xref: /openbmc/linux/fs/xfs/xfs_inode.c (revision 7211ec63)
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 #include "xfs_dir2_priv.h"
54 
55 kmem_zone_t *xfs_inode_zone;
56 
57 /*
58  * Used in xfs_itruncate_extents().  This is the maximum number of extents
59  * freed from a file in a single transaction.
60  */
61 #define	XFS_ITRUNC_MAX_EXTENTS	2
62 
63 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
64 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
65 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
66 
67 /*
68  * helper function to extract extent size hint from inode
69  */
70 xfs_extlen_t
71 xfs_get_extsz_hint(
72 	struct xfs_inode	*ip)
73 {
74 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
75 		return ip->i_d.di_extsize;
76 	if (XFS_IS_REALTIME_INODE(ip))
77 		return ip->i_mount->m_sb.sb_rextsize;
78 	return 0;
79 }
80 
81 /*
82  * Helper function to extract CoW extent size hint from inode.
83  * Between the extent size hint and the CoW extent size hint, we
84  * return the greater of the two.  If the value is zero (automatic),
85  * use the default size.
86  */
87 xfs_extlen_t
88 xfs_get_cowextsz_hint(
89 	struct xfs_inode	*ip)
90 {
91 	xfs_extlen_t		a, b;
92 
93 	a = 0;
94 	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
95 		a = ip->i_d.di_cowextsize;
96 	b = xfs_get_extsz_hint(ip);
97 
98 	a = max(a, b);
99 	if (a == 0)
100 		return XFS_DEFAULT_COWEXTSZ_HINT;
101 	return a;
102 }
103 
104 /*
105  * These two are wrapper routines around the xfs_ilock() routine used to
106  * centralize some grungy code.  They are used in places that wish to lock the
107  * inode solely for reading the extents.  The reason these places can't just
108  * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
109  * bringing in of the extents from disk for a file in b-tree format.  If the
110  * inode is in b-tree format, then we need to lock the inode exclusively until
111  * the extents are read in.  Locking it exclusively all the time would limit
112  * our parallelism unnecessarily, though.  What we do instead is check to see
113  * if the extents have been read in yet, and only lock the inode exclusively
114  * if they have not.
115  *
116  * The functions return a value which should be given to the corresponding
117  * xfs_iunlock() call.
118  */
119 uint
120 xfs_ilock_data_map_shared(
121 	struct xfs_inode	*ip)
122 {
123 	uint			lock_mode = XFS_ILOCK_SHARED;
124 
125 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
126 	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
127 		lock_mode = XFS_ILOCK_EXCL;
128 	xfs_ilock(ip, lock_mode);
129 	return lock_mode;
130 }
131 
132 uint
133 xfs_ilock_attr_map_shared(
134 	struct xfs_inode	*ip)
135 {
136 	uint			lock_mode = XFS_ILOCK_SHARED;
137 
138 	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
139 	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
140 		lock_mode = XFS_ILOCK_EXCL;
141 	xfs_ilock(ip, lock_mode);
142 	return lock_mode;
143 }
144 
145 /*
146  * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
147  * multi-reader locks: i_mmap_lock and the i_lock.  This routine allows
148  * various combinations of the locks to be obtained.
149  *
150  * The 3 locks should always be ordered so that the IO lock is obtained first,
151  * the mmap lock second and the ilock last in order to prevent deadlock.
152  *
153  * Basic locking order:
154  *
155  * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
156  *
157  * mmap_sem locking order:
158  *
159  * i_rwsem -> page lock -> mmap_sem
160  * mmap_sem -> i_mmap_lock -> page_lock
161  *
162  * The difference in mmap_sem locking order mean that we cannot hold the
163  * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
164  * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
165  * in get_user_pages() to map the user pages into the kernel address space for
166  * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
167  * page faults already hold the mmap_sem.
168  *
169  * Hence to serialise fully against both syscall and mmap based IO, we need to
170  * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
171  * taken in places where we need to invalidate the page cache in a race
172  * free manner (e.g. truncate, hole punch and other extent manipulation
173  * functions).
174  */
175 void
176 xfs_ilock(
177 	xfs_inode_t		*ip,
178 	uint			lock_flags)
179 {
180 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
181 
182 	/*
183 	 * You can't set both SHARED and EXCL for the same lock,
184 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
185 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
186 	 */
187 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
188 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
189 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
190 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
191 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
192 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
193 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
194 
195 	if (lock_flags & XFS_IOLOCK_EXCL) {
196 		down_write_nested(&VFS_I(ip)->i_rwsem,
197 				  XFS_IOLOCK_DEP(lock_flags));
198 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
199 		down_read_nested(&VFS_I(ip)->i_rwsem,
200 				 XFS_IOLOCK_DEP(lock_flags));
201 	}
202 
203 	if (lock_flags & XFS_MMAPLOCK_EXCL)
204 		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
205 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
206 		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
207 
208 	if (lock_flags & XFS_ILOCK_EXCL)
209 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
210 	else if (lock_flags & XFS_ILOCK_SHARED)
211 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 }
213 
214 /*
215  * This is just like xfs_ilock(), except that the caller
216  * is guaranteed not to sleep.  It returns 1 if it gets
217  * the requested locks and 0 otherwise.  If the IO lock is
218  * obtained but the inode lock cannot be, then the IO lock
219  * is dropped before returning.
220  *
221  * ip -- the inode being locked
222  * lock_flags -- this parameter indicates the inode's locks to be
223  *       to be locked.  See the comment for xfs_ilock() for a list
224  *	 of valid values.
225  */
226 int
227 xfs_ilock_nowait(
228 	xfs_inode_t		*ip,
229 	uint			lock_flags)
230 {
231 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
232 
233 	/*
234 	 * You can't set both SHARED and EXCL for the same lock,
235 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
236 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
237 	 */
238 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
239 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
240 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
241 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
242 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
243 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
244 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
245 
246 	if (lock_flags & XFS_IOLOCK_EXCL) {
247 		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
248 			goto out;
249 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
250 		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
251 			goto out;
252 	}
253 
254 	if (lock_flags & XFS_MMAPLOCK_EXCL) {
255 		if (!mrtryupdate(&ip->i_mmaplock))
256 			goto out_undo_iolock;
257 	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
258 		if (!mrtryaccess(&ip->i_mmaplock))
259 			goto out_undo_iolock;
260 	}
261 
262 	if (lock_flags & XFS_ILOCK_EXCL) {
263 		if (!mrtryupdate(&ip->i_lock))
264 			goto out_undo_mmaplock;
265 	} else if (lock_flags & XFS_ILOCK_SHARED) {
266 		if (!mrtryaccess(&ip->i_lock))
267 			goto out_undo_mmaplock;
268 	}
269 	return 1;
270 
271 out_undo_mmaplock:
272 	if (lock_flags & XFS_MMAPLOCK_EXCL)
273 		mrunlock_excl(&ip->i_mmaplock);
274 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
275 		mrunlock_shared(&ip->i_mmaplock);
276 out_undo_iolock:
277 	if (lock_flags & XFS_IOLOCK_EXCL)
278 		up_write(&VFS_I(ip)->i_rwsem);
279 	else if (lock_flags & XFS_IOLOCK_SHARED)
280 		up_read(&VFS_I(ip)->i_rwsem);
281 out:
282 	return 0;
283 }
284 
285 /*
286  * xfs_iunlock() is used to drop the inode locks acquired with
287  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
288  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
289  * that we know which locks to drop.
290  *
291  * ip -- the inode being unlocked
292  * lock_flags -- this parameter indicates the inode's locks to be
293  *       to be unlocked.  See the comment for xfs_ilock() for a list
294  *	 of valid values for this parameter.
295  *
296  */
297 void
298 xfs_iunlock(
299 	xfs_inode_t		*ip,
300 	uint			lock_flags)
301 {
302 	/*
303 	 * You can't set both SHARED and EXCL for the same lock,
304 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
305 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
306 	 */
307 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
308 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
309 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
310 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
311 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
312 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
313 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
314 	ASSERT(lock_flags != 0);
315 
316 	if (lock_flags & XFS_IOLOCK_EXCL)
317 		up_write(&VFS_I(ip)->i_rwsem);
318 	else if (lock_flags & XFS_IOLOCK_SHARED)
319 		up_read(&VFS_I(ip)->i_rwsem);
320 
321 	if (lock_flags & XFS_MMAPLOCK_EXCL)
322 		mrunlock_excl(&ip->i_mmaplock);
323 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
324 		mrunlock_shared(&ip->i_mmaplock);
325 
326 	if (lock_flags & XFS_ILOCK_EXCL)
327 		mrunlock_excl(&ip->i_lock);
328 	else if (lock_flags & XFS_ILOCK_SHARED)
329 		mrunlock_shared(&ip->i_lock);
330 
331 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
332 }
333 
334 /*
335  * give up write locks.  the i/o lock cannot be held nested
336  * if it is being demoted.
337  */
338 void
339 xfs_ilock_demote(
340 	xfs_inode_t		*ip,
341 	uint			lock_flags)
342 {
343 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
344 	ASSERT((lock_flags &
345 		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
346 
347 	if (lock_flags & XFS_ILOCK_EXCL)
348 		mrdemote(&ip->i_lock);
349 	if (lock_flags & XFS_MMAPLOCK_EXCL)
350 		mrdemote(&ip->i_mmaplock);
351 	if (lock_flags & XFS_IOLOCK_EXCL)
352 		downgrade_write(&VFS_I(ip)->i_rwsem);
353 
354 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
355 }
356 
357 #if defined(DEBUG) || defined(XFS_WARN)
358 int
359 xfs_isilocked(
360 	xfs_inode_t		*ip,
361 	uint			lock_flags)
362 {
363 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
364 		if (!(lock_flags & XFS_ILOCK_SHARED))
365 			return !!ip->i_lock.mr_writer;
366 		return rwsem_is_locked(&ip->i_lock.mr_lock);
367 	}
368 
369 	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
370 		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
371 			return !!ip->i_mmaplock.mr_writer;
372 		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
373 	}
374 
375 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
376 		if (!(lock_flags & XFS_IOLOCK_SHARED))
377 			return !debug_locks ||
378 				lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
379 		return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
380 	}
381 
382 	ASSERT(0);
383 	return 0;
384 }
385 #endif
386 
387 #ifdef DEBUG
388 int xfs_locked_n;
389 int xfs_small_retries;
390 int xfs_middle_retries;
391 int xfs_lots_retries;
392 int xfs_lock_delays;
393 #endif
394 
395 /*
396  * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
397  * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
398  * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
399  * errors and warnings.
400  */
401 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
402 static bool
403 xfs_lockdep_subclass_ok(
404 	int subclass)
405 {
406 	return subclass < MAX_LOCKDEP_SUBCLASSES;
407 }
408 #else
409 #define xfs_lockdep_subclass_ok(subclass)	(true)
410 #endif
411 
412 /*
413  * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
414  * value. This can be called for any type of inode lock combination, including
415  * parent locking. Care must be taken to ensure we don't overrun the subclass
416  * storage fields in the class mask we build.
417  */
418 static inline int
419 xfs_lock_inumorder(int lock_mode, int subclass)
420 {
421 	int	class = 0;
422 
423 	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
424 			      XFS_ILOCK_RTSUM)));
425 	ASSERT(xfs_lockdep_subclass_ok(subclass));
426 
427 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
428 		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
429 		class += subclass << XFS_IOLOCK_SHIFT;
430 	}
431 
432 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
433 		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
434 		class += subclass << XFS_MMAPLOCK_SHIFT;
435 	}
436 
437 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
438 		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
439 		class += subclass << XFS_ILOCK_SHIFT;
440 	}
441 
442 	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
443 }
444 
445 /*
446  * The following routine will lock n inodes in exclusive mode.  We assume the
447  * caller calls us with the inodes in i_ino order.
448  *
449  * We need to detect deadlock where an inode that we lock is in the AIL and we
450  * start waiting for another inode that is locked by a thread in a long running
451  * transaction (such as truncate). This can result in deadlock since the long
452  * running trans might need to wait for the inode we just locked in order to
453  * push the tail and free space in the log.
454  *
455  * xfs_lock_inodes() can only be used to lock one type of lock at a time -
456  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
457  * lock more than one at a time, lockdep will report false positives saying we
458  * have violated locking orders.
459  */
460 static void
461 xfs_lock_inodes(
462 	xfs_inode_t	**ips,
463 	int		inodes,
464 	uint		lock_mode)
465 {
466 	int		attempts = 0, i, j, try_lock;
467 	xfs_log_item_t	*lp;
468 
469 	/*
470 	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
471 	 * support an arbitrary depth of locking here, but absolute limits on
472 	 * inodes depend on the the type of locking and the limits placed by
473 	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
474 	 * the asserts.
475 	 */
476 	ASSERT(ips && inodes >= 2 && inodes <= 5);
477 	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
478 			    XFS_ILOCK_EXCL));
479 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
480 			      XFS_ILOCK_SHARED)));
481 	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
482 		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
483 	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
484 		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
485 
486 	if (lock_mode & XFS_IOLOCK_EXCL) {
487 		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
488 	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
489 		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
490 
491 	try_lock = 0;
492 	i = 0;
493 again:
494 	for (; i < inodes; i++) {
495 		ASSERT(ips[i]);
496 
497 		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
498 			continue;
499 
500 		/*
501 		 * If try_lock is not set yet, make sure all locked inodes are
502 		 * not in the AIL.  If any are, set try_lock to be used later.
503 		 */
504 		if (!try_lock) {
505 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
506 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
507 				if (lp && (lp->li_flags & XFS_LI_IN_AIL))
508 					try_lock++;
509 			}
510 		}
511 
512 		/*
513 		 * If any of the previous locks we have locked is in the AIL,
514 		 * we must TRY to get the second and subsequent locks. If
515 		 * we can't get any, we must release all we have
516 		 * and try again.
517 		 */
518 		if (!try_lock) {
519 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
520 			continue;
521 		}
522 
523 		/* try_lock means we have an inode locked that is in the AIL. */
524 		ASSERT(i != 0);
525 		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
526 			continue;
527 
528 		/*
529 		 * Unlock all previous guys and try again.  xfs_iunlock will try
530 		 * to push the tail if the inode is in the AIL.
531 		 */
532 		attempts++;
533 		for (j = i - 1; j >= 0; j--) {
534 			/*
535 			 * Check to see if we've already unlocked this one.  Not
536 			 * the first one going back, and the inode ptr is the
537 			 * same.
538 			 */
539 			if (j != (i - 1) && ips[j] == ips[j + 1])
540 				continue;
541 
542 			xfs_iunlock(ips[j], lock_mode);
543 		}
544 
545 		if ((attempts % 5) == 0) {
546 			delay(1); /* Don't just spin the CPU */
547 #ifdef DEBUG
548 			xfs_lock_delays++;
549 #endif
550 		}
551 		i = 0;
552 		try_lock = 0;
553 		goto again;
554 	}
555 
556 #ifdef DEBUG
557 	if (attempts) {
558 		if (attempts < 5) xfs_small_retries++;
559 		else if (attempts < 100) xfs_middle_retries++;
560 		else xfs_lots_retries++;
561 	} else {
562 		xfs_locked_n++;
563 	}
564 #endif
565 }
566 
567 /*
568  * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
569  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
570  * lock more than one at a time, lockdep will report false positives saying we
571  * have violated locking orders.
572  */
573 void
574 xfs_lock_two_inodes(
575 	xfs_inode_t		*ip0,
576 	xfs_inode_t		*ip1,
577 	uint			lock_mode)
578 {
579 	xfs_inode_t		*temp;
580 	int			attempts = 0;
581 	xfs_log_item_t		*lp;
582 
583 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
584 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
585 		ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
586 
587 	ASSERT(ip0->i_ino != ip1->i_ino);
588 
589 	if (ip0->i_ino > ip1->i_ino) {
590 		temp = ip0;
591 		ip0 = ip1;
592 		ip1 = temp;
593 	}
594 
595  again:
596 	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
597 
598 	/*
599 	 * If the first lock we have locked is in the AIL, we must TRY to get
600 	 * the second lock. If we can't get it, we must release the first one
601 	 * and try again.
602 	 */
603 	lp = (xfs_log_item_t *)ip0->i_itemp;
604 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
605 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
606 			xfs_iunlock(ip0, lock_mode);
607 			if ((++attempts % 5) == 0)
608 				delay(1); /* Don't just spin the CPU */
609 			goto again;
610 		}
611 	} else {
612 		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
613 	}
614 }
615 
616 
617 void
618 __xfs_iflock(
619 	struct xfs_inode	*ip)
620 {
621 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
622 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
623 
624 	do {
625 		prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
626 		if (xfs_isiflocked(ip))
627 			io_schedule();
628 	} while (!xfs_iflock_nowait(ip));
629 
630 	finish_wait(wq, &wait.wq_entry);
631 }
632 
633 STATIC uint
634 _xfs_dic2xflags(
635 	uint16_t		di_flags,
636 	uint64_t		di_flags2,
637 	bool			has_attr)
638 {
639 	uint			flags = 0;
640 
641 	if (di_flags & XFS_DIFLAG_ANY) {
642 		if (di_flags & XFS_DIFLAG_REALTIME)
643 			flags |= FS_XFLAG_REALTIME;
644 		if (di_flags & XFS_DIFLAG_PREALLOC)
645 			flags |= FS_XFLAG_PREALLOC;
646 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
647 			flags |= FS_XFLAG_IMMUTABLE;
648 		if (di_flags & XFS_DIFLAG_APPEND)
649 			flags |= FS_XFLAG_APPEND;
650 		if (di_flags & XFS_DIFLAG_SYNC)
651 			flags |= FS_XFLAG_SYNC;
652 		if (di_flags & XFS_DIFLAG_NOATIME)
653 			flags |= FS_XFLAG_NOATIME;
654 		if (di_flags & XFS_DIFLAG_NODUMP)
655 			flags |= FS_XFLAG_NODUMP;
656 		if (di_flags & XFS_DIFLAG_RTINHERIT)
657 			flags |= FS_XFLAG_RTINHERIT;
658 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
659 			flags |= FS_XFLAG_PROJINHERIT;
660 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
661 			flags |= FS_XFLAG_NOSYMLINKS;
662 		if (di_flags & XFS_DIFLAG_EXTSIZE)
663 			flags |= FS_XFLAG_EXTSIZE;
664 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
665 			flags |= FS_XFLAG_EXTSZINHERIT;
666 		if (di_flags & XFS_DIFLAG_NODEFRAG)
667 			flags |= FS_XFLAG_NODEFRAG;
668 		if (di_flags & XFS_DIFLAG_FILESTREAM)
669 			flags |= FS_XFLAG_FILESTREAM;
670 	}
671 
672 	if (di_flags2 & XFS_DIFLAG2_ANY) {
673 		if (di_flags2 & XFS_DIFLAG2_DAX)
674 			flags |= FS_XFLAG_DAX;
675 		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
676 			flags |= FS_XFLAG_COWEXTSIZE;
677 	}
678 
679 	if (has_attr)
680 		flags |= FS_XFLAG_HASATTR;
681 
682 	return flags;
683 }
684 
685 uint
686 xfs_ip2xflags(
687 	struct xfs_inode	*ip)
688 {
689 	struct xfs_icdinode	*dic = &ip->i_d;
690 
691 	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
692 }
693 
694 /*
695  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
696  * is allowed, otherwise it has to be an exact match. If a CI match is found,
697  * ci_name->name will point to a the actual name (caller must free) or
698  * will be set to NULL if an exact match is found.
699  */
700 int
701 xfs_lookup(
702 	xfs_inode_t		*dp,
703 	struct xfs_name		*name,
704 	xfs_inode_t		**ipp,
705 	struct xfs_name		*ci_name)
706 {
707 	xfs_ino_t		inum;
708 	int			error;
709 
710 	trace_xfs_lookup(dp, name);
711 
712 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
713 		return -EIO;
714 
715 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
716 	if (error)
717 		goto out_unlock;
718 
719 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
720 	if (error)
721 		goto out_free_name;
722 
723 	return 0;
724 
725 out_free_name:
726 	if (ci_name)
727 		kmem_free(ci_name->name);
728 out_unlock:
729 	*ipp = NULL;
730 	return error;
731 }
732 
733 /*
734  * Allocate an inode on disk and return a copy of its in-core version.
735  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
736  * appropriately within the inode.  The uid and gid for the inode are
737  * set according to the contents of the given cred structure.
738  *
739  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
740  * has a free inode available, call xfs_iget() to obtain the in-core
741  * version of the allocated inode.  Finally, fill in the inode and
742  * log its initial contents.  In this case, ialloc_context would be
743  * set to NULL.
744  *
745  * If xfs_dialloc() does not have an available inode, it will replenish
746  * its supply by doing an allocation. Since we can only do one
747  * allocation within a transaction without deadlocks, we must commit
748  * the current transaction before returning the inode itself.
749  * In this case, therefore, we will set ialloc_context and return.
750  * The caller should then commit the current transaction, start a new
751  * transaction, and call xfs_ialloc() again to actually get the inode.
752  *
753  * To ensure that some other process does not grab the inode that
754  * was allocated during the first call to xfs_ialloc(), this routine
755  * also returns the [locked] bp pointing to the head of the freelist
756  * as ialloc_context.  The caller should hold this buffer across
757  * the commit and pass it back into this routine on the second call.
758  *
759  * If we are allocating quota inodes, we do not have a parent inode
760  * to attach to or associate with (i.e. pip == NULL) because they
761  * are not linked into the directory structure - they are attached
762  * directly to the superblock - and so have no parent.
763  */
764 static int
765 xfs_ialloc(
766 	xfs_trans_t	*tp,
767 	xfs_inode_t	*pip,
768 	umode_t		mode,
769 	xfs_nlink_t	nlink,
770 	xfs_dev_t	rdev,
771 	prid_t		prid,
772 	int		okalloc,
773 	xfs_buf_t	**ialloc_context,
774 	xfs_inode_t	**ipp)
775 {
776 	struct xfs_mount *mp = tp->t_mountp;
777 	xfs_ino_t	ino;
778 	xfs_inode_t	*ip;
779 	uint		flags;
780 	int		error;
781 	struct timespec	tv;
782 	struct inode	*inode;
783 
784 	/*
785 	 * Call the space management code to pick
786 	 * the on-disk inode to be allocated.
787 	 */
788 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
789 			    ialloc_context, &ino);
790 	if (error)
791 		return error;
792 	if (*ialloc_context || ino == NULLFSINO) {
793 		*ipp = NULL;
794 		return 0;
795 	}
796 	ASSERT(*ialloc_context == NULL);
797 
798 	/*
799 	 * Get the in-core inode with the lock held exclusively.
800 	 * This is because we're setting fields here we need
801 	 * to prevent others from looking at until we're done.
802 	 */
803 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
804 			 XFS_ILOCK_EXCL, &ip);
805 	if (error)
806 		return error;
807 	ASSERT(ip != NULL);
808 	inode = VFS_I(ip);
809 
810 	/*
811 	 * We always convert v1 inodes to v2 now - we only support filesystems
812 	 * with >= v2 inode capability, so there is no reason for ever leaving
813 	 * an inode in v1 format.
814 	 */
815 	if (ip->i_d.di_version == 1)
816 		ip->i_d.di_version = 2;
817 
818 	inode->i_mode = mode;
819 	set_nlink(inode, nlink);
820 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
821 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
822 	xfs_set_projid(ip, prid);
823 
824 	if (pip && XFS_INHERIT_GID(pip)) {
825 		ip->i_d.di_gid = pip->i_d.di_gid;
826 		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
827 			inode->i_mode |= S_ISGID;
828 	}
829 
830 	/*
831 	 * If the group ID of the new file does not match the effective group
832 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
833 	 * (and only if the irix_sgid_inherit compatibility variable is set).
834 	 */
835 	if ((irix_sgid_inherit) &&
836 	    (inode->i_mode & S_ISGID) &&
837 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
838 		inode->i_mode &= ~S_ISGID;
839 
840 	ip->i_d.di_size = 0;
841 	ip->i_d.di_nextents = 0;
842 	ASSERT(ip->i_d.di_nblocks == 0);
843 
844 	tv = current_time(inode);
845 	inode->i_mtime = tv;
846 	inode->i_atime = tv;
847 	inode->i_ctime = tv;
848 
849 	ip->i_d.di_extsize = 0;
850 	ip->i_d.di_dmevmask = 0;
851 	ip->i_d.di_dmstate = 0;
852 	ip->i_d.di_flags = 0;
853 
854 	if (ip->i_d.di_version == 3) {
855 		inode->i_version = 1;
856 		ip->i_d.di_flags2 = 0;
857 		ip->i_d.di_cowextsize = 0;
858 		ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
859 		ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
860 	}
861 
862 
863 	flags = XFS_ILOG_CORE;
864 	switch (mode & S_IFMT) {
865 	case S_IFIFO:
866 	case S_IFCHR:
867 	case S_IFBLK:
868 	case S_IFSOCK:
869 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
870 		ip->i_df.if_u2.if_rdev = rdev;
871 		ip->i_df.if_flags = 0;
872 		flags |= XFS_ILOG_DEV;
873 		break;
874 	case S_IFREG:
875 	case S_IFDIR:
876 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
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 
914 			ip->i_d.di_flags |= di_flags;
915 		}
916 		if (pip &&
917 		    (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
918 		    pip->i_d.di_version == 3 &&
919 		    ip->i_d.di_version == 3) {
920 			uint64_t	di_flags2 = 0;
921 
922 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
923 				di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
924 				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
925 			}
926 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
927 				di_flags2 |= XFS_DIFLAG2_DAX;
928 
929 			ip->i_d.di_flags2 |= di_flags2;
930 		}
931 		/* FALLTHROUGH */
932 	case S_IFLNK:
933 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
934 		ip->i_df.if_flags = XFS_IFEXTENTS;
935 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
936 		ip->i_df.if_u1.if_extents = NULL;
937 		break;
938 	default:
939 		ASSERT(0);
940 	}
941 	/*
942 	 * Attribute fork settings for new inode.
943 	 */
944 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
945 	ip->i_d.di_anextents = 0;
946 
947 	/*
948 	 * Log the new values stuffed into the inode.
949 	 */
950 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
951 	xfs_trans_log_inode(tp, ip, flags);
952 
953 	/* now that we have an i_mode we can setup the inode structure */
954 	xfs_setup_inode(ip);
955 
956 	*ipp = ip;
957 	return 0;
958 }
959 
960 /*
961  * Allocates a new inode from disk and return a pointer to the
962  * incore copy. This routine will internally commit the current
963  * transaction and allocate a new one if the Space Manager needed
964  * to do an allocation to replenish the inode free-list.
965  *
966  * This routine is designed to be called from xfs_create and
967  * xfs_create_dir.
968  *
969  */
970 int
971 xfs_dir_ialloc(
972 	xfs_trans_t	**tpp,		/* input: current transaction;
973 					   output: may be a new transaction. */
974 	xfs_inode_t	*dp,		/* directory within whose allocate
975 					   the inode. */
976 	umode_t		mode,
977 	xfs_nlink_t	nlink,
978 	xfs_dev_t	rdev,
979 	prid_t		prid,		/* project id */
980 	int		okalloc,	/* ok to allocate new space */
981 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
982 					   locked. */
983 	int		*committed)
984 
985 {
986 	xfs_trans_t	*tp;
987 	xfs_inode_t	*ip;
988 	xfs_buf_t	*ialloc_context = NULL;
989 	int		code;
990 	void		*dqinfo;
991 	uint		tflags;
992 
993 	tp = *tpp;
994 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
995 
996 	/*
997 	 * xfs_ialloc will return a pointer to an incore inode if
998 	 * the Space Manager has an available inode on the free
999 	 * list. Otherwise, it will do an allocation and replenish
1000 	 * the freelist.  Since we can only do one allocation per
1001 	 * transaction without deadlocks, we will need to commit the
1002 	 * current transaction and start a new one.  We will then
1003 	 * need to call xfs_ialloc again to get the inode.
1004 	 *
1005 	 * If xfs_ialloc did an allocation to replenish the freelist,
1006 	 * it returns the bp containing the head of the freelist as
1007 	 * ialloc_context. We will hold a lock on it across the
1008 	 * transaction commit so that no other process can steal
1009 	 * the inode(s) that we've just allocated.
1010 	 */
1011 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1012 			  &ialloc_context, &ip);
1013 
1014 	/*
1015 	 * Return an error if we were unable to allocate a new inode.
1016 	 * This should only happen if we run out of space on disk or
1017 	 * encounter a disk error.
1018 	 */
1019 	if (code) {
1020 		*ipp = NULL;
1021 		return code;
1022 	}
1023 	if (!ialloc_context && !ip) {
1024 		*ipp = NULL;
1025 		return -ENOSPC;
1026 	}
1027 
1028 	/*
1029 	 * If the AGI buffer is non-NULL, then we were unable to get an
1030 	 * inode in one operation.  We need to commit the current
1031 	 * transaction and call xfs_ialloc() again.  It is guaranteed
1032 	 * to succeed the second time.
1033 	 */
1034 	if (ialloc_context) {
1035 		/*
1036 		 * Normally, xfs_trans_commit releases all the locks.
1037 		 * We call bhold to hang on to the ialloc_context across
1038 		 * the commit.  Holding this buffer prevents any other
1039 		 * processes from doing any allocations in this
1040 		 * allocation group.
1041 		 */
1042 		xfs_trans_bhold(tp, ialloc_context);
1043 
1044 		/*
1045 		 * We want the quota changes to be associated with the next
1046 		 * transaction, NOT this one. So, detach the dqinfo from this
1047 		 * and attach it to the next transaction.
1048 		 */
1049 		dqinfo = NULL;
1050 		tflags = 0;
1051 		if (tp->t_dqinfo) {
1052 			dqinfo = (void *)tp->t_dqinfo;
1053 			tp->t_dqinfo = NULL;
1054 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1055 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1056 		}
1057 
1058 		code = xfs_trans_roll(&tp);
1059 		if (committed != NULL)
1060 			*committed = 1;
1061 
1062 		/*
1063 		 * Re-attach the quota info that we detached from prev trx.
1064 		 */
1065 		if (dqinfo) {
1066 			tp->t_dqinfo = dqinfo;
1067 			tp->t_flags |= tflags;
1068 		}
1069 
1070 		if (code) {
1071 			xfs_buf_relse(ialloc_context);
1072 			*tpp = tp;
1073 			*ipp = NULL;
1074 			return code;
1075 		}
1076 		xfs_trans_bjoin(tp, ialloc_context);
1077 
1078 		/*
1079 		 * Call ialloc again. Since we've locked out all
1080 		 * other allocations in this allocation group,
1081 		 * this call should always succeed.
1082 		 */
1083 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1084 				  okalloc, &ialloc_context, &ip);
1085 
1086 		/*
1087 		 * If we get an error at this point, return to the caller
1088 		 * so that the current transaction can be aborted.
1089 		 */
1090 		if (code) {
1091 			*tpp = tp;
1092 			*ipp = NULL;
1093 			return code;
1094 		}
1095 		ASSERT(!ialloc_context && ip);
1096 
1097 	} else {
1098 		if (committed != NULL)
1099 			*committed = 0;
1100 	}
1101 
1102 	*ipp = ip;
1103 	*tpp = tp;
1104 
1105 	return 0;
1106 }
1107 
1108 /*
1109  * Decrement the link count on an inode & log the change.  If this causes the
1110  * link count to go to zero, move the inode to AGI unlinked list so that it can
1111  * be freed when the last active reference goes away via xfs_inactive().
1112  */
1113 static int			/* error */
1114 xfs_droplink(
1115 	xfs_trans_t *tp,
1116 	xfs_inode_t *ip)
1117 {
1118 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1119 
1120 	drop_nlink(VFS_I(ip));
1121 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1122 
1123 	if (VFS_I(ip)->i_nlink)
1124 		return 0;
1125 
1126 	return xfs_iunlink(tp, ip);
1127 }
1128 
1129 /*
1130  * Increment the link count on an inode & log the change.
1131  */
1132 static int
1133 xfs_bumplink(
1134 	xfs_trans_t *tp,
1135 	xfs_inode_t *ip)
1136 {
1137 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1138 
1139 	ASSERT(ip->i_d.di_version > 1);
1140 	inc_nlink(VFS_I(ip));
1141 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1142 	return 0;
1143 }
1144 
1145 int
1146 xfs_create(
1147 	xfs_inode_t		*dp,
1148 	struct xfs_name		*name,
1149 	umode_t			mode,
1150 	xfs_dev_t		rdev,
1151 	xfs_inode_t		**ipp)
1152 {
1153 	int			is_dir = S_ISDIR(mode);
1154 	struct xfs_mount	*mp = dp->i_mount;
1155 	struct xfs_inode	*ip = NULL;
1156 	struct xfs_trans	*tp = NULL;
1157 	int			error;
1158 	struct xfs_defer_ops	dfops;
1159 	xfs_fsblock_t		first_block;
1160 	bool                    unlock_dp_on_error = false;
1161 	prid_t			prid;
1162 	struct xfs_dquot	*udqp = NULL;
1163 	struct xfs_dquot	*gdqp = NULL;
1164 	struct xfs_dquot	*pdqp = NULL;
1165 	struct xfs_trans_res	*tres;
1166 	uint			resblks;
1167 
1168 	trace_xfs_create(dp, name);
1169 
1170 	if (XFS_FORCED_SHUTDOWN(mp))
1171 		return -EIO;
1172 
1173 	prid = xfs_get_initial_prid(dp);
1174 
1175 	/*
1176 	 * Make sure that we have allocated dquot(s) on disk.
1177 	 */
1178 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1179 					xfs_kgid_to_gid(current_fsgid()), prid,
1180 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1181 					&udqp, &gdqp, &pdqp);
1182 	if (error)
1183 		return error;
1184 
1185 	if (is_dir) {
1186 		rdev = 0;
1187 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1188 		tres = &M_RES(mp)->tr_mkdir;
1189 	} else {
1190 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1191 		tres = &M_RES(mp)->tr_create;
1192 	}
1193 
1194 	/*
1195 	 * Initially assume that the file does not exist and
1196 	 * reserve the resources for that case.  If that is not
1197 	 * the case we'll drop the one we have and get a more
1198 	 * appropriate transaction later.
1199 	 */
1200 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1201 	if (error == -ENOSPC) {
1202 		/* flush outstanding delalloc blocks and retry */
1203 		xfs_flush_inodes(mp);
1204 		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1205 	}
1206 	if (error == -ENOSPC) {
1207 		/* No space at all so try a "no-allocation" reservation */
1208 		resblks = 0;
1209 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1210 	}
1211 	if (error)
1212 		goto out_release_inode;
1213 
1214 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1215 	unlock_dp_on_error = true;
1216 
1217 	xfs_defer_init(&dfops, &first_block);
1218 
1219 	/*
1220 	 * Reserve disk quota and the inode.
1221 	 */
1222 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1223 						pdqp, resblks, 1, 0);
1224 	if (error)
1225 		goto out_trans_cancel;
1226 
1227 	if (!resblks) {
1228 		error = xfs_dir_canenter(tp, dp, name);
1229 		if (error)
1230 			goto out_trans_cancel;
1231 	}
1232 
1233 	/*
1234 	 * A newly created regular or special file just has one directory
1235 	 * entry pointing to them, but a directory also the "." entry
1236 	 * pointing to itself.
1237 	 */
1238 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1239 			       prid, resblks > 0, &ip, NULL);
1240 	if (error)
1241 		goto out_trans_cancel;
1242 
1243 	/*
1244 	 * Now we join the directory inode to the transaction.  We do not do it
1245 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1246 	 * (and release all the locks).  An error from here on will result in
1247 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1248 	 * error path.
1249 	 */
1250 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1251 	unlock_dp_on_error = false;
1252 
1253 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1254 					&first_block, &dfops, resblks ?
1255 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1256 	if (error) {
1257 		ASSERT(error != -ENOSPC);
1258 		goto out_trans_cancel;
1259 	}
1260 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1261 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1262 
1263 	if (is_dir) {
1264 		error = xfs_dir_init(tp, ip, dp);
1265 		if (error)
1266 			goto out_bmap_cancel;
1267 
1268 		error = xfs_bumplink(tp, dp);
1269 		if (error)
1270 			goto out_bmap_cancel;
1271 	}
1272 
1273 	/*
1274 	 * If this is a synchronous mount, make sure that the
1275 	 * create transaction goes to disk before returning to
1276 	 * the user.
1277 	 */
1278 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1279 		xfs_trans_set_sync(tp);
1280 
1281 	/*
1282 	 * Attach the dquot(s) to the inodes and modify them incore.
1283 	 * These ids of the inode couldn't have changed since the new
1284 	 * inode has been locked ever since it was created.
1285 	 */
1286 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1287 
1288 	error = xfs_defer_finish(&tp, &dfops);
1289 	if (error)
1290 		goto out_bmap_cancel;
1291 
1292 	error = xfs_trans_commit(tp);
1293 	if (error)
1294 		goto out_release_inode;
1295 
1296 	xfs_qm_dqrele(udqp);
1297 	xfs_qm_dqrele(gdqp);
1298 	xfs_qm_dqrele(pdqp);
1299 
1300 	*ipp = ip;
1301 	return 0;
1302 
1303  out_bmap_cancel:
1304 	xfs_defer_cancel(&dfops);
1305  out_trans_cancel:
1306 	xfs_trans_cancel(tp);
1307  out_release_inode:
1308 	/*
1309 	 * Wait until after the current transaction is aborted to finish the
1310 	 * setup of the inode and release the inode.  This prevents recursive
1311 	 * transactions and deadlocks from xfs_inactive.
1312 	 */
1313 	if (ip) {
1314 		xfs_finish_inode_setup(ip);
1315 		IRELE(ip);
1316 	}
1317 
1318 	xfs_qm_dqrele(udqp);
1319 	xfs_qm_dqrele(gdqp);
1320 	xfs_qm_dqrele(pdqp);
1321 
1322 	if (unlock_dp_on_error)
1323 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1324 	return error;
1325 }
1326 
1327 int
1328 xfs_create_tmpfile(
1329 	struct xfs_inode	*dp,
1330 	struct dentry		*dentry,
1331 	umode_t			mode,
1332 	struct xfs_inode	**ipp)
1333 {
1334 	struct xfs_mount	*mp = dp->i_mount;
1335 	struct xfs_inode	*ip = NULL;
1336 	struct xfs_trans	*tp = NULL;
1337 	int			error;
1338 	prid_t                  prid;
1339 	struct xfs_dquot	*udqp = NULL;
1340 	struct xfs_dquot	*gdqp = NULL;
1341 	struct xfs_dquot	*pdqp = NULL;
1342 	struct xfs_trans_res	*tres;
1343 	uint			resblks;
1344 
1345 	if (XFS_FORCED_SHUTDOWN(mp))
1346 		return -EIO;
1347 
1348 	prid = xfs_get_initial_prid(dp);
1349 
1350 	/*
1351 	 * Make sure that we have allocated dquot(s) on disk.
1352 	 */
1353 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1354 				xfs_kgid_to_gid(current_fsgid()), prid,
1355 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1356 				&udqp, &gdqp, &pdqp);
1357 	if (error)
1358 		return error;
1359 
1360 	resblks = XFS_IALLOC_SPACE_RES(mp);
1361 	tres = &M_RES(mp)->tr_create_tmpfile;
1362 
1363 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1364 	if (error == -ENOSPC) {
1365 		/* No space at all so try a "no-allocation" reservation */
1366 		resblks = 0;
1367 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1368 	}
1369 	if (error)
1370 		goto out_release_inode;
1371 
1372 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1373 						pdqp, resblks, 1, 0);
1374 	if (error)
1375 		goto out_trans_cancel;
1376 
1377 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1378 				prid, resblks > 0, &ip, NULL);
1379 	if (error)
1380 		goto out_trans_cancel;
1381 
1382 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1383 		xfs_trans_set_sync(tp);
1384 
1385 	/*
1386 	 * Attach the dquot(s) to the inodes and modify them incore.
1387 	 * These ids of the inode couldn't have changed since the new
1388 	 * inode has been locked ever since it was created.
1389 	 */
1390 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1391 
1392 	error = xfs_iunlink(tp, ip);
1393 	if (error)
1394 		goto out_trans_cancel;
1395 
1396 	error = xfs_trans_commit(tp);
1397 	if (error)
1398 		goto out_release_inode;
1399 
1400 	xfs_qm_dqrele(udqp);
1401 	xfs_qm_dqrele(gdqp);
1402 	xfs_qm_dqrele(pdqp);
1403 
1404 	*ipp = ip;
1405 	return 0;
1406 
1407  out_trans_cancel:
1408 	xfs_trans_cancel(tp);
1409  out_release_inode:
1410 	/*
1411 	 * Wait until after the current transaction is aborted to finish the
1412 	 * setup of the inode and release the inode.  This prevents recursive
1413 	 * transactions and deadlocks from xfs_inactive.
1414 	 */
1415 	if (ip) {
1416 		xfs_finish_inode_setup(ip);
1417 		IRELE(ip);
1418 	}
1419 
1420 	xfs_qm_dqrele(udqp);
1421 	xfs_qm_dqrele(gdqp);
1422 	xfs_qm_dqrele(pdqp);
1423 
1424 	return error;
1425 }
1426 
1427 int
1428 xfs_link(
1429 	xfs_inode_t		*tdp,
1430 	xfs_inode_t		*sip,
1431 	struct xfs_name		*target_name)
1432 {
1433 	xfs_mount_t		*mp = tdp->i_mount;
1434 	xfs_trans_t		*tp;
1435 	int			error;
1436 	struct xfs_defer_ops	dfops;
1437 	xfs_fsblock_t           first_block;
1438 	int			resblks;
1439 
1440 	trace_xfs_link(tdp, target_name);
1441 
1442 	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1443 
1444 	if (XFS_FORCED_SHUTDOWN(mp))
1445 		return -EIO;
1446 
1447 	error = xfs_qm_dqattach(sip, 0);
1448 	if (error)
1449 		goto std_return;
1450 
1451 	error = xfs_qm_dqattach(tdp, 0);
1452 	if (error)
1453 		goto std_return;
1454 
1455 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1456 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1457 	if (error == -ENOSPC) {
1458 		resblks = 0;
1459 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1460 	}
1461 	if (error)
1462 		goto std_return;
1463 
1464 	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1465 
1466 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1467 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1468 
1469 	/*
1470 	 * If we are using project inheritance, we only allow hard link
1471 	 * creation in our tree when the project IDs are the same; else
1472 	 * the tree quota mechanism could be circumvented.
1473 	 */
1474 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1475 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1476 		error = -EXDEV;
1477 		goto error_return;
1478 	}
1479 
1480 	if (!resblks) {
1481 		error = xfs_dir_canenter(tp, tdp, target_name);
1482 		if (error)
1483 			goto error_return;
1484 	}
1485 
1486 	xfs_defer_init(&dfops, &first_block);
1487 
1488 	/*
1489 	 * Handle initial link state of O_TMPFILE inode
1490 	 */
1491 	if (VFS_I(sip)->i_nlink == 0) {
1492 		error = xfs_iunlink_remove(tp, sip);
1493 		if (error)
1494 			goto error_return;
1495 	}
1496 
1497 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1498 					&first_block, &dfops, resblks);
1499 	if (error)
1500 		goto error_return;
1501 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1502 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1503 
1504 	error = xfs_bumplink(tp, sip);
1505 	if (error)
1506 		goto error_return;
1507 
1508 	/*
1509 	 * If this is a synchronous mount, make sure that the
1510 	 * link transaction goes to disk before returning to
1511 	 * the user.
1512 	 */
1513 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1514 		xfs_trans_set_sync(tp);
1515 
1516 	error = xfs_defer_finish(&tp, &dfops);
1517 	if (error) {
1518 		xfs_defer_cancel(&dfops);
1519 		goto error_return;
1520 	}
1521 
1522 	return xfs_trans_commit(tp);
1523 
1524  error_return:
1525 	xfs_trans_cancel(tp);
1526  std_return:
1527 	return error;
1528 }
1529 
1530 /*
1531  * Free up the underlying blocks past new_size.  The new size must be smaller
1532  * than the current size.  This routine can be used both for the attribute and
1533  * data fork, and does not modify the inode size, which is left to the caller.
1534  *
1535  * The transaction passed to this routine must have made a permanent log
1536  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1537  * given transaction and start new ones, so make sure everything involved in
1538  * the transaction is tidy before calling here.  Some transaction will be
1539  * returned to the caller to be committed.  The incoming transaction must
1540  * already include the inode, and both inode locks must be held exclusively.
1541  * The inode must also be "held" within the transaction.  On return the inode
1542  * will be "held" within the returned transaction.  This routine does NOT
1543  * require any disk space to be reserved for it within the transaction.
1544  *
1545  * If we get an error, we must return with the inode locked and linked into the
1546  * current transaction. This keeps things simple for the higher level code,
1547  * because it always knows that the inode is locked and held in the transaction
1548  * that returns to it whether errors occur or not.  We don't mark the inode
1549  * dirty on error so that transactions can be easily aborted if possible.
1550  */
1551 int
1552 xfs_itruncate_extents(
1553 	struct xfs_trans	**tpp,
1554 	struct xfs_inode	*ip,
1555 	int			whichfork,
1556 	xfs_fsize_t		new_size)
1557 {
1558 	struct xfs_mount	*mp = ip->i_mount;
1559 	struct xfs_trans	*tp = *tpp;
1560 	struct xfs_defer_ops	dfops;
1561 	xfs_fsblock_t		first_block;
1562 	xfs_fileoff_t		first_unmap_block;
1563 	xfs_fileoff_t		last_block;
1564 	xfs_filblks_t		unmap_len;
1565 	int			error = 0;
1566 	int			done = 0;
1567 
1568 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1569 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1570 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1571 	ASSERT(new_size <= XFS_ISIZE(ip));
1572 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1573 	ASSERT(ip->i_itemp != NULL);
1574 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1575 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1576 
1577 	trace_xfs_itruncate_extents_start(ip, new_size);
1578 
1579 	/*
1580 	 * Since it is possible for space to become allocated beyond
1581 	 * the end of the file (in a crash where the space is allocated
1582 	 * but the inode size is not yet updated), simply remove any
1583 	 * blocks which show up between the new EOF and the maximum
1584 	 * possible file size.  If the first block to be removed is
1585 	 * beyond the maximum file size (ie it is the same as last_block),
1586 	 * then there is nothing to do.
1587 	 */
1588 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1589 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1590 	if (first_unmap_block == last_block)
1591 		return 0;
1592 
1593 	ASSERT(first_unmap_block < last_block);
1594 	unmap_len = last_block - first_unmap_block + 1;
1595 	while (!done) {
1596 		xfs_defer_init(&dfops, &first_block);
1597 		error = xfs_bunmapi(tp, ip,
1598 				    first_unmap_block, unmap_len,
1599 				    xfs_bmapi_aflag(whichfork),
1600 				    XFS_ITRUNC_MAX_EXTENTS,
1601 				    &first_block, &dfops,
1602 				    &done);
1603 		if (error)
1604 			goto out_bmap_cancel;
1605 
1606 		/*
1607 		 * Duplicate the transaction that has the permanent
1608 		 * reservation and commit the old transaction.
1609 		 */
1610 		xfs_defer_ijoin(&dfops, ip);
1611 		error = xfs_defer_finish(&tp, &dfops);
1612 		if (error)
1613 			goto out_bmap_cancel;
1614 
1615 		error = xfs_trans_roll_inode(&tp, ip);
1616 		if (error)
1617 			goto out;
1618 	}
1619 
1620 	/* Remove all pending CoW reservations. */
1621 	error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1622 			last_block, true);
1623 	if (error)
1624 		goto out;
1625 
1626 	/*
1627 	 * Clear the reflink flag if there are no data fork blocks and
1628 	 * there are no extents staged in the cow fork.
1629 	 */
1630 	if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) {
1631 		if (ip->i_d.di_nblocks == 0)
1632 			ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1633 		xfs_inode_clear_cowblocks_tag(ip);
1634 	}
1635 
1636 	/*
1637 	 * Always re-log the inode so that our permanent transaction can keep
1638 	 * on rolling it forward in the log.
1639 	 */
1640 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1641 
1642 	trace_xfs_itruncate_extents_end(ip, new_size);
1643 
1644 out:
1645 	*tpp = tp;
1646 	return error;
1647 out_bmap_cancel:
1648 	/*
1649 	 * If the bunmapi call encounters an error, return to the caller where
1650 	 * the transaction can be properly aborted.  We just need to make sure
1651 	 * we're not holding any resources that we were not when we came in.
1652 	 */
1653 	xfs_defer_cancel(&dfops);
1654 	goto out;
1655 }
1656 
1657 int
1658 xfs_release(
1659 	xfs_inode_t	*ip)
1660 {
1661 	xfs_mount_t	*mp = ip->i_mount;
1662 	int		error;
1663 
1664 	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1665 		return 0;
1666 
1667 	/* If this is a read-only mount, don't do this (would generate I/O) */
1668 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1669 		return 0;
1670 
1671 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1672 		int truncated;
1673 
1674 		/*
1675 		 * If we previously truncated this file and removed old data
1676 		 * in the process, we want to initiate "early" writeout on
1677 		 * the last close.  This is an attempt to combat the notorious
1678 		 * NULL files problem which is particularly noticeable from a
1679 		 * truncate down, buffered (re-)write (delalloc), followed by
1680 		 * a crash.  What we are effectively doing here is
1681 		 * significantly reducing the time window where we'd otherwise
1682 		 * be exposed to that problem.
1683 		 */
1684 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1685 		if (truncated) {
1686 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1687 			if (ip->i_delayed_blks > 0) {
1688 				error = filemap_flush(VFS_I(ip)->i_mapping);
1689 				if (error)
1690 					return error;
1691 			}
1692 		}
1693 	}
1694 
1695 	if (VFS_I(ip)->i_nlink == 0)
1696 		return 0;
1697 
1698 	if (xfs_can_free_eofblocks(ip, false)) {
1699 
1700 		/*
1701 		 * Check if the inode is being opened, written and closed
1702 		 * frequently and we have delayed allocation blocks outstanding
1703 		 * (e.g. streaming writes from the NFS server), truncating the
1704 		 * blocks past EOF will cause fragmentation to occur.
1705 		 *
1706 		 * In this case don't do the truncation, but we have to be
1707 		 * careful how we detect this case. Blocks beyond EOF show up as
1708 		 * i_delayed_blks even when the inode is clean, so we need to
1709 		 * truncate them away first before checking for a dirty release.
1710 		 * Hence on the first dirty close we will still remove the
1711 		 * speculative allocation, but after that we will leave it in
1712 		 * place.
1713 		 */
1714 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1715 			return 0;
1716 		/*
1717 		 * If we can't get the iolock just skip truncating the blocks
1718 		 * past EOF because we could deadlock with the mmap_sem
1719 		 * otherwise. We'll get another chance to drop them once the
1720 		 * last reference to the inode is dropped, so we'll never leak
1721 		 * blocks permanently.
1722 		 */
1723 		if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1724 			error = xfs_free_eofblocks(ip);
1725 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1726 			if (error)
1727 				return error;
1728 		}
1729 
1730 		/* delalloc blocks after truncation means it really is dirty */
1731 		if (ip->i_delayed_blks)
1732 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1733 	}
1734 	return 0;
1735 }
1736 
1737 /*
1738  * xfs_inactive_truncate
1739  *
1740  * Called to perform a truncate when an inode becomes unlinked.
1741  */
1742 STATIC int
1743 xfs_inactive_truncate(
1744 	struct xfs_inode *ip)
1745 {
1746 	struct xfs_mount	*mp = ip->i_mount;
1747 	struct xfs_trans	*tp;
1748 	int			error;
1749 
1750 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1751 	if (error) {
1752 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1753 		return error;
1754 	}
1755 
1756 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1757 	xfs_trans_ijoin(tp, ip, 0);
1758 
1759 	/*
1760 	 * Log the inode size first to prevent stale data exposure in the event
1761 	 * of a system crash before the truncate completes. See the related
1762 	 * comment in xfs_vn_setattr_size() for details.
1763 	 */
1764 	ip->i_d.di_size = 0;
1765 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1766 
1767 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1768 	if (error)
1769 		goto error_trans_cancel;
1770 
1771 	ASSERT(ip->i_d.di_nextents == 0);
1772 
1773 	error = xfs_trans_commit(tp);
1774 	if (error)
1775 		goto error_unlock;
1776 
1777 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1778 	return 0;
1779 
1780 error_trans_cancel:
1781 	xfs_trans_cancel(tp);
1782 error_unlock:
1783 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1784 	return error;
1785 }
1786 
1787 /*
1788  * xfs_inactive_ifree()
1789  *
1790  * Perform the inode free when an inode is unlinked.
1791  */
1792 STATIC int
1793 xfs_inactive_ifree(
1794 	struct xfs_inode *ip)
1795 {
1796 	struct xfs_defer_ops	dfops;
1797 	xfs_fsblock_t		first_block;
1798 	struct xfs_mount	*mp = ip->i_mount;
1799 	struct xfs_trans	*tp;
1800 	int			error;
1801 
1802 	/*
1803 	 * We try to use a per-AG reservation for any block needed by the finobt
1804 	 * tree, but as the finobt feature predates the per-AG reservation
1805 	 * support a degraded file system might not have enough space for the
1806 	 * reservation at mount time.  In that case try to dip into the reserved
1807 	 * pool and pray.
1808 	 *
1809 	 * Send a warning if the reservation does happen to fail, as the inode
1810 	 * now remains allocated and sits on the unlinked list until the fs is
1811 	 * repaired.
1812 	 */
1813 	if (unlikely(mp->m_inotbt_nores)) {
1814 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1815 				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1816 				&tp);
1817 	} else {
1818 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1819 	}
1820 	if (error) {
1821 		if (error == -ENOSPC) {
1822 			xfs_warn_ratelimited(mp,
1823 			"Failed to remove inode(s) from unlinked list. "
1824 			"Please free space, unmount and run xfs_repair.");
1825 		} else {
1826 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1827 		}
1828 		return error;
1829 	}
1830 
1831 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1832 	xfs_trans_ijoin(tp, ip, 0);
1833 
1834 	xfs_defer_init(&dfops, &first_block);
1835 	error = xfs_ifree(tp, ip, &dfops);
1836 	if (error) {
1837 		/*
1838 		 * If we fail to free the inode, shut down.  The cancel
1839 		 * might do that, we need to make sure.  Otherwise the
1840 		 * inode might be lost for a long time or forever.
1841 		 */
1842 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1843 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1844 				__func__, error);
1845 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1846 		}
1847 		xfs_trans_cancel(tp);
1848 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1849 		return error;
1850 	}
1851 
1852 	/*
1853 	 * Credit the quota account(s). The inode is gone.
1854 	 */
1855 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1856 
1857 	/*
1858 	 * Just ignore errors at this point.  There is nothing we can do except
1859 	 * to try to keep going. Make sure it's not a silent error.
1860 	 */
1861 	error = xfs_defer_finish(&tp, &dfops);
1862 	if (error) {
1863 		xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1864 			__func__, error);
1865 		xfs_defer_cancel(&dfops);
1866 	}
1867 	error = xfs_trans_commit(tp);
1868 	if (error)
1869 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1870 			__func__, error);
1871 
1872 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1873 	return 0;
1874 }
1875 
1876 /*
1877  * xfs_inactive
1878  *
1879  * This is called when the vnode reference count for the vnode
1880  * goes to zero.  If the file has been unlinked, then it must
1881  * now be truncated.  Also, we clear all of the read-ahead state
1882  * kept for the inode here since the file is now closed.
1883  */
1884 void
1885 xfs_inactive(
1886 	xfs_inode_t	*ip)
1887 {
1888 	struct xfs_mount	*mp;
1889 	int			error;
1890 	int			truncate = 0;
1891 
1892 	/*
1893 	 * If the inode is already free, then there can be nothing
1894 	 * to clean up here.
1895 	 */
1896 	if (VFS_I(ip)->i_mode == 0) {
1897 		ASSERT(ip->i_df.if_real_bytes == 0);
1898 		ASSERT(ip->i_df.if_broot_bytes == 0);
1899 		return;
1900 	}
1901 
1902 	mp = ip->i_mount;
1903 	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1904 
1905 	/* If this is a read-only mount, don't do this (would generate I/O) */
1906 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1907 		return;
1908 
1909 	if (VFS_I(ip)->i_nlink != 0) {
1910 		/*
1911 		 * force is true because we are evicting an inode from the
1912 		 * cache. Post-eof blocks must be freed, lest we end up with
1913 		 * broken free space accounting.
1914 		 *
1915 		 * Note: don't bother with iolock here since lockdep complains
1916 		 * about acquiring it in reclaim context. We have the only
1917 		 * reference to the inode at this point anyways.
1918 		 */
1919 		if (xfs_can_free_eofblocks(ip, true))
1920 			xfs_free_eofblocks(ip);
1921 
1922 		return;
1923 	}
1924 
1925 	if (S_ISREG(VFS_I(ip)->i_mode) &&
1926 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1927 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1928 		truncate = 1;
1929 
1930 	error = xfs_qm_dqattach(ip, 0);
1931 	if (error)
1932 		return;
1933 
1934 	if (S_ISLNK(VFS_I(ip)->i_mode))
1935 		error = xfs_inactive_symlink(ip);
1936 	else if (truncate)
1937 		error = xfs_inactive_truncate(ip);
1938 	if (error)
1939 		return;
1940 
1941 	/*
1942 	 * If there are attributes associated with the file then blow them away
1943 	 * now.  The code calls a routine that recursively deconstructs the
1944 	 * attribute fork. If also blows away the in-core attribute fork.
1945 	 */
1946 	if (XFS_IFORK_Q(ip)) {
1947 		error = xfs_attr_inactive(ip);
1948 		if (error)
1949 			return;
1950 	}
1951 
1952 	ASSERT(!ip->i_afp);
1953 	ASSERT(ip->i_d.di_anextents == 0);
1954 	ASSERT(ip->i_d.di_forkoff == 0);
1955 
1956 	/*
1957 	 * Free the inode.
1958 	 */
1959 	error = xfs_inactive_ifree(ip);
1960 	if (error)
1961 		return;
1962 
1963 	/*
1964 	 * Release the dquots held by inode, if any.
1965 	 */
1966 	xfs_qm_dqdetach(ip);
1967 }
1968 
1969 /*
1970  * This is called when the inode's link count goes to 0 or we are creating a
1971  * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1972  * set to true as the link count is dropped to zero by the VFS after we've
1973  * created the file successfully, so we have to add it to the unlinked list
1974  * while the link count is non-zero.
1975  *
1976  * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1977  * list when the inode is freed.
1978  */
1979 STATIC int
1980 xfs_iunlink(
1981 	struct xfs_trans *tp,
1982 	struct xfs_inode *ip)
1983 {
1984 	xfs_mount_t	*mp = tp->t_mountp;
1985 	xfs_agi_t	*agi;
1986 	xfs_dinode_t	*dip;
1987 	xfs_buf_t	*agibp;
1988 	xfs_buf_t	*ibp;
1989 	xfs_agino_t	agino;
1990 	short		bucket_index;
1991 	int		offset;
1992 	int		error;
1993 
1994 	ASSERT(VFS_I(ip)->i_mode != 0);
1995 
1996 	/*
1997 	 * Get the agi buffer first.  It ensures lock ordering
1998 	 * on the list.
1999 	 */
2000 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
2001 	if (error)
2002 		return error;
2003 	agi = XFS_BUF_TO_AGI(agibp);
2004 
2005 	/*
2006 	 * Get the index into the agi hash table for the
2007 	 * list this inode will go on.
2008 	 */
2009 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2010 	ASSERT(agino != 0);
2011 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2012 	ASSERT(agi->agi_unlinked[bucket_index]);
2013 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2014 
2015 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2016 		/*
2017 		 * There is already another inode in the bucket we need
2018 		 * to add ourselves to.  Add us at the front of the list.
2019 		 * Here we put the head pointer into our next pointer,
2020 		 * and then we fall through to point the head at us.
2021 		 */
2022 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2023 				       0, 0);
2024 		if (error)
2025 			return error;
2026 
2027 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2028 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2029 		offset = ip->i_imap.im_boffset +
2030 			offsetof(xfs_dinode_t, di_next_unlinked);
2031 
2032 		/* need to recalc the inode CRC if appropriate */
2033 		xfs_dinode_calc_crc(mp, dip);
2034 
2035 		xfs_trans_inode_buf(tp, ibp);
2036 		xfs_trans_log_buf(tp, ibp, offset,
2037 				  (offset + sizeof(xfs_agino_t) - 1));
2038 		xfs_inobp_check(mp, ibp);
2039 	}
2040 
2041 	/*
2042 	 * Point the bucket head pointer at the inode being inserted.
2043 	 */
2044 	ASSERT(agino != 0);
2045 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2046 	offset = offsetof(xfs_agi_t, agi_unlinked) +
2047 		(sizeof(xfs_agino_t) * bucket_index);
2048 	xfs_trans_log_buf(tp, agibp, offset,
2049 			  (offset + sizeof(xfs_agino_t) - 1));
2050 	return 0;
2051 }
2052 
2053 /*
2054  * Pull the on-disk inode from the AGI unlinked list.
2055  */
2056 STATIC int
2057 xfs_iunlink_remove(
2058 	xfs_trans_t	*tp,
2059 	xfs_inode_t	*ip)
2060 {
2061 	xfs_ino_t	next_ino;
2062 	xfs_mount_t	*mp;
2063 	xfs_agi_t	*agi;
2064 	xfs_dinode_t	*dip;
2065 	xfs_buf_t	*agibp;
2066 	xfs_buf_t	*ibp;
2067 	xfs_agnumber_t	agno;
2068 	xfs_agino_t	agino;
2069 	xfs_agino_t	next_agino;
2070 	xfs_buf_t	*last_ibp;
2071 	xfs_dinode_t	*last_dip = NULL;
2072 	short		bucket_index;
2073 	int		offset, last_offset = 0;
2074 	int		error;
2075 
2076 	mp = tp->t_mountp;
2077 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2078 
2079 	/*
2080 	 * Get the agi buffer first.  It ensures lock ordering
2081 	 * on the list.
2082 	 */
2083 	error = xfs_read_agi(mp, tp, agno, &agibp);
2084 	if (error)
2085 		return error;
2086 
2087 	agi = XFS_BUF_TO_AGI(agibp);
2088 
2089 	/*
2090 	 * Get the index into the agi hash table for the
2091 	 * list this inode will go on.
2092 	 */
2093 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2094 	ASSERT(agino != 0);
2095 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2096 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2097 	ASSERT(agi->agi_unlinked[bucket_index]);
2098 
2099 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2100 		/*
2101 		 * We're at the head of the list.  Get the inode's on-disk
2102 		 * buffer to see if there is anyone after us on the list.
2103 		 * Only modify our next pointer if it is not already NULLAGINO.
2104 		 * This saves us the overhead of dealing with the buffer when
2105 		 * there is no need to change it.
2106 		 */
2107 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2108 				       0, 0);
2109 		if (error) {
2110 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2111 				__func__, error);
2112 			return error;
2113 		}
2114 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2115 		ASSERT(next_agino != 0);
2116 		if (next_agino != NULLAGINO) {
2117 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2118 			offset = ip->i_imap.im_boffset +
2119 				offsetof(xfs_dinode_t, di_next_unlinked);
2120 
2121 			/* need to recalc the inode CRC if appropriate */
2122 			xfs_dinode_calc_crc(mp, dip);
2123 
2124 			xfs_trans_inode_buf(tp, ibp);
2125 			xfs_trans_log_buf(tp, ibp, offset,
2126 					  (offset + sizeof(xfs_agino_t) - 1));
2127 			xfs_inobp_check(mp, ibp);
2128 		} else {
2129 			xfs_trans_brelse(tp, ibp);
2130 		}
2131 		/*
2132 		 * Point the bucket head pointer at the next inode.
2133 		 */
2134 		ASSERT(next_agino != 0);
2135 		ASSERT(next_agino != agino);
2136 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2137 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2138 			(sizeof(xfs_agino_t) * bucket_index);
2139 		xfs_trans_log_buf(tp, agibp, offset,
2140 				  (offset + sizeof(xfs_agino_t) - 1));
2141 	} else {
2142 		/*
2143 		 * We need to search the list for the inode being freed.
2144 		 */
2145 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2146 		last_ibp = NULL;
2147 		while (next_agino != agino) {
2148 			struct xfs_imap	imap;
2149 
2150 			if (last_ibp)
2151 				xfs_trans_brelse(tp, last_ibp);
2152 
2153 			imap.im_blkno = 0;
2154 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2155 
2156 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2157 			if (error) {
2158 				xfs_warn(mp,
2159 	"%s: xfs_imap returned error %d.",
2160 					 __func__, error);
2161 				return error;
2162 			}
2163 
2164 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2165 					       &last_ibp, 0, 0);
2166 			if (error) {
2167 				xfs_warn(mp,
2168 	"%s: xfs_imap_to_bp returned error %d.",
2169 					__func__, error);
2170 				return error;
2171 			}
2172 
2173 			last_offset = imap.im_boffset;
2174 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2175 			ASSERT(next_agino != NULLAGINO);
2176 			ASSERT(next_agino != 0);
2177 		}
2178 
2179 		/*
2180 		 * Now last_ibp points to the buffer previous to us on the
2181 		 * unlinked list.  Pull us from the list.
2182 		 */
2183 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2184 				       0, 0);
2185 		if (error) {
2186 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2187 				__func__, error);
2188 			return error;
2189 		}
2190 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2191 		ASSERT(next_agino != 0);
2192 		ASSERT(next_agino != agino);
2193 		if (next_agino != NULLAGINO) {
2194 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2195 			offset = ip->i_imap.im_boffset +
2196 				offsetof(xfs_dinode_t, di_next_unlinked);
2197 
2198 			/* need to recalc the inode CRC if appropriate */
2199 			xfs_dinode_calc_crc(mp, dip);
2200 
2201 			xfs_trans_inode_buf(tp, ibp);
2202 			xfs_trans_log_buf(tp, ibp, offset,
2203 					  (offset + sizeof(xfs_agino_t) - 1));
2204 			xfs_inobp_check(mp, ibp);
2205 		} else {
2206 			xfs_trans_brelse(tp, ibp);
2207 		}
2208 		/*
2209 		 * Point the previous inode on the list to the next inode.
2210 		 */
2211 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2212 		ASSERT(next_agino != 0);
2213 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2214 
2215 		/* need to recalc the inode CRC if appropriate */
2216 		xfs_dinode_calc_crc(mp, last_dip);
2217 
2218 		xfs_trans_inode_buf(tp, last_ibp);
2219 		xfs_trans_log_buf(tp, last_ibp, offset,
2220 				  (offset + sizeof(xfs_agino_t) - 1));
2221 		xfs_inobp_check(mp, last_ibp);
2222 	}
2223 	return 0;
2224 }
2225 
2226 /*
2227  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2228  * inodes that are in memory - they all must be marked stale and attached to
2229  * the cluster buffer.
2230  */
2231 STATIC int
2232 xfs_ifree_cluster(
2233 	xfs_inode_t		*free_ip,
2234 	xfs_trans_t		*tp,
2235 	struct xfs_icluster	*xic)
2236 {
2237 	xfs_mount_t		*mp = free_ip->i_mount;
2238 	int			blks_per_cluster;
2239 	int			inodes_per_cluster;
2240 	int			nbufs;
2241 	int			i, j;
2242 	int			ioffset;
2243 	xfs_daddr_t		blkno;
2244 	xfs_buf_t		*bp;
2245 	xfs_inode_t		*ip;
2246 	xfs_inode_log_item_t	*iip;
2247 	xfs_log_item_t		*lip;
2248 	struct xfs_perag	*pag;
2249 	xfs_ino_t		inum;
2250 
2251 	inum = xic->first_ino;
2252 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2253 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2254 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2255 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2256 
2257 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2258 		/*
2259 		 * The allocation bitmap tells us which inodes of the chunk were
2260 		 * physically allocated. Skip the cluster if an inode falls into
2261 		 * a sparse region.
2262 		 */
2263 		ioffset = inum - xic->first_ino;
2264 		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2265 			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2266 			continue;
2267 		}
2268 
2269 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2270 					 XFS_INO_TO_AGBNO(mp, inum));
2271 
2272 		/*
2273 		 * We obtain and lock the backing buffer first in the process
2274 		 * here, as we have to ensure that any dirty inode that we
2275 		 * can't get the flush lock on is attached to the buffer.
2276 		 * If we scan the in-memory inodes first, then buffer IO can
2277 		 * complete before we get a lock on it, and hence we may fail
2278 		 * to mark all the active inodes on the buffer stale.
2279 		 */
2280 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2281 					mp->m_bsize * blks_per_cluster,
2282 					XBF_UNMAPPED);
2283 
2284 		if (!bp)
2285 			return -ENOMEM;
2286 
2287 		/*
2288 		 * This buffer may not have been correctly initialised as we
2289 		 * didn't read it from disk. That's not important because we are
2290 		 * only using to mark the buffer as stale in the log, and to
2291 		 * attach stale cached inodes on it. That means it will never be
2292 		 * dispatched for IO. If it is, we want to know about it, and we
2293 		 * want it to fail. We can acheive this by adding a write
2294 		 * verifier to the buffer.
2295 		 */
2296 		 bp->b_ops = &xfs_inode_buf_ops;
2297 
2298 		/*
2299 		 * Walk the inodes already attached to the buffer and mark them
2300 		 * stale. These will all have the flush locks held, so an
2301 		 * in-memory inode walk can't lock them. By marking them all
2302 		 * stale first, we will not attempt to lock them in the loop
2303 		 * below as the XFS_ISTALE flag will be set.
2304 		 */
2305 		lip = bp->b_fspriv;
2306 		while (lip) {
2307 			if (lip->li_type == XFS_LI_INODE) {
2308 				iip = (xfs_inode_log_item_t *)lip;
2309 				ASSERT(iip->ili_logged == 1);
2310 				lip->li_cb = xfs_istale_done;
2311 				xfs_trans_ail_copy_lsn(mp->m_ail,
2312 							&iip->ili_flush_lsn,
2313 							&iip->ili_item.li_lsn);
2314 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2315 			}
2316 			lip = lip->li_bio_list;
2317 		}
2318 
2319 
2320 		/*
2321 		 * For each inode in memory attempt to add it to the inode
2322 		 * buffer and set it up for being staled on buffer IO
2323 		 * completion.  This is safe as we've locked out tail pushing
2324 		 * and flushing by locking the buffer.
2325 		 *
2326 		 * We have already marked every inode that was part of a
2327 		 * transaction stale above, which means there is no point in
2328 		 * even trying to lock them.
2329 		 */
2330 		for (i = 0; i < inodes_per_cluster; i++) {
2331 retry:
2332 			rcu_read_lock();
2333 			ip = radix_tree_lookup(&pag->pag_ici_root,
2334 					XFS_INO_TO_AGINO(mp, (inum + i)));
2335 
2336 			/* Inode not in memory, nothing to do */
2337 			if (!ip) {
2338 				rcu_read_unlock();
2339 				continue;
2340 			}
2341 
2342 			/*
2343 			 * because this is an RCU protected lookup, we could
2344 			 * find a recently freed or even reallocated inode
2345 			 * during the lookup. We need to check under the
2346 			 * i_flags_lock for a valid inode here. Skip it if it
2347 			 * is not valid, the wrong inode or stale.
2348 			 */
2349 			spin_lock(&ip->i_flags_lock);
2350 			if (ip->i_ino != inum + i ||
2351 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2352 				spin_unlock(&ip->i_flags_lock);
2353 				rcu_read_unlock();
2354 				continue;
2355 			}
2356 			spin_unlock(&ip->i_flags_lock);
2357 
2358 			/*
2359 			 * Don't try to lock/unlock the current inode, but we
2360 			 * _cannot_ skip the other inodes that we did not find
2361 			 * in the list attached to the buffer and are not
2362 			 * already marked stale. If we can't lock it, back off
2363 			 * and retry.
2364 			 */
2365 			if (ip != free_ip) {
2366 				if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2367 					rcu_read_unlock();
2368 					delay(1);
2369 					goto retry;
2370 				}
2371 
2372 				/*
2373 				 * Check the inode number again in case we're
2374 				 * racing with freeing in xfs_reclaim_inode().
2375 				 * See the comments in that function for more
2376 				 * information as to why the initial check is
2377 				 * not sufficient.
2378 				 */
2379 				if (ip->i_ino != inum + i) {
2380 					xfs_iunlock(ip, XFS_ILOCK_EXCL);
2381 					continue;
2382 				}
2383 			}
2384 			rcu_read_unlock();
2385 
2386 			xfs_iflock(ip);
2387 			xfs_iflags_set(ip, XFS_ISTALE);
2388 
2389 			/*
2390 			 * we don't need to attach clean inodes or those only
2391 			 * with unlogged changes (which we throw away, anyway).
2392 			 */
2393 			iip = ip->i_itemp;
2394 			if (!iip || xfs_inode_clean(ip)) {
2395 				ASSERT(ip != free_ip);
2396 				xfs_ifunlock(ip);
2397 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2398 				continue;
2399 			}
2400 
2401 			iip->ili_last_fields = iip->ili_fields;
2402 			iip->ili_fields = 0;
2403 			iip->ili_fsync_fields = 0;
2404 			iip->ili_logged = 1;
2405 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2406 						&iip->ili_item.li_lsn);
2407 
2408 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2409 						  &iip->ili_item);
2410 
2411 			if (ip != free_ip)
2412 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2413 		}
2414 
2415 		xfs_trans_stale_inode_buf(tp, bp);
2416 		xfs_trans_binval(tp, bp);
2417 	}
2418 
2419 	xfs_perag_put(pag);
2420 	return 0;
2421 }
2422 
2423 /*
2424  * This is called to return an inode to the inode free list.
2425  * The inode should already be truncated to 0 length and have
2426  * no pages associated with it.  This routine also assumes that
2427  * the inode is already a part of the transaction.
2428  *
2429  * The on-disk copy of the inode will have been added to the list
2430  * of unlinked inodes in the AGI. We need to remove the inode from
2431  * that list atomically with respect to freeing it here.
2432  */
2433 int
2434 xfs_ifree(
2435 	xfs_trans_t	*tp,
2436 	xfs_inode_t	*ip,
2437 	struct xfs_defer_ops	*dfops)
2438 {
2439 	int			error;
2440 	struct xfs_icluster	xic = { 0 };
2441 
2442 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2443 	ASSERT(VFS_I(ip)->i_nlink == 0);
2444 	ASSERT(ip->i_d.di_nextents == 0);
2445 	ASSERT(ip->i_d.di_anextents == 0);
2446 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2447 	ASSERT(ip->i_d.di_nblocks == 0);
2448 
2449 	/*
2450 	 * Pull the on-disk inode from the AGI unlinked list.
2451 	 */
2452 	error = xfs_iunlink_remove(tp, ip);
2453 	if (error)
2454 		return error;
2455 
2456 	error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2457 	if (error)
2458 		return error;
2459 
2460 	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2461 	ip->i_d.di_flags = 0;
2462 	ip->i_d.di_dmevmask = 0;
2463 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2464 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2465 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2466 	/*
2467 	 * Bump the generation count so no one will be confused
2468 	 * by reincarnations of this inode.
2469 	 */
2470 	VFS_I(ip)->i_generation++;
2471 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2472 
2473 	if (xic.deleted)
2474 		error = xfs_ifree_cluster(ip, tp, &xic);
2475 
2476 	return error;
2477 }
2478 
2479 /*
2480  * This is called to unpin an inode.  The caller must have the inode locked
2481  * in at least shared mode so that the buffer cannot be subsequently pinned
2482  * once someone is waiting for it to be unpinned.
2483  */
2484 static void
2485 xfs_iunpin(
2486 	struct xfs_inode	*ip)
2487 {
2488 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2489 
2490 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2491 
2492 	/* Give the log a push to start the unpinning I/O */
2493 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2494 
2495 }
2496 
2497 static void
2498 __xfs_iunpin_wait(
2499 	struct xfs_inode	*ip)
2500 {
2501 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2502 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2503 
2504 	xfs_iunpin(ip);
2505 
2506 	do {
2507 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2508 		if (xfs_ipincount(ip))
2509 			io_schedule();
2510 	} while (xfs_ipincount(ip));
2511 	finish_wait(wq, &wait.wq_entry);
2512 }
2513 
2514 void
2515 xfs_iunpin_wait(
2516 	struct xfs_inode	*ip)
2517 {
2518 	if (xfs_ipincount(ip))
2519 		__xfs_iunpin_wait(ip);
2520 }
2521 
2522 /*
2523  * Removing an inode from the namespace involves removing the directory entry
2524  * and dropping the link count on the inode. Removing the directory entry can
2525  * result in locking an AGF (directory blocks were freed) and removing a link
2526  * count can result in placing the inode on an unlinked list which results in
2527  * locking an AGI.
2528  *
2529  * The big problem here is that we have an ordering constraint on AGF and AGI
2530  * locking - inode allocation locks the AGI, then can allocate a new extent for
2531  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2532  * removes the inode from the unlinked list, requiring that we lock the AGI
2533  * first, and then freeing the inode can result in an inode chunk being freed
2534  * and hence freeing disk space requiring that we lock an AGF.
2535  *
2536  * Hence the ordering that is imposed by other parts of the code is AGI before
2537  * AGF. This means we cannot remove the directory entry before we drop the inode
2538  * reference count and put it on the unlinked list as this results in a lock
2539  * order of AGF then AGI, and this can deadlock against inode allocation and
2540  * freeing. Therefore we must drop the link counts before we remove the
2541  * directory entry.
2542  *
2543  * This is still safe from a transactional point of view - it is not until we
2544  * get to xfs_defer_finish() that we have the possibility of multiple
2545  * transactions in this operation. Hence as long as we remove the directory
2546  * entry and drop the link count in the first transaction of the remove
2547  * operation, there are no transactional constraints on the ordering here.
2548  */
2549 int
2550 xfs_remove(
2551 	xfs_inode_t             *dp,
2552 	struct xfs_name		*name,
2553 	xfs_inode_t		*ip)
2554 {
2555 	xfs_mount_t		*mp = dp->i_mount;
2556 	xfs_trans_t             *tp = NULL;
2557 	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2558 	int                     error = 0;
2559 	struct xfs_defer_ops	dfops;
2560 	xfs_fsblock_t           first_block;
2561 	uint			resblks;
2562 
2563 	trace_xfs_remove(dp, name);
2564 
2565 	if (XFS_FORCED_SHUTDOWN(mp))
2566 		return -EIO;
2567 
2568 	error = xfs_qm_dqattach(dp, 0);
2569 	if (error)
2570 		goto std_return;
2571 
2572 	error = xfs_qm_dqattach(ip, 0);
2573 	if (error)
2574 		goto std_return;
2575 
2576 	/*
2577 	 * We try to get the real space reservation first,
2578 	 * allowing for directory btree deletion(s) implying
2579 	 * possible bmap insert(s).  If we can't get the space
2580 	 * reservation then we use 0 instead, and avoid the bmap
2581 	 * btree insert(s) in the directory code by, if the bmap
2582 	 * insert tries to happen, instead trimming the LAST
2583 	 * block from the directory.
2584 	 */
2585 	resblks = XFS_REMOVE_SPACE_RES(mp);
2586 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2587 	if (error == -ENOSPC) {
2588 		resblks = 0;
2589 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2590 				&tp);
2591 	}
2592 	if (error) {
2593 		ASSERT(error != -ENOSPC);
2594 		goto std_return;
2595 	}
2596 
2597 	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2598 
2599 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2600 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2601 
2602 	/*
2603 	 * If we're removing a directory perform some additional validation.
2604 	 */
2605 	if (is_dir) {
2606 		ASSERT(VFS_I(ip)->i_nlink >= 2);
2607 		if (VFS_I(ip)->i_nlink != 2) {
2608 			error = -ENOTEMPTY;
2609 			goto out_trans_cancel;
2610 		}
2611 		if (!xfs_dir_isempty(ip)) {
2612 			error = -ENOTEMPTY;
2613 			goto out_trans_cancel;
2614 		}
2615 
2616 		/* Drop the link from ip's "..".  */
2617 		error = xfs_droplink(tp, dp);
2618 		if (error)
2619 			goto out_trans_cancel;
2620 
2621 		/* Drop the "." link from ip to self.  */
2622 		error = xfs_droplink(tp, ip);
2623 		if (error)
2624 			goto out_trans_cancel;
2625 	} else {
2626 		/*
2627 		 * When removing a non-directory we need to log the parent
2628 		 * inode here.  For a directory this is done implicitly
2629 		 * by the xfs_droplink call for the ".." entry.
2630 		 */
2631 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2632 	}
2633 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2634 
2635 	/* Drop the link from dp to ip. */
2636 	error = xfs_droplink(tp, ip);
2637 	if (error)
2638 		goto out_trans_cancel;
2639 
2640 	xfs_defer_init(&dfops, &first_block);
2641 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2642 					&first_block, &dfops, resblks);
2643 	if (error) {
2644 		ASSERT(error != -ENOENT);
2645 		goto out_bmap_cancel;
2646 	}
2647 
2648 	/*
2649 	 * If this is a synchronous mount, make sure that the
2650 	 * remove transaction goes to disk before returning to
2651 	 * the user.
2652 	 */
2653 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2654 		xfs_trans_set_sync(tp);
2655 
2656 	error = xfs_defer_finish(&tp, &dfops);
2657 	if (error)
2658 		goto out_bmap_cancel;
2659 
2660 	error = xfs_trans_commit(tp);
2661 	if (error)
2662 		goto std_return;
2663 
2664 	if (is_dir && xfs_inode_is_filestream(ip))
2665 		xfs_filestream_deassociate(ip);
2666 
2667 	return 0;
2668 
2669  out_bmap_cancel:
2670 	xfs_defer_cancel(&dfops);
2671  out_trans_cancel:
2672 	xfs_trans_cancel(tp);
2673  std_return:
2674 	return error;
2675 }
2676 
2677 /*
2678  * Enter all inodes for a rename transaction into a sorted array.
2679  */
2680 #define __XFS_SORT_INODES	5
2681 STATIC void
2682 xfs_sort_for_rename(
2683 	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2684 	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2685 	struct xfs_inode	*ip1,	/* in: inode of old entry */
2686 	struct xfs_inode	*ip2,	/* in: inode of new entry */
2687 	struct xfs_inode	*wip,	/* in: whiteout inode */
2688 	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2689 	int			*num_inodes)  /* in/out: inodes in array */
2690 {
2691 	int			i, j;
2692 
2693 	ASSERT(*num_inodes == __XFS_SORT_INODES);
2694 	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2695 
2696 	/*
2697 	 * i_tab contains a list of pointers to inodes.  We initialize
2698 	 * the table here & we'll sort it.  We will then use it to
2699 	 * order the acquisition of the inode locks.
2700 	 *
2701 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2702 	 */
2703 	i = 0;
2704 	i_tab[i++] = dp1;
2705 	i_tab[i++] = dp2;
2706 	i_tab[i++] = ip1;
2707 	if (ip2)
2708 		i_tab[i++] = ip2;
2709 	if (wip)
2710 		i_tab[i++] = wip;
2711 	*num_inodes = i;
2712 
2713 	/*
2714 	 * Sort the elements via bubble sort.  (Remember, there are at
2715 	 * most 5 elements to sort, so this is adequate.)
2716 	 */
2717 	for (i = 0; i < *num_inodes; i++) {
2718 		for (j = 1; j < *num_inodes; j++) {
2719 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2720 				struct xfs_inode *temp = i_tab[j];
2721 				i_tab[j] = i_tab[j-1];
2722 				i_tab[j-1] = temp;
2723 			}
2724 		}
2725 	}
2726 }
2727 
2728 static int
2729 xfs_finish_rename(
2730 	struct xfs_trans	*tp,
2731 	struct xfs_defer_ops	*dfops)
2732 {
2733 	int			error;
2734 
2735 	/*
2736 	 * If this is a synchronous mount, make sure that the rename transaction
2737 	 * goes to disk before returning to the user.
2738 	 */
2739 	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2740 		xfs_trans_set_sync(tp);
2741 
2742 	error = xfs_defer_finish(&tp, dfops);
2743 	if (error) {
2744 		xfs_defer_cancel(dfops);
2745 		xfs_trans_cancel(tp);
2746 		return error;
2747 	}
2748 
2749 	return xfs_trans_commit(tp);
2750 }
2751 
2752 /*
2753  * xfs_cross_rename()
2754  *
2755  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2756  */
2757 STATIC int
2758 xfs_cross_rename(
2759 	struct xfs_trans	*tp,
2760 	struct xfs_inode	*dp1,
2761 	struct xfs_name		*name1,
2762 	struct xfs_inode	*ip1,
2763 	struct xfs_inode	*dp2,
2764 	struct xfs_name		*name2,
2765 	struct xfs_inode	*ip2,
2766 	struct xfs_defer_ops	*dfops,
2767 	xfs_fsblock_t		*first_block,
2768 	int			spaceres)
2769 {
2770 	int		error = 0;
2771 	int		ip1_flags = 0;
2772 	int		ip2_flags = 0;
2773 	int		dp2_flags = 0;
2774 
2775 	/* Swap inode number for dirent in first parent */
2776 	error = xfs_dir_replace(tp, dp1, name1,
2777 				ip2->i_ino,
2778 				first_block, dfops, spaceres);
2779 	if (error)
2780 		goto out_trans_abort;
2781 
2782 	/* Swap inode number for dirent in second parent */
2783 	error = xfs_dir_replace(tp, dp2, name2,
2784 				ip1->i_ino,
2785 				first_block, dfops, spaceres);
2786 	if (error)
2787 		goto out_trans_abort;
2788 
2789 	/*
2790 	 * If we're renaming one or more directories across different parents,
2791 	 * update the respective ".." entries (and link counts) to match the new
2792 	 * parents.
2793 	 */
2794 	if (dp1 != dp2) {
2795 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2796 
2797 		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2798 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2799 						dp1->i_ino, first_block,
2800 						dfops, spaceres);
2801 			if (error)
2802 				goto out_trans_abort;
2803 
2804 			/* transfer ip2 ".." reference to dp1 */
2805 			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2806 				error = xfs_droplink(tp, dp2);
2807 				if (error)
2808 					goto out_trans_abort;
2809 				error = xfs_bumplink(tp, dp1);
2810 				if (error)
2811 					goto out_trans_abort;
2812 			}
2813 
2814 			/*
2815 			 * Although ip1 isn't changed here, userspace needs
2816 			 * to be warned about the change, so that applications
2817 			 * relying on it (like backup ones), will properly
2818 			 * notify the change
2819 			 */
2820 			ip1_flags |= XFS_ICHGTIME_CHG;
2821 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2822 		}
2823 
2824 		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2825 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2826 						dp2->i_ino, first_block,
2827 						dfops, spaceres);
2828 			if (error)
2829 				goto out_trans_abort;
2830 
2831 			/* transfer ip1 ".." reference to dp2 */
2832 			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2833 				error = xfs_droplink(tp, dp1);
2834 				if (error)
2835 					goto out_trans_abort;
2836 				error = xfs_bumplink(tp, dp2);
2837 				if (error)
2838 					goto out_trans_abort;
2839 			}
2840 
2841 			/*
2842 			 * Although ip2 isn't changed here, userspace needs
2843 			 * to be warned about the change, so that applications
2844 			 * relying on it (like backup ones), will properly
2845 			 * notify the change
2846 			 */
2847 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2848 			ip2_flags |= XFS_ICHGTIME_CHG;
2849 		}
2850 	}
2851 
2852 	if (ip1_flags) {
2853 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2854 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2855 	}
2856 	if (ip2_flags) {
2857 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2858 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2859 	}
2860 	if (dp2_flags) {
2861 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2862 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2863 	}
2864 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2865 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2866 	return xfs_finish_rename(tp, dfops);
2867 
2868 out_trans_abort:
2869 	xfs_defer_cancel(dfops);
2870 	xfs_trans_cancel(tp);
2871 	return error;
2872 }
2873 
2874 /*
2875  * xfs_rename_alloc_whiteout()
2876  *
2877  * Return a referenced, unlinked, unlocked inode that that can be used as a
2878  * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2879  * crash between allocating the inode and linking it into the rename transaction
2880  * recovery will free the inode and we won't leak it.
2881  */
2882 static int
2883 xfs_rename_alloc_whiteout(
2884 	struct xfs_inode	*dp,
2885 	struct xfs_inode	**wip)
2886 {
2887 	struct xfs_inode	*tmpfile;
2888 	int			error;
2889 
2890 	error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2891 	if (error)
2892 		return error;
2893 
2894 	/*
2895 	 * Prepare the tmpfile inode as if it were created through the VFS.
2896 	 * Otherwise, the link increment paths will complain about nlink 0->1.
2897 	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2898 	 * and flag it as linkable.
2899 	 */
2900 	drop_nlink(VFS_I(tmpfile));
2901 	xfs_setup_iops(tmpfile);
2902 	xfs_finish_inode_setup(tmpfile);
2903 	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2904 
2905 	*wip = tmpfile;
2906 	return 0;
2907 }
2908 
2909 /*
2910  * xfs_rename
2911  */
2912 int
2913 xfs_rename(
2914 	struct xfs_inode	*src_dp,
2915 	struct xfs_name		*src_name,
2916 	struct xfs_inode	*src_ip,
2917 	struct xfs_inode	*target_dp,
2918 	struct xfs_name		*target_name,
2919 	struct xfs_inode	*target_ip,
2920 	unsigned int		flags)
2921 {
2922 	struct xfs_mount	*mp = src_dp->i_mount;
2923 	struct xfs_trans	*tp;
2924 	struct xfs_defer_ops	dfops;
2925 	xfs_fsblock_t		first_block;
2926 	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2927 	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2928 	int			num_inodes = __XFS_SORT_INODES;
2929 	bool			new_parent = (src_dp != target_dp);
2930 	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2931 	int			spaceres;
2932 	int			error;
2933 
2934 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2935 
2936 	if ((flags & RENAME_EXCHANGE) && !target_ip)
2937 		return -EINVAL;
2938 
2939 	/*
2940 	 * If we are doing a whiteout operation, allocate the whiteout inode
2941 	 * we will be placing at the target and ensure the type is set
2942 	 * appropriately.
2943 	 */
2944 	if (flags & RENAME_WHITEOUT) {
2945 		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2946 		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2947 		if (error)
2948 			return error;
2949 
2950 		/* setup target dirent info as whiteout */
2951 		src_name->type = XFS_DIR3_FT_CHRDEV;
2952 	}
2953 
2954 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2955 				inodes, &num_inodes);
2956 
2957 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2958 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2959 	if (error == -ENOSPC) {
2960 		spaceres = 0;
2961 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2962 				&tp);
2963 	}
2964 	if (error)
2965 		goto out_release_wip;
2966 
2967 	/*
2968 	 * Attach the dquots to the inodes
2969 	 */
2970 	error = xfs_qm_vop_rename_dqattach(inodes);
2971 	if (error)
2972 		goto out_trans_cancel;
2973 
2974 	/*
2975 	 * Lock all the participating inodes. Depending upon whether
2976 	 * the target_name exists in the target directory, and
2977 	 * whether the target directory is the same as the source
2978 	 * directory, we can lock from 2 to 4 inodes.
2979 	 */
2980 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2981 
2982 	/*
2983 	 * Join all the inodes to the transaction. From this point on,
2984 	 * we can rely on either trans_commit or trans_cancel to unlock
2985 	 * them.
2986 	 */
2987 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2988 	if (new_parent)
2989 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2990 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2991 	if (target_ip)
2992 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2993 	if (wip)
2994 		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2995 
2996 	/*
2997 	 * If we are using project inheritance, we only allow renames
2998 	 * into our tree when the project IDs are the same; else the
2999 	 * tree quota mechanism would be circumvented.
3000 	 */
3001 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3002 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3003 		error = -EXDEV;
3004 		goto out_trans_cancel;
3005 	}
3006 
3007 	xfs_defer_init(&dfops, &first_block);
3008 
3009 	/* RENAME_EXCHANGE is unique from here on. */
3010 	if (flags & RENAME_EXCHANGE)
3011 		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3012 					target_dp, target_name, target_ip,
3013 					&dfops, &first_block, spaceres);
3014 
3015 	/*
3016 	 * Set up the target.
3017 	 */
3018 	if (target_ip == NULL) {
3019 		/*
3020 		 * If there's no space reservation, check the entry will
3021 		 * fit before actually inserting it.
3022 		 */
3023 		if (!spaceres) {
3024 			error = xfs_dir_canenter(tp, target_dp, target_name);
3025 			if (error)
3026 				goto out_trans_cancel;
3027 		}
3028 		/*
3029 		 * If target does not exist and the rename crosses
3030 		 * directories, adjust the target directory link count
3031 		 * to account for the ".." reference from the new entry.
3032 		 */
3033 		error = xfs_dir_createname(tp, target_dp, target_name,
3034 						src_ip->i_ino, &first_block,
3035 						&dfops, spaceres);
3036 		if (error)
3037 			goto out_bmap_cancel;
3038 
3039 		xfs_trans_ichgtime(tp, target_dp,
3040 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3041 
3042 		if (new_parent && src_is_directory) {
3043 			error = xfs_bumplink(tp, target_dp);
3044 			if (error)
3045 				goto out_bmap_cancel;
3046 		}
3047 	} else { /* target_ip != NULL */
3048 		/*
3049 		 * If target exists and it's a directory, check that both
3050 		 * target and source are directories and that target can be
3051 		 * destroyed, or that neither is a directory.
3052 		 */
3053 		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3054 			/*
3055 			 * Make sure target dir is empty.
3056 			 */
3057 			if (!(xfs_dir_isempty(target_ip)) ||
3058 			    (VFS_I(target_ip)->i_nlink > 2)) {
3059 				error = -EEXIST;
3060 				goto out_trans_cancel;
3061 			}
3062 		}
3063 
3064 		/*
3065 		 * Link the source inode under the target name.
3066 		 * If the source inode is a directory and we are moving
3067 		 * it across directories, its ".." entry will be
3068 		 * inconsistent until we replace that down below.
3069 		 *
3070 		 * In case there is already an entry with the same
3071 		 * name at the destination directory, remove it first.
3072 		 */
3073 		error = xfs_dir_replace(tp, target_dp, target_name,
3074 					src_ip->i_ino,
3075 					&first_block, &dfops, spaceres);
3076 		if (error)
3077 			goto out_bmap_cancel;
3078 
3079 		xfs_trans_ichgtime(tp, target_dp,
3080 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3081 
3082 		/*
3083 		 * Decrement the link count on the target since the target
3084 		 * dir no longer points to it.
3085 		 */
3086 		error = xfs_droplink(tp, target_ip);
3087 		if (error)
3088 			goto out_bmap_cancel;
3089 
3090 		if (src_is_directory) {
3091 			/*
3092 			 * Drop the link from the old "." entry.
3093 			 */
3094 			error = xfs_droplink(tp, target_ip);
3095 			if (error)
3096 				goto out_bmap_cancel;
3097 		}
3098 	} /* target_ip != NULL */
3099 
3100 	/*
3101 	 * Remove the source.
3102 	 */
3103 	if (new_parent && src_is_directory) {
3104 		/*
3105 		 * Rewrite the ".." entry to point to the new
3106 		 * directory.
3107 		 */
3108 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3109 					target_dp->i_ino,
3110 					&first_block, &dfops, spaceres);
3111 		ASSERT(error != -EEXIST);
3112 		if (error)
3113 			goto out_bmap_cancel;
3114 	}
3115 
3116 	/*
3117 	 * We always want to hit the ctime on the source inode.
3118 	 *
3119 	 * This isn't strictly required by the standards since the source
3120 	 * inode isn't really being changed, but old unix file systems did
3121 	 * it and some incremental backup programs won't work without it.
3122 	 */
3123 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3124 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3125 
3126 	/*
3127 	 * Adjust the link count on src_dp.  This is necessary when
3128 	 * renaming a directory, either within one parent when
3129 	 * the target existed, or across two parent directories.
3130 	 */
3131 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3132 
3133 		/*
3134 		 * Decrement link count on src_directory since the
3135 		 * entry that's moved no longer points to it.
3136 		 */
3137 		error = xfs_droplink(tp, src_dp);
3138 		if (error)
3139 			goto out_bmap_cancel;
3140 	}
3141 
3142 	/*
3143 	 * For whiteouts, we only need to update the source dirent with the
3144 	 * inode number of the whiteout inode rather than removing it
3145 	 * altogether.
3146 	 */
3147 	if (wip) {
3148 		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3149 					&first_block, &dfops, spaceres);
3150 	} else
3151 		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3152 					   &first_block, &dfops, spaceres);
3153 	if (error)
3154 		goto out_bmap_cancel;
3155 
3156 	/*
3157 	 * For whiteouts, we need to bump the link count on the whiteout inode.
3158 	 * This means that failures all the way up to this point leave the inode
3159 	 * on the unlinked list and so cleanup is a simple matter of dropping
3160 	 * the remaining reference to it. If we fail here after bumping the link
3161 	 * count, we're shutting down the filesystem so we'll never see the
3162 	 * intermediate state on disk.
3163 	 */
3164 	if (wip) {
3165 		ASSERT(VFS_I(wip)->i_nlink == 0);
3166 		error = xfs_bumplink(tp, wip);
3167 		if (error)
3168 			goto out_bmap_cancel;
3169 		error = xfs_iunlink_remove(tp, wip);
3170 		if (error)
3171 			goto out_bmap_cancel;
3172 		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3173 
3174 		/*
3175 		 * Now we have a real link, clear the "I'm a tmpfile" state
3176 		 * flag from the inode so it doesn't accidentally get misused in
3177 		 * future.
3178 		 */
3179 		VFS_I(wip)->i_state &= ~I_LINKABLE;
3180 	}
3181 
3182 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3183 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3184 	if (new_parent)
3185 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3186 
3187 	error = xfs_finish_rename(tp, &dfops);
3188 	if (wip)
3189 		IRELE(wip);
3190 	return error;
3191 
3192 out_bmap_cancel:
3193 	xfs_defer_cancel(&dfops);
3194 out_trans_cancel:
3195 	xfs_trans_cancel(tp);
3196 out_release_wip:
3197 	if (wip)
3198 		IRELE(wip);
3199 	return error;
3200 }
3201 
3202 STATIC int
3203 xfs_iflush_cluster(
3204 	struct xfs_inode	*ip,
3205 	struct xfs_buf		*bp)
3206 {
3207 	struct xfs_mount	*mp = ip->i_mount;
3208 	struct xfs_perag	*pag;
3209 	unsigned long		first_index, mask;
3210 	unsigned long		inodes_per_cluster;
3211 	int			cilist_size;
3212 	struct xfs_inode	**cilist;
3213 	struct xfs_inode	*cip;
3214 	int			nr_found;
3215 	int			clcount = 0;
3216 	int			bufwasdelwri;
3217 	int			i;
3218 
3219 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3220 
3221 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3222 	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3223 	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3224 	if (!cilist)
3225 		goto out_put;
3226 
3227 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3228 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3229 	rcu_read_lock();
3230 	/* really need a gang lookup range call here */
3231 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3232 					first_index, inodes_per_cluster);
3233 	if (nr_found == 0)
3234 		goto out_free;
3235 
3236 	for (i = 0; i < nr_found; i++) {
3237 		cip = cilist[i];
3238 		if (cip == ip)
3239 			continue;
3240 
3241 		/*
3242 		 * because this is an RCU protected lookup, we could find a
3243 		 * recently freed or even reallocated inode during the lookup.
3244 		 * We need to check under the i_flags_lock for a valid inode
3245 		 * here. Skip it if it is not valid or the wrong inode.
3246 		 */
3247 		spin_lock(&cip->i_flags_lock);
3248 		if (!cip->i_ino ||
3249 		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3250 			spin_unlock(&cip->i_flags_lock);
3251 			continue;
3252 		}
3253 
3254 		/*
3255 		 * Once we fall off the end of the cluster, no point checking
3256 		 * any more inodes in the list because they will also all be
3257 		 * outside the cluster.
3258 		 */
3259 		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3260 			spin_unlock(&cip->i_flags_lock);
3261 			break;
3262 		}
3263 		spin_unlock(&cip->i_flags_lock);
3264 
3265 		/*
3266 		 * Do an un-protected check to see if the inode is dirty and
3267 		 * is a candidate for flushing.  These checks will be repeated
3268 		 * later after the appropriate locks are acquired.
3269 		 */
3270 		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3271 			continue;
3272 
3273 		/*
3274 		 * Try to get locks.  If any are unavailable or it is pinned,
3275 		 * then this inode cannot be flushed and is skipped.
3276 		 */
3277 
3278 		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3279 			continue;
3280 		if (!xfs_iflock_nowait(cip)) {
3281 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3282 			continue;
3283 		}
3284 		if (xfs_ipincount(cip)) {
3285 			xfs_ifunlock(cip);
3286 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3287 			continue;
3288 		}
3289 
3290 
3291 		/*
3292 		 * Check the inode number again, just to be certain we are not
3293 		 * racing with freeing in xfs_reclaim_inode(). See the comments
3294 		 * in that function for more information as to why the initial
3295 		 * check is not sufficient.
3296 		 */
3297 		if (!cip->i_ino) {
3298 			xfs_ifunlock(cip);
3299 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3300 			continue;
3301 		}
3302 
3303 		/*
3304 		 * arriving here means that this inode can be flushed.  First
3305 		 * re-check that it's dirty before flushing.
3306 		 */
3307 		if (!xfs_inode_clean(cip)) {
3308 			int	error;
3309 			error = xfs_iflush_int(cip, bp);
3310 			if (error) {
3311 				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3312 				goto cluster_corrupt_out;
3313 			}
3314 			clcount++;
3315 		} else {
3316 			xfs_ifunlock(cip);
3317 		}
3318 		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3319 	}
3320 
3321 	if (clcount) {
3322 		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3323 		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3324 	}
3325 
3326 out_free:
3327 	rcu_read_unlock();
3328 	kmem_free(cilist);
3329 out_put:
3330 	xfs_perag_put(pag);
3331 	return 0;
3332 
3333 
3334 cluster_corrupt_out:
3335 	/*
3336 	 * Corruption detected in the clustering loop.  Invalidate the
3337 	 * inode buffer and shut down the filesystem.
3338 	 */
3339 	rcu_read_unlock();
3340 	/*
3341 	 * Clean up the buffer.  If it was delwri, just release it --
3342 	 * brelse can handle it with no problems.  If not, shut down the
3343 	 * filesystem before releasing the buffer.
3344 	 */
3345 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3346 	if (bufwasdelwri)
3347 		xfs_buf_relse(bp);
3348 
3349 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3350 
3351 	if (!bufwasdelwri) {
3352 		/*
3353 		 * Just like incore_relse: if we have b_iodone functions,
3354 		 * mark the buffer as an error and call them.  Otherwise
3355 		 * mark it as stale and brelse.
3356 		 */
3357 		if (bp->b_iodone) {
3358 			bp->b_flags &= ~XBF_DONE;
3359 			xfs_buf_stale(bp);
3360 			xfs_buf_ioerror(bp, -EIO);
3361 			xfs_buf_ioend(bp);
3362 		} else {
3363 			xfs_buf_stale(bp);
3364 			xfs_buf_relse(bp);
3365 		}
3366 	}
3367 
3368 	/*
3369 	 * Unlocks the flush lock
3370 	 */
3371 	xfs_iflush_abort(cip, false);
3372 	kmem_free(cilist);
3373 	xfs_perag_put(pag);
3374 	return -EFSCORRUPTED;
3375 }
3376 
3377 /*
3378  * Flush dirty inode metadata into the backing buffer.
3379  *
3380  * The caller must have the inode lock and the inode flush lock held.  The
3381  * inode lock will still be held upon return to the caller, and the inode
3382  * flush lock will be released after the inode has reached the disk.
3383  *
3384  * The caller must write out the buffer returned in *bpp and release it.
3385  */
3386 int
3387 xfs_iflush(
3388 	struct xfs_inode	*ip,
3389 	struct xfs_buf		**bpp)
3390 {
3391 	struct xfs_mount	*mp = ip->i_mount;
3392 	struct xfs_buf		*bp = NULL;
3393 	struct xfs_dinode	*dip;
3394 	int			error;
3395 
3396 	XFS_STATS_INC(mp, xs_iflush_count);
3397 
3398 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3399 	ASSERT(xfs_isiflocked(ip));
3400 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3401 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3402 
3403 	*bpp = NULL;
3404 
3405 	xfs_iunpin_wait(ip);
3406 
3407 	/*
3408 	 * For stale inodes we cannot rely on the backing buffer remaining
3409 	 * stale in cache for the remaining life of the stale inode and so
3410 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3411 	 * inodes below. We have to check this after ensuring the inode is
3412 	 * unpinned so that it is safe to reclaim the stale inode after the
3413 	 * flush call.
3414 	 */
3415 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3416 		xfs_ifunlock(ip);
3417 		return 0;
3418 	}
3419 
3420 	/*
3421 	 * This may have been unpinned because the filesystem is shutting
3422 	 * down forcibly. If that's the case we must not write this inode
3423 	 * to disk, because the log record didn't make it to disk.
3424 	 *
3425 	 * We also have to remove the log item from the AIL in this case,
3426 	 * as we wait for an empty AIL as part of the unmount process.
3427 	 */
3428 	if (XFS_FORCED_SHUTDOWN(mp)) {
3429 		error = -EIO;
3430 		goto abort_out;
3431 	}
3432 
3433 	/*
3434 	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3435 	 * operation here, so we may get  an EAGAIN error. In that case, we
3436 	 * simply want to return with the inode still dirty.
3437 	 *
3438 	 * If we get any other error, we effectively have a corruption situation
3439 	 * and we cannot flush the inode, so we treat it the same as failing
3440 	 * xfs_iflush_int().
3441 	 */
3442 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3443 			       0);
3444 	if (error == -EAGAIN) {
3445 		xfs_ifunlock(ip);
3446 		return error;
3447 	}
3448 	if (error)
3449 		goto corrupt_out;
3450 
3451 	/*
3452 	 * First flush out the inode that xfs_iflush was called with.
3453 	 */
3454 	error = xfs_iflush_int(ip, bp);
3455 	if (error)
3456 		goto corrupt_out;
3457 
3458 	/*
3459 	 * If the buffer is pinned then push on the log now so we won't
3460 	 * get stuck waiting in the write for too long.
3461 	 */
3462 	if (xfs_buf_ispinned(bp))
3463 		xfs_log_force(mp, 0);
3464 
3465 	/*
3466 	 * inode clustering:
3467 	 * see if other inodes can be gathered into this write
3468 	 */
3469 	error = xfs_iflush_cluster(ip, bp);
3470 	if (error)
3471 		goto cluster_corrupt_out;
3472 
3473 	*bpp = bp;
3474 	return 0;
3475 
3476 corrupt_out:
3477 	if (bp)
3478 		xfs_buf_relse(bp);
3479 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3480 cluster_corrupt_out:
3481 	error = -EFSCORRUPTED;
3482 abort_out:
3483 	/*
3484 	 * Unlocks the flush lock
3485 	 */
3486 	xfs_iflush_abort(ip, false);
3487 	return error;
3488 }
3489 
3490 STATIC int
3491 xfs_iflush_int(
3492 	struct xfs_inode	*ip,
3493 	struct xfs_buf		*bp)
3494 {
3495 	struct xfs_inode_log_item *iip = ip->i_itemp;
3496 	struct xfs_dinode	*dip;
3497 	struct xfs_mount	*mp = ip->i_mount;
3498 
3499 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3500 	ASSERT(xfs_isiflocked(ip));
3501 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3502 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3503 	ASSERT(iip != NULL && iip->ili_fields != 0);
3504 	ASSERT(ip->i_d.di_version > 1);
3505 
3506 	/* set *dip = inode's place in the buffer */
3507 	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3508 
3509 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3510 			       mp, XFS_ERRTAG_IFLUSH_1)) {
3511 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3512 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3513 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3514 		goto corrupt_out;
3515 	}
3516 	if (S_ISREG(VFS_I(ip)->i_mode)) {
3517 		if (XFS_TEST_ERROR(
3518 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3519 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3520 		    mp, XFS_ERRTAG_IFLUSH_3)) {
3521 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3522 				"%s: Bad regular inode %Lu, ptr 0x%p",
3523 				__func__, ip->i_ino, ip);
3524 			goto corrupt_out;
3525 		}
3526 	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3527 		if (XFS_TEST_ERROR(
3528 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3529 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3530 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3531 		    mp, XFS_ERRTAG_IFLUSH_4)) {
3532 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3533 				"%s: Bad directory inode %Lu, ptr 0x%p",
3534 				__func__, ip->i_ino, ip);
3535 			goto corrupt_out;
3536 		}
3537 	}
3538 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3539 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3540 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3541 			"%s: detected corrupt incore inode %Lu, "
3542 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3543 			__func__, ip->i_ino,
3544 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3545 			ip->i_d.di_nblocks, ip);
3546 		goto corrupt_out;
3547 	}
3548 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3549 				mp, XFS_ERRTAG_IFLUSH_6)) {
3550 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3551 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3552 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3553 		goto corrupt_out;
3554 	}
3555 
3556 	/*
3557 	 * Inode item log recovery for v2 inodes are dependent on the
3558 	 * di_flushiter count for correct sequencing. We bump the flush
3559 	 * iteration count so we can detect flushes which postdate a log record
3560 	 * during recovery. This is redundant as we now log every change and
3561 	 * hence this can't happen but we need to still do it to ensure
3562 	 * backwards compatibility with old kernels that predate logging all
3563 	 * inode changes.
3564 	 */
3565 	if (ip->i_d.di_version < 3)
3566 		ip->i_d.di_flushiter++;
3567 
3568 	/* Check the inline directory data. */
3569 	if (S_ISDIR(VFS_I(ip)->i_mode) &&
3570 	    ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3571 	    xfs_dir2_sf_verify(ip))
3572 		goto corrupt_out;
3573 
3574 	/*
3575 	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3576 	 * copy out the core of the inode, because if the inode is dirty at all
3577 	 * the core must be.
3578 	 */
3579 	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3580 
3581 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3582 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3583 		ip->i_d.di_flushiter = 0;
3584 
3585 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3586 	if (XFS_IFORK_Q(ip))
3587 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3588 	xfs_inobp_check(mp, bp);
3589 
3590 	/*
3591 	 * We've recorded everything logged in the inode, so we'd like to clear
3592 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3593 	 * However, we can't stop logging all this information until the data
3594 	 * we've copied into the disk buffer is written to disk.  If we did we
3595 	 * might overwrite the copy of the inode in the log with all the data
3596 	 * after re-logging only part of it, and in the face of a crash we
3597 	 * wouldn't have all the data we need to recover.
3598 	 *
3599 	 * What we do is move the bits to the ili_last_fields field.  When
3600 	 * logging the inode, these bits are moved back to the ili_fields field.
3601 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3602 	 * know that the information those bits represent is permanently on
3603 	 * disk.  As long as the flush completes before the inode is logged
3604 	 * again, then both ili_fields and ili_last_fields will be cleared.
3605 	 *
3606 	 * We can play with the ili_fields bits here, because the inode lock
3607 	 * must be held exclusively in order to set bits there and the flush
3608 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3609 	 * done routine can tell whether or not to look in the AIL.  Also, store
3610 	 * the current LSN of the inode so that we can tell whether the item has
3611 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3612 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3613 	 * atomically.
3614 	 */
3615 	iip->ili_last_fields = iip->ili_fields;
3616 	iip->ili_fields = 0;
3617 	iip->ili_fsync_fields = 0;
3618 	iip->ili_logged = 1;
3619 
3620 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3621 				&iip->ili_item.li_lsn);
3622 
3623 	/*
3624 	 * Attach the function xfs_iflush_done to the inode's
3625 	 * buffer.  This will remove the inode from the AIL
3626 	 * and unlock the inode's flush lock when the inode is
3627 	 * completely written to disk.
3628 	 */
3629 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3630 
3631 	/* generate the checksum. */
3632 	xfs_dinode_calc_crc(mp, dip);
3633 
3634 	ASSERT(bp->b_fspriv != NULL);
3635 	ASSERT(bp->b_iodone != NULL);
3636 	return 0;
3637 
3638 corrupt_out:
3639 	return -EFSCORRUPTED;
3640 }
3641