xref: /linux/fs/xfs/xfs_inode.c (revision 4cb584e0ee7df70fd0376aee60cf701855ea8c81)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
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.
8  *
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.
13  *
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
17  */
18 #include <linux/log2.h>
19 
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_sb.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"
32 #include "xfs_dir2.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_attr.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"
40 #include "xfs_bmap.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"
50 #include "xfs_log.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
53 
54 kmem_zone_t *xfs_inode_zone;
55 
56 /*
57  * Used in xfs_itruncate_extents().  This is the maximum number of extents
58  * freed from a file in a single transaction.
59  */
60 #define	XFS_ITRUNC_MAX_EXTENTS	2
61 
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 *);
65 
66 /*
67  * helper function to extract extent size hint from inode
68  */
69 xfs_extlen_t
70 xfs_get_extsz_hint(
71 	struct xfs_inode	*ip)
72 {
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;
77 	return 0;
78 }
79 
80 /*
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.
85  */
86 xfs_extlen_t
87 xfs_get_cowextsz_hint(
88 	struct xfs_inode	*ip)
89 {
90 	xfs_extlen_t		a, b;
91 
92 	a = 0;
93 	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
94 		a = ip->i_d.di_cowextsize;
95 	b = xfs_get_extsz_hint(ip);
96 
97 	a = max(a, b);
98 	if (a == 0)
99 		return XFS_DEFAULT_COWEXTSZ_HINT;
100 	return a;
101 }
102 
103 /*
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
113  * if they have not.
114  *
115  * The functions return a value which should be given to the corresponding
116  * xfs_iunlock() call.
117  */
118 uint
119 xfs_ilock_data_map_shared(
120 	struct xfs_inode	*ip)
121 {
122 	uint			lock_mode = XFS_ILOCK_SHARED;
123 
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);
128 	return lock_mode;
129 }
130 
131 uint
132 xfs_ilock_attr_map_shared(
133 	struct xfs_inode	*ip)
134 {
135 	uint			lock_mode = XFS_ILOCK_SHARED;
136 
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);
141 	return lock_mode;
142 }
143 
144 /*
145  * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
146  * multi-reader locks: i_mmap_lock and the i_lock.  This routine allows
147  * various combinations of the locks to be obtained.
148  *
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.
151  *
152  * Basic locking order:
153  *
154  * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
155  *
156  * mmap_sem locking order:
157  *
158  * i_rwsem -> page lock -> mmap_sem
159  * mmap_sem -> i_mmap_lock -> page_lock
160  *
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_rwsem cannot be taken inside a page fault because
166  * page faults already hold the mmap_sem.
167  *
168  * Hence to serialise fully against both syscall and mmap based IO, we need to
169  * take both the i_rwsem 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
172  * functions).
173  */
174 void
175 xfs_ilock(
176 	xfs_inode_t		*ip,
177 	uint			lock_flags)
178 {
179 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
180 
181 	/*
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.
185 	 */
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);
193 
194 	if (lock_flags & XFS_IOLOCK_EXCL) {
195 		down_write_nested(&VFS_I(ip)->i_rwsem,
196 				  XFS_IOLOCK_DEP(lock_flags));
197 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
198 		down_read_nested(&VFS_I(ip)->i_rwsem,
199 				 XFS_IOLOCK_DEP(lock_flags));
200 	}
201 
202 	if (lock_flags & XFS_MMAPLOCK_EXCL)
203 		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
204 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
205 		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
206 
207 	if (lock_flags & XFS_ILOCK_EXCL)
208 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
209 	else if (lock_flags & XFS_ILOCK_SHARED)
210 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
211 }
212 
213 /*
214  * This is just like xfs_ilock(), except that the caller
215  * is guaranteed not to sleep.  It returns 1 if it gets
216  * the requested locks and 0 otherwise.  If the IO lock is
217  * obtained but the inode lock cannot be, then the IO lock
218  * is dropped before returning.
219  *
220  * ip -- the inode being locked
221  * lock_flags -- this parameter indicates the inode's locks to be
222  *       to be locked.  See the comment for xfs_ilock() for a list
223  *	 of valid values.
224  */
225 int
226 xfs_ilock_nowait(
227 	xfs_inode_t		*ip,
228 	uint			lock_flags)
229 {
230 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
231 
232 	/*
233 	 * You can't set both SHARED and EXCL for the same lock,
234 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
235 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
236 	 */
237 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
238 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
239 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
240 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
241 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
242 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
243 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
244 
245 	if (lock_flags & XFS_IOLOCK_EXCL) {
246 		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
247 			goto out;
248 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
249 		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
250 			goto out;
251 	}
252 
253 	if (lock_flags & XFS_MMAPLOCK_EXCL) {
254 		if (!mrtryupdate(&ip->i_mmaplock))
255 			goto out_undo_iolock;
256 	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
257 		if (!mrtryaccess(&ip->i_mmaplock))
258 			goto out_undo_iolock;
259 	}
260 
261 	if (lock_flags & XFS_ILOCK_EXCL) {
262 		if (!mrtryupdate(&ip->i_lock))
263 			goto out_undo_mmaplock;
264 	} else if (lock_flags & XFS_ILOCK_SHARED) {
265 		if (!mrtryaccess(&ip->i_lock))
266 			goto out_undo_mmaplock;
267 	}
268 	return 1;
269 
270 out_undo_mmaplock:
271 	if (lock_flags & XFS_MMAPLOCK_EXCL)
272 		mrunlock_excl(&ip->i_mmaplock);
273 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
274 		mrunlock_shared(&ip->i_mmaplock);
275 out_undo_iolock:
276 	if (lock_flags & XFS_IOLOCK_EXCL)
277 		up_write(&VFS_I(ip)->i_rwsem);
278 	else if (lock_flags & XFS_IOLOCK_SHARED)
279 		up_read(&VFS_I(ip)->i_rwsem);
280 out:
281 	return 0;
282 }
283 
284 /*
285  * xfs_iunlock() is used to drop the inode locks acquired with
286  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
287  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
288  * that we know which locks to drop.
289  *
290  * ip -- the inode being unlocked
291  * lock_flags -- this parameter indicates the inode's locks to be
292  *       to be unlocked.  See the comment for xfs_ilock() for a list
293  *	 of valid values for this parameter.
294  *
295  */
296 void
297 xfs_iunlock(
298 	xfs_inode_t		*ip,
299 	uint			lock_flags)
300 {
301 	/*
302 	 * You can't set both SHARED and EXCL for the same lock,
303 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
304 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
305 	 */
306 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
307 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
308 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
309 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
310 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
311 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
312 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
313 	ASSERT(lock_flags != 0);
314 
315 	if (lock_flags & XFS_IOLOCK_EXCL)
316 		up_write(&VFS_I(ip)->i_rwsem);
317 	else if (lock_flags & XFS_IOLOCK_SHARED)
318 		up_read(&VFS_I(ip)->i_rwsem);
319 
320 	if (lock_flags & XFS_MMAPLOCK_EXCL)
321 		mrunlock_excl(&ip->i_mmaplock);
322 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
323 		mrunlock_shared(&ip->i_mmaplock);
324 
325 	if (lock_flags & XFS_ILOCK_EXCL)
326 		mrunlock_excl(&ip->i_lock);
327 	else if (lock_flags & XFS_ILOCK_SHARED)
328 		mrunlock_shared(&ip->i_lock);
329 
330 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
331 }
332 
333 /*
334  * give up write locks.  the i/o lock cannot be held nested
335  * if it is being demoted.
336  */
337 void
338 xfs_ilock_demote(
339 	xfs_inode_t		*ip,
340 	uint			lock_flags)
341 {
342 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
343 	ASSERT((lock_flags &
344 		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
345 
346 	if (lock_flags & XFS_ILOCK_EXCL)
347 		mrdemote(&ip->i_lock);
348 	if (lock_flags & XFS_MMAPLOCK_EXCL)
349 		mrdemote(&ip->i_mmaplock);
350 	if (lock_flags & XFS_IOLOCK_EXCL)
351 		downgrade_write(&VFS_I(ip)->i_rwsem);
352 
353 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
354 }
355 
356 #if defined(DEBUG) || defined(XFS_WARN)
357 int
358 xfs_isilocked(
359 	xfs_inode_t		*ip,
360 	uint			lock_flags)
361 {
362 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
363 		if (!(lock_flags & XFS_ILOCK_SHARED))
364 			return !!ip->i_lock.mr_writer;
365 		return rwsem_is_locked(&ip->i_lock.mr_lock);
366 	}
367 
368 	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
369 		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
370 			return !!ip->i_mmaplock.mr_writer;
371 		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
372 	}
373 
374 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
375 		if (!(lock_flags & XFS_IOLOCK_SHARED))
376 			return !debug_locks ||
377 				lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
378 		return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
379 	}
380 
381 	ASSERT(0);
382 	return 0;
383 }
384 #endif
385 
386 #ifdef DEBUG
387 int xfs_locked_n;
388 int xfs_small_retries;
389 int xfs_middle_retries;
390 int xfs_lots_retries;
391 int xfs_lock_delays;
392 #endif
393 
394 /*
395  * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
396  * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
397  * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
398  * errors and warnings.
399  */
400 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
401 static bool
402 xfs_lockdep_subclass_ok(
403 	int subclass)
404 {
405 	return subclass < MAX_LOCKDEP_SUBCLASSES;
406 }
407 #else
408 #define xfs_lockdep_subclass_ok(subclass)	(true)
409 #endif
410 
411 /*
412  * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
413  * value. This can be called for any type of inode lock combination, including
414  * parent locking. Care must be taken to ensure we don't overrun the subclass
415  * storage fields in the class mask we build.
416  */
417 static inline int
418 xfs_lock_inumorder(int lock_mode, int subclass)
419 {
420 	int	class = 0;
421 
422 	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
423 			      XFS_ILOCK_RTSUM)));
424 	ASSERT(xfs_lockdep_subclass_ok(subclass));
425 
426 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
427 		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
428 		class += subclass << XFS_IOLOCK_SHIFT;
429 	}
430 
431 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
432 		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
433 		class += subclass << XFS_MMAPLOCK_SHIFT;
434 	}
435 
436 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
437 		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
438 		class += subclass << XFS_ILOCK_SHIFT;
439 	}
440 
441 	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
442 }
443 
444 /*
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.
447  *
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.
453  *
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.
458  */
459 static void
460 xfs_lock_inodes(
461 	xfs_inode_t	**ips,
462 	int		inodes,
463 	uint		lock_mode)
464 {
465 	int		attempts = 0, i, j, try_lock;
466 	xfs_log_item_t	*lp;
467 
468 	/*
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
473 	 * the asserts.
474 	 */
475 	ASSERT(ips && inodes >= 2 && inodes <= 5);
476 	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
477 			    XFS_ILOCK_EXCL));
478 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
479 			      XFS_ILOCK_SHARED)));
480 	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
481 		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
482 	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
483 		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
484 
485 	if (lock_mode & XFS_IOLOCK_EXCL) {
486 		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
487 	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
488 		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
489 
490 	try_lock = 0;
491 	i = 0;
492 again:
493 	for (; i < inodes; i++) {
494 		ASSERT(ips[i]);
495 
496 		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
497 			continue;
498 
499 		/*
500 		 * If try_lock is not set yet, make sure all locked inodes are
501 		 * not in the AIL.  If any are, set try_lock to be used later.
502 		 */
503 		if (!try_lock) {
504 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
505 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
506 				if (lp && (lp->li_flags & XFS_LI_IN_AIL))
507 					try_lock++;
508 			}
509 		}
510 
511 		/*
512 		 * If any of the previous locks we have locked is in the AIL,
513 		 * we must TRY to get the second and subsequent locks. If
514 		 * we can't get any, we must release all we have
515 		 * and try again.
516 		 */
517 		if (!try_lock) {
518 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
519 			continue;
520 		}
521 
522 		/* try_lock means we have an inode locked that is in the AIL. */
523 		ASSERT(i != 0);
524 		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
525 			continue;
526 
527 		/*
528 		 * Unlock all previous guys and try again.  xfs_iunlock will try
529 		 * to push the tail if the inode is in the AIL.
530 		 */
531 		attempts++;
532 		for (j = i - 1; j >= 0; j--) {
533 			/*
534 			 * Check to see if we've already unlocked this one.  Not
535 			 * the first one going back, and the inode ptr is the
536 			 * same.
537 			 */
538 			if (j != (i - 1) && ips[j] == ips[j + 1])
539 				continue;
540 
541 			xfs_iunlock(ips[j], lock_mode);
542 		}
543 
544 		if ((attempts % 5) == 0) {
545 			delay(1); /* Don't just spin the CPU */
546 #ifdef DEBUG
547 			xfs_lock_delays++;
548 #endif
549 		}
550 		i = 0;
551 		try_lock = 0;
552 		goto again;
553 	}
554 
555 #ifdef DEBUG
556 	if (attempts) {
557 		if (attempts < 5) xfs_small_retries++;
558 		else if (attempts < 100) xfs_middle_retries++;
559 		else xfs_lots_retries++;
560 	} else {
561 		xfs_locked_n++;
562 	}
563 #endif
564 }
565 
566 /*
567  * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
568  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
569  * lock more than one at a time, lockdep will report false positives saying we
570  * have violated locking orders.
571  */
572 void
573 xfs_lock_two_inodes(
574 	xfs_inode_t		*ip0,
575 	xfs_inode_t		*ip1,
576 	uint			lock_mode)
577 {
578 	xfs_inode_t		*temp;
579 	int			attempts = 0;
580 	xfs_log_item_t		*lp;
581 
582 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
583 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
584 		ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
585 
586 	ASSERT(ip0->i_ino != ip1->i_ino);
587 
588 	if (ip0->i_ino > ip1->i_ino) {
589 		temp = ip0;
590 		ip0 = ip1;
591 		ip1 = temp;
592 	}
593 
594  again:
595 	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
596 
597 	/*
598 	 * If the first lock we have locked is in the AIL, we must TRY to get
599 	 * the second lock. If we can't get it, we must release the first one
600 	 * and try again.
601 	 */
602 	lp = (xfs_log_item_t *)ip0->i_itemp;
603 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
604 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
605 			xfs_iunlock(ip0, lock_mode);
606 			if ((++attempts % 5) == 0)
607 				delay(1); /* Don't just spin the CPU */
608 			goto again;
609 		}
610 	} else {
611 		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
612 	}
613 }
614 
615 
616 void
617 __xfs_iflock(
618 	struct xfs_inode	*ip)
619 {
620 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
621 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
622 
623 	do {
624 		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
625 		if (xfs_isiflocked(ip))
626 			io_schedule();
627 	} while (!xfs_iflock_nowait(ip));
628 
629 	finish_wait(wq, &wait.wait);
630 }
631 
632 STATIC uint
633 _xfs_dic2xflags(
634 	__uint16_t		di_flags,
635 	uint64_t		di_flags2,
636 	bool			has_attr)
637 {
638 	uint			flags = 0;
639 
640 	if (di_flags & XFS_DIFLAG_ANY) {
641 		if (di_flags & XFS_DIFLAG_REALTIME)
642 			flags |= FS_XFLAG_REALTIME;
643 		if (di_flags & XFS_DIFLAG_PREALLOC)
644 			flags |= FS_XFLAG_PREALLOC;
645 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
646 			flags |= FS_XFLAG_IMMUTABLE;
647 		if (di_flags & XFS_DIFLAG_APPEND)
648 			flags |= FS_XFLAG_APPEND;
649 		if (di_flags & XFS_DIFLAG_SYNC)
650 			flags |= FS_XFLAG_SYNC;
651 		if (di_flags & XFS_DIFLAG_NOATIME)
652 			flags |= FS_XFLAG_NOATIME;
653 		if (di_flags & XFS_DIFLAG_NODUMP)
654 			flags |= FS_XFLAG_NODUMP;
655 		if (di_flags & XFS_DIFLAG_RTINHERIT)
656 			flags |= FS_XFLAG_RTINHERIT;
657 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
658 			flags |= FS_XFLAG_PROJINHERIT;
659 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
660 			flags |= FS_XFLAG_NOSYMLINKS;
661 		if (di_flags & XFS_DIFLAG_EXTSIZE)
662 			flags |= FS_XFLAG_EXTSIZE;
663 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
664 			flags |= FS_XFLAG_EXTSZINHERIT;
665 		if (di_flags & XFS_DIFLAG_NODEFRAG)
666 			flags |= FS_XFLAG_NODEFRAG;
667 		if (di_flags & XFS_DIFLAG_FILESTREAM)
668 			flags |= FS_XFLAG_FILESTREAM;
669 	}
670 
671 	if (di_flags2 & XFS_DIFLAG2_ANY) {
672 		if (di_flags2 & XFS_DIFLAG2_DAX)
673 			flags |= FS_XFLAG_DAX;
674 		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
675 			flags |= FS_XFLAG_COWEXTSIZE;
676 	}
677 
678 	if (has_attr)
679 		flags |= FS_XFLAG_HASATTR;
680 
681 	return flags;
682 }
683 
684 uint
685 xfs_ip2xflags(
686 	struct xfs_inode	*ip)
687 {
688 	struct xfs_icdinode	*dic = &ip->i_d;
689 
690 	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
691 }
692 
693 /*
694  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
695  * is allowed, otherwise it has to be an exact match. If a CI match is found,
696  * ci_name->name will point to a the actual name (caller must free) or
697  * will be set to NULL if an exact match is found.
698  */
699 int
700 xfs_lookup(
701 	xfs_inode_t		*dp,
702 	struct xfs_name		*name,
703 	xfs_inode_t		**ipp,
704 	struct xfs_name		*ci_name)
705 {
706 	xfs_ino_t		inum;
707 	int			error;
708 
709 	trace_xfs_lookup(dp, name);
710 
711 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
712 		return -EIO;
713 
714 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
715 	if (error)
716 		goto out_unlock;
717 
718 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
719 	if (error)
720 		goto out_free_name;
721 
722 	return 0;
723 
724 out_free_name:
725 	if (ci_name)
726 		kmem_free(ci_name->name);
727 out_unlock:
728 	*ipp = NULL;
729 	return error;
730 }
731 
732 /*
733  * Allocate an inode on disk and return a copy of its in-core version.
734  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
735  * appropriately within the inode.  The uid and gid for the inode are
736  * set according to the contents of the given cred structure.
737  *
738  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
739  * has a free inode available, call xfs_iget() to obtain the in-core
740  * version of the allocated inode.  Finally, fill in the inode and
741  * log its initial contents.  In this case, ialloc_context would be
742  * set to NULL.
743  *
744  * If xfs_dialloc() does not have an available inode, it will replenish
745  * its supply by doing an allocation. Since we can only do one
746  * allocation within a transaction without deadlocks, we must commit
747  * the current transaction before returning the inode itself.
748  * In this case, therefore, we will set ialloc_context and return.
749  * The caller should then commit the current transaction, start a new
750  * transaction, and call xfs_ialloc() again to actually get the inode.
751  *
752  * To ensure that some other process does not grab the inode that
753  * was allocated during the first call to xfs_ialloc(), this routine
754  * also returns the [locked] bp pointing to the head of the freelist
755  * as ialloc_context.  The caller should hold this buffer across
756  * the commit and pass it back into this routine on the second call.
757  *
758  * If we are allocating quota inodes, we do not have a parent inode
759  * to attach to or associate with (i.e. pip == NULL) because they
760  * are not linked into the directory structure - they are attached
761  * directly to the superblock - and so have no parent.
762  */
763 static int
764 xfs_ialloc(
765 	xfs_trans_t	*tp,
766 	xfs_inode_t	*pip,
767 	umode_t		mode,
768 	xfs_nlink_t	nlink,
769 	xfs_dev_t	rdev,
770 	prid_t		prid,
771 	int		okalloc,
772 	xfs_buf_t	**ialloc_context,
773 	xfs_inode_t	**ipp)
774 {
775 	struct xfs_mount *mp = tp->t_mountp;
776 	xfs_ino_t	ino;
777 	xfs_inode_t	*ip;
778 	uint		flags;
779 	int		error;
780 	struct timespec	tv;
781 	struct inode	*inode;
782 
783 	/*
784 	 * Call the space management code to pick
785 	 * the on-disk inode to be allocated.
786 	 */
787 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
788 			    ialloc_context, &ino);
789 	if (error)
790 		return error;
791 	if (*ialloc_context || ino == NULLFSINO) {
792 		*ipp = NULL;
793 		return 0;
794 	}
795 	ASSERT(*ialloc_context == NULL);
796 
797 	/*
798 	 * Get the in-core inode with the lock held exclusively.
799 	 * This is because we're setting fields here we need
800 	 * to prevent others from looking at until we're done.
801 	 */
802 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
803 			 XFS_ILOCK_EXCL, &ip);
804 	if (error)
805 		return error;
806 	ASSERT(ip != NULL);
807 	inode = VFS_I(ip);
808 
809 	/*
810 	 * We always convert v1 inodes to v2 now - we only support filesystems
811 	 * with >= v2 inode capability, so there is no reason for ever leaving
812 	 * an inode in v1 format.
813 	 */
814 	if (ip->i_d.di_version == 1)
815 		ip->i_d.di_version = 2;
816 
817 	inode->i_mode = mode;
818 	set_nlink(inode, nlink);
819 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
820 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
821 	xfs_set_projid(ip, prid);
822 
823 	if (pip && XFS_INHERIT_GID(pip)) {
824 		ip->i_d.di_gid = pip->i_d.di_gid;
825 		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
826 			inode->i_mode |= S_ISGID;
827 	}
828 
829 	/*
830 	 * If the group ID of the new file does not match the effective group
831 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
832 	 * (and only if the irix_sgid_inherit compatibility variable is set).
833 	 */
834 	if ((irix_sgid_inherit) &&
835 	    (inode->i_mode & S_ISGID) &&
836 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
837 		inode->i_mode &= ~S_ISGID;
838 
839 	ip->i_d.di_size = 0;
840 	ip->i_d.di_nextents = 0;
841 	ASSERT(ip->i_d.di_nblocks == 0);
842 
843 	tv = current_time(inode);
844 	inode->i_mtime = tv;
845 	inode->i_atime = tv;
846 	inode->i_ctime = tv;
847 
848 	ip->i_d.di_extsize = 0;
849 	ip->i_d.di_dmevmask = 0;
850 	ip->i_d.di_dmstate = 0;
851 	ip->i_d.di_flags = 0;
852 
853 	if (ip->i_d.di_version == 3) {
854 		inode->i_version = 1;
855 		ip->i_d.di_flags2 = 0;
856 		ip->i_d.di_cowextsize = 0;
857 		ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
858 		ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
859 	}
860 
861 
862 	flags = XFS_ILOG_CORE;
863 	switch (mode & S_IFMT) {
864 	case S_IFIFO:
865 	case S_IFCHR:
866 	case S_IFBLK:
867 	case S_IFSOCK:
868 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
869 		ip->i_df.if_u2.if_rdev = rdev;
870 		ip->i_df.if_flags = 0;
871 		flags |= XFS_ILOG_DEV;
872 		break;
873 	case S_IFREG:
874 	case S_IFDIR:
875 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
876 			uint64_t	di_flags2 = 0;
877 			uint		di_flags = 0;
878 
879 			if (S_ISDIR(mode)) {
880 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
881 					di_flags |= XFS_DIFLAG_RTINHERIT;
882 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
883 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
884 					ip->i_d.di_extsize = pip->i_d.di_extsize;
885 				}
886 				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
887 					di_flags |= XFS_DIFLAG_PROJINHERIT;
888 			} else if (S_ISREG(mode)) {
889 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
890 					di_flags |= XFS_DIFLAG_REALTIME;
891 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
892 					di_flags |= XFS_DIFLAG_EXTSIZE;
893 					ip->i_d.di_extsize = pip->i_d.di_extsize;
894 				}
895 			}
896 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
897 			    xfs_inherit_noatime)
898 				di_flags |= XFS_DIFLAG_NOATIME;
899 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
900 			    xfs_inherit_nodump)
901 				di_flags |= XFS_DIFLAG_NODUMP;
902 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
903 			    xfs_inherit_sync)
904 				di_flags |= XFS_DIFLAG_SYNC;
905 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
906 			    xfs_inherit_nosymlinks)
907 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
908 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
909 			    xfs_inherit_nodefrag)
910 				di_flags |= XFS_DIFLAG_NODEFRAG;
911 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
912 				di_flags |= XFS_DIFLAG_FILESTREAM;
913 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
914 				di_flags2 |= XFS_DIFLAG2_DAX;
915 
916 			ip->i_d.di_flags |= di_flags;
917 			ip->i_d.di_flags2 |= di_flags2;
918 		}
919 		if (pip &&
920 		    (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
921 		    pip->i_d.di_version == 3 &&
922 		    ip->i_d.di_version == 3) {
923 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
924 				ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
925 				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
926 			}
927 		}
928 		/* FALLTHROUGH */
929 	case S_IFLNK:
930 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
931 		ip->i_df.if_flags = XFS_IFEXTENTS;
932 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
933 		ip->i_df.if_u1.if_extents = NULL;
934 		break;
935 	default:
936 		ASSERT(0);
937 	}
938 	/*
939 	 * Attribute fork settings for new inode.
940 	 */
941 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
942 	ip->i_d.di_anextents = 0;
943 
944 	/*
945 	 * Log the new values stuffed into the inode.
946 	 */
947 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
948 	xfs_trans_log_inode(tp, ip, flags);
949 
950 	/* now that we have an i_mode we can setup the inode structure */
951 	xfs_setup_inode(ip);
952 
953 	*ipp = ip;
954 	return 0;
955 }
956 
957 /*
958  * Allocates a new inode from disk and return a pointer to the
959  * incore copy. This routine will internally commit the current
960  * transaction and allocate a new one if the Space Manager needed
961  * to do an allocation to replenish the inode free-list.
962  *
963  * This routine is designed to be called from xfs_create and
964  * xfs_create_dir.
965  *
966  */
967 int
968 xfs_dir_ialloc(
969 	xfs_trans_t	**tpp,		/* input: current transaction;
970 					   output: may be a new transaction. */
971 	xfs_inode_t	*dp,		/* directory within whose allocate
972 					   the inode. */
973 	umode_t		mode,
974 	xfs_nlink_t	nlink,
975 	xfs_dev_t	rdev,
976 	prid_t		prid,		/* project id */
977 	int		okalloc,	/* ok to allocate new space */
978 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
979 					   locked. */
980 	int		*committed)
981 
982 {
983 	xfs_trans_t	*tp;
984 	xfs_inode_t	*ip;
985 	xfs_buf_t	*ialloc_context = NULL;
986 	int		code;
987 	void		*dqinfo;
988 	uint		tflags;
989 
990 	tp = *tpp;
991 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
992 
993 	/*
994 	 * xfs_ialloc will return a pointer to an incore inode if
995 	 * the Space Manager has an available inode on the free
996 	 * list. Otherwise, it will do an allocation and replenish
997 	 * the freelist.  Since we can only do one allocation per
998 	 * transaction without deadlocks, we will need to commit the
999 	 * current transaction and start a new one.  We will then
1000 	 * need to call xfs_ialloc again to get the inode.
1001 	 *
1002 	 * If xfs_ialloc did an allocation to replenish the freelist,
1003 	 * it returns the bp containing the head of the freelist as
1004 	 * ialloc_context. We will hold a lock on it across the
1005 	 * transaction commit so that no other process can steal
1006 	 * the inode(s) that we've just allocated.
1007 	 */
1008 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1009 			  &ialloc_context, &ip);
1010 
1011 	/*
1012 	 * Return an error if we were unable to allocate a new inode.
1013 	 * This should only happen if we run out of space on disk or
1014 	 * encounter a disk error.
1015 	 */
1016 	if (code) {
1017 		*ipp = NULL;
1018 		return code;
1019 	}
1020 	if (!ialloc_context && !ip) {
1021 		*ipp = NULL;
1022 		return -ENOSPC;
1023 	}
1024 
1025 	/*
1026 	 * If the AGI buffer is non-NULL, then we were unable to get an
1027 	 * inode in one operation.  We need to commit the current
1028 	 * transaction and call xfs_ialloc() again.  It is guaranteed
1029 	 * to succeed the second time.
1030 	 */
1031 	if (ialloc_context) {
1032 		/*
1033 		 * Normally, xfs_trans_commit releases all the locks.
1034 		 * We call bhold to hang on to the ialloc_context across
1035 		 * the commit.  Holding this buffer prevents any other
1036 		 * processes from doing any allocations in this
1037 		 * allocation group.
1038 		 */
1039 		xfs_trans_bhold(tp, ialloc_context);
1040 
1041 		/*
1042 		 * We want the quota changes to be associated with the next
1043 		 * transaction, NOT this one. So, detach the dqinfo from this
1044 		 * and attach it to the next transaction.
1045 		 */
1046 		dqinfo = NULL;
1047 		tflags = 0;
1048 		if (tp->t_dqinfo) {
1049 			dqinfo = (void *)tp->t_dqinfo;
1050 			tp->t_dqinfo = NULL;
1051 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1052 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1053 		}
1054 
1055 		code = xfs_trans_roll(&tp, NULL);
1056 		if (committed != NULL)
1057 			*committed = 1;
1058 
1059 		/*
1060 		 * Re-attach the quota info that we detached from prev trx.
1061 		 */
1062 		if (dqinfo) {
1063 			tp->t_dqinfo = dqinfo;
1064 			tp->t_flags |= tflags;
1065 		}
1066 
1067 		if (code) {
1068 			xfs_buf_relse(ialloc_context);
1069 			*tpp = tp;
1070 			*ipp = NULL;
1071 			return code;
1072 		}
1073 		xfs_trans_bjoin(tp, ialloc_context);
1074 
1075 		/*
1076 		 * Call ialloc again. Since we've locked out all
1077 		 * other allocations in this allocation group,
1078 		 * this call should always succeed.
1079 		 */
1080 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1081 				  okalloc, &ialloc_context, &ip);
1082 
1083 		/*
1084 		 * If we get an error at this point, return to the caller
1085 		 * so that the current transaction can be aborted.
1086 		 */
1087 		if (code) {
1088 			*tpp = tp;
1089 			*ipp = NULL;
1090 			return code;
1091 		}
1092 		ASSERT(!ialloc_context && ip);
1093 
1094 	} else {
1095 		if (committed != NULL)
1096 			*committed = 0;
1097 	}
1098 
1099 	*ipp = ip;
1100 	*tpp = tp;
1101 
1102 	return 0;
1103 }
1104 
1105 /*
1106  * Decrement the link count on an inode & log the change.  If this causes the
1107  * link count to go to zero, move the inode to AGI unlinked list so that it can
1108  * be freed when the last active reference goes away via xfs_inactive().
1109  */
1110 static int			/* error */
1111 xfs_droplink(
1112 	xfs_trans_t *tp,
1113 	xfs_inode_t *ip)
1114 {
1115 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1116 
1117 	drop_nlink(VFS_I(ip));
1118 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1119 
1120 	if (VFS_I(ip)->i_nlink)
1121 		return 0;
1122 
1123 	return xfs_iunlink(tp, ip);
1124 }
1125 
1126 /*
1127  * Increment the link count on an inode & log the change.
1128  */
1129 static int
1130 xfs_bumplink(
1131 	xfs_trans_t *tp,
1132 	xfs_inode_t *ip)
1133 {
1134 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1135 
1136 	ASSERT(ip->i_d.di_version > 1);
1137 	inc_nlink(VFS_I(ip));
1138 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1139 	return 0;
1140 }
1141 
1142 int
1143 xfs_create(
1144 	xfs_inode_t		*dp,
1145 	struct xfs_name		*name,
1146 	umode_t			mode,
1147 	xfs_dev_t		rdev,
1148 	xfs_inode_t		**ipp)
1149 {
1150 	int			is_dir = S_ISDIR(mode);
1151 	struct xfs_mount	*mp = dp->i_mount;
1152 	struct xfs_inode	*ip = NULL;
1153 	struct xfs_trans	*tp = NULL;
1154 	int			error;
1155 	struct xfs_defer_ops	dfops;
1156 	xfs_fsblock_t		first_block;
1157 	bool                    unlock_dp_on_error = false;
1158 	prid_t			prid;
1159 	struct xfs_dquot	*udqp = NULL;
1160 	struct xfs_dquot	*gdqp = NULL;
1161 	struct xfs_dquot	*pdqp = NULL;
1162 	struct xfs_trans_res	*tres;
1163 	uint			resblks;
1164 
1165 	trace_xfs_create(dp, name);
1166 
1167 	if (XFS_FORCED_SHUTDOWN(mp))
1168 		return -EIO;
1169 
1170 	prid = xfs_get_initial_prid(dp);
1171 
1172 	/*
1173 	 * Make sure that we have allocated dquot(s) on disk.
1174 	 */
1175 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1176 					xfs_kgid_to_gid(current_fsgid()), prid,
1177 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1178 					&udqp, &gdqp, &pdqp);
1179 	if (error)
1180 		return error;
1181 
1182 	if (is_dir) {
1183 		rdev = 0;
1184 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1185 		tres = &M_RES(mp)->tr_mkdir;
1186 	} else {
1187 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1188 		tres = &M_RES(mp)->tr_create;
1189 	}
1190 
1191 	/*
1192 	 * Initially assume that the file does not exist and
1193 	 * reserve the resources for that case.  If that is not
1194 	 * the case we'll drop the one we have and get a more
1195 	 * appropriate transaction later.
1196 	 */
1197 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1198 	if (error == -ENOSPC) {
1199 		/* flush outstanding delalloc blocks and retry */
1200 		xfs_flush_inodes(mp);
1201 		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1202 	}
1203 	if (error == -ENOSPC) {
1204 		/* No space at all so try a "no-allocation" reservation */
1205 		resblks = 0;
1206 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1207 	}
1208 	if (error)
1209 		goto out_release_inode;
1210 
1211 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1212 	unlock_dp_on_error = true;
1213 
1214 	xfs_defer_init(&dfops, &first_block);
1215 
1216 	/*
1217 	 * Reserve disk quota and the inode.
1218 	 */
1219 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1220 						pdqp, resblks, 1, 0);
1221 	if (error)
1222 		goto out_trans_cancel;
1223 
1224 	if (!resblks) {
1225 		error = xfs_dir_canenter(tp, dp, name);
1226 		if (error)
1227 			goto out_trans_cancel;
1228 	}
1229 
1230 	/*
1231 	 * A newly created regular or special file just has one directory
1232 	 * entry pointing to them, but a directory also the "." entry
1233 	 * pointing to itself.
1234 	 */
1235 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1236 			       prid, resblks > 0, &ip, NULL);
1237 	if (error)
1238 		goto out_trans_cancel;
1239 
1240 	/*
1241 	 * Now we join the directory inode to the transaction.  We do not do it
1242 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1243 	 * (and release all the locks).  An error from here on will result in
1244 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1245 	 * error path.
1246 	 */
1247 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1248 	unlock_dp_on_error = false;
1249 
1250 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1251 					&first_block, &dfops, resblks ?
1252 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1253 	if (error) {
1254 		ASSERT(error != -ENOSPC);
1255 		goto out_trans_cancel;
1256 	}
1257 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1258 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1259 
1260 	if (is_dir) {
1261 		error = xfs_dir_init(tp, ip, dp);
1262 		if (error)
1263 			goto out_bmap_cancel;
1264 
1265 		error = xfs_bumplink(tp, dp);
1266 		if (error)
1267 			goto out_bmap_cancel;
1268 	}
1269 
1270 	/*
1271 	 * If this is a synchronous mount, make sure that the
1272 	 * create transaction goes to disk before returning to
1273 	 * the user.
1274 	 */
1275 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1276 		xfs_trans_set_sync(tp);
1277 
1278 	/*
1279 	 * Attach the dquot(s) to the inodes and modify them incore.
1280 	 * These ids of the inode couldn't have changed since the new
1281 	 * inode has been locked ever since it was created.
1282 	 */
1283 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1284 
1285 	error = xfs_defer_finish(&tp, &dfops, NULL);
1286 	if (error)
1287 		goto out_bmap_cancel;
1288 
1289 	error = xfs_trans_commit(tp);
1290 	if (error)
1291 		goto out_release_inode;
1292 
1293 	xfs_qm_dqrele(udqp);
1294 	xfs_qm_dqrele(gdqp);
1295 	xfs_qm_dqrele(pdqp);
1296 
1297 	*ipp = ip;
1298 	return 0;
1299 
1300  out_bmap_cancel:
1301 	xfs_defer_cancel(&dfops);
1302  out_trans_cancel:
1303 	xfs_trans_cancel(tp);
1304  out_release_inode:
1305 	/*
1306 	 * Wait until after the current transaction is aborted to finish the
1307 	 * setup of the inode and release the inode.  This prevents recursive
1308 	 * transactions and deadlocks from xfs_inactive.
1309 	 */
1310 	if (ip) {
1311 		xfs_finish_inode_setup(ip);
1312 		IRELE(ip);
1313 	}
1314 
1315 	xfs_qm_dqrele(udqp);
1316 	xfs_qm_dqrele(gdqp);
1317 	xfs_qm_dqrele(pdqp);
1318 
1319 	if (unlock_dp_on_error)
1320 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1321 	return error;
1322 }
1323 
1324 int
1325 xfs_create_tmpfile(
1326 	struct xfs_inode	*dp,
1327 	struct dentry		*dentry,
1328 	umode_t			mode,
1329 	struct xfs_inode	**ipp)
1330 {
1331 	struct xfs_mount	*mp = dp->i_mount;
1332 	struct xfs_inode	*ip = NULL;
1333 	struct xfs_trans	*tp = NULL;
1334 	int			error;
1335 	prid_t                  prid;
1336 	struct xfs_dquot	*udqp = NULL;
1337 	struct xfs_dquot	*gdqp = NULL;
1338 	struct xfs_dquot	*pdqp = NULL;
1339 	struct xfs_trans_res	*tres;
1340 	uint			resblks;
1341 
1342 	if (XFS_FORCED_SHUTDOWN(mp))
1343 		return -EIO;
1344 
1345 	prid = xfs_get_initial_prid(dp);
1346 
1347 	/*
1348 	 * Make sure that we have allocated dquot(s) on disk.
1349 	 */
1350 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1351 				xfs_kgid_to_gid(current_fsgid()), prid,
1352 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1353 				&udqp, &gdqp, &pdqp);
1354 	if (error)
1355 		return error;
1356 
1357 	resblks = XFS_IALLOC_SPACE_RES(mp);
1358 	tres = &M_RES(mp)->tr_create_tmpfile;
1359 
1360 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1361 	if (error == -ENOSPC) {
1362 		/* No space at all so try a "no-allocation" reservation */
1363 		resblks = 0;
1364 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1365 	}
1366 	if (error)
1367 		goto out_release_inode;
1368 
1369 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1370 						pdqp, resblks, 1, 0);
1371 	if (error)
1372 		goto out_trans_cancel;
1373 
1374 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1375 				prid, resblks > 0, &ip, NULL);
1376 	if (error)
1377 		goto out_trans_cancel;
1378 
1379 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1380 		xfs_trans_set_sync(tp);
1381 
1382 	/*
1383 	 * Attach the dquot(s) to the inodes and modify them incore.
1384 	 * These ids of the inode couldn't have changed since the new
1385 	 * inode has been locked ever since it was created.
1386 	 */
1387 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1388 
1389 	error = xfs_iunlink(tp, ip);
1390 	if (error)
1391 		goto out_trans_cancel;
1392 
1393 	error = xfs_trans_commit(tp);
1394 	if (error)
1395 		goto out_release_inode;
1396 
1397 	xfs_qm_dqrele(udqp);
1398 	xfs_qm_dqrele(gdqp);
1399 	xfs_qm_dqrele(pdqp);
1400 
1401 	*ipp = ip;
1402 	return 0;
1403 
1404  out_trans_cancel:
1405 	xfs_trans_cancel(tp);
1406  out_release_inode:
1407 	/*
1408 	 * Wait until after the current transaction is aborted to finish the
1409 	 * setup of the inode and release the inode.  This prevents recursive
1410 	 * transactions and deadlocks from xfs_inactive.
1411 	 */
1412 	if (ip) {
1413 		xfs_finish_inode_setup(ip);
1414 		IRELE(ip);
1415 	}
1416 
1417 	xfs_qm_dqrele(udqp);
1418 	xfs_qm_dqrele(gdqp);
1419 	xfs_qm_dqrele(pdqp);
1420 
1421 	return error;
1422 }
1423 
1424 int
1425 xfs_link(
1426 	xfs_inode_t		*tdp,
1427 	xfs_inode_t		*sip,
1428 	struct xfs_name		*target_name)
1429 {
1430 	xfs_mount_t		*mp = tdp->i_mount;
1431 	xfs_trans_t		*tp;
1432 	int			error;
1433 	struct xfs_defer_ops	dfops;
1434 	xfs_fsblock_t           first_block;
1435 	int			resblks;
1436 
1437 	trace_xfs_link(tdp, target_name);
1438 
1439 	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1440 
1441 	if (XFS_FORCED_SHUTDOWN(mp))
1442 		return -EIO;
1443 
1444 	error = xfs_qm_dqattach(sip, 0);
1445 	if (error)
1446 		goto std_return;
1447 
1448 	error = xfs_qm_dqattach(tdp, 0);
1449 	if (error)
1450 		goto std_return;
1451 
1452 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1453 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1454 	if (error == -ENOSPC) {
1455 		resblks = 0;
1456 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1457 	}
1458 	if (error)
1459 		goto std_return;
1460 
1461 	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1462 
1463 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1464 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1465 
1466 	/*
1467 	 * If we are using project inheritance, we only allow hard link
1468 	 * creation in our tree when the project IDs are the same; else
1469 	 * the tree quota mechanism could be circumvented.
1470 	 */
1471 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1472 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1473 		error = -EXDEV;
1474 		goto error_return;
1475 	}
1476 
1477 	if (!resblks) {
1478 		error = xfs_dir_canenter(tp, tdp, target_name);
1479 		if (error)
1480 			goto error_return;
1481 	}
1482 
1483 	xfs_defer_init(&dfops, &first_block);
1484 
1485 	/*
1486 	 * Handle initial link state of O_TMPFILE inode
1487 	 */
1488 	if (VFS_I(sip)->i_nlink == 0) {
1489 		error = xfs_iunlink_remove(tp, sip);
1490 		if (error)
1491 			goto error_return;
1492 	}
1493 
1494 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1495 					&first_block, &dfops, resblks);
1496 	if (error)
1497 		goto error_return;
1498 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1499 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1500 
1501 	error = xfs_bumplink(tp, sip);
1502 	if (error)
1503 		goto error_return;
1504 
1505 	/*
1506 	 * If this is a synchronous mount, make sure that the
1507 	 * link transaction goes to disk before returning to
1508 	 * the user.
1509 	 */
1510 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1511 		xfs_trans_set_sync(tp);
1512 
1513 	error = xfs_defer_finish(&tp, &dfops, NULL);
1514 	if (error) {
1515 		xfs_defer_cancel(&dfops);
1516 		goto error_return;
1517 	}
1518 
1519 	return xfs_trans_commit(tp);
1520 
1521  error_return:
1522 	xfs_trans_cancel(tp);
1523  std_return:
1524 	return error;
1525 }
1526 
1527 /*
1528  * Free up the underlying blocks past new_size.  The new size must be smaller
1529  * than the current size.  This routine can be used both for the attribute and
1530  * data fork, and does not modify the inode size, which is left to the caller.
1531  *
1532  * The transaction passed to this routine must have made a permanent log
1533  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1534  * given transaction and start new ones, so make sure everything involved in
1535  * the transaction is tidy before calling here.  Some transaction will be
1536  * returned to the caller to be committed.  The incoming transaction must
1537  * already include the inode, and both inode locks must be held exclusively.
1538  * The inode must also be "held" within the transaction.  On return the inode
1539  * will be "held" within the returned transaction.  This routine does NOT
1540  * require any disk space to be reserved for it within the transaction.
1541  *
1542  * If we get an error, we must return with the inode locked and linked into the
1543  * current transaction. This keeps things simple for the higher level code,
1544  * because it always knows that the inode is locked and held in the transaction
1545  * that returns to it whether errors occur or not.  We don't mark the inode
1546  * dirty on error so that transactions can be easily aborted if possible.
1547  */
1548 int
1549 xfs_itruncate_extents(
1550 	struct xfs_trans	**tpp,
1551 	struct xfs_inode	*ip,
1552 	int			whichfork,
1553 	xfs_fsize_t		new_size)
1554 {
1555 	struct xfs_mount	*mp = ip->i_mount;
1556 	struct xfs_trans	*tp = *tpp;
1557 	struct xfs_defer_ops	dfops;
1558 	xfs_fsblock_t		first_block;
1559 	xfs_fileoff_t		first_unmap_block;
1560 	xfs_fileoff_t		last_block;
1561 	xfs_filblks_t		unmap_len;
1562 	int			error = 0;
1563 	int			done = 0;
1564 
1565 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1566 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1567 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1568 	ASSERT(new_size <= XFS_ISIZE(ip));
1569 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1570 	ASSERT(ip->i_itemp != NULL);
1571 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1572 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1573 
1574 	trace_xfs_itruncate_extents_start(ip, new_size);
1575 
1576 	/*
1577 	 * Since it is possible for space to become allocated beyond
1578 	 * the end of the file (in a crash where the space is allocated
1579 	 * but the inode size is not yet updated), simply remove any
1580 	 * blocks which show up between the new EOF and the maximum
1581 	 * possible file size.  If the first block to be removed is
1582 	 * beyond the maximum file size (ie it is the same as last_block),
1583 	 * then there is nothing to do.
1584 	 */
1585 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1586 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1587 	if (first_unmap_block == last_block)
1588 		return 0;
1589 
1590 	ASSERT(first_unmap_block < last_block);
1591 	unmap_len = last_block - first_unmap_block + 1;
1592 	while (!done) {
1593 		xfs_defer_init(&dfops, &first_block);
1594 		error = xfs_bunmapi(tp, ip,
1595 				    first_unmap_block, unmap_len,
1596 				    xfs_bmapi_aflag(whichfork),
1597 				    XFS_ITRUNC_MAX_EXTENTS,
1598 				    &first_block, &dfops,
1599 				    &done);
1600 		if (error)
1601 			goto out_bmap_cancel;
1602 
1603 		/*
1604 		 * Duplicate the transaction that has the permanent
1605 		 * reservation and commit the old transaction.
1606 		 */
1607 		error = xfs_defer_finish(&tp, &dfops, ip);
1608 		if (error)
1609 			goto out_bmap_cancel;
1610 
1611 		error = xfs_trans_roll(&tp, ip);
1612 		if (error)
1613 			goto out;
1614 	}
1615 
1616 	/* Remove all pending CoW reservations. */
1617 	error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1618 			last_block);
1619 	if (error)
1620 		goto out;
1621 
1622 	/*
1623 	 * Clear the reflink flag if we truncated everything.
1624 	 */
1625 	if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip)) {
1626 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1627 		xfs_inode_clear_cowblocks_tag(ip);
1628 	}
1629 
1630 	/*
1631 	 * Always re-log the inode so that our permanent transaction can keep
1632 	 * on rolling it forward in the log.
1633 	 */
1634 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1635 
1636 	trace_xfs_itruncate_extents_end(ip, new_size);
1637 
1638 out:
1639 	*tpp = tp;
1640 	return error;
1641 out_bmap_cancel:
1642 	/*
1643 	 * If the bunmapi call encounters an error, return to the caller where
1644 	 * the transaction can be properly aborted.  We just need to make sure
1645 	 * we're not holding any resources that we were not when we came in.
1646 	 */
1647 	xfs_defer_cancel(&dfops);
1648 	goto out;
1649 }
1650 
1651 int
1652 xfs_release(
1653 	xfs_inode_t	*ip)
1654 {
1655 	xfs_mount_t	*mp = ip->i_mount;
1656 	int		error;
1657 
1658 	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1659 		return 0;
1660 
1661 	/* If this is a read-only mount, don't do this (would generate I/O) */
1662 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1663 		return 0;
1664 
1665 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1666 		int truncated;
1667 
1668 		/*
1669 		 * If we previously truncated this file and removed old data
1670 		 * in the process, we want to initiate "early" writeout on
1671 		 * the last close.  This is an attempt to combat the notorious
1672 		 * NULL files problem which is particularly noticeable from a
1673 		 * truncate down, buffered (re-)write (delalloc), followed by
1674 		 * a crash.  What we are effectively doing here is
1675 		 * significantly reducing the time window where we'd otherwise
1676 		 * be exposed to that problem.
1677 		 */
1678 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1679 		if (truncated) {
1680 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1681 			if (ip->i_delayed_blks > 0) {
1682 				error = filemap_flush(VFS_I(ip)->i_mapping);
1683 				if (error)
1684 					return error;
1685 			}
1686 		}
1687 	}
1688 
1689 	if (VFS_I(ip)->i_nlink == 0)
1690 		return 0;
1691 
1692 	if (xfs_can_free_eofblocks(ip, false)) {
1693 
1694 		/*
1695 		 * If we can't get the iolock just skip truncating the blocks
1696 		 * past EOF because we could deadlock with the mmap_sem
1697 		 * otherwise.  We'll get another chance to drop them once the
1698 		 * last reference to the inode is dropped, so we'll never leak
1699 		 * blocks permanently.
1700 		 *
1701 		 * Further, check if the inode is being opened, written and
1702 		 * closed frequently and we have delayed allocation blocks
1703 		 * outstanding (e.g. streaming writes from the NFS server),
1704 		 * truncating the blocks past EOF will cause fragmentation to
1705 		 * occur.
1706 		 *
1707 		 * In this case don't do the truncation, either, but we have to
1708 		 * be careful how we detect this case. Blocks beyond EOF show
1709 		 * up as i_delayed_blks even when the inode is clean, so we
1710 		 * need to truncate them away first before checking for a dirty
1711 		 * release. Hence on the first dirty close we will still remove
1712 		 * the speculative allocation, but after that we will leave it
1713 		 * in place.
1714 		 */
1715 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1716 			return 0;
1717 
1718 		error = xfs_free_eofblocks(mp, ip, true);
1719 		if (error && error != -EAGAIN)
1720 			return error;
1721 
1722 		/* delalloc blocks after truncation means it really is dirty */
1723 		if (ip->i_delayed_blks)
1724 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1725 	}
1726 	return 0;
1727 }
1728 
1729 /*
1730  * xfs_inactive_truncate
1731  *
1732  * Called to perform a truncate when an inode becomes unlinked.
1733  */
1734 STATIC int
1735 xfs_inactive_truncate(
1736 	struct xfs_inode *ip)
1737 {
1738 	struct xfs_mount	*mp = ip->i_mount;
1739 	struct xfs_trans	*tp;
1740 	int			error;
1741 
1742 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1743 	if (error) {
1744 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1745 		return error;
1746 	}
1747 
1748 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1749 	xfs_trans_ijoin(tp, ip, 0);
1750 
1751 	/*
1752 	 * Log the inode size first to prevent stale data exposure in the event
1753 	 * of a system crash before the truncate completes. See the related
1754 	 * comment in xfs_vn_setattr_size() for details.
1755 	 */
1756 	ip->i_d.di_size = 0;
1757 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1758 
1759 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1760 	if (error)
1761 		goto error_trans_cancel;
1762 
1763 	ASSERT(ip->i_d.di_nextents == 0);
1764 
1765 	error = xfs_trans_commit(tp);
1766 	if (error)
1767 		goto error_unlock;
1768 
1769 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1770 	return 0;
1771 
1772 error_trans_cancel:
1773 	xfs_trans_cancel(tp);
1774 error_unlock:
1775 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1776 	return error;
1777 }
1778 
1779 /*
1780  * xfs_inactive_ifree()
1781  *
1782  * Perform the inode free when an inode is unlinked.
1783  */
1784 STATIC int
1785 xfs_inactive_ifree(
1786 	struct xfs_inode *ip)
1787 {
1788 	struct xfs_defer_ops	dfops;
1789 	xfs_fsblock_t		first_block;
1790 	struct xfs_mount	*mp = ip->i_mount;
1791 	struct xfs_trans	*tp;
1792 	int			error;
1793 
1794 	/*
1795 	 * We try to use a per-AG reservation for any block needed by the finobt
1796 	 * tree, but as the finobt feature predates the per-AG reservation
1797 	 * support a degraded file system might not have enough space for the
1798 	 * reservation at mount time.  In that case try to dip into the reserved
1799 	 * pool and pray.
1800 	 *
1801 	 * Send a warning if the reservation does happen to fail, as the inode
1802 	 * now remains allocated and sits on the unlinked list until the fs is
1803 	 * repaired.
1804 	 */
1805 	if (unlikely(mp->m_inotbt_nores)) {
1806 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1807 				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1808 				&tp);
1809 	} else {
1810 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1811 	}
1812 	if (error) {
1813 		if (error == -ENOSPC) {
1814 			xfs_warn_ratelimited(mp,
1815 			"Failed to remove inode(s) from unlinked list. "
1816 			"Please free space, unmount and run xfs_repair.");
1817 		} else {
1818 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1819 		}
1820 		return error;
1821 	}
1822 
1823 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1824 	xfs_trans_ijoin(tp, ip, 0);
1825 
1826 	xfs_defer_init(&dfops, &first_block);
1827 	error = xfs_ifree(tp, ip, &dfops);
1828 	if (error) {
1829 		/*
1830 		 * If we fail to free the inode, shut down.  The cancel
1831 		 * might do that, we need to make sure.  Otherwise the
1832 		 * inode might be lost for a long time or forever.
1833 		 */
1834 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1835 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1836 				__func__, error);
1837 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1838 		}
1839 		xfs_trans_cancel(tp);
1840 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1841 		return error;
1842 	}
1843 
1844 	/*
1845 	 * Credit the quota account(s). The inode is gone.
1846 	 */
1847 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1848 
1849 	/*
1850 	 * Just ignore errors at this point.  There is nothing we can do except
1851 	 * to try to keep going. Make sure it's not a silent error.
1852 	 */
1853 	error = xfs_defer_finish(&tp, &dfops, NULL);
1854 	if (error) {
1855 		xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1856 			__func__, error);
1857 		xfs_defer_cancel(&dfops);
1858 	}
1859 	error = xfs_trans_commit(tp);
1860 	if (error)
1861 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1862 			__func__, error);
1863 
1864 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1865 	return 0;
1866 }
1867 
1868 /*
1869  * xfs_inactive
1870  *
1871  * This is called when the vnode reference count for the vnode
1872  * goes to zero.  If the file has been unlinked, then it must
1873  * now be truncated.  Also, we clear all of the read-ahead state
1874  * kept for the inode here since the file is now closed.
1875  */
1876 void
1877 xfs_inactive(
1878 	xfs_inode_t	*ip)
1879 {
1880 	struct xfs_mount	*mp;
1881 	int			error;
1882 	int			truncate = 0;
1883 
1884 	/*
1885 	 * If the inode is already free, then there can be nothing
1886 	 * to clean up here.
1887 	 */
1888 	if (VFS_I(ip)->i_mode == 0) {
1889 		ASSERT(ip->i_df.if_real_bytes == 0);
1890 		ASSERT(ip->i_df.if_broot_bytes == 0);
1891 		return;
1892 	}
1893 
1894 	mp = ip->i_mount;
1895 	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1896 
1897 	/* If this is a read-only mount, don't do this (would generate I/O) */
1898 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1899 		return;
1900 
1901 	if (VFS_I(ip)->i_nlink != 0) {
1902 		/*
1903 		 * force is true because we are evicting an inode from the
1904 		 * cache. Post-eof blocks must be freed, lest we end up with
1905 		 * broken free space accounting.
1906 		 */
1907 		if (xfs_can_free_eofblocks(ip, true))
1908 			xfs_free_eofblocks(mp, ip, false);
1909 
1910 		return;
1911 	}
1912 
1913 	if (S_ISREG(VFS_I(ip)->i_mode) &&
1914 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1915 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1916 		truncate = 1;
1917 
1918 	error = xfs_qm_dqattach(ip, 0);
1919 	if (error)
1920 		return;
1921 
1922 	if (S_ISLNK(VFS_I(ip)->i_mode))
1923 		error = xfs_inactive_symlink(ip);
1924 	else if (truncate)
1925 		error = xfs_inactive_truncate(ip);
1926 	if (error)
1927 		return;
1928 
1929 	/*
1930 	 * If there are attributes associated with the file then blow them away
1931 	 * now.  The code calls a routine that recursively deconstructs the
1932 	 * attribute fork. If also blows away the in-core attribute fork.
1933 	 */
1934 	if (XFS_IFORK_Q(ip)) {
1935 		error = xfs_attr_inactive(ip);
1936 		if (error)
1937 			return;
1938 	}
1939 
1940 	ASSERT(!ip->i_afp);
1941 	ASSERT(ip->i_d.di_anextents == 0);
1942 	ASSERT(ip->i_d.di_forkoff == 0);
1943 
1944 	/*
1945 	 * Free the inode.
1946 	 */
1947 	error = xfs_inactive_ifree(ip);
1948 	if (error)
1949 		return;
1950 
1951 	/*
1952 	 * Release the dquots held by inode, if any.
1953 	 */
1954 	xfs_qm_dqdetach(ip);
1955 }
1956 
1957 /*
1958  * This is called when the inode's link count goes to 0 or we are creating a
1959  * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1960  * set to true as the link count is dropped to zero by the VFS after we've
1961  * created the file successfully, so we have to add it to the unlinked list
1962  * while the link count is non-zero.
1963  *
1964  * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1965  * list when the inode is freed.
1966  */
1967 STATIC int
1968 xfs_iunlink(
1969 	struct xfs_trans *tp,
1970 	struct xfs_inode *ip)
1971 {
1972 	xfs_mount_t	*mp = tp->t_mountp;
1973 	xfs_agi_t	*agi;
1974 	xfs_dinode_t	*dip;
1975 	xfs_buf_t	*agibp;
1976 	xfs_buf_t	*ibp;
1977 	xfs_agino_t	agino;
1978 	short		bucket_index;
1979 	int		offset;
1980 	int		error;
1981 
1982 	ASSERT(VFS_I(ip)->i_mode != 0);
1983 
1984 	/*
1985 	 * Get the agi buffer first.  It ensures lock ordering
1986 	 * on the list.
1987 	 */
1988 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1989 	if (error)
1990 		return error;
1991 	agi = XFS_BUF_TO_AGI(agibp);
1992 
1993 	/*
1994 	 * Get the index into the agi hash table for the
1995 	 * list this inode will go on.
1996 	 */
1997 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1998 	ASSERT(agino != 0);
1999 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2000 	ASSERT(agi->agi_unlinked[bucket_index]);
2001 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2002 
2003 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2004 		/*
2005 		 * There is already another inode in the bucket we need
2006 		 * to add ourselves to.  Add us at the front of the list.
2007 		 * Here we put the head pointer into our next pointer,
2008 		 * and then we fall through to point the head at us.
2009 		 */
2010 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2011 				       0, 0);
2012 		if (error)
2013 			return error;
2014 
2015 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2016 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2017 		offset = ip->i_imap.im_boffset +
2018 			offsetof(xfs_dinode_t, di_next_unlinked);
2019 
2020 		/* need to recalc the inode CRC if appropriate */
2021 		xfs_dinode_calc_crc(mp, dip);
2022 
2023 		xfs_trans_inode_buf(tp, ibp);
2024 		xfs_trans_log_buf(tp, ibp, offset,
2025 				  (offset + sizeof(xfs_agino_t) - 1));
2026 		xfs_inobp_check(mp, ibp);
2027 	}
2028 
2029 	/*
2030 	 * Point the bucket head pointer at the inode being inserted.
2031 	 */
2032 	ASSERT(agino != 0);
2033 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2034 	offset = offsetof(xfs_agi_t, agi_unlinked) +
2035 		(sizeof(xfs_agino_t) * bucket_index);
2036 	xfs_trans_log_buf(tp, agibp, offset,
2037 			  (offset + sizeof(xfs_agino_t) - 1));
2038 	return 0;
2039 }
2040 
2041 /*
2042  * Pull the on-disk inode from the AGI unlinked list.
2043  */
2044 STATIC int
2045 xfs_iunlink_remove(
2046 	xfs_trans_t	*tp,
2047 	xfs_inode_t	*ip)
2048 {
2049 	xfs_ino_t	next_ino;
2050 	xfs_mount_t	*mp;
2051 	xfs_agi_t	*agi;
2052 	xfs_dinode_t	*dip;
2053 	xfs_buf_t	*agibp;
2054 	xfs_buf_t	*ibp;
2055 	xfs_agnumber_t	agno;
2056 	xfs_agino_t	agino;
2057 	xfs_agino_t	next_agino;
2058 	xfs_buf_t	*last_ibp;
2059 	xfs_dinode_t	*last_dip = NULL;
2060 	short		bucket_index;
2061 	int		offset, last_offset = 0;
2062 	int		error;
2063 
2064 	mp = tp->t_mountp;
2065 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2066 
2067 	/*
2068 	 * Get the agi buffer first.  It ensures lock ordering
2069 	 * on the list.
2070 	 */
2071 	error = xfs_read_agi(mp, tp, agno, &agibp);
2072 	if (error)
2073 		return error;
2074 
2075 	agi = XFS_BUF_TO_AGI(agibp);
2076 
2077 	/*
2078 	 * Get the index into the agi hash table for the
2079 	 * list this inode will go on.
2080 	 */
2081 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2082 	ASSERT(agino != 0);
2083 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2084 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2085 	ASSERT(agi->agi_unlinked[bucket_index]);
2086 
2087 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2088 		/*
2089 		 * We're at the head of the list.  Get the inode's on-disk
2090 		 * buffer to see if there is anyone after us on the list.
2091 		 * Only modify our next pointer if it is not already NULLAGINO.
2092 		 * This saves us the overhead of dealing with the buffer when
2093 		 * there is no need to change it.
2094 		 */
2095 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2096 				       0, 0);
2097 		if (error) {
2098 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2099 				__func__, error);
2100 			return error;
2101 		}
2102 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2103 		ASSERT(next_agino != 0);
2104 		if (next_agino != NULLAGINO) {
2105 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2106 			offset = ip->i_imap.im_boffset +
2107 				offsetof(xfs_dinode_t, di_next_unlinked);
2108 
2109 			/* need to recalc the inode CRC if appropriate */
2110 			xfs_dinode_calc_crc(mp, dip);
2111 
2112 			xfs_trans_inode_buf(tp, ibp);
2113 			xfs_trans_log_buf(tp, ibp, offset,
2114 					  (offset + sizeof(xfs_agino_t) - 1));
2115 			xfs_inobp_check(mp, ibp);
2116 		} else {
2117 			xfs_trans_brelse(tp, ibp);
2118 		}
2119 		/*
2120 		 * Point the bucket head pointer at the next inode.
2121 		 */
2122 		ASSERT(next_agino != 0);
2123 		ASSERT(next_agino != agino);
2124 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2125 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2126 			(sizeof(xfs_agino_t) * bucket_index);
2127 		xfs_trans_log_buf(tp, agibp, offset,
2128 				  (offset + sizeof(xfs_agino_t) - 1));
2129 	} else {
2130 		/*
2131 		 * We need to search the list for the inode being freed.
2132 		 */
2133 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2134 		last_ibp = NULL;
2135 		while (next_agino != agino) {
2136 			struct xfs_imap	imap;
2137 
2138 			if (last_ibp)
2139 				xfs_trans_brelse(tp, last_ibp);
2140 
2141 			imap.im_blkno = 0;
2142 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2143 
2144 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2145 			if (error) {
2146 				xfs_warn(mp,
2147 	"%s: xfs_imap returned error %d.",
2148 					 __func__, error);
2149 				return error;
2150 			}
2151 
2152 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2153 					       &last_ibp, 0, 0);
2154 			if (error) {
2155 				xfs_warn(mp,
2156 	"%s: xfs_imap_to_bp returned error %d.",
2157 					__func__, error);
2158 				return error;
2159 			}
2160 
2161 			last_offset = imap.im_boffset;
2162 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2163 			ASSERT(next_agino != NULLAGINO);
2164 			ASSERT(next_agino != 0);
2165 		}
2166 
2167 		/*
2168 		 * Now last_ibp points to the buffer previous to us on the
2169 		 * unlinked list.  Pull us from the list.
2170 		 */
2171 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2172 				       0, 0);
2173 		if (error) {
2174 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2175 				__func__, error);
2176 			return error;
2177 		}
2178 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2179 		ASSERT(next_agino != 0);
2180 		ASSERT(next_agino != agino);
2181 		if (next_agino != NULLAGINO) {
2182 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2183 			offset = ip->i_imap.im_boffset +
2184 				offsetof(xfs_dinode_t, di_next_unlinked);
2185 
2186 			/* need to recalc the inode CRC if appropriate */
2187 			xfs_dinode_calc_crc(mp, dip);
2188 
2189 			xfs_trans_inode_buf(tp, ibp);
2190 			xfs_trans_log_buf(tp, ibp, offset,
2191 					  (offset + sizeof(xfs_agino_t) - 1));
2192 			xfs_inobp_check(mp, ibp);
2193 		} else {
2194 			xfs_trans_brelse(tp, ibp);
2195 		}
2196 		/*
2197 		 * Point the previous inode on the list to the next inode.
2198 		 */
2199 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2200 		ASSERT(next_agino != 0);
2201 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2202 
2203 		/* need to recalc the inode CRC if appropriate */
2204 		xfs_dinode_calc_crc(mp, last_dip);
2205 
2206 		xfs_trans_inode_buf(tp, last_ibp);
2207 		xfs_trans_log_buf(tp, last_ibp, offset,
2208 				  (offset + sizeof(xfs_agino_t) - 1));
2209 		xfs_inobp_check(mp, last_ibp);
2210 	}
2211 	return 0;
2212 }
2213 
2214 /*
2215  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2216  * inodes that are in memory - they all must be marked stale and attached to
2217  * the cluster buffer.
2218  */
2219 STATIC int
2220 xfs_ifree_cluster(
2221 	xfs_inode_t		*free_ip,
2222 	xfs_trans_t		*tp,
2223 	struct xfs_icluster	*xic)
2224 {
2225 	xfs_mount_t		*mp = free_ip->i_mount;
2226 	int			blks_per_cluster;
2227 	int			inodes_per_cluster;
2228 	int			nbufs;
2229 	int			i, j;
2230 	int			ioffset;
2231 	xfs_daddr_t		blkno;
2232 	xfs_buf_t		*bp;
2233 	xfs_inode_t		*ip;
2234 	xfs_inode_log_item_t	*iip;
2235 	xfs_log_item_t		*lip;
2236 	struct xfs_perag	*pag;
2237 	xfs_ino_t		inum;
2238 
2239 	inum = xic->first_ino;
2240 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2241 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2242 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2243 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2244 
2245 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2246 		/*
2247 		 * The allocation bitmap tells us which inodes of the chunk were
2248 		 * physically allocated. Skip the cluster if an inode falls into
2249 		 * a sparse region.
2250 		 */
2251 		ioffset = inum - xic->first_ino;
2252 		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2253 			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2254 			continue;
2255 		}
2256 
2257 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2258 					 XFS_INO_TO_AGBNO(mp, inum));
2259 
2260 		/*
2261 		 * We obtain and lock the backing buffer first in the process
2262 		 * here, as we have to ensure that any dirty inode that we
2263 		 * can't get the flush lock on is attached to the buffer.
2264 		 * If we scan the in-memory inodes first, then buffer IO can
2265 		 * complete before we get a lock on it, and hence we may fail
2266 		 * to mark all the active inodes on the buffer stale.
2267 		 */
2268 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2269 					mp->m_bsize * blks_per_cluster,
2270 					XBF_UNMAPPED);
2271 
2272 		if (!bp)
2273 			return -ENOMEM;
2274 
2275 		/*
2276 		 * This buffer may not have been correctly initialised as we
2277 		 * didn't read it from disk. That's not important because we are
2278 		 * only using to mark the buffer as stale in the log, and to
2279 		 * attach stale cached inodes on it. That means it will never be
2280 		 * dispatched for IO. If it is, we want to know about it, and we
2281 		 * want it to fail. We can acheive this by adding a write
2282 		 * verifier to the buffer.
2283 		 */
2284 		 bp->b_ops = &xfs_inode_buf_ops;
2285 
2286 		/*
2287 		 * Walk the inodes already attached to the buffer and mark them
2288 		 * stale. These will all have the flush locks held, so an
2289 		 * in-memory inode walk can't lock them. By marking them all
2290 		 * stale first, we will not attempt to lock them in the loop
2291 		 * below as the XFS_ISTALE flag will be set.
2292 		 */
2293 		lip = bp->b_fspriv;
2294 		while (lip) {
2295 			if (lip->li_type == XFS_LI_INODE) {
2296 				iip = (xfs_inode_log_item_t *)lip;
2297 				ASSERT(iip->ili_logged == 1);
2298 				lip->li_cb = xfs_istale_done;
2299 				xfs_trans_ail_copy_lsn(mp->m_ail,
2300 							&iip->ili_flush_lsn,
2301 							&iip->ili_item.li_lsn);
2302 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2303 			}
2304 			lip = lip->li_bio_list;
2305 		}
2306 
2307 
2308 		/*
2309 		 * For each inode in memory attempt to add it to the inode
2310 		 * buffer and set it up for being staled on buffer IO
2311 		 * completion.  This is safe as we've locked out tail pushing
2312 		 * and flushing by locking the buffer.
2313 		 *
2314 		 * We have already marked every inode that was part of a
2315 		 * transaction stale above, which means there is no point in
2316 		 * even trying to lock them.
2317 		 */
2318 		for (i = 0; i < inodes_per_cluster; i++) {
2319 retry:
2320 			rcu_read_lock();
2321 			ip = radix_tree_lookup(&pag->pag_ici_root,
2322 					XFS_INO_TO_AGINO(mp, (inum + i)));
2323 
2324 			/* Inode not in memory, nothing to do */
2325 			if (!ip) {
2326 				rcu_read_unlock();
2327 				continue;
2328 			}
2329 
2330 			/*
2331 			 * because this is an RCU protected lookup, we could
2332 			 * find a recently freed or even reallocated inode
2333 			 * during the lookup. We need to check under the
2334 			 * i_flags_lock for a valid inode here. Skip it if it
2335 			 * is not valid, the wrong inode or stale.
2336 			 */
2337 			spin_lock(&ip->i_flags_lock);
2338 			if (ip->i_ino != inum + i ||
2339 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2340 				spin_unlock(&ip->i_flags_lock);
2341 				rcu_read_unlock();
2342 				continue;
2343 			}
2344 			spin_unlock(&ip->i_flags_lock);
2345 
2346 			/*
2347 			 * Don't try to lock/unlock the current inode, but we
2348 			 * _cannot_ skip the other inodes that we did not find
2349 			 * in the list attached to the buffer and are not
2350 			 * already marked stale. If we can't lock it, back off
2351 			 * and retry.
2352 			 */
2353 			if (ip != free_ip &&
2354 			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2355 				rcu_read_unlock();
2356 				delay(1);
2357 				goto retry;
2358 			}
2359 			rcu_read_unlock();
2360 
2361 			xfs_iflock(ip);
2362 			xfs_iflags_set(ip, XFS_ISTALE);
2363 
2364 			/*
2365 			 * we don't need to attach clean inodes or those only
2366 			 * with unlogged changes (which we throw away, anyway).
2367 			 */
2368 			iip = ip->i_itemp;
2369 			if (!iip || xfs_inode_clean(ip)) {
2370 				ASSERT(ip != free_ip);
2371 				xfs_ifunlock(ip);
2372 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2373 				continue;
2374 			}
2375 
2376 			iip->ili_last_fields = iip->ili_fields;
2377 			iip->ili_fields = 0;
2378 			iip->ili_fsync_fields = 0;
2379 			iip->ili_logged = 1;
2380 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2381 						&iip->ili_item.li_lsn);
2382 
2383 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2384 						  &iip->ili_item);
2385 
2386 			if (ip != free_ip)
2387 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2388 		}
2389 
2390 		xfs_trans_stale_inode_buf(tp, bp);
2391 		xfs_trans_binval(tp, bp);
2392 	}
2393 
2394 	xfs_perag_put(pag);
2395 	return 0;
2396 }
2397 
2398 /*
2399  * This is called to return an inode to the inode free list.
2400  * The inode should already be truncated to 0 length and have
2401  * no pages associated with it.  This routine also assumes that
2402  * the inode is already a part of the transaction.
2403  *
2404  * The on-disk copy of the inode will have been added to the list
2405  * of unlinked inodes in the AGI. We need to remove the inode from
2406  * that list atomically with respect to freeing it here.
2407  */
2408 int
2409 xfs_ifree(
2410 	xfs_trans_t	*tp,
2411 	xfs_inode_t	*ip,
2412 	struct xfs_defer_ops	*dfops)
2413 {
2414 	int			error;
2415 	struct xfs_icluster	xic = { 0 };
2416 
2417 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2418 	ASSERT(VFS_I(ip)->i_nlink == 0);
2419 	ASSERT(ip->i_d.di_nextents == 0);
2420 	ASSERT(ip->i_d.di_anextents == 0);
2421 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2422 	ASSERT(ip->i_d.di_nblocks == 0);
2423 
2424 	/*
2425 	 * Pull the on-disk inode from the AGI unlinked list.
2426 	 */
2427 	error = xfs_iunlink_remove(tp, ip);
2428 	if (error)
2429 		return error;
2430 
2431 	error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2432 	if (error)
2433 		return error;
2434 
2435 	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2436 	ip->i_d.di_flags = 0;
2437 	ip->i_d.di_dmevmask = 0;
2438 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2439 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2440 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2441 	/*
2442 	 * Bump the generation count so no one will be confused
2443 	 * by reincarnations of this inode.
2444 	 */
2445 	VFS_I(ip)->i_generation++;
2446 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2447 
2448 	if (xic.deleted)
2449 		error = xfs_ifree_cluster(ip, tp, &xic);
2450 
2451 	return error;
2452 }
2453 
2454 /*
2455  * This is called to unpin an inode.  The caller must have the inode locked
2456  * in at least shared mode so that the buffer cannot be subsequently pinned
2457  * once someone is waiting for it to be unpinned.
2458  */
2459 static void
2460 xfs_iunpin(
2461 	struct xfs_inode	*ip)
2462 {
2463 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2464 
2465 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2466 
2467 	/* Give the log a push to start the unpinning I/O */
2468 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2469 
2470 }
2471 
2472 static void
2473 __xfs_iunpin_wait(
2474 	struct xfs_inode	*ip)
2475 {
2476 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2477 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2478 
2479 	xfs_iunpin(ip);
2480 
2481 	do {
2482 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2483 		if (xfs_ipincount(ip))
2484 			io_schedule();
2485 	} while (xfs_ipincount(ip));
2486 	finish_wait(wq, &wait.wait);
2487 }
2488 
2489 void
2490 xfs_iunpin_wait(
2491 	struct xfs_inode	*ip)
2492 {
2493 	if (xfs_ipincount(ip))
2494 		__xfs_iunpin_wait(ip);
2495 }
2496 
2497 /*
2498  * Removing an inode from the namespace involves removing the directory entry
2499  * and dropping the link count on the inode. Removing the directory entry can
2500  * result in locking an AGF (directory blocks were freed) and removing a link
2501  * count can result in placing the inode on an unlinked list which results in
2502  * locking an AGI.
2503  *
2504  * The big problem here is that we have an ordering constraint on AGF and AGI
2505  * locking - inode allocation locks the AGI, then can allocate a new extent for
2506  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2507  * removes the inode from the unlinked list, requiring that we lock the AGI
2508  * first, and then freeing the inode can result in an inode chunk being freed
2509  * and hence freeing disk space requiring that we lock an AGF.
2510  *
2511  * Hence the ordering that is imposed by other parts of the code is AGI before
2512  * AGF. This means we cannot remove the directory entry before we drop the inode
2513  * reference count and put it on the unlinked list as this results in a lock
2514  * order of AGF then AGI, and this can deadlock against inode allocation and
2515  * freeing. Therefore we must drop the link counts before we remove the
2516  * directory entry.
2517  *
2518  * This is still safe from a transactional point of view - it is not until we
2519  * get to xfs_defer_finish() that we have the possibility of multiple
2520  * transactions in this operation. Hence as long as we remove the directory
2521  * entry and drop the link count in the first transaction of the remove
2522  * operation, there are no transactional constraints on the ordering here.
2523  */
2524 int
2525 xfs_remove(
2526 	xfs_inode_t             *dp,
2527 	struct xfs_name		*name,
2528 	xfs_inode_t		*ip)
2529 {
2530 	xfs_mount_t		*mp = dp->i_mount;
2531 	xfs_trans_t             *tp = NULL;
2532 	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2533 	int                     error = 0;
2534 	struct xfs_defer_ops	dfops;
2535 	xfs_fsblock_t           first_block;
2536 	uint			resblks;
2537 
2538 	trace_xfs_remove(dp, name);
2539 
2540 	if (XFS_FORCED_SHUTDOWN(mp))
2541 		return -EIO;
2542 
2543 	error = xfs_qm_dqattach(dp, 0);
2544 	if (error)
2545 		goto std_return;
2546 
2547 	error = xfs_qm_dqattach(ip, 0);
2548 	if (error)
2549 		goto std_return;
2550 
2551 	/*
2552 	 * We try to get the real space reservation first,
2553 	 * allowing for directory btree deletion(s) implying
2554 	 * possible bmap insert(s).  If we can't get the space
2555 	 * reservation then we use 0 instead, and avoid the bmap
2556 	 * btree insert(s) in the directory code by, if the bmap
2557 	 * insert tries to happen, instead trimming the LAST
2558 	 * block from the directory.
2559 	 */
2560 	resblks = XFS_REMOVE_SPACE_RES(mp);
2561 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2562 	if (error == -ENOSPC) {
2563 		resblks = 0;
2564 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2565 				&tp);
2566 	}
2567 	if (error) {
2568 		ASSERT(error != -ENOSPC);
2569 		goto std_return;
2570 	}
2571 
2572 	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2573 
2574 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2575 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2576 
2577 	/*
2578 	 * If we're removing a directory perform some additional validation.
2579 	 */
2580 	if (is_dir) {
2581 		ASSERT(VFS_I(ip)->i_nlink >= 2);
2582 		if (VFS_I(ip)->i_nlink != 2) {
2583 			error = -ENOTEMPTY;
2584 			goto out_trans_cancel;
2585 		}
2586 		if (!xfs_dir_isempty(ip)) {
2587 			error = -ENOTEMPTY;
2588 			goto out_trans_cancel;
2589 		}
2590 
2591 		/* Drop the link from ip's "..".  */
2592 		error = xfs_droplink(tp, dp);
2593 		if (error)
2594 			goto out_trans_cancel;
2595 
2596 		/* Drop the "." link from ip to self.  */
2597 		error = xfs_droplink(tp, ip);
2598 		if (error)
2599 			goto out_trans_cancel;
2600 	} else {
2601 		/*
2602 		 * When removing a non-directory we need to log the parent
2603 		 * inode here.  For a directory this is done implicitly
2604 		 * by the xfs_droplink call for the ".." entry.
2605 		 */
2606 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2607 	}
2608 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2609 
2610 	/* Drop the link from dp to ip. */
2611 	error = xfs_droplink(tp, ip);
2612 	if (error)
2613 		goto out_trans_cancel;
2614 
2615 	xfs_defer_init(&dfops, &first_block);
2616 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2617 					&first_block, &dfops, resblks);
2618 	if (error) {
2619 		ASSERT(error != -ENOENT);
2620 		goto out_bmap_cancel;
2621 	}
2622 
2623 	/*
2624 	 * If this is a synchronous mount, make sure that the
2625 	 * remove transaction goes to disk before returning to
2626 	 * the user.
2627 	 */
2628 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2629 		xfs_trans_set_sync(tp);
2630 
2631 	error = xfs_defer_finish(&tp, &dfops, NULL);
2632 	if (error)
2633 		goto out_bmap_cancel;
2634 
2635 	error = xfs_trans_commit(tp);
2636 	if (error)
2637 		goto std_return;
2638 
2639 	if (is_dir && xfs_inode_is_filestream(ip))
2640 		xfs_filestream_deassociate(ip);
2641 
2642 	return 0;
2643 
2644  out_bmap_cancel:
2645 	xfs_defer_cancel(&dfops);
2646  out_trans_cancel:
2647 	xfs_trans_cancel(tp);
2648  std_return:
2649 	return error;
2650 }
2651 
2652 /*
2653  * Enter all inodes for a rename transaction into a sorted array.
2654  */
2655 #define __XFS_SORT_INODES	5
2656 STATIC void
2657 xfs_sort_for_rename(
2658 	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2659 	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2660 	struct xfs_inode	*ip1,	/* in: inode of old entry */
2661 	struct xfs_inode	*ip2,	/* in: inode of new entry */
2662 	struct xfs_inode	*wip,	/* in: whiteout inode */
2663 	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2664 	int			*num_inodes)  /* in/out: inodes in array */
2665 {
2666 	int			i, j;
2667 
2668 	ASSERT(*num_inodes == __XFS_SORT_INODES);
2669 	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2670 
2671 	/*
2672 	 * i_tab contains a list of pointers to inodes.  We initialize
2673 	 * the table here & we'll sort it.  We will then use it to
2674 	 * order the acquisition of the inode locks.
2675 	 *
2676 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2677 	 */
2678 	i = 0;
2679 	i_tab[i++] = dp1;
2680 	i_tab[i++] = dp2;
2681 	i_tab[i++] = ip1;
2682 	if (ip2)
2683 		i_tab[i++] = ip2;
2684 	if (wip)
2685 		i_tab[i++] = wip;
2686 	*num_inodes = i;
2687 
2688 	/*
2689 	 * Sort the elements via bubble sort.  (Remember, there are at
2690 	 * most 5 elements to sort, so this is adequate.)
2691 	 */
2692 	for (i = 0; i < *num_inodes; i++) {
2693 		for (j = 1; j < *num_inodes; j++) {
2694 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2695 				struct xfs_inode *temp = i_tab[j];
2696 				i_tab[j] = i_tab[j-1];
2697 				i_tab[j-1] = temp;
2698 			}
2699 		}
2700 	}
2701 }
2702 
2703 static int
2704 xfs_finish_rename(
2705 	struct xfs_trans	*tp,
2706 	struct xfs_defer_ops	*dfops)
2707 {
2708 	int			error;
2709 
2710 	/*
2711 	 * If this is a synchronous mount, make sure that the rename transaction
2712 	 * goes to disk before returning to the user.
2713 	 */
2714 	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2715 		xfs_trans_set_sync(tp);
2716 
2717 	error = xfs_defer_finish(&tp, dfops, NULL);
2718 	if (error) {
2719 		xfs_defer_cancel(dfops);
2720 		xfs_trans_cancel(tp);
2721 		return error;
2722 	}
2723 
2724 	return xfs_trans_commit(tp);
2725 }
2726 
2727 /*
2728  * xfs_cross_rename()
2729  *
2730  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2731  */
2732 STATIC int
2733 xfs_cross_rename(
2734 	struct xfs_trans	*tp,
2735 	struct xfs_inode	*dp1,
2736 	struct xfs_name		*name1,
2737 	struct xfs_inode	*ip1,
2738 	struct xfs_inode	*dp2,
2739 	struct xfs_name		*name2,
2740 	struct xfs_inode	*ip2,
2741 	struct xfs_defer_ops	*dfops,
2742 	xfs_fsblock_t		*first_block,
2743 	int			spaceres)
2744 {
2745 	int		error = 0;
2746 	int		ip1_flags = 0;
2747 	int		ip2_flags = 0;
2748 	int		dp2_flags = 0;
2749 
2750 	/* Swap inode number for dirent in first parent */
2751 	error = xfs_dir_replace(tp, dp1, name1,
2752 				ip2->i_ino,
2753 				first_block, dfops, spaceres);
2754 	if (error)
2755 		goto out_trans_abort;
2756 
2757 	/* Swap inode number for dirent in second parent */
2758 	error = xfs_dir_replace(tp, dp2, name2,
2759 				ip1->i_ino,
2760 				first_block, dfops, spaceres);
2761 	if (error)
2762 		goto out_trans_abort;
2763 
2764 	/*
2765 	 * If we're renaming one or more directories across different parents,
2766 	 * update the respective ".." entries (and link counts) to match the new
2767 	 * parents.
2768 	 */
2769 	if (dp1 != dp2) {
2770 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2771 
2772 		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2773 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2774 						dp1->i_ino, first_block,
2775 						dfops, spaceres);
2776 			if (error)
2777 				goto out_trans_abort;
2778 
2779 			/* transfer ip2 ".." reference to dp1 */
2780 			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2781 				error = xfs_droplink(tp, dp2);
2782 				if (error)
2783 					goto out_trans_abort;
2784 				error = xfs_bumplink(tp, dp1);
2785 				if (error)
2786 					goto out_trans_abort;
2787 			}
2788 
2789 			/*
2790 			 * Although ip1 isn't changed here, userspace needs
2791 			 * to be warned about the change, so that applications
2792 			 * relying on it (like backup ones), will properly
2793 			 * notify the change
2794 			 */
2795 			ip1_flags |= XFS_ICHGTIME_CHG;
2796 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2797 		}
2798 
2799 		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2800 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2801 						dp2->i_ino, first_block,
2802 						dfops, spaceres);
2803 			if (error)
2804 				goto out_trans_abort;
2805 
2806 			/* transfer ip1 ".." reference to dp2 */
2807 			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2808 				error = xfs_droplink(tp, dp1);
2809 				if (error)
2810 					goto out_trans_abort;
2811 				error = xfs_bumplink(tp, dp2);
2812 				if (error)
2813 					goto out_trans_abort;
2814 			}
2815 
2816 			/*
2817 			 * Although ip2 isn't changed here, userspace needs
2818 			 * to be warned about the change, so that applications
2819 			 * relying on it (like backup ones), will properly
2820 			 * notify the change
2821 			 */
2822 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2823 			ip2_flags |= XFS_ICHGTIME_CHG;
2824 		}
2825 	}
2826 
2827 	if (ip1_flags) {
2828 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2829 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2830 	}
2831 	if (ip2_flags) {
2832 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2833 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2834 	}
2835 	if (dp2_flags) {
2836 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2837 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2838 	}
2839 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2840 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2841 	return xfs_finish_rename(tp, dfops);
2842 
2843 out_trans_abort:
2844 	xfs_defer_cancel(dfops);
2845 	xfs_trans_cancel(tp);
2846 	return error;
2847 }
2848 
2849 /*
2850  * xfs_rename_alloc_whiteout()
2851  *
2852  * Return a referenced, unlinked, unlocked inode that that can be used as a
2853  * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2854  * crash between allocating the inode and linking it into the rename transaction
2855  * recovery will free the inode and we won't leak it.
2856  */
2857 static int
2858 xfs_rename_alloc_whiteout(
2859 	struct xfs_inode	*dp,
2860 	struct xfs_inode	**wip)
2861 {
2862 	struct xfs_inode	*tmpfile;
2863 	int			error;
2864 
2865 	error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2866 	if (error)
2867 		return error;
2868 
2869 	/*
2870 	 * Prepare the tmpfile inode as if it were created through the VFS.
2871 	 * Otherwise, the link increment paths will complain about nlink 0->1.
2872 	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2873 	 * and flag it as linkable.
2874 	 */
2875 	drop_nlink(VFS_I(tmpfile));
2876 	xfs_setup_iops(tmpfile);
2877 	xfs_finish_inode_setup(tmpfile);
2878 	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2879 
2880 	*wip = tmpfile;
2881 	return 0;
2882 }
2883 
2884 /*
2885  * xfs_rename
2886  */
2887 int
2888 xfs_rename(
2889 	struct xfs_inode	*src_dp,
2890 	struct xfs_name		*src_name,
2891 	struct xfs_inode	*src_ip,
2892 	struct xfs_inode	*target_dp,
2893 	struct xfs_name		*target_name,
2894 	struct xfs_inode	*target_ip,
2895 	unsigned int		flags)
2896 {
2897 	struct xfs_mount	*mp = src_dp->i_mount;
2898 	struct xfs_trans	*tp;
2899 	struct xfs_defer_ops	dfops;
2900 	xfs_fsblock_t		first_block;
2901 	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2902 	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2903 	int			num_inodes = __XFS_SORT_INODES;
2904 	bool			new_parent = (src_dp != target_dp);
2905 	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2906 	int			spaceres;
2907 	int			error;
2908 
2909 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2910 
2911 	if ((flags & RENAME_EXCHANGE) && !target_ip)
2912 		return -EINVAL;
2913 
2914 	/*
2915 	 * If we are doing a whiteout operation, allocate the whiteout inode
2916 	 * we will be placing at the target and ensure the type is set
2917 	 * appropriately.
2918 	 */
2919 	if (flags & RENAME_WHITEOUT) {
2920 		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2921 		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2922 		if (error)
2923 			return error;
2924 
2925 		/* setup target dirent info as whiteout */
2926 		src_name->type = XFS_DIR3_FT_CHRDEV;
2927 	}
2928 
2929 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2930 				inodes, &num_inodes);
2931 
2932 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2933 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2934 	if (error == -ENOSPC) {
2935 		spaceres = 0;
2936 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2937 				&tp);
2938 	}
2939 	if (error)
2940 		goto out_release_wip;
2941 
2942 	/*
2943 	 * Attach the dquots to the inodes
2944 	 */
2945 	error = xfs_qm_vop_rename_dqattach(inodes);
2946 	if (error)
2947 		goto out_trans_cancel;
2948 
2949 	/*
2950 	 * Lock all the participating inodes. Depending upon whether
2951 	 * the target_name exists in the target directory, and
2952 	 * whether the target directory is the same as the source
2953 	 * directory, we can lock from 2 to 4 inodes.
2954 	 */
2955 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2956 
2957 	/*
2958 	 * Join all the inodes to the transaction. From this point on,
2959 	 * we can rely on either trans_commit or trans_cancel to unlock
2960 	 * them.
2961 	 */
2962 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2963 	if (new_parent)
2964 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2965 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2966 	if (target_ip)
2967 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2968 	if (wip)
2969 		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2970 
2971 	/*
2972 	 * If we are using project inheritance, we only allow renames
2973 	 * into our tree when the project IDs are the same; else the
2974 	 * tree quota mechanism would be circumvented.
2975 	 */
2976 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2977 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2978 		error = -EXDEV;
2979 		goto out_trans_cancel;
2980 	}
2981 
2982 	xfs_defer_init(&dfops, &first_block);
2983 
2984 	/* RENAME_EXCHANGE is unique from here on. */
2985 	if (flags & RENAME_EXCHANGE)
2986 		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2987 					target_dp, target_name, target_ip,
2988 					&dfops, &first_block, spaceres);
2989 
2990 	/*
2991 	 * Set up the target.
2992 	 */
2993 	if (target_ip == NULL) {
2994 		/*
2995 		 * If there's no space reservation, check the entry will
2996 		 * fit before actually inserting it.
2997 		 */
2998 		if (!spaceres) {
2999 			error = xfs_dir_canenter(tp, target_dp, target_name);
3000 			if (error)
3001 				goto out_trans_cancel;
3002 		}
3003 		/*
3004 		 * If target does not exist and the rename crosses
3005 		 * directories, adjust the target directory link count
3006 		 * to account for the ".." reference from the new entry.
3007 		 */
3008 		error = xfs_dir_createname(tp, target_dp, target_name,
3009 						src_ip->i_ino, &first_block,
3010 						&dfops, spaceres);
3011 		if (error)
3012 			goto out_bmap_cancel;
3013 
3014 		xfs_trans_ichgtime(tp, target_dp,
3015 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3016 
3017 		if (new_parent && src_is_directory) {
3018 			error = xfs_bumplink(tp, target_dp);
3019 			if (error)
3020 				goto out_bmap_cancel;
3021 		}
3022 	} else { /* target_ip != NULL */
3023 		/*
3024 		 * If target exists and it's a directory, check that both
3025 		 * target and source are directories and that target can be
3026 		 * destroyed, or that neither is a directory.
3027 		 */
3028 		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3029 			/*
3030 			 * Make sure target dir is empty.
3031 			 */
3032 			if (!(xfs_dir_isempty(target_ip)) ||
3033 			    (VFS_I(target_ip)->i_nlink > 2)) {
3034 				error = -EEXIST;
3035 				goto out_trans_cancel;
3036 			}
3037 		}
3038 
3039 		/*
3040 		 * Link the source inode under the target name.
3041 		 * If the source inode is a directory and we are moving
3042 		 * it across directories, its ".." entry will be
3043 		 * inconsistent until we replace that down below.
3044 		 *
3045 		 * In case there is already an entry with the same
3046 		 * name at the destination directory, remove it first.
3047 		 */
3048 		error = xfs_dir_replace(tp, target_dp, target_name,
3049 					src_ip->i_ino,
3050 					&first_block, &dfops, spaceres);
3051 		if (error)
3052 			goto out_bmap_cancel;
3053 
3054 		xfs_trans_ichgtime(tp, target_dp,
3055 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3056 
3057 		/*
3058 		 * Decrement the link count on the target since the target
3059 		 * dir no longer points to it.
3060 		 */
3061 		error = xfs_droplink(tp, target_ip);
3062 		if (error)
3063 			goto out_bmap_cancel;
3064 
3065 		if (src_is_directory) {
3066 			/*
3067 			 * Drop the link from the old "." entry.
3068 			 */
3069 			error = xfs_droplink(tp, target_ip);
3070 			if (error)
3071 				goto out_bmap_cancel;
3072 		}
3073 	} /* target_ip != NULL */
3074 
3075 	/*
3076 	 * Remove the source.
3077 	 */
3078 	if (new_parent && src_is_directory) {
3079 		/*
3080 		 * Rewrite the ".." entry to point to the new
3081 		 * directory.
3082 		 */
3083 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3084 					target_dp->i_ino,
3085 					&first_block, &dfops, spaceres);
3086 		ASSERT(error != -EEXIST);
3087 		if (error)
3088 			goto out_bmap_cancel;
3089 	}
3090 
3091 	/*
3092 	 * We always want to hit the ctime on the source inode.
3093 	 *
3094 	 * This isn't strictly required by the standards since the source
3095 	 * inode isn't really being changed, but old unix file systems did
3096 	 * it and some incremental backup programs won't work without it.
3097 	 */
3098 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3099 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3100 
3101 	/*
3102 	 * Adjust the link count on src_dp.  This is necessary when
3103 	 * renaming a directory, either within one parent when
3104 	 * the target existed, or across two parent directories.
3105 	 */
3106 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3107 
3108 		/*
3109 		 * Decrement link count on src_directory since the
3110 		 * entry that's moved no longer points to it.
3111 		 */
3112 		error = xfs_droplink(tp, src_dp);
3113 		if (error)
3114 			goto out_bmap_cancel;
3115 	}
3116 
3117 	/*
3118 	 * For whiteouts, we only need to update the source dirent with the
3119 	 * inode number of the whiteout inode rather than removing it
3120 	 * altogether.
3121 	 */
3122 	if (wip) {
3123 		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3124 					&first_block, &dfops, spaceres);
3125 	} else
3126 		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3127 					   &first_block, &dfops, spaceres);
3128 	if (error)
3129 		goto out_bmap_cancel;
3130 
3131 	/*
3132 	 * For whiteouts, we need to bump the link count on the whiteout inode.
3133 	 * This means that failures all the way up to this point leave the inode
3134 	 * on the unlinked list and so cleanup is a simple matter of dropping
3135 	 * the remaining reference to it. If we fail here after bumping the link
3136 	 * count, we're shutting down the filesystem so we'll never see the
3137 	 * intermediate state on disk.
3138 	 */
3139 	if (wip) {
3140 		ASSERT(VFS_I(wip)->i_nlink == 0);
3141 		error = xfs_bumplink(tp, wip);
3142 		if (error)
3143 			goto out_bmap_cancel;
3144 		error = xfs_iunlink_remove(tp, wip);
3145 		if (error)
3146 			goto out_bmap_cancel;
3147 		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3148 
3149 		/*
3150 		 * Now we have a real link, clear the "I'm a tmpfile" state
3151 		 * flag from the inode so it doesn't accidentally get misused in
3152 		 * future.
3153 		 */
3154 		VFS_I(wip)->i_state &= ~I_LINKABLE;
3155 	}
3156 
3157 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3158 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3159 	if (new_parent)
3160 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3161 
3162 	error = xfs_finish_rename(tp, &dfops);
3163 	if (wip)
3164 		IRELE(wip);
3165 	return error;
3166 
3167 out_bmap_cancel:
3168 	xfs_defer_cancel(&dfops);
3169 out_trans_cancel:
3170 	xfs_trans_cancel(tp);
3171 out_release_wip:
3172 	if (wip)
3173 		IRELE(wip);
3174 	return error;
3175 }
3176 
3177 STATIC int
3178 xfs_iflush_cluster(
3179 	struct xfs_inode	*ip,
3180 	struct xfs_buf		*bp)
3181 {
3182 	struct xfs_mount	*mp = ip->i_mount;
3183 	struct xfs_perag	*pag;
3184 	unsigned long		first_index, mask;
3185 	unsigned long		inodes_per_cluster;
3186 	int			cilist_size;
3187 	struct xfs_inode	**cilist;
3188 	struct xfs_inode	*cip;
3189 	int			nr_found;
3190 	int			clcount = 0;
3191 	int			bufwasdelwri;
3192 	int			i;
3193 
3194 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3195 
3196 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3197 	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3198 	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3199 	if (!cilist)
3200 		goto out_put;
3201 
3202 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3203 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3204 	rcu_read_lock();
3205 	/* really need a gang lookup range call here */
3206 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3207 					first_index, inodes_per_cluster);
3208 	if (nr_found == 0)
3209 		goto out_free;
3210 
3211 	for (i = 0; i < nr_found; i++) {
3212 		cip = cilist[i];
3213 		if (cip == ip)
3214 			continue;
3215 
3216 		/*
3217 		 * because this is an RCU protected lookup, we could find a
3218 		 * recently freed or even reallocated inode during the lookup.
3219 		 * We need to check under the i_flags_lock for a valid inode
3220 		 * here. Skip it if it is not valid or the wrong inode.
3221 		 */
3222 		spin_lock(&cip->i_flags_lock);
3223 		if (!cip->i_ino ||
3224 		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3225 			spin_unlock(&cip->i_flags_lock);
3226 			continue;
3227 		}
3228 
3229 		/*
3230 		 * Once we fall off the end of the cluster, no point checking
3231 		 * any more inodes in the list because they will also all be
3232 		 * outside the cluster.
3233 		 */
3234 		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3235 			spin_unlock(&cip->i_flags_lock);
3236 			break;
3237 		}
3238 		spin_unlock(&cip->i_flags_lock);
3239 
3240 		/*
3241 		 * Do an un-protected check to see if the inode is dirty and
3242 		 * is a candidate for flushing.  These checks will be repeated
3243 		 * later after the appropriate locks are acquired.
3244 		 */
3245 		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3246 			continue;
3247 
3248 		/*
3249 		 * Try to get locks.  If any are unavailable or it is pinned,
3250 		 * then this inode cannot be flushed and is skipped.
3251 		 */
3252 
3253 		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3254 			continue;
3255 		if (!xfs_iflock_nowait(cip)) {
3256 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3257 			continue;
3258 		}
3259 		if (xfs_ipincount(cip)) {
3260 			xfs_ifunlock(cip);
3261 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3262 			continue;
3263 		}
3264 
3265 
3266 		/*
3267 		 * Check the inode number again, just to be certain we are not
3268 		 * racing with freeing in xfs_reclaim_inode(). See the comments
3269 		 * in that function for more information as to why the initial
3270 		 * check is not sufficient.
3271 		 */
3272 		if (!cip->i_ino) {
3273 			xfs_ifunlock(cip);
3274 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3275 			continue;
3276 		}
3277 
3278 		/*
3279 		 * arriving here means that this inode can be flushed.  First
3280 		 * re-check that it's dirty before flushing.
3281 		 */
3282 		if (!xfs_inode_clean(cip)) {
3283 			int	error;
3284 			error = xfs_iflush_int(cip, bp);
3285 			if (error) {
3286 				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3287 				goto cluster_corrupt_out;
3288 			}
3289 			clcount++;
3290 		} else {
3291 			xfs_ifunlock(cip);
3292 		}
3293 		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3294 	}
3295 
3296 	if (clcount) {
3297 		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3298 		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3299 	}
3300 
3301 out_free:
3302 	rcu_read_unlock();
3303 	kmem_free(cilist);
3304 out_put:
3305 	xfs_perag_put(pag);
3306 	return 0;
3307 
3308 
3309 cluster_corrupt_out:
3310 	/*
3311 	 * Corruption detected in the clustering loop.  Invalidate the
3312 	 * inode buffer and shut down the filesystem.
3313 	 */
3314 	rcu_read_unlock();
3315 	/*
3316 	 * Clean up the buffer.  If it was delwri, just release it --
3317 	 * brelse can handle it with no problems.  If not, shut down the
3318 	 * filesystem before releasing the buffer.
3319 	 */
3320 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3321 	if (bufwasdelwri)
3322 		xfs_buf_relse(bp);
3323 
3324 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3325 
3326 	if (!bufwasdelwri) {
3327 		/*
3328 		 * Just like incore_relse: if we have b_iodone functions,
3329 		 * mark the buffer as an error and call them.  Otherwise
3330 		 * mark it as stale and brelse.
3331 		 */
3332 		if (bp->b_iodone) {
3333 			bp->b_flags &= ~XBF_DONE;
3334 			xfs_buf_stale(bp);
3335 			xfs_buf_ioerror(bp, -EIO);
3336 			xfs_buf_ioend(bp);
3337 		} else {
3338 			xfs_buf_stale(bp);
3339 			xfs_buf_relse(bp);
3340 		}
3341 	}
3342 
3343 	/*
3344 	 * Unlocks the flush lock
3345 	 */
3346 	xfs_iflush_abort(cip, false);
3347 	kmem_free(cilist);
3348 	xfs_perag_put(pag);
3349 	return -EFSCORRUPTED;
3350 }
3351 
3352 /*
3353  * Flush dirty inode metadata into the backing buffer.
3354  *
3355  * The caller must have the inode lock and the inode flush lock held.  The
3356  * inode lock will still be held upon return to the caller, and the inode
3357  * flush lock will be released after the inode has reached the disk.
3358  *
3359  * The caller must write out the buffer returned in *bpp and release it.
3360  */
3361 int
3362 xfs_iflush(
3363 	struct xfs_inode	*ip,
3364 	struct xfs_buf		**bpp)
3365 {
3366 	struct xfs_mount	*mp = ip->i_mount;
3367 	struct xfs_buf		*bp = NULL;
3368 	struct xfs_dinode	*dip;
3369 	int			error;
3370 
3371 	XFS_STATS_INC(mp, xs_iflush_count);
3372 
3373 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3374 	ASSERT(xfs_isiflocked(ip));
3375 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3376 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3377 
3378 	*bpp = NULL;
3379 
3380 	xfs_iunpin_wait(ip);
3381 
3382 	/*
3383 	 * For stale inodes we cannot rely on the backing buffer remaining
3384 	 * stale in cache for the remaining life of the stale inode and so
3385 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3386 	 * inodes below. We have to check this after ensuring the inode is
3387 	 * unpinned so that it is safe to reclaim the stale inode after the
3388 	 * flush call.
3389 	 */
3390 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3391 		xfs_ifunlock(ip);
3392 		return 0;
3393 	}
3394 
3395 	/*
3396 	 * This may have been unpinned because the filesystem is shutting
3397 	 * down forcibly. If that's the case we must not write this inode
3398 	 * to disk, because the log record didn't make it to disk.
3399 	 *
3400 	 * We also have to remove the log item from the AIL in this case,
3401 	 * as we wait for an empty AIL as part of the unmount process.
3402 	 */
3403 	if (XFS_FORCED_SHUTDOWN(mp)) {
3404 		error = -EIO;
3405 		goto abort_out;
3406 	}
3407 
3408 	/*
3409 	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3410 	 * operation here, so we may get  an EAGAIN error. In that case, we
3411 	 * simply want to return with the inode still dirty.
3412 	 *
3413 	 * If we get any other error, we effectively have a corruption situation
3414 	 * and we cannot flush the inode, so we treat it the same as failing
3415 	 * xfs_iflush_int().
3416 	 */
3417 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3418 			       0);
3419 	if (error == -EAGAIN) {
3420 		xfs_ifunlock(ip);
3421 		return error;
3422 	}
3423 	if (error)
3424 		goto corrupt_out;
3425 
3426 	/*
3427 	 * First flush out the inode that xfs_iflush was called with.
3428 	 */
3429 	error = xfs_iflush_int(ip, bp);
3430 	if (error)
3431 		goto corrupt_out;
3432 
3433 	/*
3434 	 * If the buffer is pinned then push on the log now so we won't
3435 	 * get stuck waiting in the write for too long.
3436 	 */
3437 	if (xfs_buf_ispinned(bp))
3438 		xfs_log_force(mp, 0);
3439 
3440 	/*
3441 	 * inode clustering:
3442 	 * see if other inodes can be gathered into this write
3443 	 */
3444 	error = xfs_iflush_cluster(ip, bp);
3445 	if (error)
3446 		goto cluster_corrupt_out;
3447 
3448 	*bpp = bp;
3449 	return 0;
3450 
3451 corrupt_out:
3452 	if (bp)
3453 		xfs_buf_relse(bp);
3454 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3455 cluster_corrupt_out:
3456 	error = -EFSCORRUPTED;
3457 abort_out:
3458 	/*
3459 	 * Unlocks the flush lock
3460 	 */
3461 	xfs_iflush_abort(ip, false);
3462 	return error;
3463 }
3464 
3465 STATIC int
3466 xfs_iflush_int(
3467 	struct xfs_inode	*ip,
3468 	struct xfs_buf		*bp)
3469 {
3470 	struct xfs_inode_log_item *iip = ip->i_itemp;
3471 	struct xfs_dinode	*dip;
3472 	struct xfs_mount	*mp = ip->i_mount;
3473 
3474 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3475 	ASSERT(xfs_isiflocked(ip));
3476 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3477 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3478 	ASSERT(iip != NULL && iip->ili_fields != 0);
3479 	ASSERT(ip->i_d.di_version > 1);
3480 
3481 	/* set *dip = inode's place in the buffer */
3482 	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3483 
3484 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3485 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3486 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3487 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3488 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3489 		goto corrupt_out;
3490 	}
3491 	if (S_ISREG(VFS_I(ip)->i_mode)) {
3492 		if (XFS_TEST_ERROR(
3493 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3494 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3495 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3496 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3497 				"%s: Bad regular inode %Lu, ptr 0x%p",
3498 				__func__, ip->i_ino, ip);
3499 			goto corrupt_out;
3500 		}
3501 	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3502 		if (XFS_TEST_ERROR(
3503 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3504 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3505 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3506 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3507 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3508 				"%s: Bad directory inode %Lu, ptr 0x%p",
3509 				__func__, ip->i_ino, ip);
3510 			goto corrupt_out;
3511 		}
3512 	}
3513 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3514 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3515 				XFS_RANDOM_IFLUSH_5)) {
3516 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3517 			"%s: detected corrupt incore inode %Lu, "
3518 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3519 			__func__, ip->i_ino,
3520 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3521 			ip->i_d.di_nblocks, ip);
3522 		goto corrupt_out;
3523 	}
3524 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3525 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3526 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3527 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3528 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3529 		goto corrupt_out;
3530 	}
3531 
3532 	/*
3533 	 * Inode item log recovery for v2 inodes are dependent on the
3534 	 * di_flushiter count for correct sequencing. We bump the flush
3535 	 * iteration count so we can detect flushes which postdate a log record
3536 	 * during recovery. This is redundant as we now log every change and
3537 	 * hence this can't happen but we need to still do it to ensure
3538 	 * backwards compatibility with old kernels that predate logging all
3539 	 * inode changes.
3540 	 */
3541 	if (ip->i_d.di_version < 3)
3542 		ip->i_d.di_flushiter++;
3543 
3544 	/*
3545 	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3546 	 * copy out the core of the inode, because if the inode is dirty at all
3547 	 * the core must be.
3548 	 */
3549 	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3550 
3551 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3552 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3553 		ip->i_d.di_flushiter = 0;
3554 
3555 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3556 	if (XFS_IFORK_Q(ip))
3557 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3558 	xfs_inobp_check(mp, bp);
3559 
3560 	/*
3561 	 * We've recorded everything logged in the inode, so we'd like to clear
3562 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3563 	 * However, we can't stop logging all this information until the data
3564 	 * we've copied into the disk buffer is written to disk.  If we did we
3565 	 * might overwrite the copy of the inode in the log with all the data
3566 	 * after re-logging only part of it, and in the face of a crash we
3567 	 * wouldn't have all the data we need to recover.
3568 	 *
3569 	 * What we do is move the bits to the ili_last_fields field.  When
3570 	 * logging the inode, these bits are moved back to the ili_fields field.
3571 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3572 	 * know that the information those bits represent is permanently on
3573 	 * disk.  As long as the flush completes before the inode is logged
3574 	 * again, then both ili_fields and ili_last_fields will be cleared.
3575 	 *
3576 	 * We can play with the ili_fields bits here, because the inode lock
3577 	 * must be held exclusively in order to set bits there and the flush
3578 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3579 	 * done routine can tell whether or not to look in the AIL.  Also, store
3580 	 * the current LSN of the inode so that we can tell whether the item has
3581 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3582 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3583 	 * atomically.
3584 	 */
3585 	iip->ili_last_fields = iip->ili_fields;
3586 	iip->ili_fields = 0;
3587 	iip->ili_fsync_fields = 0;
3588 	iip->ili_logged = 1;
3589 
3590 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3591 				&iip->ili_item.li_lsn);
3592 
3593 	/*
3594 	 * Attach the function xfs_iflush_done to the inode's
3595 	 * buffer.  This will remove the inode from the AIL
3596 	 * and unlock the inode's flush lock when the inode is
3597 	 * completely written to disk.
3598 	 */
3599 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3600 
3601 	/* generate the checksum. */
3602 	xfs_dinode_calc_crc(mp, dip);
3603 
3604 	ASSERT(bp->b_fspriv != NULL);
3605 	ASSERT(bp->b_iodone != NULL);
3606 	return 0;
3607 
3608 corrupt_out:
3609 	return -EFSCORRUPTED;
3610 }
3611