2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
33 #include "xfs_attr_sf.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
54 kmem_zone_t *xfs_inode_zone;
57 * Used in xfs_itruncate_extents(). This is the maximum number of extents
58 * freed from a file in a single transaction.
60 #define XFS_ITRUNC_MAX_EXTENTS 2
62 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
63 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
64 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
67 * helper function to extract extent size hint from inode
73 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
74 return ip->i_d.di_extsize;
75 if (XFS_IS_REALTIME_INODE(ip))
76 return ip->i_mount->m_sb.sb_rextsize;
81 * Helper function to extract CoW extent size hint from inode.
82 * Between the extent size hint and the CoW extent size hint, we
83 * return the greater of the two. If the value is zero (automatic),
84 * use the default size.
87 xfs_get_cowextsz_hint(
93 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
94 a = ip->i_d.di_cowextsize;
95 b = xfs_get_extsz_hint(ip);
99 return XFS_DEFAULT_COWEXTSZ_HINT;
104 * These two are wrapper routines around the xfs_ilock() routine used to
105 * centralize some grungy code. They are used in places that wish to lock the
106 * inode solely for reading the extents. The reason these places can't just
107 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
108 * bringing in of the extents from disk for a file in b-tree format. If the
109 * inode is in b-tree format, then we need to lock the inode exclusively until
110 * the extents are read in. Locking it exclusively all the time would limit
111 * our parallelism unnecessarily, though. What we do instead is check to see
112 * if the extents have been read in yet, and only lock the inode exclusively
115 * The functions return a value which should be given to the corresponding
116 * xfs_iunlock() call.
119 xfs_ilock_data_map_shared(
120 struct xfs_inode *ip)
122 uint lock_mode = XFS_ILOCK_SHARED;
124 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
125 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
126 lock_mode = XFS_ILOCK_EXCL;
127 xfs_ilock(ip, lock_mode);
132 xfs_ilock_attr_map_shared(
133 struct xfs_inode *ip)
135 uint lock_mode = XFS_ILOCK_SHARED;
137 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
138 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
139 lock_mode = XFS_ILOCK_EXCL;
140 xfs_ilock(ip, lock_mode);
145 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
146 * the i_lock. This routine allows various combinations of the locks to be
149 * The 3 locks should always be ordered so that the IO lock is obtained first,
150 * the mmap lock second and the ilock last in order to prevent deadlock.
152 * Basic locking order:
154 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
156 * mmap_sem locking order:
158 * i_iolock -> page lock -> mmap_sem
159 * mmap_sem -> i_mmap_lock -> page_lock
161 * The difference in mmap_sem locking order mean that we cannot hold the
162 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
163 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
164 * in get_user_pages() to map the user pages into the kernel address space for
165 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
166 * page faults already hold the mmap_sem.
168 * Hence to serialise fully against both syscall and mmap based IO, we need to
169 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
170 * taken in places where we need to invalidate the page cache in a race
171 * free manner (e.g. truncate, hole punch and other extent manipulation
179 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
182 * You can't set both SHARED and EXCL for the same lock,
183 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
184 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
186 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
187 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
188 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
189 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
190 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
191 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
192 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
194 if (lock_flags & XFS_IOLOCK_EXCL)
195 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
196 else if (lock_flags & XFS_IOLOCK_SHARED)
197 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
199 if (lock_flags & XFS_MMAPLOCK_EXCL)
200 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
201 else if (lock_flags & XFS_MMAPLOCK_SHARED)
202 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
204 if (lock_flags & XFS_ILOCK_EXCL)
205 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
206 else if (lock_flags & XFS_ILOCK_SHARED)
207 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
211 * This is just like xfs_ilock(), except that the caller
212 * is guaranteed not to sleep. It returns 1 if it gets
213 * the requested locks and 0 otherwise. If the IO lock is
214 * obtained but the inode lock cannot be, then the IO lock
215 * is dropped before returning.
217 * ip -- the inode being locked
218 * lock_flags -- this parameter indicates the inode's locks to be
219 * to be locked. See the comment for xfs_ilock() for a list
227 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
230 * You can't set both SHARED and EXCL for the same lock,
231 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
232 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
234 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
235 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
236 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
237 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
238 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
239 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
240 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
242 if (lock_flags & XFS_IOLOCK_EXCL) {
243 if (!mrtryupdate(&ip->i_iolock))
245 } else if (lock_flags & XFS_IOLOCK_SHARED) {
246 if (!mrtryaccess(&ip->i_iolock))
250 if (lock_flags & XFS_MMAPLOCK_EXCL) {
251 if (!mrtryupdate(&ip->i_mmaplock))
252 goto out_undo_iolock;
253 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
254 if (!mrtryaccess(&ip->i_mmaplock))
255 goto out_undo_iolock;
258 if (lock_flags & XFS_ILOCK_EXCL) {
259 if (!mrtryupdate(&ip->i_lock))
260 goto out_undo_mmaplock;
261 } else if (lock_flags & XFS_ILOCK_SHARED) {
262 if (!mrtryaccess(&ip->i_lock))
263 goto out_undo_mmaplock;
268 if (lock_flags & XFS_MMAPLOCK_EXCL)
269 mrunlock_excl(&ip->i_mmaplock);
270 else if (lock_flags & XFS_MMAPLOCK_SHARED)
271 mrunlock_shared(&ip->i_mmaplock);
273 if (lock_flags & XFS_IOLOCK_EXCL)
274 mrunlock_excl(&ip->i_iolock);
275 else if (lock_flags & XFS_IOLOCK_SHARED)
276 mrunlock_shared(&ip->i_iolock);
282 * xfs_iunlock() is used to drop the inode locks acquired with
283 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
284 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
285 * that we know which locks to drop.
287 * ip -- the inode being unlocked
288 * lock_flags -- this parameter indicates the inode's locks to be
289 * to be unlocked. See the comment for xfs_ilock() for a list
290 * of valid values for this parameter.
299 * You can't set both SHARED and EXCL for the same lock,
300 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
301 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
303 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
304 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
305 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
306 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
307 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
308 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
309 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
310 ASSERT(lock_flags != 0);
312 if (lock_flags & XFS_IOLOCK_EXCL)
313 mrunlock_excl(&ip->i_iolock);
314 else if (lock_flags & XFS_IOLOCK_SHARED)
315 mrunlock_shared(&ip->i_iolock);
317 if (lock_flags & XFS_MMAPLOCK_EXCL)
318 mrunlock_excl(&ip->i_mmaplock);
319 else if (lock_flags & XFS_MMAPLOCK_SHARED)
320 mrunlock_shared(&ip->i_mmaplock);
322 if (lock_flags & XFS_ILOCK_EXCL)
323 mrunlock_excl(&ip->i_lock);
324 else if (lock_flags & XFS_ILOCK_SHARED)
325 mrunlock_shared(&ip->i_lock);
327 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
331 * give up write locks. the i/o lock cannot be held nested
332 * if it is being demoted.
339 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
341 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
343 if (lock_flags & XFS_ILOCK_EXCL)
344 mrdemote(&ip->i_lock);
345 if (lock_flags & XFS_MMAPLOCK_EXCL)
346 mrdemote(&ip->i_mmaplock);
347 if (lock_flags & XFS_IOLOCK_EXCL)
348 mrdemote(&ip->i_iolock);
350 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
353 #if defined(DEBUG) || defined(XFS_WARN)
359 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
360 if (!(lock_flags & XFS_ILOCK_SHARED))
361 return !!ip->i_lock.mr_writer;
362 return rwsem_is_locked(&ip->i_lock.mr_lock);
365 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
366 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
367 return !!ip->i_mmaplock.mr_writer;
368 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
371 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
372 if (!(lock_flags & XFS_IOLOCK_SHARED))
373 return !!ip->i_iolock.mr_writer;
374 return rwsem_is_locked(&ip->i_iolock.mr_lock);
384 int xfs_small_retries;
385 int xfs_middle_retries;
386 int xfs_lots_retries;
391 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
392 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
393 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
394 * errors and warnings.
396 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
398 xfs_lockdep_subclass_ok(
401 return subclass < MAX_LOCKDEP_SUBCLASSES;
404 #define xfs_lockdep_subclass_ok(subclass) (true)
408 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
409 * value. This can be called for any type of inode lock combination, including
410 * parent locking. Care must be taken to ensure we don't overrun the subclass
411 * storage fields in the class mask we build.
414 xfs_lock_inumorder(int lock_mode, int subclass)
418 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
420 ASSERT(xfs_lockdep_subclass_ok(subclass));
422 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
423 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
424 ASSERT(xfs_lockdep_subclass_ok(subclass +
425 XFS_IOLOCK_PARENT_VAL));
426 class += subclass << XFS_IOLOCK_SHIFT;
427 if (lock_mode & XFS_IOLOCK_PARENT)
428 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
431 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
432 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
433 class += subclass << XFS_MMAPLOCK_SHIFT;
436 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
437 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
438 class += subclass << XFS_ILOCK_SHIFT;
441 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
445 * The following routine will lock n inodes in exclusive mode. We assume the
446 * caller calls us with the inodes in i_ino order.
448 * We need to detect deadlock where an inode that we lock is in the AIL and we
449 * start waiting for another inode that is locked by a thread in a long running
450 * transaction (such as truncate). This can result in deadlock since the long
451 * running trans might need to wait for the inode we just locked in order to
452 * push the tail and free space in the log.
454 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
455 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
456 * lock more than one at a time, lockdep will report false positives saying we
457 * have violated locking orders.
465 int attempts = 0, i, j, try_lock;
469 * Currently supports between 2 and 5 inodes with exclusive locking. We
470 * support an arbitrary depth of locking here, but absolute limits on
471 * inodes depend on the the type of locking and the limits placed by
472 * lockdep annotations in xfs_lock_inumorder. These are all checked by
475 ASSERT(ips && inodes >= 2 && inodes <= 5);
476 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
478 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
480 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
481 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
482 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
483 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
484 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
485 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
487 if (lock_mode & XFS_IOLOCK_EXCL) {
488 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
489 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
490 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
495 for (; i < inodes; i++) {
498 if (i && (ips[i] == ips[i - 1])) /* Already locked */
502 * If try_lock is not set yet, make sure all locked inodes are
503 * not in the AIL. If any are, set try_lock to be used later.
506 for (j = (i - 1); j >= 0 && !try_lock; j--) {
507 lp = (xfs_log_item_t *)ips[j]->i_itemp;
508 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
514 * If any of the previous locks we have locked is in the AIL,
515 * we must TRY to get the second and subsequent locks. If
516 * we can't get any, we must release all we have
520 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
524 /* try_lock means we have an inode locked that is in the AIL. */
526 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
530 * Unlock all previous guys and try again. xfs_iunlock will try
531 * to push the tail if the inode is in the AIL.
534 for (j = i - 1; j >= 0; j--) {
536 * Check to see if we've already unlocked this one. Not
537 * the first one going back, and the inode ptr is the
540 if (j != (i - 1) && ips[j] == ips[j + 1])
543 xfs_iunlock(ips[j], lock_mode);
546 if ((attempts % 5) == 0) {
547 delay(1); /* Don't just spin the CPU */
559 if (attempts < 5) xfs_small_retries++;
560 else if (attempts < 100) xfs_middle_retries++;
561 else xfs_lots_retries++;
569 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
570 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
571 * lock more than one at a time, lockdep will report false positives saying we
572 * have violated locking orders.
584 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
585 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
586 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
587 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
588 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
590 ASSERT(ip0->i_ino != ip1->i_ino);
592 if (ip0->i_ino > ip1->i_ino) {
599 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
602 * If the first lock we have locked is in the AIL, we must TRY to get
603 * the second lock. If we can't get it, we must release the first one
606 lp = (xfs_log_item_t *)ip0->i_itemp;
607 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
608 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
609 xfs_iunlock(ip0, lock_mode);
610 if ((++attempts % 5) == 0)
611 delay(1); /* Don't just spin the CPU */
615 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
622 struct xfs_inode *ip)
624 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
625 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
628 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
629 if (xfs_isiflocked(ip))
631 } while (!xfs_iflock_nowait(ip));
633 finish_wait(wq, &wait.wait);
644 if (di_flags & XFS_DIFLAG_ANY) {
645 if (di_flags & XFS_DIFLAG_REALTIME)
646 flags |= FS_XFLAG_REALTIME;
647 if (di_flags & XFS_DIFLAG_PREALLOC)
648 flags |= FS_XFLAG_PREALLOC;
649 if (di_flags & XFS_DIFLAG_IMMUTABLE)
650 flags |= FS_XFLAG_IMMUTABLE;
651 if (di_flags & XFS_DIFLAG_APPEND)
652 flags |= FS_XFLAG_APPEND;
653 if (di_flags & XFS_DIFLAG_SYNC)
654 flags |= FS_XFLAG_SYNC;
655 if (di_flags & XFS_DIFLAG_NOATIME)
656 flags |= FS_XFLAG_NOATIME;
657 if (di_flags & XFS_DIFLAG_NODUMP)
658 flags |= FS_XFLAG_NODUMP;
659 if (di_flags & XFS_DIFLAG_RTINHERIT)
660 flags |= FS_XFLAG_RTINHERIT;
661 if (di_flags & XFS_DIFLAG_PROJINHERIT)
662 flags |= FS_XFLAG_PROJINHERIT;
663 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
664 flags |= FS_XFLAG_NOSYMLINKS;
665 if (di_flags & XFS_DIFLAG_EXTSIZE)
666 flags |= FS_XFLAG_EXTSIZE;
667 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
668 flags |= FS_XFLAG_EXTSZINHERIT;
669 if (di_flags & XFS_DIFLAG_NODEFRAG)
670 flags |= FS_XFLAG_NODEFRAG;
671 if (di_flags & XFS_DIFLAG_FILESTREAM)
672 flags |= FS_XFLAG_FILESTREAM;
675 if (di_flags2 & XFS_DIFLAG2_ANY) {
676 if (di_flags2 & XFS_DIFLAG2_DAX)
677 flags |= FS_XFLAG_DAX;
678 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
679 flags |= FS_XFLAG_COWEXTSIZE;
683 flags |= FS_XFLAG_HASATTR;
690 struct xfs_inode *ip)
692 struct xfs_icdinode *dic = &ip->i_d;
694 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
698 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
699 * is allowed, otherwise it has to be an exact match. If a CI match is found,
700 * ci_name->name will point to a the actual name (caller must free) or
701 * will be set to NULL if an exact match is found.
706 struct xfs_name *name,
708 struct xfs_name *ci_name)
713 trace_xfs_lookup(dp, name);
715 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
718 xfs_ilock(dp, XFS_IOLOCK_SHARED);
719 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
723 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
727 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
732 kmem_free(ci_name->name);
734 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
740 * Allocate an inode on disk and return a copy of its in-core version.
741 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
742 * appropriately within the inode. The uid and gid for the inode are
743 * set according to the contents of the given cred structure.
745 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
746 * has a free inode available, call xfs_iget() to obtain the in-core
747 * version of the allocated inode. Finally, fill in the inode and
748 * log its initial contents. In this case, ialloc_context would be
751 * If xfs_dialloc() does not have an available inode, it will replenish
752 * its supply by doing an allocation. Since we can only do one
753 * allocation within a transaction without deadlocks, we must commit
754 * the current transaction before returning the inode itself.
755 * In this case, therefore, we will set ialloc_context and return.
756 * The caller should then commit the current transaction, start a new
757 * transaction, and call xfs_ialloc() again to actually get the inode.
759 * To ensure that some other process does not grab the inode that
760 * was allocated during the first call to xfs_ialloc(), this routine
761 * also returns the [locked] bp pointing to the head of the freelist
762 * as ialloc_context. The caller should hold this buffer across
763 * the commit and pass it back into this routine on the second call.
765 * If we are allocating quota inodes, we do not have a parent inode
766 * to attach to or associate with (i.e. pip == NULL) because they
767 * are not linked into the directory structure - they are attached
768 * directly to the superblock - and so have no parent.
779 xfs_buf_t **ialloc_context,
782 struct xfs_mount *mp = tp->t_mountp;
791 * Call the space management code to pick
792 * the on-disk inode to be allocated.
794 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
795 ialloc_context, &ino);
798 if (*ialloc_context || ino == NULLFSINO) {
802 ASSERT(*ialloc_context == NULL);
805 * Get the in-core inode with the lock held exclusively.
806 * This is because we're setting fields here we need
807 * to prevent others from looking at until we're done.
809 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
810 XFS_ILOCK_EXCL, &ip);
817 * We always convert v1 inodes to v2 now - we only support filesystems
818 * with >= v2 inode capability, so there is no reason for ever leaving
819 * an inode in v1 format.
821 if (ip->i_d.di_version == 1)
822 ip->i_d.di_version = 2;
824 inode->i_mode = mode;
825 set_nlink(inode, nlink);
826 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
827 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
828 xfs_set_projid(ip, prid);
830 if (pip && XFS_INHERIT_GID(pip)) {
831 ip->i_d.di_gid = pip->i_d.di_gid;
832 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
833 inode->i_mode |= S_ISGID;
837 * If the group ID of the new file does not match the effective group
838 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
839 * (and only if the irix_sgid_inherit compatibility variable is set).
841 if ((irix_sgid_inherit) &&
842 (inode->i_mode & S_ISGID) &&
843 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
844 inode->i_mode &= ~S_ISGID;
847 ip->i_d.di_nextents = 0;
848 ASSERT(ip->i_d.di_nblocks == 0);
850 tv = current_time(inode);
855 ip->i_d.di_extsize = 0;
856 ip->i_d.di_dmevmask = 0;
857 ip->i_d.di_dmstate = 0;
858 ip->i_d.di_flags = 0;
860 if (ip->i_d.di_version == 3) {
861 inode->i_version = 1;
862 ip->i_d.di_flags2 = 0;
863 ip->i_d.di_cowextsize = 0;
864 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
865 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
869 flags = XFS_ILOG_CORE;
870 switch (mode & S_IFMT) {
875 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
876 ip->i_df.if_u2.if_rdev = rdev;
877 ip->i_df.if_flags = 0;
878 flags |= XFS_ILOG_DEV;
882 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
883 uint64_t di_flags2 = 0;
887 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
888 di_flags |= XFS_DIFLAG_RTINHERIT;
889 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
890 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
891 ip->i_d.di_extsize = pip->i_d.di_extsize;
893 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
894 di_flags |= XFS_DIFLAG_PROJINHERIT;
895 } else if (S_ISREG(mode)) {
896 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
897 di_flags |= XFS_DIFLAG_REALTIME;
898 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
899 di_flags |= XFS_DIFLAG_EXTSIZE;
900 ip->i_d.di_extsize = pip->i_d.di_extsize;
903 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
905 di_flags |= XFS_DIFLAG_NOATIME;
906 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
908 di_flags |= XFS_DIFLAG_NODUMP;
909 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
911 di_flags |= XFS_DIFLAG_SYNC;
912 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
913 xfs_inherit_nosymlinks)
914 di_flags |= XFS_DIFLAG_NOSYMLINKS;
915 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
916 xfs_inherit_nodefrag)
917 di_flags |= XFS_DIFLAG_NODEFRAG;
918 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
919 di_flags |= XFS_DIFLAG_FILESTREAM;
920 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
921 di_flags2 |= XFS_DIFLAG2_DAX;
923 ip->i_d.di_flags |= di_flags;
924 ip->i_d.di_flags2 |= di_flags2;
927 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
928 pip->i_d.di_version == 3 &&
929 ip->i_d.di_version == 3) {
930 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
931 ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
932 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
937 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
938 ip->i_df.if_flags = XFS_IFEXTENTS;
939 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
940 ip->i_df.if_u1.if_extents = NULL;
946 * Attribute fork settings for new inode.
948 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
949 ip->i_d.di_anextents = 0;
952 * Log the new values stuffed into the inode.
954 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
955 xfs_trans_log_inode(tp, ip, flags);
957 /* now that we have an i_mode we can setup the inode structure */
965 * Allocates a new inode from disk and return a pointer to the
966 * incore copy. This routine will internally commit the current
967 * transaction and allocate a new one if the Space Manager needed
968 * to do an allocation to replenish the inode free-list.
970 * This routine is designed to be called from xfs_create and
976 xfs_trans_t **tpp, /* input: current transaction;
977 output: may be a new transaction. */
978 xfs_inode_t *dp, /* directory within whose allocate
983 prid_t prid, /* project id */
984 int okalloc, /* ok to allocate new space */
985 xfs_inode_t **ipp, /* pointer to inode; it will be
992 xfs_buf_t *ialloc_context = NULL;
998 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1001 * xfs_ialloc will return a pointer to an incore inode if
1002 * the Space Manager has an available inode on the free
1003 * list. Otherwise, it will do an allocation and replenish
1004 * the freelist. Since we can only do one allocation per
1005 * transaction without deadlocks, we will need to commit the
1006 * current transaction and start a new one. We will then
1007 * need to call xfs_ialloc again to get the inode.
1009 * If xfs_ialloc did an allocation to replenish the freelist,
1010 * it returns the bp containing the head of the freelist as
1011 * ialloc_context. We will hold a lock on it across the
1012 * transaction commit so that no other process can steal
1013 * the inode(s) that we've just allocated.
1015 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1016 &ialloc_context, &ip);
1019 * Return an error if we were unable to allocate a new inode.
1020 * This should only happen if we run out of space on disk or
1021 * encounter a disk error.
1027 if (!ialloc_context && !ip) {
1033 * If the AGI buffer is non-NULL, then we were unable to get an
1034 * inode in one operation. We need to commit the current
1035 * transaction and call xfs_ialloc() again. It is guaranteed
1036 * to succeed the second time.
1038 if (ialloc_context) {
1040 * Normally, xfs_trans_commit releases all the locks.
1041 * We call bhold to hang on to the ialloc_context across
1042 * the commit. Holding this buffer prevents any other
1043 * processes from doing any allocations in this
1046 xfs_trans_bhold(tp, ialloc_context);
1049 * We want the quota changes to be associated with the next
1050 * transaction, NOT this one. So, detach the dqinfo from this
1051 * and attach it to the next transaction.
1056 dqinfo = (void *)tp->t_dqinfo;
1057 tp->t_dqinfo = NULL;
1058 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1059 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1062 code = xfs_trans_roll(&tp, NULL);
1063 if (committed != NULL)
1067 * Re-attach the quota info that we detached from prev trx.
1070 tp->t_dqinfo = dqinfo;
1071 tp->t_flags |= tflags;
1075 xfs_buf_relse(ialloc_context);
1080 xfs_trans_bjoin(tp, ialloc_context);
1083 * Call ialloc again. Since we've locked out all
1084 * other allocations in this allocation group,
1085 * this call should always succeed.
1087 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1088 okalloc, &ialloc_context, &ip);
1091 * If we get an error at this point, return to the caller
1092 * so that the current transaction can be aborted.
1099 ASSERT(!ialloc_context && ip);
1102 if (committed != NULL)
1113 * Decrement the link count on an inode & log the change. If this causes the
1114 * link count to go to zero, move the inode to AGI unlinked list so that it can
1115 * be freed when the last active reference goes away via xfs_inactive().
1117 static int /* error */
1122 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1124 drop_nlink(VFS_I(ip));
1125 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1127 if (VFS_I(ip)->i_nlink)
1130 return xfs_iunlink(tp, ip);
1134 * Increment the link count on an inode & log the change.
1141 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1143 ASSERT(ip->i_d.di_version > 1);
1144 inc_nlink(VFS_I(ip));
1145 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1152 struct xfs_name *name,
1157 int is_dir = S_ISDIR(mode);
1158 struct xfs_mount *mp = dp->i_mount;
1159 struct xfs_inode *ip = NULL;
1160 struct xfs_trans *tp = NULL;
1162 struct xfs_defer_ops dfops;
1163 xfs_fsblock_t first_block;
1164 bool unlock_dp_on_error = false;
1166 struct xfs_dquot *udqp = NULL;
1167 struct xfs_dquot *gdqp = NULL;
1168 struct xfs_dquot *pdqp = NULL;
1169 struct xfs_trans_res *tres;
1172 trace_xfs_create(dp, name);
1174 if (XFS_FORCED_SHUTDOWN(mp))
1177 prid = xfs_get_initial_prid(dp);
1180 * Make sure that we have allocated dquot(s) on disk.
1182 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1183 xfs_kgid_to_gid(current_fsgid()), prid,
1184 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1185 &udqp, &gdqp, &pdqp);
1191 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1192 tres = &M_RES(mp)->tr_mkdir;
1194 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1195 tres = &M_RES(mp)->tr_create;
1199 * Initially assume that the file does not exist and
1200 * reserve the resources for that case. If that is not
1201 * the case we'll drop the one we have and get a more
1202 * appropriate transaction later.
1204 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1205 if (error == -ENOSPC) {
1206 /* flush outstanding delalloc blocks and retry */
1207 xfs_flush_inodes(mp);
1208 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1210 if (error == -ENOSPC) {
1211 /* No space at all so try a "no-allocation" reservation */
1213 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1216 goto out_release_inode;
1218 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1219 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
1220 unlock_dp_on_error = true;
1222 xfs_defer_init(&dfops, &first_block);
1225 * Reserve disk quota and the inode.
1227 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1228 pdqp, resblks, 1, 0);
1230 goto out_trans_cancel;
1233 error = xfs_dir_canenter(tp, dp, name);
1235 goto out_trans_cancel;
1239 * A newly created regular or special file just has one directory
1240 * entry pointing to them, but a directory also the "." entry
1241 * pointing to itself.
1243 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1244 prid, resblks > 0, &ip, NULL);
1246 goto out_trans_cancel;
1249 * Now we join the directory inode to the transaction. We do not do it
1250 * earlier because xfs_dir_ialloc might commit the previous transaction
1251 * (and release all the locks). An error from here on will result in
1252 * the transaction cancel unlocking dp so don't do it explicitly in the
1255 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1256 unlock_dp_on_error = false;
1258 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1259 &first_block, &dfops, resblks ?
1260 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1262 ASSERT(error != -ENOSPC);
1263 goto out_trans_cancel;
1265 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1266 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1269 error = xfs_dir_init(tp, ip, dp);
1271 goto out_bmap_cancel;
1273 error = xfs_bumplink(tp, dp);
1275 goto out_bmap_cancel;
1279 * If this is a synchronous mount, make sure that the
1280 * create transaction goes to disk before returning to
1283 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1284 xfs_trans_set_sync(tp);
1287 * Attach the dquot(s) to the inodes and modify them incore.
1288 * These ids of the inode couldn't have changed since the new
1289 * inode has been locked ever since it was created.
1291 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1293 error = xfs_defer_finish(&tp, &dfops, NULL);
1295 goto out_bmap_cancel;
1297 error = xfs_trans_commit(tp);
1299 goto out_release_inode;
1301 xfs_qm_dqrele(udqp);
1302 xfs_qm_dqrele(gdqp);
1303 xfs_qm_dqrele(pdqp);
1309 xfs_defer_cancel(&dfops);
1311 xfs_trans_cancel(tp);
1314 * Wait until after the current transaction is aborted to finish the
1315 * setup of the inode and release the inode. This prevents recursive
1316 * transactions and deadlocks from xfs_inactive.
1319 xfs_finish_inode_setup(ip);
1323 xfs_qm_dqrele(udqp);
1324 xfs_qm_dqrele(gdqp);
1325 xfs_qm_dqrele(pdqp);
1327 if (unlock_dp_on_error)
1328 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1334 struct xfs_inode *dp,
1335 struct dentry *dentry,
1337 struct xfs_inode **ipp)
1339 struct xfs_mount *mp = dp->i_mount;
1340 struct xfs_inode *ip = NULL;
1341 struct xfs_trans *tp = NULL;
1344 struct xfs_dquot *udqp = NULL;
1345 struct xfs_dquot *gdqp = NULL;
1346 struct xfs_dquot *pdqp = NULL;
1347 struct xfs_trans_res *tres;
1350 if (XFS_FORCED_SHUTDOWN(mp))
1353 prid = xfs_get_initial_prid(dp);
1356 * Make sure that we have allocated dquot(s) on disk.
1358 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1359 xfs_kgid_to_gid(current_fsgid()), prid,
1360 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1361 &udqp, &gdqp, &pdqp);
1365 resblks = XFS_IALLOC_SPACE_RES(mp);
1366 tres = &M_RES(mp)->tr_create_tmpfile;
1368 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1369 if (error == -ENOSPC) {
1370 /* No space at all so try a "no-allocation" reservation */
1372 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1375 goto out_release_inode;
1377 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1378 pdqp, resblks, 1, 0);
1380 goto out_trans_cancel;
1382 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1383 prid, resblks > 0, &ip, NULL);
1385 goto out_trans_cancel;
1387 if (mp->m_flags & XFS_MOUNT_WSYNC)
1388 xfs_trans_set_sync(tp);
1391 * Attach the dquot(s) to the inodes and modify them incore.
1392 * These ids of the inode couldn't have changed since the new
1393 * inode has been locked ever since it was created.
1395 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1397 error = xfs_iunlink(tp, ip);
1399 goto out_trans_cancel;
1401 error = xfs_trans_commit(tp);
1403 goto out_release_inode;
1405 xfs_qm_dqrele(udqp);
1406 xfs_qm_dqrele(gdqp);
1407 xfs_qm_dqrele(pdqp);
1413 xfs_trans_cancel(tp);
1416 * Wait until after the current transaction is aborted to finish the
1417 * setup of the inode and release the inode. This prevents recursive
1418 * transactions and deadlocks from xfs_inactive.
1421 xfs_finish_inode_setup(ip);
1425 xfs_qm_dqrele(udqp);
1426 xfs_qm_dqrele(gdqp);
1427 xfs_qm_dqrele(pdqp);
1436 struct xfs_name *target_name)
1438 xfs_mount_t *mp = tdp->i_mount;
1441 struct xfs_defer_ops dfops;
1442 xfs_fsblock_t first_block;
1445 trace_xfs_link(tdp, target_name);
1447 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1449 if (XFS_FORCED_SHUTDOWN(mp))
1452 error = xfs_qm_dqattach(sip, 0);
1456 error = xfs_qm_dqattach(tdp, 0);
1460 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1461 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1462 if (error == -ENOSPC) {
1464 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1469 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
1470 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1472 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1473 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1476 * If we are using project inheritance, we only allow hard link
1477 * creation in our tree when the project IDs are the same; else
1478 * the tree quota mechanism could be circumvented.
1480 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1481 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1487 error = xfs_dir_canenter(tp, tdp, target_name);
1492 xfs_defer_init(&dfops, &first_block);
1495 * Handle initial link state of O_TMPFILE inode
1497 if (VFS_I(sip)->i_nlink == 0) {
1498 error = xfs_iunlink_remove(tp, sip);
1503 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1504 &first_block, &dfops, resblks);
1507 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1508 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1510 error = xfs_bumplink(tp, sip);
1515 * If this is a synchronous mount, make sure that the
1516 * link transaction goes to disk before returning to
1519 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1520 xfs_trans_set_sync(tp);
1522 error = xfs_defer_finish(&tp, &dfops, NULL);
1524 xfs_defer_cancel(&dfops);
1528 return xfs_trans_commit(tp);
1531 xfs_trans_cancel(tp);
1537 * Free up the underlying blocks past new_size. The new size must be smaller
1538 * than the current size. This routine can be used both for the attribute and
1539 * data fork, and does not modify the inode size, which is left to the caller.
1541 * The transaction passed to this routine must have made a permanent log
1542 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1543 * given transaction and start new ones, so make sure everything involved in
1544 * the transaction is tidy before calling here. Some transaction will be
1545 * returned to the caller to be committed. The incoming transaction must
1546 * already include the inode, and both inode locks must be held exclusively.
1547 * The inode must also be "held" within the transaction. On return the inode
1548 * will be "held" within the returned transaction. This routine does NOT
1549 * require any disk space to be reserved for it within the transaction.
1551 * If we get an error, we must return with the inode locked and linked into the
1552 * current transaction. This keeps things simple for the higher level code,
1553 * because it always knows that the inode is locked and held in the transaction
1554 * that returns to it whether errors occur or not. We don't mark the inode
1555 * dirty on error so that transactions can be easily aborted if possible.
1558 xfs_itruncate_extents(
1559 struct xfs_trans **tpp,
1560 struct xfs_inode *ip,
1562 xfs_fsize_t new_size)
1564 struct xfs_mount *mp = ip->i_mount;
1565 struct xfs_trans *tp = *tpp;
1566 struct xfs_defer_ops dfops;
1567 xfs_fsblock_t first_block;
1568 xfs_fileoff_t first_unmap_block;
1569 xfs_fileoff_t last_block;
1570 xfs_filblks_t unmap_len;
1574 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1575 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1576 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1577 ASSERT(new_size <= XFS_ISIZE(ip));
1578 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1579 ASSERT(ip->i_itemp != NULL);
1580 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1581 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1583 trace_xfs_itruncate_extents_start(ip, new_size);
1586 * Since it is possible for space to become allocated beyond
1587 * the end of the file (in a crash where the space is allocated
1588 * but the inode size is not yet updated), simply remove any
1589 * blocks which show up between the new EOF and the maximum
1590 * possible file size. If the first block to be removed is
1591 * beyond the maximum file size (ie it is the same as last_block),
1592 * then there is nothing to do.
1594 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1595 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1596 if (first_unmap_block == last_block)
1599 ASSERT(first_unmap_block < last_block);
1600 unmap_len = last_block - first_unmap_block + 1;
1602 xfs_defer_init(&dfops, &first_block);
1603 error = xfs_bunmapi(tp, ip,
1604 first_unmap_block, unmap_len,
1605 xfs_bmapi_aflag(whichfork),
1606 XFS_ITRUNC_MAX_EXTENTS,
1607 &first_block, &dfops,
1610 goto out_bmap_cancel;
1613 * Duplicate the transaction that has the permanent
1614 * reservation and commit the old transaction.
1616 error = xfs_defer_finish(&tp, &dfops, ip);
1618 goto out_bmap_cancel;
1620 error = xfs_trans_roll(&tp, ip);
1625 /* Remove all pending CoW reservations. */
1626 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1632 * Clear the reflink flag if we truncated everything.
1634 if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip)) {
1635 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1636 xfs_inode_clear_cowblocks_tag(ip);
1640 * Always re-log the inode so that our permanent transaction can keep
1641 * on rolling it forward in the log.
1643 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1645 trace_xfs_itruncate_extents_end(ip, new_size);
1652 * If the bunmapi call encounters an error, return to the caller where
1653 * the transaction can be properly aborted. We just need to make sure
1654 * we're not holding any resources that we were not when we came in.
1656 xfs_defer_cancel(&dfops);
1664 xfs_mount_t *mp = ip->i_mount;
1667 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1670 /* If this is a read-only mount, don't do this (would generate I/O) */
1671 if (mp->m_flags & XFS_MOUNT_RDONLY)
1674 if (!XFS_FORCED_SHUTDOWN(mp)) {
1678 * If we previously truncated this file and removed old data
1679 * in the process, we want to initiate "early" writeout on
1680 * the last close. This is an attempt to combat the notorious
1681 * NULL files problem which is particularly noticeable from a
1682 * truncate down, buffered (re-)write (delalloc), followed by
1683 * a crash. What we are effectively doing here is
1684 * significantly reducing the time window where we'd otherwise
1685 * be exposed to that problem.
1687 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1689 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1690 if (ip->i_delayed_blks > 0) {
1691 error = filemap_flush(VFS_I(ip)->i_mapping);
1698 if (VFS_I(ip)->i_nlink == 0)
1701 if (xfs_can_free_eofblocks(ip, false)) {
1704 * Check if the inode is being opened, written and closed
1705 * frequently and we have delayed allocation blocks outstanding
1706 * (e.g. streaming writes from the NFS server), truncating the
1707 * blocks past EOF will cause fragmentation to occur.
1709 * In this case don't do the truncation, but we have to be
1710 * careful how we detect this case. Blocks beyond EOF show up as
1711 * i_delayed_blks even when the inode is clean, so we need to
1712 * truncate them away first before checking for a dirty release.
1713 * Hence on the first dirty close we will still remove the
1714 * speculative allocation, but after that we will leave it in
1717 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1720 * If we can't get the iolock just skip truncating the blocks
1721 * past EOF because we could deadlock with the mmap_sem
1722 * otherwise. We'll get another chance to drop them once the
1723 * last reference to the inode is dropped, so we'll never leak
1724 * blocks permanently.
1726 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1727 error = xfs_free_eofblocks(ip);
1728 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1733 /* delalloc blocks after truncation means it really is dirty */
1734 if (ip->i_delayed_blks)
1735 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1741 * xfs_inactive_truncate
1743 * Called to perform a truncate when an inode becomes unlinked.
1746 xfs_inactive_truncate(
1747 struct xfs_inode *ip)
1749 struct xfs_mount *mp = ip->i_mount;
1750 struct xfs_trans *tp;
1753 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1755 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1759 xfs_ilock(ip, XFS_ILOCK_EXCL);
1760 xfs_trans_ijoin(tp, ip, 0);
1763 * Log the inode size first to prevent stale data exposure in the event
1764 * of a system crash before the truncate completes. See the related
1765 * comment in xfs_vn_setattr_size() for details.
1767 ip->i_d.di_size = 0;
1768 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1770 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1772 goto error_trans_cancel;
1774 ASSERT(ip->i_d.di_nextents == 0);
1776 error = xfs_trans_commit(tp);
1780 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1784 xfs_trans_cancel(tp);
1786 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1791 * xfs_inactive_ifree()
1793 * Perform the inode free when an inode is unlinked.
1797 struct xfs_inode *ip)
1799 struct xfs_defer_ops dfops;
1800 xfs_fsblock_t first_block;
1801 struct xfs_mount *mp = ip->i_mount;
1802 struct xfs_trans *tp;
1806 * We try to use a per-AG reservation for any block needed by the finobt
1807 * tree, but as the finobt feature predates the per-AG reservation
1808 * support a degraded file system might not have enough space for the
1809 * reservation at mount time. In that case try to dip into the reserved
1812 * Send a warning if the reservation does happen to fail, as the inode
1813 * now remains allocated and sits on the unlinked list until the fs is
1816 if (unlikely(mp->m_inotbt_nores)) {
1817 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1818 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1821 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1824 if (error == -ENOSPC) {
1825 xfs_warn_ratelimited(mp,
1826 "Failed to remove inode(s) from unlinked list. "
1827 "Please free space, unmount and run xfs_repair.");
1829 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1834 xfs_ilock(ip, XFS_ILOCK_EXCL);
1835 xfs_trans_ijoin(tp, ip, 0);
1837 xfs_defer_init(&dfops, &first_block);
1838 error = xfs_ifree(tp, ip, &dfops);
1841 * If we fail to free the inode, shut down. The cancel
1842 * might do that, we need to make sure. Otherwise the
1843 * inode might be lost for a long time or forever.
1845 if (!XFS_FORCED_SHUTDOWN(mp)) {
1846 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1848 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1850 xfs_trans_cancel(tp);
1851 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1856 * Credit the quota account(s). The inode is gone.
1858 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1861 * Just ignore errors at this point. There is nothing we can do except
1862 * to try to keep going. Make sure it's not a silent error.
1864 error = xfs_defer_finish(&tp, &dfops, NULL);
1866 xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1868 xfs_defer_cancel(&dfops);
1870 error = xfs_trans_commit(tp);
1872 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1875 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1882 * This is called when the vnode reference count for the vnode
1883 * goes to zero. If the file has been unlinked, then it must
1884 * now be truncated. Also, we clear all of the read-ahead state
1885 * kept for the inode here since the file is now closed.
1891 struct xfs_mount *mp;
1896 * If the inode is already free, then there can be nothing
1899 if (VFS_I(ip)->i_mode == 0) {
1900 ASSERT(ip->i_df.if_real_bytes == 0);
1901 ASSERT(ip->i_df.if_broot_bytes == 0);
1906 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1908 /* If this is a read-only mount, don't do this (would generate I/O) */
1909 if (mp->m_flags & XFS_MOUNT_RDONLY)
1912 if (VFS_I(ip)->i_nlink != 0) {
1914 * force is true because we are evicting an inode from the
1915 * cache. Post-eof blocks must be freed, lest we end up with
1916 * broken free space accounting.
1918 if (xfs_can_free_eofblocks(ip, true)) {
1919 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1920 xfs_free_eofblocks(ip);
1921 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1927 if (S_ISREG(VFS_I(ip)->i_mode) &&
1928 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1929 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1932 error = xfs_qm_dqattach(ip, 0);
1936 if (S_ISLNK(VFS_I(ip)->i_mode))
1937 error = xfs_inactive_symlink(ip);
1939 error = xfs_inactive_truncate(ip);
1944 * If there are attributes associated with the file then blow them away
1945 * now. The code calls a routine that recursively deconstructs the
1946 * attribute fork. If also blows away the in-core attribute fork.
1948 if (XFS_IFORK_Q(ip)) {
1949 error = xfs_attr_inactive(ip);
1955 ASSERT(ip->i_d.di_anextents == 0);
1956 ASSERT(ip->i_d.di_forkoff == 0);
1961 error = xfs_inactive_ifree(ip);
1966 * Release the dquots held by inode, if any.
1968 xfs_qm_dqdetach(ip);
1972 * This is called when the inode's link count goes to 0 or we are creating a
1973 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1974 * set to true as the link count is dropped to zero by the VFS after we've
1975 * created the file successfully, so we have to add it to the unlinked list
1976 * while the link count is non-zero.
1978 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1979 * list when the inode is freed.
1983 struct xfs_trans *tp,
1984 struct xfs_inode *ip)
1986 xfs_mount_t *mp = tp->t_mountp;
1996 ASSERT(VFS_I(ip)->i_mode != 0);
1999 * Get the agi buffer first. It ensures lock ordering
2002 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
2005 agi = XFS_BUF_TO_AGI(agibp);
2008 * Get the index into the agi hash table for the
2009 * list this inode will go on.
2011 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2013 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2014 ASSERT(agi->agi_unlinked[bucket_index]);
2015 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2017 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2019 * There is already another inode in the bucket we need
2020 * to add ourselves to. Add us at the front of the list.
2021 * Here we put the head pointer into our next pointer,
2022 * and then we fall through to point the head at us.
2024 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2029 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2030 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2031 offset = ip->i_imap.im_boffset +
2032 offsetof(xfs_dinode_t, di_next_unlinked);
2034 /* need to recalc the inode CRC if appropriate */
2035 xfs_dinode_calc_crc(mp, dip);
2037 xfs_trans_inode_buf(tp, ibp);
2038 xfs_trans_log_buf(tp, ibp, offset,
2039 (offset + sizeof(xfs_agino_t) - 1));
2040 xfs_inobp_check(mp, ibp);
2044 * Point the bucket head pointer at the inode being inserted.
2047 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2048 offset = offsetof(xfs_agi_t, agi_unlinked) +
2049 (sizeof(xfs_agino_t) * bucket_index);
2050 xfs_trans_log_buf(tp, agibp, offset,
2051 (offset + sizeof(xfs_agino_t) - 1));
2056 * Pull the on-disk inode from the AGI unlinked list.
2069 xfs_agnumber_t agno;
2071 xfs_agino_t next_agino;
2072 xfs_buf_t *last_ibp;
2073 xfs_dinode_t *last_dip = NULL;
2075 int offset, last_offset = 0;
2079 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2082 * Get the agi buffer first. It ensures lock ordering
2085 error = xfs_read_agi(mp, tp, agno, &agibp);
2089 agi = XFS_BUF_TO_AGI(agibp);
2092 * Get the index into the agi hash table for the
2093 * list this inode will go on.
2095 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2097 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2098 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2099 ASSERT(agi->agi_unlinked[bucket_index]);
2101 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2103 * We're at the head of the list. Get the inode's on-disk
2104 * buffer to see if there is anyone after us on the list.
2105 * Only modify our next pointer if it is not already NULLAGINO.
2106 * This saves us the overhead of dealing with the buffer when
2107 * there is no need to change it.
2109 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2112 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2116 next_agino = be32_to_cpu(dip->di_next_unlinked);
2117 ASSERT(next_agino != 0);
2118 if (next_agino != NULLAGINO) {
2119 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2120 offset = ip->i_imap.im_boffset +
2121 offsetof(xfs_dinode_t, di_next_unlinked);
2123 /* need to recalc the inode CRC if appropriate */
2124 xfs_dinode_calc_crc(mp, dip);
2126 xfs_trans_inode_buf(tp, ibp);
2127 xfs_trans_log_buf(tp, ibp, offset,
2128 (offset + sizeof(xfs_agino_t) - 1));
2129 xfs_inobp_check(mp, ibp);
2131 xfs_trans_brelse(tp, ibp);
2134 * Point the bucket head pointer at the next inode.
2136 ASSERT(next_agino != 0);
2137 ASSERT(next_agino != agino);
2138 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2139 offset = offsetof(xfs_agi_t, agi_unlinked) +
2140 (sizeof(xfs_agino_t) * bucket_index);
2141 xfs_trans_log_buf(tp, agibp, offset,
2142 (offset + sizeof(xfs_agino_t) - 1));
2145 * We need to search the list for the inode being freed.
2147 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2149 while (next_agino != agino) {
2150 struct xfs_imap imap;
2153 xfs_trans_brelse(tp, last_ibp);
2156 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2158 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2161 "%s: xfs_imap returned error %d.",
2166 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2170 "%s: xfs_imap_to_bp returned error %d.",
2175 last_offset = imap.im_boffset;
2176 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2177 ASSERT(next_agino != NULLAGINO);
2178 ASSERT(next_agino != 0);
2182 * Now last_ibp points to the buffer previous to us on the
2183 * unlinked list. Pull us from the list.
2185 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2188 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2192 next_agino = be32_to_cpu(dip->di_next_unlinked);
2193 ASSERT(next_agino != 0);
2194 ASSERT(next_agino != agino);
2195 if (next_agino != NULLAGINO) {
2196 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2197 offset = ip->i_imap.im_boffset +
2198 offsetof(xfs_dinode_t, di_next_unlinked);
2200 /* need to recalc the inode CRC if appropriate */
2201 xfs_dinode_calc_crc(mp, dip);
2203 xfs_trans_inode_buf(tp, ibp);
2204 xfs_trans_log_buf(tp, ibp, offset,
2205 (offset + sizeof(xfs_agino_t) - 1));
2206 xfs_inobp_check(mp, ibp);
2208 xfs_trans_brelse(tp, ibp);
2211 * Point the previous inode on the list to the next inode.
2213 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2214 ASSERT(next_agino != 0);
2215 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2217 /* need to recalc the inode CRC if appropriate */
2218 xfs_dinode_calc_crc(mp, last_dip);
2220 xfs_trans_inode_buf(tp, last_ibp);
2221 xfs_trans_log_buf(tp, last_ibp, offset,
2222 (offset + sizeof(xfs_agino_t) - 1));
2223 xfs_inobp_check(mp, last_ibp);
2229 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2230 * inodes that are in memory - they all must be marked stale and attached to
2231 * the cluster buffer.
2235 xfs_inode_t *free_ip,
2237 struct xfs_icluster *xic)
2239 xfs_mount_t *mp = free_ip->i_mount;
2240 int blks_per_cluster;
2241 int inodes_per_cluster;
2248 xfs_inode_log_item_t *iip;
2249 xfs_log_item_t *lip;
2250 struct xfs_perag *pag;
2253 inum = xic->first_ino;
2254 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2255 blks_per_cluster = xfs_icluster_size_fsb(mp);
2256 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2257 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2259 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2261 * The allocation bitmap tells us which inodes of the chunk were
2262 * physically allocated. Skip the cluster if an inode falls into
2265 ioffset = inum - xic->first_ino;
2266 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2267 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2271 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2272 XFS_INO_TO_AGBNO(mp, inum));
2275 * We obtain and lock the backing buffer first in the process
2276 * here, as we have to ensure that any dirty inode that we
2277 * can't get the flush lock on is attached to the buffer.
2278 * If we scan the in-memory inodes first, then buffer IO can
2279 * complete before we get a lock on it, and hence we may fail
2280 * to mark all the active inodes on the buffer stale.
2282 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2283 mp->m_bsize * blks_per_cluster,
2290 * This buffer may not have been correctly initialised as we
2291 * didn't read it from disk. That's not important because we are
2292 * only using to mark the buffer as stale in the log, and to
2293 * attach stale cached inodes on it. That means it will never be
2294 * dispatched for IO. If it is, we want to know about it, and we
2295 * want it to fail. We can acheive this by adding a write
2296 * verifier to the buffer.
2298 bp->b_ops = &xfs_inode_buf_ops;
2301 * Walk the inodes already attached to the buffer and mark them
2302 * stale. These will all have the flush locks held, so an
2303 * in-memory inode walk can't lock them. By marking them all
2304 * stale first, we will not attempt to lock them in the loop
2305 * below as the XFS_ISTALE flag will be set.
2309 if (lip->li_type == XFS_LI_INODE) {
2310 iip = (xfs_inode_log_item_t *)lip;
2311 ASSERT(iip->ili_logged == 1);
2312 lip->li_cb = xfs_istale_done;
2313 xfs_trans_ail_copy_lsn(mp->m_ail,
2314 &iip->ili_flush_lsn,
2315 &iip->ili_item.li_lsn);
2316 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2318 lip = lip->li_bio_list;
2323 * For each inode in memory attempt to add it to the inode
2324 * buffer and set it up for being staled on buffer IO
2325 * completion. This is safe as we've locked out tail pushing
2326 * and flushing by locking the buffer.
2328 * We have already marked every inode that was part of a
2329 * transaction stale above, which means there is no point in
2330 * even trying to lock them.
2332 for (i = 0; i < inodes_per_cluster; i++) {
2335 ip = radix_tree_lookup(&pag->pag_ici_root,
2336 XFS_INO_TO_AGINO(mp, (inum + i)));
2338 /* Inode not in memory, nothing to do */
2345 * because this is an RCU protected lookup, we could
2346 * find a recently freed or even reallocated inode
2347 * during the lookup. We need to check under the
2348 * i_flags_lock for a valid inode here. Skip it if it
2349 * is not valid, the wrong inode or stale.
2351 spin_lock(&ip->i_flags_lock);
2352 if (ip->i_ino != inum + i ||
2353 __xfs_iflags_test(ip, XFS_ISTALE)) {
2354 spin_unlock(&ip->i_flags_lock);
2358 spin_unlock(&ip->i_flags_lock);
2361 * Don't try to lock/unlock the current inode, but we
2362 * _cannot_ skip the other inodes that we did not find
2363 * in the list attached to the buffer and are not
2364 * already marked stale. If we can't lock it, back off
2367 if (ip != free_ip &&
2368 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2376 xfs_iflags_set(ip, XFS_ISTALE);
2379 * we don't need to attach clean inodes or those only
2380 * with unlogged changes (which we throw away, anyway).
2383 if (!iip || xfs_inode_clean(ip)) {
2384 ASSERT(ip != free_ip);
2386 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2390 iip->ili_last_fields = iip->ili_fields;
2391 iip->ili_fields = 0;
2392 iip->ili_fsync_fields = 0;
2393 iip->ili_logged = 1;
2394 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2395 &iip->ili_item.li_lsn);
2397 xfs_buf_attach_iodone(bp, xfs_istale_done,
2401 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2404 xfs_trans_stale_inode_buf(tp, bp);
2405 xfs_trans_binval(tp, bp);
2413 * This is called to return an inode to the inode free list.
2414 * The inode should already be truncated to 0 length and have
2415 * no pages associated with it. This routine also assumes that
2416 * the inode is already a part of the transaction.
2418 * The on-disk copy of the inode will have been added to the list
2419 * of unlinked inodes in the AGI. We need to remove the inode from
2420 * that list atomically with respect to freeing it here.
2426 struct xfs_defer_ops *dfops)
2429 struct xfs_icluster xic = { 0 };
2431 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2432 ASSERT(VFS_I(ip)->i_nlink == 0);
2433 ASSERT(ip->i_d.di_nextents == 0);
2434 ASSERT(ip->i_d.di_anextents == 0);
2435 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2436 ASSERT(ip->i_d.di_nblocks == 0);
2439 * Pull the on-disk inode from the AGI unlinked list.
2441 error = xfs_iunlink_remove(tp, ip);
2445 error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2449 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2450 ip->i_d.di_flags = 0;
2451 ip->i_d.di_dmevmask = 0;
2452 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2453 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2454 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2456 * Bump the generation count so no one will be confused
2457 * by reincarnations of this inode.
2459 VFS_I(ip)->i_generation++;
2460 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2463 error = xfs_ifree_cluster(ip, tp, &xic);
2469 * This is called to unpin an inode. The caller must have the inode locked
2470 * in at least shared mode so that the buffer cannot be subsequently pinned
2471 * once someone is waiting for it to be unpinned.
2475 struct xfs_inode *ip)
2477 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2479 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2481 /* Give the log a push to start the unpinning I/O */
2482 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2488 struct xfs_inode *ip)
2490 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2491 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2496 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2497 if (xfs_ipincount(ip))
2499 } while (xfs_ipincount(ip));
2500 finish_wait(wq, &wait.wait);
2505 struct xfs_inode *ip)
2507 if (xfs_ipincount(ip))
2508 __xfs_iunpin_wait(ip);
2512 * Removing an inode from the namespace involves removing the directory entry
2513 * and dropping the link count on the inode. Removing the directory entry can
2514 * result in locking an AGF (directory blocks were freed) and removing a link
2515 * count can result in placing the inode on an unlinked list which results in
2518 * The big problem here is that we have an ordering constraint on AGF and AGI
2519 * locking - inode allocation locks the AGI, then can allocate a new extent for
2520 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2521 * removes the inode from the unlinked list, requiring that we lock the AGI
2522 * first, and then freeing the inode can result in an inode chunk being freed
2523 * and hence freeing disk space requiring that we lock an AGF.
2525 * Hence the ordering that is imposed by other parts of the code is AGI before
2526 * AGF. This means we cannot remove the directory entry before we drop the inode
2527 * reference count and put it on the unlinked list as this results in a lock
2528 * order of AGF then AGI, and this can deadlock against inode allocation and
2529 * freeing. Therefore we must drop the link counts before we remove the
2532 * This is still safe from a transactional point of view - it is not until we
2533 * get to xfs_defer_finish() that we have the possibility of multiple
2534 * transactions in this operation. Hence as long as we remove the directory
2535 * entry and drop the link count in the first transaction of the remove
2536 * operation, there are no transactional constraints on the ordering here.
2541 struct xfs_name *name,
2544 xfs_mount_t *mp = dp->i_mount;
2545 xfs_trans_t *tp = NULL;
2546 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2548 struct xfs_defer_ops dfops;
2549 xfs_fsblock_t first_block;
2552 trace_xfs_remove(dp, name);
2554 if (XFS_FORCED_SHUTDOWN(mp))
2557 error = xfs_qm_dqattach(dp, 0);
2561 error = xfs_qm_dqattach(ip, 0);
2566 * We try to get the real space reservation first,
2567 * allowing for directory btree deletion(s) implying
2568 * possible bmap insert(s). If we can't get the space
2569 * reservation then we use 0 instead, and avoid the bmap
2570 * btree insert(s) in the directory code by, if the bmap
2571 * insert tries to happen, instead trimming the LAST
2572 * block from the directory.
2574 resblks = XFS_REMOVE_SPACE_RES(mp);
2575 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2576 if (error == -ENOSPC) {
2578 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2582 ASSERT(error != -ENOSPC);
2586 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2587 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2589 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2590 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2593 * If we're removing a directory perform some additional validation.
2596 ASSERT(VFS_I(ip)->i_nlink >= 2);
2597 if (VFS_I(ip)->i_nlink != 2) {
2599 goto out_trans_cancel;
2601 if (!xfs_dir_isempty(ip)) {
2603 goto out_trans_cancel;
2606 /* Drop the link from ip's "..". */
2607 error = xfs_droplink(tp, dp);
2609 goto out_trans_cancel;
2611 /* Drop the "." link from ip to self. */
2612 error = xfs_droplink(tp, ip);
2614 goto out_trans_cancel;
2617 * When removing a non-directory we need to log the parent
2618 * inode here. For a directory this is done implicitly
2619 * by the xfs_droplink call for the ".." entry.
2621 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2623 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2625 /* Drop the link from dp to ip. */
2626 error = xfs_droplink(tp, ip);
2628 goto out_trans_cancel;
2630 xfs_defer_init(&dfops, &first_block);
2631 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2632 &first_block, &dfops, resblks);
2634 ASSERT(error != -ENOENT);
2635 goto out_bmap_cancel;
2639 * If this is a synchronous mount, make sure that the
2640 * remove transaction goes to disk before returning to
2643 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2644 xfs_trans_set_sync(tp);
2646 error = xfs_defer_finish(&tp, &dfops, NULL);
2648 goto out_bmap_cancel;
2650 error = xfs_trans_commit(tp);
2654 if (is_dir && xfs_inode_is_filestream(ip))
2655 xfs_filestream_deassociate(ip);
2660 xfs_defer_cancel(&dfops);
2662 xfs_trans_cancel(tp);
2668 * Enter all inodes for a rename transaction into a sorted array.
2670 #define __XFS_SORT_INODES 5
2672 xfs_sort_for_rename(
2673 struct xfs_inode *dp1, /* in: old (source) directory inode */
2674 struct xfs_inode *dp2, /* in: new (target) directory inode */
2675 struct xfs_inode *ip1, /* in: inode of old entry */
2676 struct xfs_inode *ip2, /* in: inode of new entry */
2677 struct xfs_inode *wip, /* in: whiteout inode */
2678 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2679 int *num_inodes) /* in/out: inodes in array */
2683 ASSERT(*num_inodes == __XFS_SORT_INODES);
2684 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2687 * i_tab contains a list of pointers to inodes. We initialize
2688 * the table here & we'll sort it. We will then use it to
2689 * order the acquisition of the inode locks.
2691 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2704 * Sort the elements via bubble sort. (Remember, there are at
2705 * most 5 elements to sort, so this is adequate.)
2707 for (i = 0; i < *num_inodes; i++) {
2708 for (j = 1; j < *num_inodes; j++) {
2709 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2710 struct xfs_inode *temp = i_tab[j];
2711 i_tab[j] = i_tab[j-1];
2720 struct xfs_trans *tp,
2721 struct xfs_defer_ops *dfops)
2726 * If this is a synchronous mount, make sure that the rename transaction
2727 * goes to disk before returning to the user.
2729 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2730 xfs_trans_set_sync(tp);
2732 error = xfs_defer_finish(&tp, dfops, NULL);
2734 xfs_defer_cancel(dfops);
2735 xfs_trans_cancel(tp);
2739 return xfs_trans_commit(tp);
2743 * xfs_cross_rename()
2745 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2749 struct xfs_trans *tp,
2750 struct xfs_inode *dp1,
2751 struct xfs_name *name1,
2752 struct xfs_inode *ip1,
2753 struct xfs_inode *dp2,
2754 struct xfs_name *name2,
2755 struct xfs_inode *ip2,
2756 struct xfs_defer_ops *dfops,
2757 xfs_fsblock_t *first_block,
2765 /* Swap inode number for dirent in first parent */
2766 error = xfs_dir_replace(tp, dp1, name1,
2768 first_block, dfops, spaceres);
2770 goto out_trans_abort;
2772 /* Swap inode number for dirent in second parent */
2773 error = xfs_dir_replace(tp, dp2, name2,
2775 first_block, dfops, spaceres);
2777 goto out_trans_abort;
2780 * If we're renaming one or more directories across different parents,
2781 * update the respective ".." entries (and link counts) to match the new
2785 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2787 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2788 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2789 dp1->i_ino, first_block,
2792 goto out_trans_abort;
2794 /* transfer ip2 ".." reference to dp1 */
2795 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2796 error = xfs_droplink(tp, dp2);
2798 goto out_trans_abort;
2799 error = xfs_bumplink(tp, dp1);
2801 goto out_trans_abort;
2805 * Although ip1 isn't changed here, userspace needs
2806 * to be warned about the change, so that applications
2807 * relying on it (like backup ones), will properly
2810 ip1_flags |= XFS_ICHGTIME_CHG;
2811 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2814 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2815 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2816 dp2->i_ino, first_block,
2819 goto out_trans_abort;
2821 /* transfer ip1 ".." reference to dp2 */
2822 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2823 error = xfs_droplink(tp, dp1);
2825 goto out_trans_abort;
2826 error = xfs_bumplink(tp, dp2);
2828 goto out_trans_abort;
2832 * Although ip2 isn't changed here, userspace needs
2833 * to be warned about the change, so that applications
2834 * relying on it (like backup ones), will properly
2837 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2838 ip2_flags |= XFS_ICHGTIME_CHG;
2843 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2844 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2847 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2848 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2851 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2852 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2854 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2855 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2856 return xfs_finish_rename(tp, dfops);
2859 xfs_defer_cancel(dfops);
2860 xfs_trans_cancel(tp);
2865 * xfs_rename_alloc_whiteout()
2867 * Return a referenced, unlinked, unlocked inode that that can be used as a
2868 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2869 * crash between allocating the inode and linking it into the rename transaction
2870 * recovery will free the inode and we won't leak it.
2873 xfs_rename_alloc_whiteout(
2874 struct xfs_inode *dp,
2875 struct xfs_inode **wip)
2877 struct xfs_inode *tmpfile;
2880 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2885 * Prepare the tmpfile inode as if it were created through the VFS.
2886 * Otherwise, the link increment paths will complain about nlink 0->1.
2887 * Drop the link count as done by d_tmpfile(), complete the inode setup
2888 * and flag it as linkable.
2890 drop_nlink(VFS_I(tmpfile));
2891 xfs_setup_iops(tmpfile);
2892 xfs_finish_inode_setup(tmpfile);
2893 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2904 struct xfs_inode *src_dp,
2905 struct xfs_name *src_name,
2906 struct xfs_inode *src_ip,
2907 struct xfs_inode *target_dp,
2908 struct xfs_name *target_name,
2909 struct xfs_inode *target_ip,
2912 struct xfs_mount *mp = src_dp->i_mount;
2913 struct xfs_trans *tp;
2914 struct xfs_defer_ops dfops;
2915 xfs_fsblock_t first_block;
2916 struct xfs_inode *wip = NULL; /* whiteout inode */
2917 struct xfs_inode *inodes[__XFS_SORT_INODES];
2918 int num_inodes = __XFS_SORT_INODES;
2919 bool new_parent = (src_dp != target_dp);
2920 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2924 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2926 if ((flags & RENAME_EXCHANGE) && !target_ip)
2930 * If we are doing a whiteout operation, allocate the whiteout inode
2931 * we will be placing at the target and ensure the type is set
2934 if (flags & RENAME_WHITEOUT) {
2935 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2936 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2940 /* setup target dirent info as whiteout */
2941 src_name->type = XFS_DIR3_FT_CHRDEV;
2944 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2945 inodes, &num_inodes);
2947 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2948 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2949 if (error == -ENOSPC) {
2951 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2955 goto out_release_wip;
2958 * Attach the dquots to the inodes
2960 error = xfs_qm_vop_rename_dqattach(inodes);
2962 goto out_trans_cancel;
2965 * Lock all the participating inodes. Depending upon whether
2966 * the target_name exists in the target directory, and
2967 * whether the target directory is the same as the source
2968 * directory, we can lock from 2 to 4 inodes.
2971 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2973 xfs_lock_two_inodes(src_dp, target_dp,
2974 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2976 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2979 * Join all the inodes to the transaction. From this point on,
2980 * we can rely on either trans_commit or trans_cancel to unlock
2983 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2985 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2986 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2988 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2990 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2993 * If we are using project inheritance, we only allow renames
2994 * into our tree when the project IDs are the same; else the
2995 * tree quota mechanism would be circumvented.
2997 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2998 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3000 goto out_trans_cancel;
3003 xfs_defer_init(&dfops, &first_block);
3005 /* RENAME_EXCHANGE is unique from here on. */
3006 if (flags & RENAME_EXCHANGE)
3007 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3008 target_dp, target_name, target_ip,
3009 &dfops, &first_block, spaceres);
3012 * Set up the target.
3014 if (target_ip == NULL) {
3016 * If there's no space reservation, check the entry will
3017 * fit before actually inserting it.
3020 error = xfs_dir_canenter(tp, target_dp, target_name);
3022 goto out_trans_cancel;
3025 * If target does not exist and the rename crosses
3026 * directories, adjust the target directory link count
3027 * to account for the ".." reference from the new entry.
3029 error = xfs_dir_createname(tp, target_dp, target_name,
3030 src_ip->i_ino, &first_block,
3033 goto out_bmap_cancel;
3035 xfs_trans_ichgtime(tp, target_dp,
3036 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3038 if (new_parent && src_is_directory) {
3039 error = xfs_bumplink(tp, target_dp);
3041 goto out_bmap_cancel;
3043 } else { /* target_ip != NULL */
3045 * If target exists and it's a directory, check that both
3046 * target and source are directories and that target can be
3047 * destroyed, or that neither is a directory.
3049 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3051 * Make sure target dir is empty.
3053 if (!(xfs_dir_isempty(target_ip)) ||
3054 (VFS_I(target_ip)->i_nlink > 2)) {
3056 goto out_trans_cancel;
3061 * Link the source inode under the target name.
3062 * If the source inode is a directory and we are moving
3063 * it across directories, its ".." entry will be
3064 * inconsistent until we replace that down below.
3066 * In case there is already an entry with the same
3067 * name at the destination directory, remove it first.
3069 error = xfs_dir_replace(tp, target_dp, target_name,
3071 &first_block, &dfops, spaceres);
3073 goto out_bmap_cancel;
3075 xfs_trans_ichgtime(tp, target_dp,
3076 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3079 * Decrement the link count on the target since the target
3080 * dir no longer points to it.
3082 error = xfs_droplink(tp, target_ip);
3084 goto out_bmap_cancel;
3086 if (src_is_directory) {
3088 * Drop the link from the old "." entry.
3090 error = xfs_droplink(tp, target_ip);
3092 goto out_bmap_cancel;
3094 } /* target_ip != NULL */
3097 * Remove the source.
3099 if (new_parent && src_is_directory) {
3101 * Rewrite the ".." entry to point to the new
3104 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3106 &first_block, &dfops, spaceres);
3107 ASSERT(error != -EEXIST);
3109 goto out_bmap_cancel;
3113 * We always want to hit the ctime on the source inode.
3115 * This isn't strictly required by the standards since the source
3116 * inode isn't really being changed, but old unix file systems did
3117 * it and some incremental backup programs won't work without it.
3119 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3120 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3123 * Adjust the link count on src_dp. This is necessary when
3124 * renaming a directory, either within one parent when
3125 * the target existed, or across two parent directories.
3127 if (src_is_directory && (new_parent || target_ip != NULL)) {
3130 * Decrement link count on src_directory since the
3131 * entry that's moved no longer points to it.
3133 error = xfs_droplink(tp, src_dp);
3135 goto out_bmap_cancel;
3139 * For whiteouts, we only need to update the source dirent with the
3140 * inode number of the whiteout inode rather than removing it
3144 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3145 &first_block, &dfops, spaceres);
3147 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3148 &first_block, &dfops, spaceres);
3150 goto out_bmap_cancel;
3153 * For whiteouts, we need to bump the link count on the whiteout inode.
3154 * This means that failures all the way up to this point leave the inode
3155 * on the unlinked list and so cleanup is a simple matter of dropping
3156 * the remaining reference to it. If we fail here after bumping the link
3157 * count, we're shutting down the filesystem so we'll never see the
3158 * intermediate state on disk.
3161 ASSERT(VFS_I(wip)->i_nlink == 0);
3162 error = xfs_bumplink(tp, wip);
3164 goto out_bmap_cancel;
3165 error = xfs_iunlink_remove(tp, wip);
3167 goto out_bmap_cancel;
3168 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3171 * Now we have a real link, clear the "I'm a tmpfile" state
3172 * flag from the inode so it doesn't accidentally get misused in
3175 VFS_I(wip)->i_state &= ~I_LINKABLE;
3178 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3179 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3181 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3183 error = xfs_finish_rename(tp, &dfops);
3189 xfs_defer_cancel(&dfops);
3191 xfs_trans_cancel(tp);
3200 struct xfs_inode *ip,
3203 struct xfs_mount *mp = ip->i_mount;
3204 struct xfs_perag *pag;
3205 unsigned long first_index, mask;
3206 unsigned long inodes_per_cluster;
3208 struct xfs_inode **cilist;
3209 struct xfs_inode *cip;
3215 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3217 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3218 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3219 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3223 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3224 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3226 /* really need a gang lookup range call here */
3227 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3228 first_index, inodes_per_cluster);
3232 for (i = 0; i < nr_found; i++) {
3238 * because this is an RCU protected lookup, we could find a
3239 * recently freed or even reallocated inode during the lookup.
3240 * We need to check under the i_flags_lock for a valid inode
3241 * here. Skip it if it is not valid or the wrong inode.
3243 spin_lock(&cip->i_flags_lock);
3245 __xfs_iflags_test(cip, XFS_ISTALE)) {
3246 spin_unlock(&cip->i_flags_lock);
3251 * Once we fall off the end of the cluster, no point checking
3252 * any more inodes in the list because they will also all be
3253 * outside the cluster.
3255 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3256 spin_unlock(&cip->i_flags_lock);
3259 spin_unlock(&cip->i_flags_lock);
3262 * Do an un-protected check to see if the inode is dirty and
3263 * is a candidate for flushing. These checks will be repeated
3264 * later after the appropriate locks are acquired.
3266 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3270 * Try to get locks. If any are unavailable or it is pinned,
3271 * then this inode cannot be flushed and is skipped.
3274 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3276 if (!xfs_iflock_nowait(cip)) {
3277 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3280 if (xfs_ipincount(cip)) {
3282 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3288 * Check the inode number again, just to be certain we are not
3289 * racing with freeing in xfs_reclaim_inode(). See the comments
3290 * in that function for more information as to why the initial
3291 * check is not sufficient.
3295 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3300 * arriving here means that this inode can be flushed. First
3301 * re-check that it's dirty before flushing.
3303 if (!xfs_inode_clean(cip)) {
3305 error = xfs_iflush_int(cip, bp);
3307 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3308 goto cluster_corrupt_out;
3314 xfs_iunlock(cip, XFS_ILOCK_SHARED);
3318 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3319 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3330 cluster_corrupt_out:
3332 * Corruption detected in the clustering loop. Invalidate the
3333 * inode buffer and shut down the filesystem.
3337 * Clean up the buffer. If it was delwri, just release it --
3338 * brelse can handle it with no problems. If not, shut down the
3339 * filesystem before releasing the buffer.
3341 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3345 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3347 if (!bufwasdelwri) {
3349 * Just like incore_relse: if we have b_iodone functions,
3350 * mark the buffer as an error and call them. Otherwise
3351 * mark it as stale and brelse.
3354 bp->b_flags &= ~XBF_DONE;
3356 xfs_buf_ioerror(bp, -EIO);
3365 * Unlocks the flush lock
3367 xfs_iflush_abort(cip, false);
3370 return -EFSCORRUPTED;
3374 * Flush dirty inode metadata into the backing buffer.
3376 * The caller must have the inode lock and the inode flush lock held. The
3377 * inode lock will still be held upon return to the caller, and the inode
3378 * flush lock will be released after the inode has reached the disk.
3380 * The caller must write out the buffer returned in *bpp and release it.
3384 struct xfs_inode *ip,
3385 struct xfs_buf **bpp)
3387 struct xfs_mount *mp = ip->i_mount;
3388 struct xfs_buf *bp = NULL;
3389 struct xfs_dinode *dip;
3392 XFS_STATS_INC(mp, xs_iflush_count);
3394 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3395 ASSERT(xfs_isiflocked(ip));
3396 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3397 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3401 xfs_iunpin_wait(ip);
3404 * For stale inodes we cannot rely on the backing buffer remaining
3405 * stale in cache for the remaining life of the stale inode and so
3406 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3407 * inodes below. We have to check this after ensuring the inode is
3408 * unpinned so that it is safe to reclaim the stale inode after the
3411 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3417 * This may have been unpinned because the filesystem is shutting
3418 * down forcibly. If that's the case we must not write this inode
3419 * to disk, because the log record didn't make it to disk.
3421 * We also have to remove the log item from the AIL in this case,
3422 * as we wait for an empty AIL as part of the unmount process.
3424 if (XFS_FORCED_SHUTDOWN(mp)) {
3430 * Get the buffer containing the on-disk inode. We are doing a try-lock
3431 * operation here, so we may get an EAGAIN error. In that case, we
3432 * simply want to return with the inode still dirty.
3434 * If we get any other error, we effectively have a corruption situation
3435 * and we cannot flush the inode, so we treat it the same as failing
3438 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3440 if (error == -EAGAIN) {
3448 * First flush out the inode that xfs_iflush was called with.
3450 error = xfs_iflush_int(ip, bp);
3455 * If the buffer is pinned then push on the log now so we won't
3456 * get stuck waiting in the write for too long.
3458 if (xfs_buf_ispinned(bp))
3459 xfs_log_force(mp, 0);
3463 * see if other inodes can be gathered into this write
3465 error = xfs_iflush_cluster(ip, bp);
3467 goto cluster_corrupt_out;
3475 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3476 cluster_corrupt_out:
3477 error = -EFSCORRUPTED;
3480 * Unlocks the flush lock
3482 xfs_iflush_abort(ip, false);
3488 struct xfs_inode *ip,
3491 struct xfs_inode_log_item *iip = ip->i_itemp;
3492 struct xfs_dinode *dip;
3493 struct xfs_mount *mp = ip->i_mount;
3495 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3496 ASSERT(xfs_isiflocked(ip));
3497 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3498 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3499 ASSERT(iip != NULL && iip->ili_fields != 0);
3500 ASSERT(ip->i_d.di_version > 1);
3502 /* set *dip = inode's place in the buffer */
3503 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3505 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3506 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3507 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3508 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3509 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3512 if (S_ISREG(VFS_I(ip)->i_mode)) {
3514 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3515 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3516 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3517 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3518 "%s: Bad regular inode %Lu, ptr 0x%p",
3519 __func__, ip->i_ino, ip);
3522 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3524 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3525 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3526 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3527 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3528 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3529 "%s: Bad directory inode %Lu, ptr 0x%p",
3530 __func__, ip->i_ino, ip);
3534 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3535 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3536 XFS_RANDOM_IFLUSH_5)) {
3537 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3538 "%s: detected corrupt incore inode %Lu, "
3539 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3540 __func__, ip->i_ino,
3541 ip->i_d.di_nextents + ip->i_d.di_anextents,
3542 ip->i_d.di_nblocks, ip);
3545 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3546 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3547 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3548 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3549 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3554 * Inode item log recovery for v2 inodes are dependent on the
3555 * di_flushiter count for correct sequencing. We bump the flush
3556 * iteration count so we can detect flushes which postdate a log record
3557 * during recovery. This is redundant as we now log every change and
3558 * hence this can't happen but we need to still do it to ensure
3559 * backwards compatibility with old kernels that predate logging all
3562 if (ip->i_d.di_version < 3)
3563 ip->i_d.di_flushiter++;
3566 * Copy the dirty parts of the inode into the on-disk inode. We always
3567 * copy out the core of the inode, because if the inode is dirty at all
3570 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3572 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3573 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3574 ip->i_d.di_flushiter = 0;
3576 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3577 if (XFS_IFORK_Q(ip))
3578 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3579 xfs_inobp_check(mp, bp);
3582 * We've recorded everything logged in the inode, so we'd like to clear
3583 * the ili_fields bits so we don't log and flush things unnecessarily.
3584 * However, we can't stop logging all this information until the data
3585 * we've copied into the disk buffer is written to disk. If we did we
3586 * might overwrite the copy of the inode in the log with all the data
3587 * after re-logging only part of it, and in the face of a crash we
3588 * wouldn't have all the data we need to recover.
3590 * What we do is move the bits to the ili_last_fields field. When
3591 * logging the inode, these bits are moved back to the ili_fields field.
3592 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3593 * know that the information those bits represent is permanently on
3594 * disk. As long as the flush completes before the inode is logged
3595 * again, then both ili_fields and ili_last_fields will be cleared.
3597 * We can play with the ili_fields bits here, because the inode lock
3598 * must be held exclusively in order to set bits there and the flush
3599 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3600 * done routine can tell whether or not to look in the AIL. Also, store
3601 * the current LSN of the inode so that we can tell whether the item has
3602 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3603 * need the AIL lock, because it is a 64 bit value that cannot be read
3606 iip->ili_last_fields = iip->ili_fields;
3607 iip->ili_fields = 0;
3608 iip->ili_fsync_fields = 0;
3609 iip->ili_logged = 1;
3611 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3612 &iip->ili_item.li_lsn);
3615 * Attach the function xfs_iflush_done to the inode's
3616 * buffer. This will remove the inode from the AIL
3617 * and unlock the inode's flush lock when the inode is
3618 * completely written to disk.
3620 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3622 /* generate the checksum. */
3623 xfs_dinode_calc_crc(mp, dip);
3625 ASSERT(bp->b_fspriv != NULL);
3626 ASSERT(bp->b_iodone != NULL);
3630 return -EFSCORRUPTED;