel;
1243 error = xfs_bumplink(tp, dp);
1245 goto out_bmap_cancel;
1249 * If this is a synchronous mount, make sure that the
1250 * create transaction goes to disk before returning to
1253 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1254 xfs_trans_set_sync(tp);
1257 * Attach the dquot(s) to the inodes and modify them incore.
1258 * These ids of the inode couldn't have changed since the new
1259 * inode has been locked ever since it was created.
1261 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1263 error = xfs_bmap_finish(&tp, &free_list, &committed);
1265 goto out_bmap_cancel;
1267 error = xfs_trans_commit(tp);
1269 goto out_release_inode;
1271 xfs_qm_dqrele(udqp);
1272 xfs_qm_dqrele(gdqp);
1273 xfs_qm_dqrele(pdqp);
1279 xfs_bmap_cancel(&free_list);
1281 xfs_trans_cancel(tp);
1284 * Wait until after the current transaction is aborted to finish the
1285 * setup of the inode and release the inode. This prevents recursive
1286 * transactions and deadlocks from xfs_inactive.
1289 xfs_finish_inode_setup(ip);
1293 xfs_qm_dqrele(udqp);
1294 xfs_qm_dqrele(gdqp);
1295 xfs_qm_dqrele(pdqp);
1297 if (unlock_dp_on_error)
1298 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1304 struct xfs_inode *dp,
1305 struct dentry *dentry,
1307 struct xfs_inode **ipp)
1309 struct xfs_mount *mp = dp->i_mount;
1310 struct xfs_inode *ip = NULL;
1311 struct xfs_trans *tp = NULL;
1314 struct xfs_dquot *udqp = NULL;
1315 struct xfs_dquot *gdqp = NULL;
1316 struct xfs_dquot *pdqp = NULL;
1317 struct xfs_trans_res *tres;
1320 if (XFS_FORCED_SHUTDOWN(mp))
1323 prid = xfs_get_initial_prid(dp);
1326 * Make sure that we have allocated dquot(s) on disk.
1328 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1329 xfs_kgid_to_gid(current_fsgid()), prid,
1330 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1331 &udqp, &gdqp, &pdqp);
1335 resblks = XFS_IALLOC_SPACE_RES(mp);
1336 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1338 tres = &M_RES(mp)->tr_create_tmpfile;
1339 error = xfs_trans_reserve(tp, tres, resblks, 0);
1340 if (error == -ENOSPC) {
1341 /* No space at all so try a "no-allocation" reservation */
1343 error = xfs_trans_reserve(tp, tres, 0, 0);
1346 goto out_trans_cancel;
1348 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1349 pdqp, resblks, 1, 0);
1351 goto out_trans_cancel;
1353 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1354 prid, resblks > 0, &ip, NULL);
1356 if (error == -ENOSPC)
1357 goto out_trans_cancel;
1358 goto out_trans_cancel;
1361 if (mp->m_flags & XFS_MOUNT_WSYNC)
1362 xfs_trans_set_sync(tp);
1365 * Attach the dquot(s) to the inodes and modify them incore.
1366 * These ids of the inode couldn't have changed since the new
1367 * inode has been locked ever since it was created.
1369 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1372 error = xfs_iunlink(tp, ip);
1374 goto out_trans_cancel;
1376 error = xfs_trans_commit(tp);
1378 goto out_release_inode;
1380 xfs_qm_dqrele(udqp);
1381 xfs_qm_dqrele(gdqp);
1382 xfs_qm_dqrele(pdqp);
1388 xfs_trans_cancel(tp);
1391 * Wait until after the current transaction is aborted to finish the
1392 * setup of the inode and release the inode. This prevents recursive
1393 * transactions and deadlocks from xfs_inactive.
1396 xfs_finish_inode_setup(ip);
1400 xfs_qm_dqrele(udqp);
1401 xfs_qm_dqrele(gdqp);
1402 xfs_qm_dqrele(pdqp);
1411 struct xfs_name *target_name)
1413 xfs_mount_t *mp = tdp->i_mount;
1416 xfs_bmap_free_t free_list;
1417 xfs_fsblock_t first_block;
1421 trace_xfs_link(tdp, target_name);
1423 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1425 if (XFS_FORCED_SHUTDOWN(mp))
1428 error = xfs_qm_dqattach(sip, 0);
1432 error = xfs_qm_dqattach(tdp, 0);
1436 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1437 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1438 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1439 if (error == -ENOSPC) {
1441 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1446 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1448 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1449 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1452 * If we are using project inheritance, we only allow hard link
1453 * creation in our tree when the project IDs are the same; else
1454 * the tree quota mechanism could be circumvented.
1456 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1457 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1463 error = xfs_dir_canenter(tp, tdp, target_name);
1468 xfs_bmap_init(&free_list, &first_block);
1470 if (sip->i_d.di_nlink == 0) {
1471 error = xfs_iunlink_remove(tp, sip);
1476 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1477 &first_block, &free_list, resblks);
1480 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1481 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1483 error = xfs_bumplink(tp, sip);
1488 * If this is a synchronous mount, make sure that the
1489 * link transaction goes to disk before returning to
1492 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1493 xfs_trans_set_sync(tp);
1496 error = xfs_bmap_finish (&tp, &free_list, &committed);
1498 xfs_bmap_cancel(&free_list);
1502 return xfs_trans_commit(tp);
1505 xfs_trans_cancel(tp);
1511 * Free up the underlying blocks past new_size. The new size must be smaller
1512 * than the current size. This routine can be used both for the attribute and
1513 * data fork, and does not modify the inode size, which is left to the caller.
1515 * The transaction passed to this routine must have made a permanent log
1516 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1517 * given transaction and start new ones, so make sure everything involved in
1518 * the transaction is tidy before calling here. Some transaction will be
1519 * returned to the caller to be committed. The incoming transaction must
1520 * already include the inode, and both inode locks must be held exclusively.
1521 * The inode must also be "held" within the transaction. On return the inode
1522 * will be "held" within the returned transaction. This routine does NOT
1523 * require any disk space to be reserved for it within the transaction.
1525 * If we get an error, we must return with the inode locked and linked into the
1526 * current transaction. This keeps things simple for the higher level code,
1527 * because it always knows that the inode is locked and held in the transaction
1528 * that returns to it whether errors occur or not. We don't mark the inode
1529 * dirty on error so that transactions can be easily aborted if possible.
1532 xfs_itruncate_extents(
1533 struct xfs_trans **tpp,
1534 struct xfs_inode *ip,
1536 xfs_fsize_t new_size)
1538 struct xfs_mount *mp = ip->i_mount;
1539 struct xfs_trans *tp = *tpp;
1540 xfs_bmap_free_t free_list;
1541 xfs_fsblock_t first_block;
1542 xfs_fileoff_t first_unmap_block;
1543 xfs_fileoff_t last_block;
1544 xfs_filblks_t unmap_len;
1549 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1550 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1551 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1552 ASSERT(new_size <= XFS_ISIZE(ip));
1553 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1554 ASSERT(ip->i_itemp != NULL);
1555 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1556 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1558 trace_xfs_itruncate_extents_start(ip, new_size);
1561 * Since it is possible for space to become allocated beyond
1562 * the end of the file (in a crash where the space is allocated
1563 * but the inode size is not yet updated), simply remove any
1564 * blocks which show up between the new EOF and the maximum
1565 * possible file size. If the first block to be removed is
1566 * beyond the maximum file size (ie it is the same as last_block),
1567 * then there is nothing to do.
1569 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1570 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1571 if (first_unmap_block == last_block)
1574 ASSERT(first_unmap_block < last_block);
1575 unmap_len = last_block - first_unmap_block + 1;
1577 xfs_bmap_init(&free_list, &first_block);
1578 error = xfs_bunmapi(tp, ip,
1579 first_unmap_block, unmap_len,
1580 xfs_bmapi_aflag(whichfork),
1581 XFS_ITRUNC_MAX_EXTENTS,
1582 &first_block, &free_list,
1585 goto out_bmap_cancel;
1588 * Duplicate the transaction that has the permanent
1589 * reservation and commit the old transaction.
1591 error = xfs_bmap_finish(&tp, &free_list, &committed);
1593 xfs_trans_ijoin(tp, ip, 0);
1595 goto out_bmap_cancel;
1597 error = xfs_trans_roll(&tp, ip);
1603 * Always re-log the inode so that our permanent transaction can keep
1604 * on rolling it forward in the log.
1606 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1608 trace_xfs_itruncate_extents_end(ip, new_size);
1615 * If the bunmapi call encounters an error, return to the caller where
1616 * the transaction can be properly aborted. We just need to make sure
1617 * we're not holding any resources that we were not when we came in.
1619 xfs_bmap_cancel(&free_list);
1627 xfs_mount_t *mp = ip->i_mount;
1630 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1633 /* If this is a read-only mount, don't do this (would generate I/O) */
1634 if (mp->m_flags & XFS_MOUNT_RDONLY)
1637 if (!XFS_FORCED_SHUTDOWN(mp)) {
1641 * If we previously truncated this file and removed old data
1642 * in the process, we want to initiate "early" writeout on
1643 * the last close. This is an attempt to combat the notorious
1644 * NULL files problem which is particularly noticeable from a
1645 * truncate down, buffered (re-)write (delalloc), followed by
1646 * a crash. What we are effectively doing here is
1647 * significantly reducing the time window where we'd otherwise
1648 * be exposed to that problem.
1650 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1652 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1653 if (ip->i_delayed_blks > 0) {
1654 error = filemap_flush(VFS_I(ip)->i_mapping);
1661 if (ip->i_d.di_nlink == 0)
1664 if (xfs_can_free_eofblocks(ip, false)) {
1667 * If we can't get the iolock just skip truncating the blocks
1668 * past EOF because we could deadlock with the mmap_sem
1669 * otherwise. We'll get another chance to drop them once the
1670 * last reference to the inode is dropped, so we'll never leak
1671 * blocks permanently.
1673 * Further, check if the inode is being opened, written and
1674 * closed frequently and we have delayed allocation blocks
1675 * outstanding (e.g. streaming writes from the NFS server),
1676 * truncating the blocks past EOF will cause fragmentation to
1679 * In this case don't do the truncation, either, but we have to
1680 * be careful how we detect this case. Blocks beyond EOF show
1681 * up as i_delayed_blks even when the inode is clean, so we
1682 * need to truncate them away first before checking for a dirty
1683 * release. Hence on the first dirty close we will still remove
1684 * the speculative allocation, but after that we will leave it
1687 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1690 error = xfs_free_eofblocks(mp, ip, true);
1691 if (error && error != -EAGAIN)
1694 /* delalloc blocks after truncation means it really is dirty */
1695 if (ip->i_delayed_blks)
1696 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1702 * xfs_inactive_truncate
1704 * Called to perform a truncate when an inode becomes unlinked.
1707 xfs_inactive_truncate(
1708 struct xfs_inode *ip)
1710 struct xfs_mount *mp = ip->i_mount;
1711 struct xfs_trans *tp;
1714 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1715 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1717 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1718 xfs_trans_cancel(tp);
1722 xfs_ilock(ip, XFS_ILOCK_EXCL);
1723 xfs_trans_ijoin(tp, ip, 0);
1726 * Log the inode size first to prevent stale data exposure in the event
1727 * of a system crash before the truncate completes. See the related
1728 * comment in xfs_setattr_size() for details.
1730 ip->i_d.di_size = 0;
1731 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1733 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1735 goto error_trans_cancel;
1737 ASSERT(ip->i_d.di_nextents == 0);
1739 error = xfs_trans_commit(tp);
1743 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1747 xfs_trans_cancel(tp);
1749 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1754 * xfs_inactive_ifree()
1756 * Perform the inode free when an inode is unlinked.
1760 struct xfs_inode *ip)
1762 xfs_bmap_free_t free_list;
1763 xfs_fsblock_t first_block;
1765 struct xfs_mount *mp = ip->i_mount;
1766 struct xfs_trans *tp;
1769 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1772 * The ifree transaction might need to allocate blocks for record
1773 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1774 * allow ifree to dip into the reserved block pool if necessary.
1776 * Freeing large sets of inodes generally means freeing inode chunks,
1777 * directory and file data blocks, so this should be relatively safe.
1778 * Only under severe circumstances should it be possible to free enough
1779 * inodes to exhaust the reserve block pool via finobt expansion while
1780 * at the same time not creating free space in the filesystem.
1782 * Send a warning if the reservation does happen to fail, as the inode
1783 * now remains allocated and sits on the unlinked list until the fs is
1786 tp->t_flags |= XFS_TRANS_RESERVE;
1787 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1788 XFS_IFREE_SPACE_RES(mp), 0);
1790 if (error == -ENOSPC) {
1791 xfs_warn_ratelimited(mp,
1792 "Failed to remove inode(s) from unlinked list. "
1793 "Please free space, unmount and run xfs_repair.");
1795 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1797 xfs_trans_cancel(tp);
1801 xfs_ilock(ip, XFS_ILOCK_EXCL);
1802 xfs_trans_ijoin(tp, ip, 0);
1804 xfs_bmap_init(&free_list, &first_block);
1805 error = xfs_ifree(tp, ip, &free_list);
1808 * If we fail to free the inode, shut down. The cancel
1809 * might do that, we need to make sure. Otherwise the
1810 * inode might be lost for a long time or forever.
1812 if (!XFS_FORCED_SHUTDOWN(mp)) {
1813 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1815 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1817 xfs_trans_cancel(tp);
1818 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1823 * Credit the quota account(s). The inode is gone.
1825 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1828 * Just ignore errors at this point. There is nothing we can
1829 * do except to try to keep going. Make sure it's not a silent
1832 error = xfs_bmap_finish(&tp, &free_list, &committed);
1834 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1836 error = xfs_trans_commit(tp);
1838 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1841 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1848 * This is called when the vnode reference count for the vnode
1849 * goes to zero. If the file has been unlinked, then it must
1850 * now be truncated. Also, we clear all of the read-ahead state
1851 * kept for the inode here since the file is now closed.
1857 struct xfs_mount *mp;
1862 * If the inode is already free, then there can be nothing
1865 if (ip->i_d.di_mode == 0) {
1866 ASSERT(ip->i_df.if_real_bytes == 0);
1867 ASSERT(ip->i_df.if_broot_bytes == 0);
1873 /* If this is a read-only mount, don't do this (would generate I/O) */
1874 if (mp->m_flags & XFS_MOUNT_RDONLY)
1877 if (ip->i_d.di_nlink != 0) {
1879 * force is true because we are evicting an inode from the
1880 * cache. Post-eof blocks must be freed, lest we end up with
1881 * broken free space accounting.
1883 if (xfs_can_free_eofblocks(ip, true))
1884 xfs_free_eofblocks(mp, ip, false);
1889 if (S_ISREG(ip->i_d.di_mode) &&
1890 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1891 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1894 error = xfs_qm_dqattach(ip, 0);
1898 if (S_ISLNK(ip->i_d.di_mode))
1899 error = xfs_inactive_symlink(ip);
1901 error = xfs_inactive_truncate(ip);
1906 * If there are attributes associated with the file then blow them away
1907 * now. The code calls a routine that recursively deconstructs the
1908 * attribute fork. If also blows away the in-core attribute fork.
1910 if (XFS_IFORK_Q(ip)) {
1911 error = xfs_attr_inactive(ip);
1917 ASSERT(ip->i_d.di_anextents == 0);
1918 ASSERT(ip->i_d.di_forkoff == 0);
1923 error = xfs_inactive_ifree(ip);
1928 * Release the dquots held by inode, if any.
1930 xfs_qm_dqdetach(ip);
1934 * This is called when the inode's link count goes to 0.
1935 * We place the on-disk inode on a list in the AGI. It
1936 * will be pulled from this list when the inode is freed.
1953 ASSERT(ip->i_d.di_nlink == 0);
1954 ASSERT(ip->i_d.di_mode != 0);
1959 * Get the agi buffer first. It ensures lock ordering
1962 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1965 agi = XFS_BUF_TO_AGI(agibp);
1968 * Get the index into the agi hash table for the
1969 * list this inode will go on.
1971 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1973 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1974 ASSERT(agi->agi_unlinked[bucket_index]);
1975 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1977 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1979 * There is already another inode in the bucket we need
1980 * to add ourselves to. Add us at the front of the list.
1981 * Here we put the head pointer into our next pointer,
1982 * and then we fall through to point the head at us.
1984 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1989 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1990 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1991 offset = ip->i_imap.im_boffset +
1992 offsetof(xfs_dinode_t, di_next_unlinked);
1994 /* need to recalc the inode CRC if appropriate */
1995 xfs_dinode_calc_crc(mp, dip);
1997 xfs_trans_inode_buf(tp, ibp);
1998 xfs_trans_log_buf(tp, ibp, offset,
1999 (offset + sizeof(xfs_agino_t) - 1));
2000 xfs_inobp_check(mp, ibp);
2004 * Point the bucket head pointer at the inode being inserted.
2007 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2008 offset = offsetof(xfs_agi_t, agi_unlinked) +
2009 (sizeof(xfs_agino_t) * bucket_index);
2010 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2011 xfs_trans_log_buf(tp, agibp, offset,
2012 (offset + sizeof(xfs_agino_t) - 1));
2017 * Pull the on-disk inode from the AGI unlinked list.
2030 xfs_agnumber_t agno;
2032 xfs_agino_t next_agino;
2033 xfs_buf_t *last_ibp;
2034 xfs_dinode_t *last_dip = NULL;
2036 int offset, last_offset = 0;
2040 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2043 * Get the agi buffer first. It ensures lock ordering
2046 error = xfs_read_agi(mp, tp, agno, &agibp);
2050 agi = XFS_BUF_TO_AGI(agibp);
2053 * Get the index into the agi hash table for the
2054 * list this inode will go on.
2056 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2058 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2059 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2060 ASSERT(agi->agi_unlinked[bucket_index]);
2062 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2064 * We're at the head of the list. Get the inode's on-disk
2065 * buffer to see if there is anyone after us on the list.
2066 * Only modify our next pointer if it is not already NULLAGINO.
2067 * This saves us the overhead of dealing with the buffer when
2068 * there is no need to change it.
2070 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2073 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2077 next_agino = be32_to_cpu(dip->di_next_unlinked);
2078 ASSERT(next_agino != 0);
2079 if (next_agino != NULLAGINO) {
2080 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2081 offset = ip->i_imap.im_boffset +
2082 offsetof(xfs_dinode_t, di_next_unlinked);
2084 /* need to recalc the inode CRC if appropriate */
2085 xfs_dinode_calc_crc(mp, dip);
2087 xfs_trans_inode_buf(tp, ibp);
2088 xfs_trans_log_buf(tp, ibp, offset,
2089 (offset + sizeof(xfs_agino_t) - 1));
2090 xfs_inobp_check(mp, ibp);
2092 xfs_trans_brelse(tp, ibp);
2095 * Point the bucket head pointer at the next inode.
2097 ASSERT(next_agino != 0);
2098 ASSERT(next_agino != agino);
2099 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2100 offset = offsetof(xfs_agi_t, agi_unlinked) +
2101 (sizeof(xfs_agino_t) * bucket_index);
2102 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2103 xfs_trans_log_buf(tp, agibp, offset,
2104 (offset + sizeof(xfs_agino_t) - 1));
2107 * We need to search the list for the inode being freed.
2109 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2111 while (next_agino != agino) {
2112 struct xfs_imap imap;
2115 xfs_trans_brelse(tp, last_ibp);
2118 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2120 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2123 "%s: xfs_imap returned error %d.",
2128 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2132 "%s: xfs_imap_to_bp returned error %d.",
2137 last_offset = imap.im_boffset;
2138 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2139 ASSERT(next_agino != NULLAGINO);
2140 ASSERT(next_agino != 0);
2144 * Now last_ibp points to the buffer previous to us on the
2145 * unlinked list. Pull us from the list.
2147 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2150 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2154 next_agino = be32_to_cpu(dip->di_next_unlinked);
2155 ASSERT(next_agino != 0);
2156 ASSERT(next_agino != agino);
2157 if (next_agino != NULLAGINO) {
2158 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2159 offset = ip->i_imap.im_boffset +
2160 offsetof(xfs_dinode_t, di_next_unlinked);
2162 /* need to recalc the inode CRC if appropriate */
2163 xfs_dinode_calc_crc(mp, dip);
2165 xfs_trans_inode_buf(tp, ibp);
2166 xfs_trans_log_buf(tp, ibp, offset,
2167 (offset + sizeof(xfs_agino_t) - 1));
2168 xfs_inobp_check(mp, ibp);
2170 xfs_trans_brelse(tp, ibp);
2173 * Point the previous inode on the list to the next inode.
2175 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2176 ASSERT(next_agino != 0);
2177 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2179 /* need to recalc the inode CRC if appropriate */
2180 xfs_dinode_calc_crc(mp, last_dip);
2182 xfs_trans_inode_buf(tp, last_ibp);
2183 xfs_trans_log_buf(tp, last_ibp, offset,
2184 (offset + sizeof(xfs_agino_t) - 1));
2185 xfs_inobp_check(mp, last_ibp);
2191 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2192 * inodes that are in memory - they all must be marked stale and attached to
2193 * the cluster buffer.
2197 xfs_inode_t *free_ip,
2199 struct xfs_icluster *xic)
2201 xfs_mount_t *mp = free_ip->i_mount;
2202 int blks_per_cluster;
2203 int inodes_per_cluster;
2210 xfs_inode_log_item_t *iip;
2211 xfs_log_item_t *lip;
2212 struct xfs_perag *pag;
2215 inum = xic->first_ino;
2216 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2217 blks_per_cluster = xfs_icluster_size_fsb(mp);
2218 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2219 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2221 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2223 * The allocation bitmap tells us which inodes of the chunk were
2224 * physically allocated. Skip the cluster if an inode falls into
2227 ioffset = inum - xic->first_ino;
2228 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2229 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2233 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2234 XFS_INO_TO_AGBNO(mp, inum));
2237 * We obtain and lock the backing buffer first in the process
2238 * here, as we have to ensure that any dirty inode that we
2239 * can't get the flush lock on is attached to the buffer.
2240 * If we scan the in-memory inodes first, then buffer IO can
2241 * complete before we get a lock on it, and hence we may fail
2242 * to mark all the active inodes on the buffer stale.
2244 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2245 mp->m_bsize * blks_per_cluster,
2252 * This buffer may not have been correctly initialised as we
2253 * didn't read it from disk. That's not important because we are
2254 * only using to mark the buffer as stale in the log, and to
2255 * attach stale cached inodes on it. That means it will never be
2256 * dispatched for IO. If it is, we want to know about it, and we
2257 * want it to fail. We can acheive this by adding a write
2258 * verifier to the buffer.
2260 bp->b_ops = &xfs_inode_buf_ops;
2263 * Walk the inodes already attached to the buffer and mark them
2264 * stale. These will all have the flush locks held, so an
2265 * in-memory inode walk can't lock them. By marking them all
2266 * stale first, we will not attempt to lock them in the loop
2267 * below as the XFS_ISTALE flag will be set.
2271 if (lip->li_type == XFS_LI_INODE) {
2272 iip = (xfs_inode_log_item_t *)lip;
2273 ASSERT(iip->ili_logged == 1);
2274 lip->li_cb = xfs_istale_done;
2275 xfs_trans_ail_copy_lsn(mp->m_ail,
2276 &iip->ili_flush_lsn,
2277 &iip->ili_item.li_lsn);
2278 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2280 lip = lip->li_bio_list;
2285 * For each inode in memory attempt to add it to the inode
2286 * buffer and set it up for being staled on buffer IO
2287 * completion. This is safe as we've locked out tail pushing
2288 * and flushing by locking the buffer.
2290 * We have already marked every inode that was part of a
2291 * transaction stale above, which means there is no point in
2292 * even trying to lock them.
2294 for (i = 0; i < inodes_per_cluster; i++) {
2297 ip = radix_tree_lookup(&pag->pag_ici_root,
2298 XFS_INO_TO_AGINO(mp, (inum + i)));
2300 /* Inode not in memory, nothing to do */
2307 * because this is an RCU protected lookup, we could
2308 * find a recently freed or even reallocated inode
2309 * during the lookup. We need to check under the
2310 * i_flags_lock for a valid inode here. Skip it if it
2311 * is not valid, the wrong inode or stale.
2313 spin_lock(&ip->i_flags_lock);
2314 if (ip->i_ino != inum + i ||
2315 __xfs_iflags_test(ip, XFS_ISTALE)) {
2316 spin_unlock(&ip->i_flags_lock);
2320 spin_unlock(&ip->i_flags_lock);
2323 * Don't try to lock/unlock the current inode, but we
2324 * _cannot_ skip the other inodes that we did not find
2325 * in the list attached to the buffer and are not
2326 * already marked stale. If we can't lock it, back off
2329 if (ip != free_ip &&
2330 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2338 xfs_iflags_set(ip, XFS_ISTALE);
2341 * we don't need to attach clean inodes or those only
2342 * with unlogged changes (which we throw away, anyway).
2345 if (!iip || xfs_inode_clean(ip)) {
2346 ASSERT(ip != free_ip);
2348 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2352 iip->ili_last_fields = iip->ili_fields;
2353 iip->ili_fields = 0;
2354 iip->ili_logged = 1;
2355 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2356 &iip->ili_item.li_lsn);
2358 xfs_buf_attach_iodone(bp, xfs_istale_done,
2362 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2365 xfs_trans_stale_inode_buf(tp, bp);
2366 xfs_trans_binval(tp, bp);
2374 * This is called to return an inode to the inode free list.
2375 * The inode should already be truncated to 0 length and have
2376 * no pages associated with it. This routine also assumes that
2377 * the inode is already a part of the transaction.
2379 * The on-disk copy of the inode will have been added to the list
2380 * of unlinked inodes in the AGI. We need to remove the inode from
2381 * that list atomically with respect to freeing it here.
2387 xfs_bmap_free_t *flist)
2390 struct xfs_icluster xic = { 0 };
2392 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2393 ASSERT(ip->i_d.di_nlink == 0);
2394 ASSERT(ip->i_d.di_nextents == 0);
2395 ASSERT(ip->i_d.di_anextents == 0);
2396 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2397 ASSERT(ip->i_d.di_nblocks == 0);
2400 * Pull the on-disk inode from the AGI unlinked list.
2402 error = xfs_iunlink_remove(tp, ip);
2406 error = xfs_difree(tp, ip->i_ino, flist, &xic);
2410 ip->i_d.di_mode = 0; /* mark incore inode as free */
2411 ip->i_d.di_flags = 0;
2412 ip->i_d.di_dmevmask = 0;
2413 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2414 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2415 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2417 * Bump the generation count so no one will be confused
2418 * by reincarnations of this inode.
2421 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2424 error = xfs_ifree_cluster(ip, tp, &xic);
2430 * This is called to unpin an inode. The caller must have the inode locked
2431 * in at least shared mode so that the buffer cannot be subsequently pinned
2432 * once someone is waiting for it to be unpinned.
2436 struct xfs_inode *ip)
2438 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2440 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2442 /* Give the log a push to start the unpinning I/O */
2443 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2449 struct xfs_inode *ip)
2451 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2452 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2457 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2458 if (xfs_ipincount(ip))
2460 } while (xfs_ipincount(ip));
2461 finish_wait(wq, &wait.wait);
2466 struct xfs_inode *ip)
2468 if (xfs_ipincount(ip))
2469 __xfs_iunpin_wait(ip);
2473 * Removing an inode from the namespace involves removing the directory entry
2474 * and dropping the link count on the inode. Removing the directory entry can
2475 * result in locking an AGF (directory blocks were freed) and removing a link
2476 * count can result in placing the inode on an unlinked list which results in
2479 * The big problem here is that we have an ordering constraint on AGF and AGI
2480 * locking - inode allocation locks the AGI, then can allocate a new extent for
2481 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2482 * removes the inode from the unlinked list, requiring that we lock the AGI
2483 * first, and then freeing the inode can result in an inode chunk being freed
2484 * and hence freeing disk space requiring that we lock an AGF.
2486 * Hence the ordering that is imposed by other parts of the code is AGI before
2487 * AGF. This means we cannot remove the directory entry before we drop the inode
2488 * reference count and put it on the unlinked list as this results in a lock
2489 * order of AGF then AGI, and this can deadlock against inode allocation and
2490 * freeing. Therefore we must drop the link counts before we remove the
2493 * This is still safe from a transactional point of view - it is not until we
2494 * get to xfs_bmap_finish() that we have the possibility of multiple
2495 * transactions in this operation. Hence as long as we remove the directory
2496 * entry and drop the link count in the first transaction of the remove
2497 * operation, there are no transactional constraints on the ordering here.
2502 struct xfs_name *name,
2505 xfs_mount_t *mp = dp->i_mount;
2506 xfs_trans_t *tp = NULL;
2507 int is_dir = S_ISDIR(ip->i_d.di_mode);
2509 xfs_bmap_free_t free_list;
2510 xfs_fsblock_t first_block;
2514 trace_xfs_remove(dp, name);
2516 if (XFS_FORCED_SHUTDOWN(mp))
2519 error = xfs_qm_dqattach(dp, 0);
2523 error = xfs_qm_dqattach(ip, 0);
2528 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2530 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2533 * We try to get the real space reservation first,
2534 * allowing for directory btree deletion(s) implying
2535 * possible bmap insert(s). If we can't get the space
2536 * reservation then we use 0 instead, and avoid the bmap
2537 * btree insert(s) in the directory code by, if the bmap
2538 * insert tries to happen, instead trimming the LAST
2539 * block from the directory.
2541 resblks = XFS_REMOVE_SPACE_RES(mp);
2542 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2543 if (error == -ENOSPC) {
2545 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2548 ASSERT(error != -ENOSPC);
2549 goto out_trans_cancel;
2552 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2554 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2555 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2558 * If we're removing a directory perform some additional validation.
2561 ASSERT(ip->i_d.di_nlink >= 2);
2562 if (ip->i_d.di_nlink != 2) {
2564 goto out_trans_cancel;
2566 if (!xfs_dir_isempty(ip)) {
2568 goto out_trans_cancel;
2571 /* Drop the link from ip's "..". */
2572 error = xfs_droplink(tp, dp);
2574 goto out_trans_cancel;
2576 /* Drop the "." link from ip to self. */
2577 error = xfs_droplink(tp, ip);
2579 goto out_trans_cancel;
2582 * When removing a non-directory we need to log the parent
2583 * inode here. For a directory this is done implicitly
2584 * by the xfs_droplink call for the ".." entry.
2586 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2588 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2590 /* Drop the link from dp to ip. */
2591 error = xfs_droplink(tp, ip);
2593 goto out_trans_cancel;
2595 xfs_bmap_init(&free_list, &first_block);
2596 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2597 &first_block, &free_list, resblks);
2599 ASSERT(error != -ENOENT);
2600 goto out_bmap_cancel;
2604 * If this is a synchronous mount, make sure that the
2605 * remove transaction goes to disk before returning to
2608 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2609 xfs_trans_set_sync(tp);
2611 error = xfs_bmap_finish(&tp, &free_list, &committed);
2613 goto out_bmap_cancel;
2615 error = xfs_trans_commit(tp);
2619 if (is_dir && xfs_inode_is_filestream(ip))
2620 xfs_filestream_deassociate(ip);
2625 xfs_bmap_cancel(&free_list);
2627 xfs_trans_cancel(tp);
2633 * Enter all inodes for a rename transaction into a sorted array.
2635 #define __XFS_SORT_INODES 5
2637 xfs_sort_for_rename(
2638 struct xfs_inode *dp1, /* in: old (source) directory inode */
2639 struct xfs_inode *dp2, /* in: new (target) directory inode */
2640 struct xfs_inode *ip1, /* in: inode of old entry */
2641 struct xfs_inode *ip2, /* in: inode of new entry */
2642 struct xfs_inode *wip, /* in: whiteout inode */
2643 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2644 int *num_inodes) /* in/out: inodes in array */
2648 ASSERT(*num_inodes == __XFS_SORT_INODES);
2649 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2652 * i_tab contains a list of pointers to inodes. We initialize
2653 * the table here & we'll sort it. We will then use it to
2654 * order the acquisition of the inode locks.
2656 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2669 * Sort the elements via bubble sort. (Remember, there are at
2670 * most 5 elements to sort, so this is adequate.)
2672 for (i = 0; i < *num_inodes; i++) {
2673 for (j = 1; j < *num_inodes; j++) {
2674 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2675 struct xfs_inode *temp = i_tab[j];
2676 i_tab[j] = i_tab[j-1];
2685 struct xfs_trans *tp,
2686 struct xfs_bmap_free *free_list)
2692 * If this is a synchronous mount, make sure that the rename transaction
2693 * goes to disk before returning to the user.
2695 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2696 xfs_trans_set_sync(tp);
2698 error = xfs_bmap_finish(&tp, free_list, &committed);
2700 xfs_bmap_cancel(free_list);
2701 xfs_trans_cancel(tp);
2705 return xfs_trans_commit(tp);
2709 * xfs_cross_rename()
2711 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2715 struct xfs_trans *tp,
2716 struct xfs_inode *dp1,
2717 struct xfs_name *name1,
2718 struct xfs_inode *ip1,
2719 struct xfs_inode *dp2,
2720 struct xfs_name *name2,
2721 struct xfs_inode *ip2,
2722 struct xfs_bmap_free *free_list,
2723 xfs_fsblock_t *first_block,
2731 /* Swap inode number for dirent in first parent */
2732 error = xfs_dir_replace(tp, dp1, name1,
2734 first_block, free_list, spaceres);
2736 goto out_trans_abort;
2738 /* Swap inode number for dirent in second parent */
2739 error = xfs_dir_replace(tp, dp2, name2,
2741 first_block, free_list, spaceres);
2743 goto out_trans_abort;
2746 * If we're renaming one or more directories across different parents,
2747 * update the respective ".." entries (and link counts) to match the new
2751 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2753 if (S_ISDIR(ip2->i_d.di_mode)) {
2754 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2755 dp1->i_ino, first_block,
2756 free_list, spaceres);
2758 goto out_trans_abort;
2760 /* transfer ip2 ".." reference to dp1 */
2761 if (!S_ISDIR(ip1->i_d.di_mode)) {
2762 error = xfs_droplink(tp, dp2);
2764 goto out_trans_abort;
2765 error = xfs_bumplink(tp, dp1);
2767 goto out_trans_abort;
2771 * Although ip1 isn't changed here, userspace needs
2772 * to be warned about the change, so that applications
2773 * relying on it (like backup ones), will properly
2776 ip1_flags |= XFS_ICHGTIME_CHG;
2777 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2780 if (S_ISDIR(ip1->i_d.di_mode)) {
2781 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2782 dp2->i_ino, first_block,
2783 free_list, spaceres);
2785 goto out_trans_abort;
2787 /* transfer ip1 ".." reference to dp2 */
2788 if (!S_ISDIR(ip2->i_d.di_mode)) {
2789 error = xfs_droplink(tp, dp1);
2791 goto out_trans_abort;
2792 error = xfs_bumplink(tp, dp2);
2794 goto out_trans_abort;
2798 * Although ip2 isn't changed here, userspace needs
2799 * to be warned about the change, so that applications
2800 * relying on it (like backup ones), will properly
2803 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2804 ip2_flags |= XFS_ICHGTIME_CHG;
2809 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2810 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2813 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2814 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2817 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2818 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2820 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2821 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2822 return xfs_finish_rename(tp, free_list);
2825 xfs_bmap_cancel(free_list);
2826 xfs_trans_cancel(tp);
2831 * xfs_rename_alloc_whiteout()
2833 * Return a referenced, unlinked, unlocked inode that that can be used as a
2834 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2835 * crash between allocating the inode and linking it into the rename transaction
2836 * recovery will free the inode and we won't leak it.
2839 xfs_rename_alloc_whiteout(
2840 struct xfs_inode *dp,
2841 struct xfs_inode **wip)
2843 struct xfs_inode *tmpfile;
2846 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2851 * Prepare the tmpfile inode as if it were created through the VFS.
2852 * Otherwise, the link increment paths will complain about nlink 0->1.
2853 * Drop the link count as done by d_tmpfile(), complete the inode setup
2854 * and flag it as linkable.
2856 drop_nlink(VFS_I(tmpfile));
2857 xfs_finish_inode_setup(tmpfile);
2858 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2869 struct xfs_inode *src_dp,
2870 struct xfs_name *src_name,
2871 struct xfs_inode *src_ip,
2872 struct xfs_inode *target_dp,
2873 struct xfs_name *target_name,
2874 struct xfs_inode *target_ip,
2877 struct xfs_mount *mp = src_dp->i_mount;
2878 struct xfs_trans *tp;
2879 struct xfs_bmap_free free_list;
2880 xfs_fsblock_t first_block;
2881 struct xfs_inode *wip = NULL; /* whiteout inode */
2882 struct xfs_inode *inodes[__XFS_SORT_INODES];
2883 int num_inodes = __XFS_SORT_INODES;
2884 bool new_parent = (src_dp != target_dp);
2885 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2889 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2891 if ((flags & RENAME_EXCHANGE) && !target_ip)
2895 * If we are doing a whiteout operation, allocate the whiteout inode
2896 * we will be placing at the target and ensure the type is set
2899 if (flags & RENAME_WHITEOUT) {
2900 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2901 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2905 /* setup target dirent info as whiteout */
2906 src_name->type = XFS_DIR3_FT_CHRDEV;
2909 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2910 inodes, &num_inodes);
2912 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2913 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2914 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2915 if (error == -ENOSPC) {
2917 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2920 goto out_trans_cancel;
2923 * Attach the dquots to the inodes
2925 error = xfs_qm_vop_rename_dqattach(inodes);
2927 goto out_trans_cancel;
2930 * Lock all the participating inodes. Depending upon whether
2931 * the target_name exists in the target directory, and
2932 * whether the target directory is the same as the source
2933 * directory, we can lock from 2 to 4 inodes.
2935 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2938 * Join all the inodes to the transaction. From this point on,
2939 * we can rely on either trans_commit or trans_cancel to unlock
2942 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2944 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2945 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2947 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2949 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2952 * If we are using project inheritance, we only allow renames
2953 * into our tree when the project IDs are the same; else the
2954 * tree quota mechanism would be circumvented.
2956 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2957 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2959 goto out_trans_cancel;
2962 xfs_bmap_init(&free_list, &first_block);
2964 /* RENAME_EXCHANGE is unique from here on. */
2965 if (flags & RENAME_EXCHANGE)
2966 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2967 target_dp, target_name, target_ip,
2968 &free_list, &first_block, spaceres);
2971 * Set up the target.
2973 if (target_ip == NULL) {
2975 * If there's no space reservation, check the entry will
2976 * fit before actually inserting it.
2979 error = xfs_dir_canenter(tp, target_dp, target_name);
2981 goto out_trans_cancel;
2984 * If target does not exist and the rename crosses
2985 * directories, adjust the target directory link count
2986 * to account for the ".." reference from the new entry.
2988 error = xfs_dir_createname(tp, target_dp, target_name,
2989 src_ip->i_ino, &first_block,
2990 &free_list, spaceres);
2992 goto out_bmap_cancel;
2994 xfs_trans_ichgtime(tp, target_dp,
2995 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2997 if (new_parent && src_is_directory) {
2998 error = xfs_bumplink(tp, target_dp);
3000 goto out_bmap_cancel;
3002 } else { /* target_ip != NULL */
3004 * If target exists and it's a directory, check that both
3005 * target and source are directories and that target can be
3006 * destroyed, or that neither is a directory.
3008 if (S_ISDIR(target_ip->i_d.di_mode)) {
3010 * Make sure target dir is empty.
3012 if (!(xfs_dir_isempty(target_ip)) ||
3013 (target_ip->i_d.di_nlink > 2)) {
3015 goto out_trans_cancel;
3020 * Link the source inode under the target name.
3021 * If the source inode is a directory and we are moving
3022 * it across directories, its ".." entry will be
3023 * inconsistent until we replace that down below.
3025 * In case there is already an entry with the same
3026 * name at the destination directory, remove it first.
3028 error = xfs_dir_replace(tp, target_dp, target_name,
3030 &first_block, &free_list, spaceres);
3032 goto out_bmap_cancel;
3034 xfs_trans_ichgtime(tp, target_dp,
3035 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3038 * Decrement the link count on the target since the target
3039 * dir no longer points to it.
3041 error = xfs_droplink(tp, target_ip);
3043 goto out_bmap_cancel;
3045 if (src_is_directory) {
3047 * Drop the link from the old "." entry.
3049 error = xfs_droplink(tp, target_ip);
3051 goto out_bmap_cancel;
3053 } /* target_ip != NULL */
3056 * Remove the source.
3058 if (new_parent && src_is_directory) {
3060 * Rewrite the ".." entry to point to the new
3063 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3065 &first_block, &free_list, spaceres);
3066 ASSERT(error != -EEXIST);
3068 goto out_bmap_cancel;
3072 * We always want to hit the ctime on the source inode.
3074 * This isn't strictly required by the standards since the source
3075 * inode isn't really being changed, but old unix file systems did
3076 * it and some incremental backup programs won't work without it.
3078 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3079 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3082 * Adjust the link count on src_dp. This is necessary when
3083 * renaming a directory, either within one parent when
3084 * the target existed, or across two parent directories.
3086 if (src_is_directory && (new_parent || target_ip != NULL)) {
3089 * Decrement link count on src_directory since the
3090 * entry that's moved no longer points to it.
3092 error = xfs_droplink(tp, src_dp);
3094 goto out_bmap_cancel;
3098 * For whiteouts, we only need to update the source dirent with the
3099 * inode number of the whiteout inode rather than removing it
3103 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3104 &first_block, &free_list, spaceres);
3106 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3107 &first_block, &free_list, spaceres);
3109 goto out_bmap_cancel;
3112 * For whiteouts, we need to bump the link count on the whiteout inode.
3113 * This means that failures all the way up to this point leave the inode
3114 * on the unlinked list and so cleanup is a simple matter of dropping
3115 * the remaining reference to it. If we fail here after bumping the link
3116 * count, we're shutting down the filesystem so we'll never see the
3117 * intermediate state on disk.
3120 ASSERT(VFS_I(wip)->i_nlink == 0 && wip->i_d.di_nlink == 0);
3121 error = xfs_bumplink(tp, wip);
3123 goto out_bmap_cancel;
3124 error = xfs_iunlink_remove(tp, wip);
3126 goto out_bmap_cancel;
3127 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3130 * Now we have a real link, clear the "I'm a tmpfile" state
3131 * flag from the inode so it doesn't accidentally get misused in
3134 VFS_I(wip)->i_state &= ~I_LINKABLE;
3137 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3138 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3140 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3142 error = xfs_finish_rename(tp, &free_list);
3148 xfs_bmap_cancel(&free_list);
3150 xfs_trans_cancel(tp);
3161 xfs_mount_t *mp = ip->i_mount;
3162 struct xfs_perag *pag;
3163 unsigned long first_index, mask;
3164 unsigned long inodes_per_cluster;
3166 xfs_inode_t **ilist;
3173 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3175 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3176 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3177 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
3181 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3182 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3184 /* really need a gang lookup range call here */
3185 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3186 first_index, inodes_per_cluster);
3190 for (i = 0; i < nr_found; i++) {
3196 * because this is an RCU protected lookup, we could find a
3197 * recently freed or even reallocated inode during the lookup.
3198 * We need to check under the i_flags_lock for a valid inode
3199 * here. Skip it if it is not valid or the wrong inode.
3201 spin_lock(&ip->i_flags_lock);
3203 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3204 spin_unlock(&ip->i_flags_lock);
3207 spin_unlock(&ip->i_flags_lock);
3210 * Do an un-protected check to see if the inode is dirty and
3211 * is a candidate for flushing. These checks will be repeated
3212 * later after the appropriate locks are acquired.
3214 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3218 * Try to get locks. If any are unavailable or it is pinned,
3219 * then this inode cannot be flushed and is skipped.
3222 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3224 if (!xfs_iflock_nowait(iq)) {
3225 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3228 if (xfs_ipincount(iq)) {
3230 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3235 * arriving here means that this inode can be flushed. First
3236 * re-check that it's dirty before flushing.
3238 if (!xfs_inode_clean(iq)) {
3240 error = xfs_iflush_int(iq, bp);
3242 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3243 goto cluster_corrupt_out;
3249 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3253 XFS_STATS_INC(xs_icluster_flushcnt);
3254 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3265 cluster_corrupt_out:
3267 * Corruption detected in the clustering loop. Invalidate the
3268 * inode buffer and shut down the filesystem.
3272 * Clean up the buffer. If it was delwri, just release it --
3273 * brelse can handle it with no problems. If not, shut down the
3274 * filesystem before releasing the buffer.
3276 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3280 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3282 if (!bufwasdelwri) {
3284 * Just like incore_relse: if we have b_iodone functions,
3285 * mark the buffer as an error and call them. Otherwise
3286 * mark it as stale and brelse.
3291 xfs_buf_ioerror(bp, -EIO);
3300 * Unlocks the flush lock
3302 xfs_iflush_abort(iq, false);
3305 return -EFSCORRUPTED;
3309 * Flush dirty inode metadata into the backing buffer.
3311 * The caller must have the inode lock and the inode flush lock held. The
3312 * inode lock will still be held upon return to the caller, and the inode
3313 * flush lock will be released after the inode has reached the disk.
3315 * The caller must write out the buffer returned in *bpp and release it.
3319 struct xfs_inode *ip,
3320 struct xfs_buf **bpp)
3322 struct xfs_mount *mp = ip->i_mount;
3324 struct xfs_dinode *dip;
3327 XFS_STATS_INC(xs_iflush_count);
3329 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3330 ASSERT(xfs_isiflocked(ip));
3331 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3332 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3336 xfs_iunpin_wait(ip);
3339 * For stale inodes we cannot rely on the backing buffer remaining
3340 * stale in cache for the remaining life of the stale inode and so
3341 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3342 * inodes below. We have to check this after ensuring the inode is
3343 * unpinned so that it is safe to reclaim the stale inode after the
3346 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3352 * This may have been unpinned because the filesystem is shutting
3353 * down forcibly. If that's the case we must not write this inode
3354 * to disk, because the log record didn't make it to disk.
3356 * We also have to remove the log item from the AIL in this case,
3357 * as we wait for an empty AIL as part of the unmount process.
3359 if (XFS_FORCED_SHUTDOWN(mp)) {
3365 * Get the buffer containing the on-disk inode.
3367 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3375 * First flush out the inode that xfs_iflush was called with.
3377 error = xfs_iflush_int(ip, bp);
3382 * If the buffer is pinned then push on the log now so we won't
3383 * get stuck waiting in the write for too long.
3385 if (xfs_buf_ispinned(bp))
3386 xfs_log_force(mp, 0);
3390 * see if other inodes can be gathered into this write
3392 error = xfs_iflush_cluster(ip, bp);
3394 goto cluster_corrupt_out;
3401 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3402 cluster_corrupt_out:
3403 error = -EFSCORRUPTED;
3406 * Unlocks the flush lock
3408 xfs_iflush_abort(ip, false);
3414 struct xfs_inode *ip,
3417 struct xfs_inode_log_item *iip = ip->i_itemp;
3418 struct xfs_dinode *dip;
3419 struct xfs_mount *mp = ip->i_mount;
3421 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3422 ASSERT(xfs_isiflocked(ip));
3423 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3424 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3425 ASSERT(iip != NULL && iip->ili_fields != 0);
3426 ASSERT(ip->i_d.di_version > 1);
3428 /* set *dip = inode's place in the buffer */
3429 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3431 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3432 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3433 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3434 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3435 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3438 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3439 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3440 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3441 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3442 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3445 if (S_ISREG(ip->i_d.di_mode)) {
3447 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3448 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3449 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3450 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3451 "%s: Bad regular inode %Lu, ptr 0x%p",
3452 __func__, ip->i_ino, ip);
3455 } else if (S_ISDIR(ip->i_d.di_mode)) {
3457 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3458 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3459 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3460 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3461 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3462 "%s: Bad directory inode %Lu, ptr 0x%p",
3463 __func__, ip->i_ino, ip);
3467 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3468 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3469 XFS_RANDOM_IFLUSH_5)) {
3470 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3471 "%s: detected corrupt incore inode %Lu, "
3472 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3473 __func__, ip->i_ino,
3474 ip->i_d.di_nextents + ip->i_d.di_anextents,
3475 ip->i_d.di_nblocks, ip);
3478 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3479 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3480 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3481 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3482 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3487 * Inode item log recovery for v2 inodes are dependent on the
3488 * di_flushiter count for correct sequencing. We bump the flush
3489 * iteration count so we can detect flushes which postdate a log record
3490 * during recovery. This is redundant as we now log every change and
3491 * hence this can't happen but we need to still do it to ensure
3492 * backwards compatibility with old kernels that predate logging all
3495 if (ip->i_d.di_version < 3)
3496 ip->i_d.di_flushiter++;
3499 * Copy the dirty parts of the inode into the on-disk
3500 * inode. We always copy out the core of the inode,
3501 * because if the inode is dirty at all the core must
3504 xfs_dinode_to_disk(dip, &ip->i_d);
3506 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3507 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3508 ip->i_d.di_flushiter = 0;
3510 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3511 if (XFS_IFORK_Q(ip))
3512 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3513 xfs_inobp_check(mp, bp);
3516 * We've recorded everything logged in the inode, so we'd like to clear
3517 * the ili_fields bits so we don't log and flush things unnecessarily.
3518 * However, we can't stop logging all this information until the data
3519 * we've copied into the disk buffer is written to disk. If we did we
3520 * might overwrite the copy of the inode in the log with all the data
3521 * after re-logging only part of it, and in the face of a crash we
3522 * wouldn't have all the data we need to recover.
3524 * What we do is move the bits to the ili_last_fields field. When
3525 * logging the inode, these bits are moved back to the ili_fields field.
3526 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3527 * know that the information those bits represent is permanently on
3528 * disk. As long as the flush completes before the inode is logged
3529 * again, then both ili_fields and ili_last_fields will be cleared.
3531 * We can play with the ili_fields bits here, because the inode lock
3532 * must be held exclusively in order to set bits there and the flush
3533 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3534 * done routine can tell whether or not to look in the AIL. Also, store
3535 * the current LSN of the inode so that we can tell whether the item has
3536 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3537 * need the AIL lock, because it is a 64 bit value that cannot be read
3540 iip->ili_last_fields = iip->ili_fields;
3541 iip->ili_fields = 0;
3542 iip->ili_logged = 1;
3544 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3545 &iip->ili_item.li_lsn);
3548 * Attach the function xfs_iflush_done to the inode's
3549 * buffer. This will remove the inode from the AIL
3550 * and unlock the inode's flush lock when the inode is
3551 * completely written to disk.
3553 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3555 /* update the lsn in the on disk inode if required */
3556 if (ip->i_d.di_version == 3)
3557 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3559 /* generate the checksum. */
3560 xfs_dinode_calc_crc(mp, dip);
3562 ASSERT(bp->b_fspriv != NULL);
3563 ASSERT(bp->b_iodone != NULL);
3567 return -EFSCORRUPTED;