xref: /linux/fs/xfs/xfs_inode.c (revision ed780686de61ab27e65f1cfedeccd7b45667bd70)
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_inode.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_dir2.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_attr.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_bmap.h"
40 #include "xfs_bmap_util.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_filestream.h"
44 #include "xfs_cksum.h"
45 #include "xfs_trace.h"
46 #include "xfs_icache.h"
47 #include "xfs_symlink.h"
48 #include "xfs_trans_priv.h"
49 #include "xfs_log.h"
50 #include "xfs_bmap_btree.h"
51 
52 kmem_zone_t *xfs_inode_zone;
53 
54 /*
55  * Used in xfs_itruncate_extents().  This is the maximum number of extents
56  * freed from a file in a single transaction.
57  */
58 #define	XFS_ITRUNC_MAX_EXTENTS	2
59 
60 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
61 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
62 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
63 
64 /*
65  * helper function to extract extent size hint from inode
66  */
67 xfs_extlen_t
68 xfs_get_extsz_hint(
69 	struct xfs_inode	*ip)
70 {
71 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
72 		return ip->i_d.di_extsize;
73 	if (XFS_IS_REALTIME_INODE(ip))
74 		return ip->i_mount->m_sb.sb_rextsize;
75 	return 0;
76 }
77 
78 /*
79  * These two are wrapper routines around the xfs_ilock() routine used to
80  * centralize some grungy code.  They are used in places that wish to lock the
81  * inode solely for reading the extents.  The reason these places can't just
82  * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
83  * bringing in of the extents from disk for a file in b-tree format.  If the
84  * inode is in b-tree format, then we need to lock the inode exclusively until
85  * the extents are read in.  Locking it exclusively all the time would limit
86  * our parallelism unnecessarily, though.  What we do instead is check to see
87  * if the extents have been read in yet, and only lock the inode exclusively
88  * if they have not.
89  *
90  * The functions return a value which should be given to the corresponding
91  * xfs_iunlock() call.
92  */
93 uint
94 xfs_ilock_data_map_shared(
95 	struct xfs_inode	*ip)
96 {
97 	uint			lock_mode = XFS_ILOCK_SHARED;
98 
99 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
100 	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
101 		lock_mode = XFS_ILOCK_EXCL;
102 	xfs_ilock(ip, lock_mode);
103 	return lock_mode;
104 }
105 
106 uint
107 xfs_ilock_attr_map_shared(
108 	struct xfs_inode	*ip)
109 {
110 	uint			lock_mode = XFS_ILOCK_SHARED;
111 
112 	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
113 	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
114 		lock_mode = XFS_ILOCK_EXCL;
115 	xfs_ilock(ip, lock_mode);
116 	return lock_mode;
117 }
118 
119 /*
120  * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
121  * the i_lock.  This routine allows various combinations of the locks to be
122  * obtained.
123  *
124  * The 3 locks should always be ordered so that the IO lock is obtained first,
125  * the mmap lock second and the ilock last in order to prevent deadlock.
126  *
127  * Basic locking order:
128  *
129  * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
130  *
131  * mmap_sem locking order:
132  *
133  * i_iolock -> page lock -> mmap_sem
134  * mmap_sem -> i_mmap_lock -> page_lock
135  *
136  * The difference in mmap_sem locking order mean that we cannot hold the
137  * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
138  * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
139  * in get_user_pages() to map the user pages into the kernel address space for
140  * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
141  * page faults already hold the mmap_sem.
142  *
143  * Hence to serialise fully against both syscall and mmap based IO, we need to
144  * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
145  * taken in places where we need to invalidate the page cache in a race
146  * free manner (e.g. truncate, hole punch and other extent manipulation
147  * functions).
148  */
149 void
150 xfs_ilock(
151 	xfs_inode_t		*ip,
152 	uint			lock_flags)
153 {
154 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
155 
156 	/*
157 	 * You can't set both SHARED and EXCL for the same lock,
158 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
159 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
160 	 */
161 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
162 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
163 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
164 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
165 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
166 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
167 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
168 
169 	if (lock_flags & XFS_IOLOCK_EXCL)
170 		mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
171 	else if (lock_flags & XFS_IOLOCK_SHARED)
172 		mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
173 
174 	if (lock_flags & XFS_MMAPLOCK_EXCL)
175 		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
176 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
177 		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
178 
179 	if (lock_flags & XFS_ILOCK_EXCL)
180 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
181 	else if (lock_flags & XFS_ILOCK_SHARED)
182 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
183 }
184 
185 /*
186  * This is just like xfs_ilock(), except that the caller
187  * is guaranteed not to sleep.  It returns 1 if it gets
188  * the requested locks and 0 otherwise.  If the IO lock is
189  * obtained but the inode lock cannot be, then the IO lock
190  * is dropped before returning.
191  *
192  * ip -- the inode being locked
193  * lock_flags -- this parameter indicates the inode's locks to be
194  *       to be locked.  See the comment for xfs_ilock() for a list
195  *	 of valid values.
196  */
197 int
198 xfs_ilock_nowait(
199 	xfs_inode_t		*ip,
200 	uint			lock_flags)
201 {
202 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
203 
204 	/*
205 	 * You can't set both SHARED and EXCL for the same lock,
206 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
207 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
208 	 */
209 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
210 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
211 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
212 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
213 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
214 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
215 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
216 
217 	if (lock_flags & XFS_IOLOCK_EXCL) {
218 		if (!mrtryupdate(&ip->i_iolock))
219 			goto out;
220 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
221 		if (!mrtryaccess(&ip->i_iolock))
222 			goto out;
223 	}
224 
225 	if (lock_flags & XFS_MMAPLOCK_EXCL) {
226 		if (!mrtryupdate(&ip->i_mmaplock))
227 			goto out_undo_iolock;
228 	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
229 		if (!mrtryaccess(&ip->i_mmaplock))
230 			goto out_undo_iolock;
231 	}
232 
233 	if (lock_flags & XFS_ILOCK_EXCL) {
234 		if (!mrtryupdate(&ip->i_lock))
235 			goto out_undo_mmaplock;
236 	} else if (lock_flags & XFS_ILOCK_SHARED) {
237 		if (!mrtryaccess(&ip->i_lock))
238 			goto out_undo_mmaplock;
239 	}
240 	return 1;
241 
242 out_undo_mmaplock:
243 	if (lock_flags & XFS_MMAPLOCK_EXCL)
244 		mrunlock_excl(&ip->i_mmaplock);
245 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
246 		mrunlock_shared(&ip->i_mmaplock);
247 out_undo_iolock:
248 	if (lock_flags & XFS_IOLOCK_EXCL)
249 		mrunlock_excl(&ip->i_iolock);
250 	else if (lock_flags & XFS_IOLOCK_SHARED)
251 		mrunlock_shared(&ip->i_iolock);
252 out:
253 	return 0;
254 }
255 
256 /*
257  * xfs_iunlock() is used to drop the inode locks acquired with
258  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
259  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
260  * that we know which locks to drop.
261  *
262  * ip -- the inode being unlocked
263  * lock_flags -- this parameter indicates the inode's locks to be
264  *       to be unlocked.  See the comment for xfs_ilock() for a list
265  *	 of valid values for this parameter.
266  *
267  */
268 void
269 xfs_iunlock(
270 	xfs_inode_t		*ip,
271 	uint			lock_flags)
272 {
273 	/*
274 	 * You can't set both SHARED and EXCL for the same lock,
275 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
276 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
277 	 */
278 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
279 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
280 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
281 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
282 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
283 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
284 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
285 	ASSERT(lock_flags != 0);
286 
287 	if (lock_flags & XFS_IOLOCK_EXCL)
288 		mrunlock_excl(&ip->i_iolock);
289 	else if (lock_flags & XFS_IOLOCK_SHARED)
290 		mrunlock_shared(&ip->i_iolock);
291 
292 	if (lock_flags & XFS_MMAPLOCK_EXCL)
293 		mrunlock_excl(&ip->i_mmaplock);
294 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
295 		mrunlock_shared(&ip->i_mmaplock);
296 
297 	if (lock_flags & XFS_ILOCK_EXCL)
298 		mrunlock_excl(&ip->i_lock);
299 	else if (lock_flags & XFS_ILOCK_SHARED)
300 		mrunlock_shared(&ip->i_lock);
301 
302 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
303 }
304 
305 /*
306  * give up write locks.  the i/o lock cannot be held nested
307  * if it is being demoted.
308  */
309 void
310 xfs_ilock_demote(
311 	xfs_inode_t		*ip,
312 	uint			lock_flags)
313 {
314 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
315 	ASSERT((lock_flags &
316 		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
317 
318 	if (lock_flags & XFS_ILOCK_EXCL)
319 		mrdemote(&ip->i_lock);
320 	if (lock_flags & XFS_MMAPLOCK_EXCL)
321 		mrdemote(&ip->i_mmaplock);
322 	if (lock_flags & XFS_IOLOCK_EXCL)
323 		mrdemote(&ip->i_iolock);
324 
325 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
326 }
327 
328 #if defined(DEBUG) || defined(XFS_WARN)
329 int
330 xfs_isilocked(
331 	xfs_inode_t		*ip,
332 	uint			lock_flags)
333 {
334 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
335 		if (!(lock_flags & XFS_ILOCK_SHARED))
336 			return !!ip->i_lock.mr_writer;
337 		return rwsem_is_locked(&ip->i_lock.mr_lock);
338 	}
339 
340 	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
341 		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
342 			return !!ip->i_mmaplock.mr_writer;
343 		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
344 	}
345 
346 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
347 		if (!(lock_flags & XFS_IOLOCK_SHARED))
348 			return !!ip->i_iolock.mr_writer;
349 		return rwsem_is_locked(&ip->i_iolock.mr_lock);
350 	}
351 
352 	ASSERT(0);
353 	return 0;
354 }
355 #endif
356 
357 #ifdef DEBUG
358 int xfs_locked_n;
359 int xfs_small_retries;
360 int xfs_middle_retries;
361 int xfs_lots_retries;
362 int xfs_lock_delays;
363 #endif
364 
365 /*
366  * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
367  * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
368  * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
369  * errors and warnings.
370  */
371 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
372 static bool
373 xfs_lockdep_subclass_ok(
374 	int subclass)
375 {
376 	return subclass < MAX_LOCKDEP_SUBCLASSES;
377 }
378 #else
379 #define xfs_lockdep_subclass_ok(subclass)	(true)
380 #endif
381 
382 /*
383  * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
384  * value. This can be called for any type of inode lock combination, including
385  * parent locking. Care must be taken to ensure we don't overrun the subclass
386  * storage fields in the class mask we build.
387  */
388 static inline int
389 xfs_lock_inumorder(int lock_mode, int subclass)
390 {
391 	int	class = 0;
392 
393 	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
394 			      XFS_ILOCK_RTSUM)));
395 	ASSERT(xfs_lockdep_subclass_ok(subclass));
396 
397 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
398 		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
399 		ASSERT(xfs_lockdep_subclass_ok(subclass +
400 						XFS_IOLOCK_PARENT_VAL));
401 		class += subclass << XFS_IOLOCK_SHIFT;
402 		if (lock_mode & XFS_IOLOCK_PARENT)
403 			class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
404 	}
405 
406 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
407 		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
408 		class += subclass << XFS_MMAPLOCK_SHIFT;
409 	}
410 
411 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
412 		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
413 		class += subclass << XFS_ILOCK_SHIFT;
414 	}
415 
416 	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
417 }
418 
419 /*
420  * The following routine will lock n inodes in exclusive mode.  We assume the
421  * caller calls us with the inodes in i_ino order.
422  *
423  * We need to detect deadlock where an inode that we lock is in the AIL and we
424  * start waiting for another inode that is locked by a thread in a long running
425  * transaction (such as truncate). This can result in deadlock since the long
426  * running trans might need to wait for the inode we just locked in order to
427  * push the tail and free space in the log.
428  *
429  * xfs_lock_inodes() can only be used to lock one type of lock at a time -
430  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
431  * lock more than one at a time, lockdep will report false positives saying we
432  * have violated locking orders.
433  */
434 static void
435 xfs_lock_inodes(
436 	xfs_inode_t	**ips,
437 	int		inodes,
438 	uint		lock_mode)
439 {
440 	int		attempts = 0, i, j, try_lock;
441 	xfs_log_item_t	*lp;
442 
443 	/*
444 	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
445 	 * support an arbitrary depth of locking here, but absolute limits on
446 	 * inodes depend on the the type of locking and the limits placed by
447 	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
448 	 * the asserts.
449 	 */
450 	ASSERT(ips && inodes >= 2 && inodes <= 5);
451 	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
452 			    XFS_ILOCK_EXCL));
453 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
454 			      XFS_ILOCK_SHARED)));
455 	ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
456 		inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
457 	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
458 		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
459 	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
460 		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
461 
462 	if (lock_mode & XFS_IOLOCK_EXCL) {
463 		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
464 	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
465 		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
466 
467 	try_lock = 0;
468 	i = 0;
469 again:
470 	for (; i < inodes; i++) {
471 		ASSERT(ips[i]);
472 
473 		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
474 			continue;
475 
476 		/*
477 		 * If try_lock is not set yet, make sure all locked inodes are
478 		 * not in the AIL.  If any are, set try_lock to be used later.
479 		 */
480 		if (!try_lock) {
481 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
482 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
483 				if (lp && (lp->li_flags & XFS_LI_IN_AIL))
484 					try_lock++;
485 			}
486 		}
487 
488 		/*
489 		 * If any of the previous locks we have locked is in the AIL,
490 		 * we must TRY to get the second and subsequent locks. If
491 		 * we can't get any, we must release all we have
492 		 * and try again.
493 		 */
494 		if (!try_lock) {
495 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
496 			continue;
497 		}
498 
499 		/* try_lock means we have an inode locked that is in the AIL. */
500 		ASSERT(i != 0);
501 		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
502 			continue;
503 
504 		/*
505 		 * Unlock all previous guys and try again.  xfs_iunlock will try
506 		 * to push the tail if the inode is in the AIL.
507 		 */
508 		attempts++;
509 		for (j = i - 1; j >= 0; j--) {
510 			/*
511 			 * Check to see if we've already unlocked this one.  Not
512 			 * the first one going back, and the inode ptr is the
513 			 * same.
514 			 */
515 			if (j != (i - 1) && ips[j] == ips[j + 1])
516 				continue;
517 
518 			xfs_iunlock(ips[j], lock_mode);
519 		}
520 
521 		if ((attempts % 5) == 0) {
522 			delay(1); /* Don't just spin the CPU */
523 #ifdef DEBUG
524 			xfs_lock_delays++;
525 #endif
526 		}
527 		i = 0;
528 		try_lock = 0;
529 		goto again;
530 	}
531 
532 #ifdef DEBUG
533 	if (attempts) {
534 		if (attempts < 5) xfs_small_retries++;
535 		else if (attempts < 100) xfs_middle_retries++;
536 		else xfs_lots_retries++;
537 	} else {
538 		xfs_locked_n++;
539 	}
540 #endif
541 }
542 
543 /*
544  * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
545  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
546  * lock more than one at a time, lockdep will report false positives saying we
547  * have violated locking orders.
548  */
549 void
550 xfs_lock_two_inodes(
551 	xfs_inode_t		*ip0,
552 	xfs_inode_t		*ip1,
553 	uint			lock_mode)
554 {
555 	xfs_inode_t		*temp;
556 	int			attempts = 0;
557 	xfs_log_item_t		*lp;
558 
559 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
560 		ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
561 		ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 	} else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
563 		ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564 
565 	ASSERT(ip0->i_ino != ip1->i_ino);
566 
567 	if (ip0->i_ino > ip1->i_ino) {
568 		temp = ip0;
569 		ip0 = ip1;
570 		ip1 = temp;
571 	}
572 
573  again:
574 	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
575 
576 	/*
577 	 * If the first lock we have locked is in the AIL, we must TRY to get
578 	 * the second lock. If we can't get it, we must release the first one
579 	 * and try again.
580 	 */
581 	lp = (xfs_log_item_t *)ip0->i_itemp;
582 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
583 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
584 			xfs_iunlock(ip0, lock_mode);
585 			if ((++attempts % 5) == 0)
586 				delay(1); /* Don't just spin the CPU */
587 			goto again;
588 		}
589 	} else {
590 		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
591 	}
592 }
593 
594 
595 void
596 __xfs_iflock(
597 	struct xfs_inode	*ip)
598 {
599 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
600 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
601 
602 	do {
603 		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
604 		if (xfs_isiflocked(ip))
605 			io_schedule();
606 	} while (!xfs_iflock_nowait(ip));
607 
608 	finish_wait(wq, &wait.wait);
609 }
610 
611 STATIC uint
612 _xfs_dic2xflags(
613 	__uint16_t		di_flags,
614 	uint64_t		di_flags2,
615 	bool			has_attr)
616 {
617 	uint			flags = 0;
618 
619 	if (di_flags & XFS_DIFLAG_ANY) {
620 		if (di_flags & XFS_DIFLAG_REALTIME)
621 			flags |= FS_XFLAG_REALTIME;
622 		if (di_flags & XFS_DIFLAG_PREALLOC)
623 			flags |= FS_XFLAG_PREALLOC;
624 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
625 			flags |= FS_XFLAG_IMMUTABLE;
626 		if (di_flags & XFS_DIFLAG_APPEND)
627 			flags |= FS_XFLAG_APPEND;
628 		if (di_flags & XFS_DIFLAG_SYNC)
629 			flags |= FS_XFLAG_SYNC;
630 		if (di_flags & XFS_DIFLAG_NOATIME)
631 			flags |= FS_XFLAG_NOATIME;
632 		if (di_flags & XFS_DIFLAG_NODUMP)
633 			flags |= FS_XFLAG_NODUMP;
634 		if (di_flags & XFS_DIFLAG_RTINHERIT)
635 			flags |= FS_XFLAG_RTINHERIT;
636 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
637 			flags |= FS_XFLAG_PROJINHERIT;
638 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
639 			flags |= FS_XFLAG_NOSYMLINKS;
640 		if (di_flags & XFS_DIFLAG_EXTSIZE)
641 			flags |= FS_XFLAG_EXTSIZE;
642 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
643 			flags |= FS_XFLAG_EXTSZINHERIT;
644 		if (di_flags & XFS_DIFLAG_NODEFRAG)
645 			flags |= FS_XFLAG_NODEFRAG;
646 		if (di_flags & XFS_DIFLAG_FILESTREAM)
647 			flags |= FS_XFLAG_FILESTREAM;
648 	}
649 
650 	if (di_flags2 & XFS_DIFLAG2_ANY) {
651 		if (di_flags2 & XFS_DIFLAG2_DAX)
652 			flags |= FS_XFLAG_DAX;
653 	}
654 
655 	if (has_attr)
656 		flags |= FS_XFLAG_HASATTR;
657 
658 	return flags;
659 }
660 
661 uint
662 xfs_ip2xflags(
663 	struct xfs_inode	*ip)
664 {
665 	struct xfs_icdinode	*dic = &ip->i_d;
666 
667 	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
668 }
669 
670 /*
671  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
672  * is allowed, otherwise it has to be an exact match. If a CI match is found,
673  * ci_name->name will point to a the actual name (caller must free) or
674  * will be set to NULL if an exact match is found.
675  */
676 int
677 xfs_lookup(
678 	xfs_inode_t		*dp,
679 	struct xfs_name		*name,
680 	xfs_inode_t		**ipp,
681 	struct xfs_name		*ci_name)
682 {
683 	xfs_ino_t		inum;
684 	int			error;
685 
686 	trace_xfs_lookup(dp, name);
687 
688 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
689 		return -EIO;
690 
691 	xfs_ilock(dp, XFS_IOLOCK_SHARED);
692 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
693 	if (error)
694 		goto out_unlock;
695 
696 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
697 	if (error)
698 		goto out_free_name;
699 
700 	xfs_iunlock(dp, XFS_IOLOCK_SHARED);
701 	return 0;
702 
703 out_free_name:
704 	if (ci_name)
705 		kmem_free(ci_name->name);
706 out_unlock:
707 	xfs_iunlock(dp, XFS_IOLOCK_SHARED);
708 	*ipp = NULL;
709 	return error;
710 }
711 
712 /*
713  * Allocate an inode on disk and return a copy of its in-core version.
714  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
715  * appropriately within the inode.  The uid and gid for the inode are
716  * set according to the contents of the given cred structure.
717  *
718  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
719  * has a free inode available, call xfs_iget() to obtain the in-core
720  * version of the allocated inode.  Finally, fill in the inode and
721  * log its initial contents.  In this case, ialloc_context would be
722  * set to NULL.
723  *
724  * If xfs_dialloc() does not have an available inode, it will replenish
725  * its supply by doing an allocation. Since we can only do one
726  * allocation within a transaction without deadlocks, we must commit
727  * the current transaction before returning the inode itself.
728  * In this case, therefore, we will set ialloc_context and return.
729  * The caller should then commit the current transaction, start a new
730  * transaction, and call xfs_ialloc() again to actually get the inode.
731  *
732  * To ensure that some other process does not grab the inode that
733  * was allocated during the first call to xfs_ialloc(), this routine
734  * also returns the [locked] bp pointing to the head of the freelist
735  * as ialloc_context.  The caller should hold this buffer across
736  * the commit and pass it back into this routine on the second call.
737  *
738  * If we are allocating quota inodes, we do not have a parent inode
739  * to attach to or associate with (i.e. pip == NULL) because they
740  * are not linked into the directory structure - they are attached
741  * directly to the superblock - and so have no parent.
742  */
743 static int
744 xfs_ialloc(
745 	xfs_trans_t	*tp,
746 	xfs_inode_t	*pip,
747 	umode_t		mode,
748 	xfs_nlink_t	nlink,
749 	xfs_dev_t	rdev,
750 	prid_t		prid,
751 	int		okalloc,
752 	xfs_buf_t	**ialloc_context,
753 	xfs_inode_t	**ipp)
754 {
755 	struct xfs_mount *mp = tp->t_mountp;
756 	xfs_ino_t	ino;
757 	xfs_inode_t	*ip;
758 	uint		flags;
759 	int		error;
760 	struct timespec	tv;
761 	struct inode	*inode;
762 
763 	/*
764 	 * Call the space management code to pick
765 	 * the on-disk inode to be allocated.
766 	 */
767 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
768 			    ialloc_context, &ino);
769 	if (error)
770 		return error;
771 	if (*ialloc_context || ino == NULLFSINO) {
772 		*ipp = NULL;
773 		return 0;
774 	}
775 	ASSERT(*ialloc_context == NULL);
776 
777 	/*
778 	 * Get the in-core inode with the lock held exclusively.
779 	 * This is because we're setting fields here we need
780 	 * to prevent others from looking at until we're done.
781 	 */
782 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
783 			 XFS_ILOCK_EXCL, &ip);
784 	if (error)
785 		return error;
786 	ASSERT(ip != NULL);
787 	inode = VFS_I(ip);
788 
789 	/*
790 	 * We always convert v1 inodes to v2 now - we only support filesystems
791 	 * with >= v2 inode capability, so there is no reason for ever leaving
792 	 * an inode in v1 format.
793 	 */
794 	if (ip->i_d.di_version == 1)
795 		ip->i_d.di_version = 2;
796 
797 	inode->i_mode = mode;
798 	set_nlink(inode, nlink);
799 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
800 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
801 	xfs_set_projid(ip, prid);
802 
803 	if (pip && XFS_INHERIT_GID(pip)) {
804 		ip->i_d.di_gid = pip->i_d.di_gid;
805 		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
806 			inode->i_mode |= S_ISGID;
807 	}
808 
809 	/*
810 	 * If the group ID of the new file does not match the effective group
811 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
812 	 * (and only if the irix_sgid_inherit compatibility variable is set).
813 	 */
814 	if ((irix_sgid_inherit) &&
815 	    (inode->i_mode & S_ISGID) &&
816 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
817 		inode->i_mode &= ~S_ISGID;
818 
819 	ip->i_d.di_size = 0;
820 	ip->i_d.di_nextents = 0;
821 	ASSERT(ip->i_d.di_nblocks == 0);
822 
823 	tv = current_fs_time(mp->m_super);
824 	inode->i_mtime = tv;
825 	inode->i_atime = tv;
826 	inode->i_ctime = tv;
827 
828 	ip->i_d.di_extsize = 0;
829 	ip->i_d.di_dmevmask = 0;
830 	ip->i_d.di_dmstate = 0;
831 	ip->i_d.di_flags = 0;
832 
833 	if (ip->i_d.di_version == 3) {
834 		inode->i_version = 1;
835 		ip->i_d.di_flags2 = 0;
836 		ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
837 		ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
838 	}
839 
840 
841 	flags = XFS_ILOG_CORE;
842 	switch (mode & S_IFMT) {
843 	case S_IFIFO:
844 	case S_IFCHR:
845 	case S_IFBLK:
846 	case S_IFSOCK:
847 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
848 		ip->i_df.if_u2.if_rdev = rdev;
849 		ip->i_df.if_flags = 0;
850 		flags |= XFS_ILOG_DEV;
851 		break;
852 	case S_IFREG:
853 	case S_IFDIR:
854 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
855 			uint64_t	di_flags2 = 0;
856 			uint		di_flags = 0;
857 
858 			if (S_ISDIR(mode)) {
859 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
860 					di_flags |= XFS_DIFLAG_RTINHERIT;
861 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
862 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
863 					ip->i_d.di_extsize = pip->i_d.di_extsize;
864 				}
865 				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
866 					di_flags |= XFS_DIFLAG_PROJINHERIT;
867 			} else if (S_ISREG(mode)) {
868 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
869 					di_flags |= XFS_DIFLAG_REALTIME;
870 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
871 					di_flags |= XFS_DIFLAG_EXTSIZE;
872 					ip->i_d.di_extsize = pip->i_d.di_extsize;
873 				}
874 			}
875 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
876 			    xfs_inherit_noatime)
877 				di_flags |= XFS_DIFLAG_NOATIME;
878 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
879 			    xfs_inherit_nodump)
880 				di_flags |= XFS_DIFLAG_NODUMP;
881 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
882 			    xfs_inherit_sync)
883 				di_flags |= XFS_DIFLAG_SYNC;
884 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
885 			    xfs_inherit_nosymlinks)
886 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
887 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
888 			    xfs_inherit_nodefrag)
889 				di_flags |= XFS_DIFLAG_NODEFRAG;
890 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
891 				di_flags |= XFS_DIFLAG_FILESTREAM;
892 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
893 				di_flags2 |= XFS_DIFLAG2_DAX;
894 
895 			ip->i_d.di_flags |= di_flags;
896 			ip->i_d.di_flags2 |= di_flags2;
897 		}
898 		/* FALLTHROUGH */
899 	case S_IFLNK:
900 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
901 		ip->i_df.if_flags = XFS_IFEXTENTS;
902 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
903 		ip->i_df.if_u1.if_extents = NULL;
904 		break;
905 	default:
906 		ASSERT(0);
907 	}
908 	/*
909 	 * Attribute fork settings for new inode.
910 	 */
911 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
912 	ip->i_d.di_anextents = 0;
913 
914 	/*
915 	 * Log the new values stuffed into the inode.
916 	 */
917 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
918 	xfs_trans_log_inode(tp, ip, flags);
919 
920 	/* now that we have an i_mode we can setup the inode structure */
921 	xfs_setup_inode(ip);
922 
923 	*ipp = ip;
924 	return 0;
925 }
926 
927 /*
928  * Allocates a new inode from disk and return a pointer to the
929  * incore copy. This routine will internally commit the current
930  * transaction and allocate a new one if the Space Manager needed
931  * to do an allocation to replenish the inode free-list.
932  *
933  * This routine is designed to be called from xfs_create and
934  * xfs_create_dir.
935  *
936  */
937 int
938 xfs_dir_ialloc(
939 	xfs_trans_t	**tpp,		/* input: current transaction;
940 					   output: may be a new transaction. */
941 	xfs_inode_t	*dp,		/* directory within whose allocate
942 					   the inode. */
943 	umode_t		mode,
944 	xfs_nlink_t	nlink,
945 	xfs_dev_t	rdev,
946 	prid_t		prid,		/* project id */
947 	int		okalloc,	/* ok to allocate new space */
948 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
949 					   locked. */
950 	int		*committed)
951 
952 {
953 	xfs_trans_t	*tp;
954 	xfs_inode_t	*ip;
955 	xfs_buf_t	*ialloc_context = NULL;
956 	int		code;
957 	void		*dqinfo;
958 	uint		tflags;
959 
960 	tp = *tpp;
961 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
962 
963 	/*
964 	 * xfs_ialloc will return a pointer to an incore inode if
965 	 * the Space Manager has an available inode on the free
966 	 * list. Otherwise, it will do an allocation and replenish
967 	 * the freelist.  Since we can only do one allocation per
968 	 * transaction without deadlocks, we will need to commit the
969 	 * current transaction and start a new one.  We will then
970 	 * need to call xfs_ialloc again to get the inode.
971 	 *
972 	 * If xfs_ialloc did an allocation to replenish the freelist,
973 	 * it returns the bp containing the head of the freelist as
974 	 * ialloc_context. We will hold a lock on it across the
975 	 * transaction commit so that no other process can steal
976 	 * the inode(s) that we've just allocated.
977 	 */
978 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
979 			  &ialloc_context, &ip);
980 
981 	/*
982 	 * Return an error if we were unable to allocate a new inode.
983 	 * This should only happen if we run out of space on disk or
984 	 * encounter a disk error.
985 	 */
986 	if (code) {
987 		*ipp = NULL;
988 		return code;
989 	}
990 	if (!ialloc_context && !ip) {
991 		*ipp = NULL;
992 		return -ENOSPC;
993 	}
994 
995 	/*
996 	 * If the AGI buffer is non-NULL, then we were unable to get an
997 	 * inode in one operation.  We need to commit the current
998 	 * transaction and call xfs_ialloc() again.  It is guaranteed
999 	 * to succeed the second time.
1000 	 */
1001 	if (ialloc_context) {
1002 		/*
1003 		 * Normally, xfs_trans_commit releases all the locks.
1004 		 * We call bhold to hang on to the ialloc_context across
1005 		 * the commit.  Holding this buffer prevents any other
1006 		 * processes from doing any allocations in this
1007 		 * allocation group.
1008 		 */
1009 		xfs_trans_bhold(tp, ialloc_context);
1010 
1011 		/*
1012 		 * We want the quota changes to be associated with the next
1013 		 * transaction, NOT this one. So, detach the dqinfo from this
1014 		 * and attach it to the next transaction.
1015 		 */
1016 		dqinfo = NULL;
1017 		tflags = 0;
1018 		if (tp->t_dqinfo) {
1019 			dqinfo = (void *)tp->t_dqinfo;
1020 			tp->t_dqinfo = NULL;
1021 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1022 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1023 		}
1024 
1025 		code = xfs_trans_roll(&tp, NULL);
1026 		if (committed != NULL)
1027 			*committed = 1;
1028 
1029 		/*
1030 		 * Re-attach the quota info that we detached from prev trx.
1031 		 */
1032 		if (dqinfo) {
1033 			tp->t_dqinfo = dqinfo;
1034 			tp->t_flags |= tflags;
1035 		}
1036 
1037 		if (code) {
1038 			xfs_buf_relse(ialloc_context);
1039 			*tpp = tp;
1040 			*ipp = NULL;
1041 			return code;
1042 		}
1043 		xfs_trans_bjoin(tp, ialloc_context);
1044 
1045 		/*
1046 		 * Call ialloc again. Since we've locked out all
1047 		 * other allocations in this allocation group,
1048 		 * this call should always succeed.
1049 		 */
1050 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1051 				  okalloc, &ialloc_context, &ip);
1052 
1053 		/*
1054 		 * If we get an error at this point, return to the caller
1055 		 * so that the current transaction can be aborted.
1056 		 */
1057 		if (code) {
1058 			*tpp = tp;
1059 			*ipp = NULL;
1060 			return code;
1061 		}
1062 		ASSERT(!ialloc_context && ip);
1063 
1064 	} else {
1065 		if (committed != NULL)
1066 			*committed = 0;
1067 	}
1068 
1069 	*ipp = ip;
1070 	*tpp = tp;
1071 
1072 	return 0;
1073 }
1074 
1075 /*
1076  * Decrement the link count on an inode & log the change.  If this causes the
1077  * link count to go to zero, move the inode to AGI unlinked list so that it can
1078  * be freed when the last active reference goes away via xfs_inactive().
1079  */
1080 static int			/* error */
1081 xfs_droplink(
1082 	xfs_trans_t *tp,
1083 	xfs_inode_t *ip)
1084 {
1085 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1086 
1087 	drop_nlink(VFS_I(ip));
1088 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1089 
1090 	if (VFS_I(ip)->i_nlink)
1091 		return 0;
1092 
1093 	return xfs_iunlink(tp, ip);
1094 }
1095 
1096 /*
1097  * Increment the link count on an inode & log the change.
1098  */
1099 static int
1100 xfs_bumplink(
1101 	xfs_trans_t *tp,
1102 	xfs_inode_t *ip)
1103 {
1104 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1105 
1106 	ASSERT(ip->i_d.di_version > 1);
1107 	inc_nlink(VFS_I(ip));
1108 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1109 	return 0;
1110 }
1111 
1112 int
1113 xfs_create(
1114 	xfs_inode_t		*dp,
1115 	struct xfs_name		*name,
1116 	umode_t			mode,
1117 	xfs_dev_t		rdev,
1118 	xfs_inode_t		**ipp)
1119 {
1120 	int			is_dir = S_ISDIR(mode);
1121 	struct xfs_mount	*mp = dp->i_mount;
1122 	struct xfs_inode	*ip = NULL;
1123 	struct xfs_trans	*tp = NULL;
1124 	int			error;
1125 	xfs_bmap_free_t		free_list;
1126 	xfs_fsblock_t		first_block;
1127 	bool                    unlock_dp_on_error = false;
1128 	prid_t			prid;
1129 	struct xfs_dquot	*udqp = NULL;
1130 	struct xfs_dquot	*gdqp = NULL;
1131 	struct xfs_dquot	*pdqp = NULL;
1132 	struct xfs_trans_res	*tres;
1133 	uint			resblks;
1134 
1135 	trace_xfs_create(dp, name);
1136 
1137 	if (XFS_FORCED_SHUTDOWN(mp))
1138 		return -EIO;
1139 
1140 	prid = xfs_get_initial_prid(dp);
1141 
1142 	/*
1143 	 * Make sure that we have allocated dquot(s) on disk.
1144 	 */
1145 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1146 					xfs_kgid_to_gid(current_fsgid()), prid,
1147 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1148 					&udqp, &gdqp, &pdqp);
1149 	if (error)
1150 		return error;
1151 
1152 	if (is_dir) {
1153 		rdev = 0;
1154 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1155 		tres = &M_RES(mp)->tr_mkdir;
1156 	} else {
1157 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1158 		tres = &M_RES(mp)->tr_create;
1159 	}
1160 
1161 	/*
1162 	 * Initially assume that the file does not exist and
1163 	 * reserve the resources for that case.  If that is not
1164 	 * the case we'll drop the one we have and get a more
1165 	 * appropriate transaction later.
1166 	 */
1167 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1168 	if (error == -ENOSPC) {
1169 		/* flush outstanding delalloc blocks and retry */
1170 		xfs_flush_inodes(mp);
1171 		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1172 	}
1173 	if (error == -ENOSPC) {
1174 		/* No space at all so try a "no-allocation" reservation */
1175 		resblks = 0;
1176 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1177 	}
1178 	if (error)
1179 		goto out_release_inode;
1180 
1181 	xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1182 		      XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
1183 	unlock_dp_on_error = true;
1184 
1185 	xfs_bmap_init(&free_list, &first_block);
1186 
1187 	/*
1188 	 * Reserve disk quota and the inode.
1189 	 */
1190 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1191 						pdqp, resblks, 1, 0);
1192 	if (error)
1193 		goto out_trans_cancel;
1194 
1195 	if (!resblks) {
1196 		error = xfs_dir_canenter(tp, dp, name);
1197 		if (error)
1198 			goto out_trans_cancel;
1199 	}
1200 
1201 	/*
1202 	 * A newly created regular or special file just has one directory
1203 	 * entry pointing to them, but a directory also the "." entry
1204 	 * pointing to itself.
1205 	 */
1206 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1207 			       prid, resblks > 0, &ip, NULL);
1208 	if (error)
1209 		goto out_trans_cancel;
1210 
1211 	/*
1212 	 * Now we join the directory inode to the transaction.  We do not do it
1213 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1214 	 * (and release all the locks).  An error from here on will result in
1215 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1216 	 * error path.
1217 	 */
1218 	xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1219 	unlock_dp_on_error = false;
1220 
1221 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1222 					&first_block, &free_list, resblks ?
1223 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1224 	if (error) {
1225 		ASSERT(error != -ENOSPC);
1226 		goto out_trans_cancel;
1227 	}
1228 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1229 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1230 
1231 	if (is_dir) {
1232 		error = xfs_dir_init(tp, ip, dp);
1233 		if (error)
1234 			goto out_bmap_cancel;
1235 
1236 		error = xfs_bumplink(tp, dp);
1237 		if (error)
1238 			goto out_bmap_cancel;
1239 	}
1240 
1241 	/*
1242 	 * If this is a synchronous mount, make sure that the
1243 	 * create transaction goes to disk before returning to
1244 	 * the user.
1245 	 */
1246 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1247 		xfs_trans_set_sync(tp);
1248 
1249 	/*
1250 	 * Attach the dquot(s) to the inodes and modify them incore.
1251 	 * These ids of the inode couldn't have changed since the new
1252 	 * inode has been locked ever since it was created.
1253 	 */
1254 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1255 
1256 	error = xfs_bmap_finish(&tp, &free_list, NULL);
1257 	if (error)
1258 		goto out_bmap_cancel;
1259 
1260 	error = xfs_trans_commit(tp);
1261 	if (error)
1262 		goto out_release_inode;
1263 
1264 	xfs_qm_dqrele(udqp);
1265 	xfs_qm_dqrele(gdqp);
1266 	xfs_qm_dqrele(pdqp);
1267 
1268 	*ipp = ip;
1269 	return 0;
1270 
1271  out_bmap_cancel:
1272 	xfs_bmap_cancel(&free_list);
1273  out_trans_cancel:
1274 	xfs_trans_cancel(tp);
1275  out_release_inode:
1276 	/*
1277 	 * Wait until after the current transaction is aborted to finish the
1278 	 * setup of the inode and release the inode.  This prevents recursive
1279 	 * transactions and deadlocks from xfs_inactive.
1280 	 */
1281 	if (ip) {
1282 		xfs_finish_inode_setup(ip);
1283 		IRELE(ip);
1284 	}
1285 
1286 	xfs_qm_dqrele(udqp);
1287 	xfs_qm_dqrele(gdqp);
1288 	xfs_qm_dqrele(pdqp);
1289 
1290 	if (unlock_dp_on_error)
1291 		xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1292 	return error;
1293 }
1294 
1295 int
1296 xfs_create_tmpfile(
1297 	struct xfs_inode	*dp,
1298 	struct dentry		*dentry,
1299 	umode_t			mode,
1300 	struct xfs_inode	**ipp)
1301 {
1302 	struct xfs_mount	*mp = dp->i_mount;
1303 	struct xfs_inode	*ip = NULL;
1304 	struct xfs_trans	*tp = NULL;
1305 	int			error;
1306 	prid_t                  prid;
1307 	struct xfs_dquot	*udqp = NULL;
1308 	struct xfs_dquot	*gdqp = NULL;
1309 	struct xfs_dquot	*pdqp = NULL;
1310 	struct xfs_trans_res	*tres;
1311 	uint			resblks;
1312 
1313 	if (XFS_FORCED_SHUTDOWN(mp))
1314 		return -EIO;
1315 
1316 	prid = xfs_get_initial_prid(dp);
1317 
1318 	/*
1319 	 * Make sure that we have allocated dquot(s) on disk.
1320 	 */
1321 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1322 				xfs_kgid_to_gid(current_fsgid()), prid,
1323 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1324 				&udqp, &gdqp, &pdqp);
1325 	if (error)
1326 		return error;
1327 
1328 	resblks = XFS_IALLOC_SPACE_RES(mp);
1329 	tres = &M_RES(mp)->tr_create_tmpfile;
1330 
1331 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1332 	if (error == -ENOSPC) {
1333 		/* No space at all so try a "no-allocation" reservation */
1334 		resblks = 0;
1335 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1336 	}
1337 	if (error)
1338 		goto out_release_inode;
1339 
1340 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1341 						pdqp, resblks, 1, 0);
1342 	if (error)
1343 		goto out_trans_cancel;
1344 
1345 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1346 				prid, resblks > 0, &ip, NULL);
1347 	if (error)
1348 		goto out_trans_cancel;
1349 
1350 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1351 		xfs_trans_set_sync(tp);
1352 
1353 	/*
1354 	 * Attach the dquot(s) to the inodes and modify them incore.
1355 	 * These ids of the inode couldn't have changed since the new
1356 	 * inode has been locked ever since it was created.
1357 	 */
1358 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1359 
1360 	error = xfs_iunlink(tp, ip);
1361 	if (error)
1362 		goto out_trans_cancel;
1363 
1364 	error = xfs_trans_commit(tp);
1365 	if (error)
1366 		goto out_release_inode;
1367 
1368 	xfs_qm_dqrele(udqp);
1369 	xfs_qm_dqrele(gdqp);
1370 	xfs_qm_dqrele(pdqp);
1371 
1372 	*ipp = ip;
1373 	return 0;
1374 
1375  out_trans_cancel:
1376 	xfs_trans_cancel(tp);
1377  out_release_inode:
1378 	/*
1379 	 * Wait until after the current transaction is aborted to finish the
1380 	 * setup of the inode and release the inode.  This prevents recursive
1381 	 * transactions and deadlocks from xfs_inactive.
1382 	 */
1383 	if (ip) {
1384 		xfs_finish_inode_setup(ip);
1385 		IRELE(ip);
1386 	}
1387 
1388 	xfs_qm_dqrele(udqp);
1389 	xfs_qm_dqrele(gdqp);
1390 	xfs_qm_dqrele(pdqp);
1391 
1392 	return error;
1393 }
1394 
1395 int
1396 xfs_link(
1397 	xfs_inode_t		*tdp,
1398 	xfs_inode_t		*sip,
1399 	struct xfs_name		*target_name)
1400 {
1401 	xfs_mount_t		*mp = tdp->i_mount;
1402 	xfs_trans_t		*tp;
1403 	int			error;
1404 	xfs_bmap_free_t         free_list;
1405 	xfs_fsblock_t           first_block;
1406 	int			resblks;
1407 
1408 	trace_xfs_link(tdp, target_name);
1409 
1410 	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1411 
1412 	if (XFS_FORCED_SHUTDOWN(mp))
1413 		return -EIO;
1414 
1415 	error = xfs_qm_dqattach(sip, 0);
1416 	if (error)
1417 		goto std_return;
1418 
1419 	error = xfs_qm_dqattach(tdp, 0);
1420 	if (error)
1421 		goto std_return;
1422 
1423 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1424 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1425 	if (error == -ENOSPC) {
1426 		resblks = 0;
1427 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1428 	}
1429 	if (error)
1430 		goto std_return;
1431 
1432 	xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
1433 	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1434 
1435 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1436 	xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1437 
1438 	/*
1439 	 * If we are using project inheritance, we only allow hard link
1440 	 * creation in our tree when the project IDs are the same; else
1441 	 * the tree quota mechanism could be circumvented.
1442 	 */
1443 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1444 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1445 		error = -EXDEV;
1446 		goto error_return;
1447 	}
1448 
1449 	if (!resblks) {
1450 		error = xfs_dir_canenter(tp, tdp, target_name);
1451 		if (error)
1452 			goto error_return;
1453 	}
1454 
1455 	xfs_bmap_init(&free_list, &first_block);
1456 
1457 	/*
1458 	 * Handle initial link state of O_TMPFILE inode
1459 	 */
1460 	if (VFS_I(sip)->i_nlink == 0) {
1461 		error = xfs_iunlink_remove(tp, sip);
1462 		if (error)
1463 			goto error_return;
1464 	}
1465 
1466 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1467 					&first_block, &free_list, resblks);
1468 	if (error)
1469 		goto error_return;
1470 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1471 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1472 
1473 	error = xfs_bumplink(tp, sip);
1474 	if (error)
1475 		goto error_return;
1476 
1477 	/*
1478 	 * If this is a synchronous mount, make sure that the
1479 	 * link transaction goes to disk before returning to
1480 	 * the user.
1481 	 */
1482 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1483 		xfs_trans_set_sync(tp);
1484 
1485 	error = xfs_bmap_finish(&tp, &free_list, NULL);
1486 	if (error) {
1487 		xfs_bmap_cancel(&free_list);
1488 		goto error_return;
1489 	}
1490 
1491 	return xfs_trans_commit(tp);
1492 
1493  error_return:
1494 	xfs_trans_cancel(tp);
1495  std_return:
1496 	return error;
1497 }
1498 
1499 /*
1500  * Free up the underlying blocks past new_size.  The new size must be smaller
1501  * than the current size.  This routine can be used both for the attribute and
1502  * data fork, and does not modify the inode size, which is left to the caller.
1503  *
1504  * The transaction passed to this routine must have made a permanent log
1505  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1506  * given transaction and start new ones, so make sure everything involved in
1507  * the transaction is tidy before calling here.  Some transaction will be
1508  * returned to the caller to be committed.  The incoming transaction must
1509  * already include the inode, and both inode locks must be held exclusively.
1510  * The inode must also be "held" within the transaction.  On return the inode
1511  * will be "held" within the returned transaction.  This routine does NOT
1512  * require any disk space to be reserved for it within the transaction.
1513  *
1514  * If we get an error, we must return with the inode locked and linked into the
1515  * current transaction. This keeps things simple for the higher level code,
1516  * because it always knows that the inode is locked and held in the transaction
1517  * that returns to it whether errors occur or not.  We don't mark the inode
1518  * dirty on error so that transactions can be easily aborted if possible.
1519  */
1520 int
1521 xfs_itruncate_extents(
1522 	struct xfs_trans	**tpp,
1523 	struct xfs_inode	*ip,
1524 	int			whichfork,
1525 	xfs_fsize_t		new_size)
1526 {
1527 	struct xfs_mount	*mp = ip->i_mount;
1528 	struct xfs_trans	*tp = *tpp;
1529 	xfs_bmap_free_t		free_list;
1530 	xfs_fsblock_t		first_block;
1531 	xfs_fileoff_t		first_unmap_block;
1532 	xfs_fileoff_t		last_block;
1533 	xfs_filblks_t		unmap_len;
1534 	int			error = 0;
1535 	int			done = 0;
1536 
1537 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1538 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1539 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1540 	ASSERT(new_size <= XFS_ISIZE(ip));
1541 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1542 	ASSERT(ip->i_itemp != NULL);
1543 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1544 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1545 
1546 	trace_xfs_itruncate_extents_start(ip, new_size);
1547 
1548 	/*
1549 	 * Since it is possible for space to become allocated beyond
1550 	 * the end of the file (in a crash where the space is allocated
1551 	 * but the inode size is not yet updated), simply remove any
1552 	 * blocks which show up between the new EOF and the maximum
1553 	 * possible file size.  If the first block to be removed is
1554 	 * beyond the maximum file size (ie it is the same as last_block),
1555 	 * then there is nothing to do.
1556 	 */
1557 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1558 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1559 	if (first_unmap_block == last_block)
1560 		return 0;
1561 
1562 	ASSERT(first_unmap_block < last_block);
1563 	unmap_len = last_block - first_unmap_block + 1;
1564 	while (!done) {
1565 		xfs_bmap_init(&free_list, &first_block);
1566 		error = xfs_bunmapi(tp, ip,
1567 				    first_unmap_block, unmap_len,
1568 				    xfs_bmapi_aflag(whichfork),
1569 				    XFS_ITRUNC_MAX_EXTENTS,
1570 				    &first_block, &free_list,
1571 				    &done);
1572 		if (error)
1573 			goto out_bmap_cancel;
1574 
1575 		/*
1576 		 * Duplicate the transaction that has the permanent
1577 		 * reservation and commit the old transaction.
1578 		 */
1579 		error = xfs_bmap_finish(&tp, &free_list, ip);
1580 		if (error)
1581 			goto out_bmap_cancel;
1582 
1583 		error = xfs_trans_roll(&tp, ip);
1584 		if (error)
1585 			goto out;
1586 	}
1587 
1588 	/*
1589 	 * Always re-log the inode so that our permanent transaction can keep
1590 	 * on rolling it forward in the log.
1591 	 */
1592 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1593 
1594 	trace_xfs_itruncate_extents_end(ip, new_size);
1595 
1596 out:
1597 	*tpp = tp;
1598 	return error;
1599 out_bmap_cancel:
1600 	/*
1601 	 * If the bunmapi call encounters an error, return to the caller where
1602 	 * the transaction can be properly aborted.  We just need to make sure
1603 	 * we're not holding any resources that we were not when we came in.
1604 	 */
1605 	xfs_bmap_cancel(&free_list);
1606 	goto out;
1607 }
1608 
1609 int
1610 xfs_release(
1611 	xfs_inode_t	*ip)
1612 {
1613 	xfs_mount_t	*mp = ip->i_mount;
1614 	int		error;
1615 
1616 	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1617 		return 0;
1618 
1619 	/* If this is a read-only mount, don't do this (would generate I/O) */
1620 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1621 		return 0;
1622 
1623 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1624 		int truncated;
1625 
1626 		/*
1627 		 * If we previously truncated this file and removed old data
1628 		 * in the process, we want to initiate "early" writeout on
1629 		 * the last close.  This is an attempt to combat the notorious
1630 		 * NULL files problem which is particularly noticeable from a
1631 		 * truncate down, buffered (re-)write (delalloc), followed by
1632 		 * a crash.  What we are effectively doing here is
1633 		 * significantly reducing the time window where we'd otherwise
1634 		 * be exposed to that problem.
1635 		 */
1636 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1637 		if (truncated) {
1638 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1639 			if (ip->i_delayed_blks > 0) {
1640 				error = filemap_flush(VFS_I(ip)->i_mapping);
1641 				if (error)
1642 					return error;
1643 			}
1644 		}
1645 	}
1646 
1647 	if (VFS_I(ip)->i_nlink == 0)
1648 		return 0;
1649 
1650 	if (xfs_can_free_eofblocks(ip, false)) {
1651 
1652 		/*
1653 		 * If we can't get the iolock just skip truncating the blocks
1654 		 * past EOF because we could deadlock with the mmap_sem
1655 		 * otherwise.  We'll get another chance to drop them once the
1656 		 * last reference to the inode is dropped, so we'll never leak
1657 		 * blocks permanently.
1658 		 *
1659 		 * Further, check if the inode is being opened, written and
1660 		 * closed frequently and we have delayed allocation blocks
1661 		 * outstanding (e.g. streaming writes from the NFS server),
1662 		 * truncating the blocks past EOF will cause fragmentation to
1663 		 * occur.
1664 		 *
1665 		 * In this case don't do the truncation, either, but we have to
1666 		 * be careful how we detect this case. Blocks beyond EOF show
1667 		 * up as i_delayed_blks even when the inode is clean, so we
1668 		 * need to truncate them away first before checking for a dirty
1669 		 * release. Hence on the first dirty close we will still remove
1670 		 * the speculative allocation, but after that we will leave it
1671 		 * in place.
1672 		 */
1673 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1674 			return 0;
1675 
1676 		error = xfs_free_eofblocks(mp, ip, true);
1677 		if (error && error != -EAGAIN)
1678 			return error;
1679 
1680 		/* delalloc blocks after truncation means it really is dirty */
1681 		if (ip->i_delayed_blks)
1682 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1683 	}
1684 	return 0;
1685 }
1686 
1687 /*
1688  * xfs_inactive_truncate
1689  *
1690  * Called to perform a truncate when an inode becomes unlinked.
1691  */
1692 STATIC int
1693 xfs_inactive_truncate(
1694 	struct xfs_inode *ip)
1695 {
1696 	struct xfs_mount	*mp = ip->i_mount;
1697 	struct xfs_trans	*tp;
1698 	int			error;
1699 
1700 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1701 	if (error) {
1702 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1703 		return error;
1704 	}
1705 
1706 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1707 	xfs_trans_ijoin(tp, ip, 0);
1708 
1709 	/*
1710 	 * Log the inode size first to prevent stale data exposure in the event
1711 	 * of a system crash before the truncate completes. See the related
1712 	 * comment in xfs_setattr_size() for details.
1713 	 */
1714 	ip->i_d.di_size = 0;
1715 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1716 
1717 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1718 	if (error)
1719 		goto error_trans_cancel;
1720 
1721 	ASSERT(ip->i_d.di_nextents == 0);
1722 
1723 	error = xfs_trans_commit(tp);
1724 	if (error)
1725 		goto error_unlock;
1726 
1727 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1728 	return 0;
1729 
1730 error_trans_cancel:
1731 	xfs_trans_cancel(tp);
1732 error_unlock:
1733 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1734 	return error;
1735 }
1736 
1737 /*
1738  * xfs_inactive_ifree()
1739  *
1740  * Perform the inode free when an inode is unlinked.
1741  */
1742 STATIC int
1743 xfs_inactive_ifree(
1744 	struct xfs_inode *ip)
1745 {
1746 	xfs_bmap_free_t		free_list;
1747 	xfs_fsblock_t		first_block;
1748 	struct xfs_mount	*mp = ip->i_mount;
1749 	struct xfs_trans	*tp;
1750 	int			error;
1751 
1752 	/*
1753 	 * The ifree transaction might need to allocate blocks for record
1754 	 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1755 	 * allow ifree to dip into the reserved block pool if necessary.
1756 	 *
1757 	 * Freeing large sets of inodes generally means freeing inode chunks,
1758 	 * directory and file data blocks, so this should be relatively safe.
1759 	 * Only under severe circumstances should it be possible to free enough
1760 	 * inodes to exhaust the reserve block pool via finobt expansion while
1761 	 * at the same time not creating free space in the filesystem.
1762 	 *
1763 	 * Send a warning if the reservation does happen to fail, as the inode
1764 	 * now remains allocated and sits on the unlinked list until the fs is
1765 	 * repaired.
1766 	 */
1767 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1768 			XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
1769 	if (error) {
1770 		if (error == -ENOSPC) {
1771 			xfs_warn_ratelimited(mp,
1772 			"Failed to remove inode(s) from unlinked list. "
1773 			"Please free space, unmount and run xfs_repair.");
1774 		} else {
1775 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1776 		}
1777 		return error;
1778 	}
1779 
1780 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1781 	xfs_trans_ijoin(tp, ip, 0);
1782 
1783 	xfs_bmap_init(&free_list, &first_block);
1784 	error = xfs_ifree(tp, ip, &free_list);
1785 	if (error) {
1786 		/*
1787 		 * If we fail to free the inode, shut down.  The cancel
1788 		 * might do that, we need to make sure.  Otherwise the
1789 		 * inode might be lost for a long time or forever.
1790 		 */
1791 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1792 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1793 				__func__, error);
1794 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1795 		}
1796 		xfs_trans_cancel(tp);
1797 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1798 		return error;
1799 	}
1800 
1801 	/*
1802 	 * Credit the quota account(s). The inode is gone.
1803 	 */
1804 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1805 
1806 	/*
1807 	 * Just ignore errors at this point.  There is nothing we can do except
1808 	 * to try to keep going. Make sure it's not a silent error.
1809 	 */
1810 	error = xfs_bmap_finish(&tp, &free_list, NULL);
1811 	if (error) {
1812 		xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1813 			__func__, error);
1814 		xfs_bmap_cancel(&free_list);
1815 	}
1816 	error = xfs_trans_commit(tp);
1817 	if (error)
1818 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1819 			__func__, error);
1820 
1821 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1822 	return 0;
1823 }
1824 
1825 /*
1826  * xfs_inactive
1827  *
1828  * This is called when the vnode reference count for the vnode
1829  * goes to zero.  If the file has been unlinked, then it must
1830  * now be truncated.  Also, we clear all of the read-ahead state
1831  * kept for the inode here since the file is now closed.
1832  */
1833 void
1834 xfs_inactive(
1835 	xfs_inode_t	*ip)
1836 {
1837 	struct xfs_mount	*mp;
1838 	int			error;
1839 	int			truncate = 0;
1840 
1841 	/*
1842 	 * If the inode is already free, then there can be nothing
1843 	 * to clean up here.
1844 	 */
1845 	if (VFS_I(ip)->i_mode == 0) {
1846 		ASSERT(ip->i_df.if_real_bytes == 0);
1847 		ASSERT(ip->i_df.if_broot_bytes == 0);
1848 		return;
1849 	}
1850 
1851 	mp = ip->i_mount;
1852 
1853 	/* If this is a read-only mount, don't do this (would generate I/O) */
1854 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1855 		return;
1856 
1857 	if (VFS_I(ip)->i_nlink != 0) {
1858 		/*
1859 		 * force is true because we are evicting an inode from the
1860 		 * cache. Post-eof blocks must be freed, lest we end up with
1861 		 * broken free space accounting.
1862 		 */
1863 		if (xfs_can_free_eofblocks(ip, true))
1864 			xfs_free_eofblocks(mp, ip, false);
1865 
1866 		return;
1867 	}
1868 
1869 	if (S_ISREG(VFS_I(ip)->i_mode) &&
1870 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1871 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1872 		truncate = 1;
1873 
1874 	error = xfs_qm_dqattach(ip, 0);
1875 	if (error)
1876 		return;
1877 
1878 	if (S_ISLNK(VFS_I(ip)->i_mode))
1879 		error = xfs_inactive_symlink(ip);
1880 	else if (truncate)
1881 		error = xfs_inactive_truncate(ip);
1882 	if (error)
1883 		return;
1884 
1885 	/*
1886 	 * If there are attributes associated with the file then blow them away
1887 	 * now.  The code calls a routine that recursively deconstructs the
1888 	 * attribute fork. If also blows away the in-core attribute fork.
1889 	 */
1890 	if (XFS_IFORK_Q(ip)) {
1891 		error = xfs_attr_inactive(ip);
1892 		if (error)
1893 			return;
1894 	}
1895 
1896 	ASSERT(!ip->i_afp);
1897 	ASSERT(ip->i_d.di_anextents == 0);
1898 	ASSERT(ip->i_d.di_forkoff == 0);
1899 
1900 	/*
1901 	 * Free the inode.
1902 	 */
1903 	error = xfs_inactive_ifree(ip);
1904 	if (error)
1905 		return;
1906 
1907 	/*
1908 	 * Release the dquots held by inode, if any.
1909 	 */
1910 	xfs_qm_dqdetach(ip);
1911 }
1912 
1913 /*
1914  * This is called when the inode's link count goes to 0 or we are creating a
1915  * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1916  * set to true as the link count is dropped to zero by the VFS after we've
1917  * created the file successfully, so we have to add it to the unlinked list
1918  * while the link count is non-zero.
1919  *
1920  * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1921  * list when the inode is freed.
1922  */
1923 STATIC int
1924 xfs_iunlink(
1925 	struct xfs_trans *tp,
1926 	struct xfs_inode *ip)
1927 {
1928 	xfs_mount_t	*mp = tp->t_mountp;
1929 	xfs_agi_t	*agi;
1930 	xfs_dinode_t	*dip;
1931 	xfs_buf_t	*agibp;
1932 	xfs_buf_t	*ibp;
1933 	xfs_agino_t	agino;
1934 	short		bucket_index;
1935 	int		offset;
1936 	int		error;
1937 
1938 	ASSERT(VFS_I(ip)->i_mode != 0);
1939 
1940 	/*
1941 	 * Get the agi buffer first.  It ensures lock ordering
1942 	 * on the list.
1943 	 */
1944 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1945 	if (error)
1946 		return error;
1947 	agi = XFS_BUF_TO_AGI(agibp);
1948 
1949 	/*
1950 	 * Get the index into the agi hash table for the
1951 	 * list this inode will go on.
1952 	 */
1953 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1954 	ASSERT(agino != 0);
1955 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1956 	ASSERT(agi->agi_unlinked[bucket_index]);
1957 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1958 
1959 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1960 		/*
1961 		 * There is already another inode in the bucket we need
1962 		 * to add ourselves to.  Add us at the front of the list.
1963 		 * Here we put the head pointer into our next pointer,
1964 		 * and then we fall through to point the head at us.
1965 		 */
1966 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1967 				       0, 0);
1968 		if (error)
1969 			return error;
1970 
1971 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1972 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1973 		offset = ip->i_imap.im_boffset +
1974 			offsetof(xfs_dinode_t, di_next_unlinked);
1975 
1976 		/* need to recalc the inode CRC if appropriate */
1977 		xfs_dinode_calc_crc(mp, dip);
1978 
1979 		xfs_trans_inode_buf(tp, ibp);
1980 		xfs_trans_log_buf(tp, ibp, offset,
1981 				  (offset + sizeof(xfs_agino_t) - 1));
1982 		xfs_inobp_check(mp, ibp);
1983 	}
1984 
1985 	/*
1986 	 * Point the bucket head pointer at the inode being inserted.
1987 	 */
1988 	ASSERT(agino != 0);
1989 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1990 	offset = offsetof(xfs_agi_t, agi_unlinked) +
1991 		(sizeof(xfs_agino_t) * bucket_index);
1992 	xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
1993 	xfs_trans_log_buf(tp, agibp, offset,
1994 			  (offset + sizeof(xfs_agino_t) - 1));
1995 	return 0;
1996 }
1997 
1998 /*
1999  * Pull the on-disk inode from the AGI unlinked list.
2000  */
2001 STATIC int
2002 xfs_iunlink_remove(
2003 	xfs_trans_t	*tp,
2004 	xfs_inode_t	*ip)
2005 {
2006 	xfs_ino_t	next_ino;
2007 	xfs_mount_t	*mp;
2008 	xfs_agi_t	*agi;
2009 	xfs_dinode_t	*dip;
2010 	xfs_buf_t	*agibp;
2011 	xfs_buf_t	*ibp;
2012 	xfs_agnumber_t	agno;
2013 	xfs_agino_t	agino;
2014 	xfs_agino_t	next_agino;
2015 	xfs_buf_t	*last_ibp;
2016 	xfs_dinode_t	*last_dip = NULL;
2017 	short		bucket_index;
2018 	int		offset, last_offset = 0;
2019 	int		error;
2020 
2021 	mp = tp->t_mountp;
2022 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2023 
2024 	/*
2025 	 * Get the agi buffer first.  It ensures lock ordering
2026 	 * on the list.
2027 	 */
2028 	error = xfs_read_agi(mp, tp, agno, &agibp);
2029 	if (error)
2030 		return error;
2031 
2032 	agi = XFS_BUF_TO_AGI(agibp);
2033 
2034 	/*
2035 	 * Get the index into the agi hash table for the
2036 	 * list this inode will go on.
2037 	 */
2038 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2039 	ASSERT(agino != 0);
2040 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2041 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2042 	ASSERT(agi->agi_unlinked[bucket_index]);
2043 
2044 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2045 		/*
2046 		 * We're at the head of the list.  Get the inode's on-disk
2047 		 * buffer to see if there is anyone after us on the list.
2048 		 * Only modify our next pointer if it is not already NULLAGINO.
2049 		 * This saves us the overhead of dealing with the buffer when
2050 		 * there is no need to change it.
2051 		 */
2052 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2053 				       0, 0);
2054 		if (error) {
2055 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2056 				__func__, error);
2057 			return error;
2058 		}
2059 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2060 		ASSERT(next_agino != 0);
2061 		if (next_agino != NULLAGINO) {
2062 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2063 			offset = ip->i_imap.im_boffset +
2064 				offsetof(xfs_dinode_t, di_next_unlinked);
2065 
2066 			/* need to recalc the inode CRC if appropriate */
2067 			xfs_dinode_calc_crc(mp, dip);
2068 
2069 			xfs_trans_inode_buf(tp, ibp);
2070 			xfs_trans_log_buf(tp, ibp, offset,
2071 					  (offset + sizeof(xfs_agino_t) - 1));
2072 			xfs_inobp_check(mp, ibp);
2073 		} else {
2074 			xfs_trans_brelse(tp, ibp);
2075 		}
2076 		/*
2077 		 * Point the bucket head pointer at the next inode.
2078 		 */
2079 		ASSERT(next_agino != 0);
2080 		ASSERT(next_agino != agino);
2081 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2082 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2083 			(sizeof(xfs_agino_t) * bucket_index);
2084 		xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2085 		xfs_trans_log_buf(tp, agibp, offset,
2086 				  (offset + sizeof(xfs_agino_t) - 1));
2087 	} else {
2088 		/*
2089 		 * We need to search the list for the inode being freed.
2090 		 */
2091 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2092 		last_ibp = NULL;
2093 		while (next_agino != agino) {
2094 			struct xfs_imap	imap;
2095 
2096 			if (last_ibp)
2097 				xfs_trans_brelse(tp, last_ibp);
2098 
2099 			imap.im_blkno = 0;
2100 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2101 
2102 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2103 			if (error) {
2104 				xfs_warn(mp,
2105 	"%s: xfs_imap returned error %d.",
2106 					 __func__, error);
2107 				return error;
2108 			}
2109 
2110 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2111 					       &last_ibp, 0, 0);
2112 			if (error) {
2113 				xfs_warn(mp,
2114 	"%s: xfs_imap_to_bp returned error %d.",
2115 					__func__, error);
2116 				return error;
2117 			}
2118 
2119 			last_offset = imap.im_boffset;
2120 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2121 			ASSERT(next_agino != NULLAGINO);
2122 			ASSERT(next_agino != 0);
2123 		}
2124 
2125 		/*
2126 		 * Now last_ibp points to the buffer previous to us on the
2127 		 * unlinked list.  Pull us from the list.
2128 		 */
2129 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2130 				       0, 0);
2131 		if (error) {
2132 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2133 				__func__, error);
2134 			return error;
2135 		}
2136 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2137 		ASSERT(next_agino != 0);
2138 		ASSERT(next_agino != agino);
2139 		if (next_agino != NULLAGINO) {
2140 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2141 			offset = ip->i_imap.im_boffset +
2142 				offsetof(xfs_dinode_t, di_next_unlinked);
2143 
2144 			/* need to recalc the inode CRC if appropriate */
2145 			xfs_dinode_calc_crc(mp, dip);
2146 
2147 			xfs_trans_inode_buf(tp, ibp);
2148 			xfs_trans_log_buf(tp, ibp, offset,
2149 					  (offset + sizeof(xfs_agino_t) - 1));
2150 			xfs_inobp_check(mp, ibp);
2151 		} else {
2152 			xfs_trans_brelse(tp, ibp);
2153 		}
2154 		/*
2155 		 * Point the previous inode on the list to the next inode.
2156 		 */
2157 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2158 		ASSERT(next_agino != 0);
2159 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2160 
2161 		/* need to recalc the inode CRC if appropriate */
2162 		xfs_dinode_calc_crc(mp, last_dip);
2163 
2164 		xfs_trans_inode_buf(tp, last_ibp);
2165 		xfs_trans_log_buf(tp, last_ibp, offset,
2166 				  (offset + sizeof(xfs_agino_t) - 1));
2167 		xfs_inobp_check(mp, last_ibp);
2168 	}
2169 	return 0;
2170 }
2171 
2172 /*
2173  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2174  * inodes that are in memory - they all must be marked stale and attached to
2175  * the cluster buffer.
2176  */
2177 STATIC int
2178 xfs_ifree_cluster(
2179 	xfs_inode_t		*free_ip,
2180 	xfs_trans_t		*tp,
2181 	struct xfs_icluster	*xic)
2182 {
2183 	xfs_mount_t		*mp = free_ip->i_mount;
2184 	int			blks_per_cluster;
2185 	int			inodes_per_cluster;
2186 	int			nbufs;
2187 	int			i, j;
2188 	int			ioffset;
2189 	xfs_daddr_t		blkno;
2190 	xfs_buf_t		*bp;
2191 	xfs_inode_t		*ip;
2192 	xfs_inode_log_item_t	*iip;
2193 	xfs_log_item_t		*lip;
2194 	struct xfs_perag	*pag;
2195 	xfs_ino_t		inum;
2196 
2197 	inum = xic->first_ino;
2198 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2199 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2200 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2201 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2202 
2203 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2204 		/*
2205 		 * The allocation bitmap tells us which inodes of the chunk were
2206 		 * physically allocated. Skip the cluster if an inode falls into
2207 		 * a sparse region.
2208 		 */
2209 		ioffset = inum - xic->first_ino;
2210 		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2211 			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2212 			continue;
2213 		}
2214 
2215 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2216 					 XFS_INO_TO_AGBNO(mp, inum));
2217 
2218 		/*
2219 		 * We obtain and lock the backing buffer first in the process
2220 		 * here, as we have to ensure that any dirty inode that we
2221 		 * can't get the flush lock on is attached to the buffer.
2222 		 * If we scan the in-memory inodes first, then buffer IO can
2223 		 * complete before we get a lock on it, and hence we may fail
2224 		 * to mark all the active inodes on the buffer stale.
2225 		 */
2226 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2227 					mp->m_bsize * blks_per_cluster,
2228 					XBF_UNMAPPED);
2229 
2230 		if (!bp)
2231 			return -ENOMEM;
2232 
2233 		/*
2234 		 * This buffer may not have been correctly initialised as we
2235 		 * didn't read it from disk. That's not important because we are
2236 		 * only using to mark the buffer as stale in the log, and to
2237 		 * attach stale cached inodes on it. That means it will never be
2238 		 * dispatched for IO. If it is, we want to know about it, and we
2239 		 * want it to fail. We can acheive this by adding a write
2240 		 * verifier to the buffer.
2241 		 */
2242 		 bp->b_ops = &xfs_inode_buf_ops;
2243 
2244 		/*
2245 		 * Walk the inodes already attached to the buffer and mark them
2246 		 * stale. These will all have the flush locks held, so an
2247 		 * in-memory inode walk can't lock them. By marking them all
2248 		 * stale first, we will not attempt to lock them in the loop
2249 		 * below as the XFS_ISTALE flag will be set.
2250 		 */
2251 		lip = bp->b_fspriv;
2252 		while (lip) {
2253 			if (lip->li_type == XFS_LI_INODE) {
2254 				iip = (xfs_inode_log_item_t *)lip;
2255 				ASSERT(iip->ili_logged == 1);
2256 				lip->li_cb = xfs_istale_done;
2257 				xfs_trans_ail_copy_lsn(mp->m_ail,
2258 							&iip->ili_flush_lsn,
2259 							&iip->ili_item.li_lsn);
2260 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2261 			}
2262 			lip = lip->li_bio_list;
2263 		}
2264 
2265 
2266 		/*
2267 		 * For each inode in memory attempt to add it to the inode
2268 		 * buffer and set it up for being staled on buffer IO
2269 		 * completion.  This is safe as we've locked out tail pushing
2270 		 * and flushing by locking the buffer.
2271 		 *
2272 		 * We have already marked every inode that was part of a
2273 		 * transaction stale above, which means there is no point in
2274 		 * even trying to lock them.
2275 		 */
2276 		for (i = 0; i < inodes_per_cluster; i++) {
2277 retry:
2278 			rcu_read_lock();
2279 			ip = radix_tree_lookup(&pag->pag_ici_root,
2280 					XFS_INO_TO_AGINO(mp, (inum + i)));
2281 
2282 			/* Inode not in memory, nothing to do */
2283 			if (!ip) {
2284 				rcu_read_unlock();
2285 				continue;
2286 			}
2287 
2288 			/*
2289 			 * because this is an RCU protected lookup, we could
2290 			 * find a recently freed or even reallocated inode
2291 			 * during the lookup. We need to check under the
2292 			 * i_flags_lock for a valid inode here. Skip it if it
2293 			 * is not valid, the wrong inode or stale.
2294 			 */
2295 			spin_lock(&ip->i_flags_lock);
2296 			if (ip->i_ino != inum + i ||
2297 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2298 				spin_unlock(&ip->i_flags_lock);
2299 				rcu_read_unlock();
2300 				continue;
2301 			}
2302 			spin_unlock(&ip->i_flags_lock);
2303 
2304 			/*
2305 			 * Don't try to lock/unlock the current inode, but we
2306 			 * _cannot_ skip the other inodes that we did not find
2307 			 * in the list attached to the buffer and are not
2308 			 * already marked stale. If we can't lock it, back off
2309 			 * and retry.
2310 			 */
2311 			if (ip != free_ip &&
2312 			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2313 				rcu_read_unlock();
2314 				delay(1);
2315 				goto retry;
2316 			}
2317 			rcu_read_unlock();
2318 
2319 			xfs_iflock(ip);
2320 			xfs_iflags_set(ip, XFS_ISTALE);
2321 
2322 			/*
2323 			 * we don't need to attach clean inodes or those only
2324 			 * with unlogged changes (which we throw away, anyway).
2325 			 */
2326 			iip = ip->i_itemp;
2327 			if (!iip || xfs_inode_clean(ip)) {
2328 				ASSERT(ip != free_ip);
2329 				xfs_ifunlock(ip);
2330 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2331 				continue;
2332 			}
2333 
2334 			iip->ili_last_fields = iip->ili_fields;
2335 			iip->ili_fields = 0;
2336 			iip->ili_fsync_fields = 0;
2337 			iip->ili_logged = 1;
2338 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2339 						&iip->ili_item.li_lsn);
2340 
2341 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2342 						  &iip->ili_item);
2343 
2344 			if (ip != free_ip)
2345 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2346 		}
2347 
2348 		xfs_trans_stale_inode_buf(tp, bp);
2349 		xfs_trans_binval(tp, bp);
2350 	}
2351 
2352 	xfs_perag_put(pag);
2353 	return 0;
2354 }
2355 
2356 /*
2357  * This is called to return an inode to the inode free list.
2358  * The inode should already be truncated to 0 length and have
2359  * no pages associated with it.  This routine also assumes that
2360  * the inode is already a part of the transaction.
2361  *
2362  * The on-disk copy of the inode will have been added to the list
2363  * of unlinked inodes in the AGI. We need to remove the inode from
2364  * that list atomically with respect to freeing it here.
2365  */
2366 int
2367 xfs_ifree(
2368 	xfs_trans_t	*tp,
2369 	xfs_inode_t	*ip,
2370 	xfs_bmap_free_t	*flist)
2371 {
2372 	int			error;
2373 	struct xfs_icluster	xic = { 0 };
2374 
2375 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2376 	ASSERT(VFS_I(ip)->i_nlink == 0);
2377 	ASSERT(ip->i_d.di_nextents == 0);
2378 	ASSERT(ip->i_d.di_anextents == 0);
2379 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2380 	ASSERT(ip->i_d.di_nblocks == 0);
2381 
2382 	/*
2383 	 * Pull the on-disk inode from the AGI unlinked list.
2384 	 */
2385 	error = xfs_iunlink_remove(tp, ip);
2386 	if (error)
2387 		return error;
2388 
2389 	error = xfs_difree(tp, ip->i_ino, flist, &xic);
2390 	if (error)
2391 		return error;
2392 
2393 	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2394 	ip->i_d.di_flags = 0;
2395 	ip->i_d.di_dmevmask = 0;
2396 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2397 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2398 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2399 	/*
2400 	 * Bump the generation count so no one will be confused
2401 	 * by reincarnations of this inode.
2402 	 */
2403 	VFS_I(ip)->i_generation++;
2404 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2405 
2406 	if (xic.deleted)
2407 		error = xfs_ifree_cluster(ip, tp, &xic);
2408 
2409 	return error;
2410 }
2411 
2412 /*
2413  * This is called to unpin an inode.  The caller must have the inode locked
2414  * in at least shared mode so that the buffer cannot be subsequently pinned
2415  * once someone is waiting for it to be unpinned.
2416  */
2417 static void
2418 xfs_iunpin(
2419 	struct xfs_inode	*ip)
2420 {
2421 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2422 
2423 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2424 
2425 	/* Give the log a push to start the unpinning I/O */
2426 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2427 
2428 }
2429 
2430 static void
2431 __xfs_iunpin_wait(
2432 	struct xfs_inode	*ip)
2433 {
2434 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2435 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2436 
2437 	xfs_iunpin(ip);
2438 
2439 	do {
2440 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2441 		if (xfs_ipincount(ip))
2442 			io_schedule();
2443 	} while (xfs_ipincount(ip));
2444 	finish_wait(wq, &wait.wait);
2445 }
2446 
2447 void
2448 xfs_iunpin_wait(
2449 	struct xfs_inode	*ip)
2450 {
2451 	if (xfs_ipincount(ip))
2452 		__xfs_iunpin_wait(ip);
2453 }
2454 
2455 /*
2456  * Removing an inode from the namespace involves removing the directory entry
2457  * and dropping the link count on the inode. Removing the directory entry can
2458  * result in locking an AGF (directory blocks were freed) and removing a link
2459  * count can result in placing the inode on an unlinked list which results in
2460  * locking an AGI.
2461  *
2462  * The big problem here is that we have an ordering constraint on AGF and AGI
2463  * locking - inode allocation locks the AGI, then can allocate a new extent for
2464  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2465  * removes the inode from the unlinked list, requiring that we lock the AGI
2466  * first, and then freeing the inode can result in an inode chunk being freed
2467  * and hence freeing disk space requiring that we lock an AGF.
2468  *
2469  * Hence the ordering that is imposed by other parts of the code is AGI before
2470  * AGF. This means we cannot remove the directory entry before we drop the inode
2471  * reference count and put it on the unlinked list as this results in a lock
2472  * order of AGF then AGI, and this can deadlock against inode allocation and
2473  * freeing. Therefore we must drop the link counts before we remove the
2474  * directory entry.
2475  *
2476  * This is still safe from a transactional point of view - it is not until we
2477  * get to xfs_bmap_finish() that we have the possibility of multiple
2478  * transactions in this operation. Hence as long as we remove the directory
2479  * entry and drop the link count in the first transaction of the remove
2480  * operation, there are no transactional constraints on the ordering here.
2481  */
2482 int
2483 xfs_remove(
2484 	xfs_inode_t             *dp,
2485 	struct xfs_name		*name,
2486 	xfs_inode_t		*ip)
2487 {
2488 	xfs_mount_t		*mp = dp->i_mount;
2489 	xfs_trans_t             *tp = NULL;
2490 	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2491 	int                     error = 0;
2492 	xfs_bmap_free_t         free_list;
2493 	xfs_fsblock_t           first_block;
2494 	uint			resblks;
2495 
2496 	trace_xfs_remove(dp, name);
2497 
2498 	if (XFS_FORCED_SHUTDOWN(mp))
2499 		return -EIO;
2500 
2501 	error = xfs_qm_dqattach(dp, 0);
2502 	if (error)
2503 		goto std_return;
2504 
2505 	error = xfs_qm_dqattach(ip, 0);
2506 	if (error)
2507 		goto std_return;
2508 
2509 	/*
2510 	 * We try to get the real space reservation first,
2511 	 * allowing for directory btree deletion(s) implying
2512 	 * possible bmap insert(s).  If we can't get the space
2513 	 * reservation then we use 0 instead, and avoid the bmap
2514 	 * btree insert(s) in the directory code by, if the bmap
2515 	 * insert tries to happen, instead trimming the LAST
2516 	 * block from the directory.
2517 	 */
2518 	resblks = XFS_REMOVE_SPACE_RES(mp);
2519 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2520 	if (error == -ENOSPC) {
2521 		resblks = 0;
2522 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2523 				&tp);
2524 	}
2525 	if (error) {
2526 		ASSERT(error != -ENOSPC);
2527 		goto std_return;
2528 	}
2529 
2530 	xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2531 	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2532 
2533 	xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2534 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2535 
2536 	/*
2537 	 * If we're removing a directory perform some additional validation.
2538 	 */
2539 	if (is_dir) {
2540 		ASSERT(VFS_I(ip)->i_nlink >= 2);
2541 		if (VFS_I(ip)->i_nlink != 2) {
2542 			error = -ENOTEMPTY;
2543 			goto out_trans_cancel;
2544 		}
2545 		if (!xfs_dir_isempty(ip)) {
2546 			error = -ENOTEMPTY;
2547 			goto out_trans_cancel;
2548 		}
2549 
2550 		/* Drop the link from ip's "..".  */
2551 		error = xfs_droplink(tp, dp);
2552 		if (error)
2553 			goto out_trans_cancel;
2554 
2555 		/* Drop the "." link from ip to self.  */
2556 		error = xfs_droplink(tp, ip);
2557 		if (error)
2558 			goto out_trans_cancel;
2559 	} else {
2560 		/*
2561 		 * When removing a non-directory we need to log the parent
2562 		 * inode here.  For a directory this is done implicitly
2563 		 * by the xfs_droplink call for the ".." entry.
2564 		 */
2565 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2566 	}
2567 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2568 
2569 	/* Drop the link from dp to ip. */
2570 	error = xfs_droplink(tp, ip);
2571 	if (error)
2572 		goto out_trans_cancel;
2573 
2574 	xfs_bmap_init(&free_list, &first_block);
2575 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2576 					&first_block, &free_list, resblks);
2577 	if (error) {
2578 		ASSERT(error != -ENOENT);
2579 		goto out_bmap_cancel;
2580 	}
2581 
2582 	/*
2583 	 * If this is a synchronous mount, make sure that the
2584 	 * remove transaction goes to disk before returning to
2585 	 * the user.
2586 	 */
2587 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2588 		xfs_trans_set_sync(tp);
2589 
2590 	error = xfs_bmap_finish(&tp, &free_list, NULL);
2591 	if (error)
2592 		goto out_bmap_cancel;
2593 
2594 	error = xfs_trans_commit(tp);
2595 	if (error)
2596 		goto std_return;
2597 
2598 	if (is_dir && xfs_inode_is_filestream(ip))
2599 		xfs_filestream_deassociate(ip);
2600 
2601 	return 0;
2602 
2603  out_bmap_cancel:
2604 	xfs_bmap_cancel(&free_list);
2605  out_trans_cancel:
2606 	xfs_trans_cancel(tp);
2607  std_return:
2608 	return error;
2609 }
2610 
2611 /*
2612  * Enter all inodes for a rename transaction into a sorted array.
2613  */
2614 #define __XFS_SORT_INODES	5
2615 STATIC void
2616 xfs_sort_for_rename(
2617 	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2618 	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2619 	struct xfs_inode	*ip1,	/* in: inode of old entry */
2620 	struct xfs_inode	*ip2,	/* in: inode of new entry */
2621 	struct xfs_inode	*wip,	/* in: whiteout inode */
2622 	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2623 	int			*num_inodes)  /* in/out: inodes in array */
2624 {
2625 	int			i, j;
2626 
2627 	ASSERT(*num_inodes == __XFS_SORT_INODES);
2628 	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2629 
2630 	/*
2631 	 * i_tab contains a list of pointers to inodes.  We initialize
2632 	 * the table here & we'll sort it.  We will then use it to
2633 	 * order the acquisition of the inode locks.
2634 	 *
2635 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2636 	 */
2637 	i = 0;
2638 	i_tab[i++] = dp1;
2639 	i_tab[i++] = dp2;
2640 	i_tab[i++] = ip1;
2641 	if (ip2)
2642 		i_tab[i++] = ip2;
2643 	if (wip)
2644 		i_tab[i++] = wip;
2645 	*num_inodes = i;
2646 
2647 	/*
2648 	 * Sort the elements via bubble sort.  (Remember, there are at
2649 	 * most 5 elements to sort, so this is adequate.)
2650 	 */
2651 	for (i = 0; i < *num_inodes; i++) {
2652 		for (j = 1; j < *num_inodes; j++) {
2653 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2654 				struct xfs_inode *temp = i_tab[j];
2655 				i_tab[j] = i_tab[j-1];
2656 				i_tab[j-1] = temp;
2657 			}
2658 		}
2659 	}
2660 }
2661 
2662 static int
2663 xfs_finish_rename(
2664 	struct xfs_trans	*tp,
2665 	struct xfs_bmap_free	*free_list)
2666 {
2667 	int			error;
2668 
2669 	/*
2670 	 * If this is a synchronous mount, make sure that the rename transaction
2671 	 * goes to disk before returning to the user.
2672 	 */
2673 	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2674 		xfs_trans_set_sync(tp);
2675 
2676 	error = xfs_bmap_finish(&tp, free_list, NULL);
2677 	if (error) {
2678 		xfs_bmap_cancel(free_list);
2679 		xfs_trans_cancel(tp);
2680 		return error;
2681 	}
2682 
2683 	return xfs_trans_commit(tp);
2684 }
2685 
2686 /*
2687  * xfs_cross_rename()
2688  *
2689  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2690  */
2691 STATIC int
2692 xfs_cross_rename(
2693 	struct xfs_trans	*tp,
2694 	struct xfs_inode	*dp1,
2695 	struct xfs_name		*name1,
2696 	struct xfs_inode	*ip1,
2697 	struct xfs_inode	*dp2,
2698 	struct xfs_name		*name2,
2699 	struct xfs_inode	*ip2,
2700 	struct xfs_bmap_free	*free_list,
2701 	xfs_fsblock_t		*first_block,
2702 	int			spaceres)
2703 {
2704 	int		error = 0;
2705 	int		ip1_flags = 0;
2706 	int		ip2_flags = 0;
2707 	int		dp2_flags = 0;
2708 
2709 	/* Swap inode number for dirent in first parent */
2710 	error = xfs_dir_replace(tp, dp1, name1,
2711 				ip2->i_ino,
2712 				first_block, free_list, spaceres);
2713 	if (error)
2714 		goto out_trans_abort;
2715 
2716 	/* Swap inode number for dirent in second parent */
2717 	error = xfs_dir_replace(tp, dp2, name2,
2718 				ip1->i_ino,
2719 				first_block, free_list, spaceres);
2720 	if (error)
2721 		goto out_trans_abort;
2722 
2723 	/*
2724 	 * If we're renaming one or more directories across different parents,
2725 	 * update the respective ".." entries (and link counts) to match the new
2726 	 * parents.
2727 	 */
2728 	if (dp1 != dp2) {
2729 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2730 
2731 		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2732 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2733 						dp1->i_ino, first_block,
2734 						free_list, spaceres);
2735 			if (error)
2736 				goto out_trans_abort;
2737 
2738 			/* transfer ip2 ".." reference to dp1 */
2739 			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2740 				error = xfs_droplink(tp, dp2);
2741 				if (error)
2742 					goto out_trans_abort;
2743 				error = xfs_bumplink(tp, dp1);
2744 				if (error)
2745 					goto out_trans_abort;
2746 			}
2747 
2748 			/*
2749 			 * Although ip1 isn't changed here, userspace needs
2750 			 * to be warned about the change, so that applications
2751 			 * relying on it (like backup ones), will properly
2752 			 * notify the change
2753 			 */
2754 			ip1_flags |= XFS_ICHGTIME_CHG;
2755 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2756 		}
2757 
2758 		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2759 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2760 						dp2->i_ino, first_block,
2761 						free_list, spaceres);
2762 			if (error)
2763 				goto out_trans_abort;
2764 
2765 			/* transfer ip1 ".." reference to dp2 */
2766 			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2767 				error = xfs_droplink(tp, dp1);
2768 				if (error)
2769 					goto out_trans_abort;
2770 				error = xfs_bumplink(tp, dp2);
2771 				if (error)
2772 					goto out_trans_abort;
2773 			}
2774 
2775 			/*
2776 			 * Although ip2 isn't changed here, userspace needs
2777 			 * to be warned about the change, so that applications
2778 			 * relying on it (like backup ones), will properly
2779 			 * notify the change
2780 			 */
2781 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2782 			ip2_flags |= XFS_ICHGTIME_CHG;
2783 		}
2784 	}
2785 
2786 	if (ip1_flags) {
2787 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2788 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2789 	}
2790 	if (ip2_flags) {
2791 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2792 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2793 	}
2794 	if (dp2_flags) {
2795 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2796 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2797 	}
2798 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2799 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2800 	return xfs_finish_rename(tp, free_list);
2801 
2802 out_trans_abort:
2803 	xfs_bmap_cancel(free_list);
2804 	xfs_trans_cancel(tp);
2805 	return error;
2806 }
2807 
2808 /*
2809  * xfs_rename_alloc_whiteout()
2810  *
2811  * Return a referenced, unlinked, unlocked inode that that can be used as a
2812  * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2813  * crash between allocating the inode and linking it into the rename transaction
2814  * recovery will free the inode and we won't leak it.
2815  */
2816 static int
2817 xfs_rename_alloc_whiteout(
2818 	struct xfs_inode	*dp,
2819 	struct xfs_inode	**wip)
2820 {
2821 	struct xfs_inode	*tmpfile;
2822 	int			error;
2823 
2824 	error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2825 	if (error)
2826 		return error;
2827 
2828 	/*
2829 	 * Prepare the tmpfile inode as if it were created through the VFS.
2830 	 * Otherwise, the link increment paths will complain about nlink 0->1.
2831 	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2832 	 * and flag it as linkable.
2833 	 */
2834 	drop_nlink(VFS_I(tmpfile));
2835 	xfs_setup_iops(tmpfile);
2836 	xfs_finish_inode_setup(tmpfile);
2837 	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2838 
2839 	*wip = tmpfile;
2840 	return 0;
2841 }
2842 
2843 /*
2844  * xfs_rename
2845  */
2846 int
2847 xfs_rename(
2848 	struct xfs_inode	*src_dp,
2849 	struct xfs_name		*src_name,
2850 	struct xfs_inode	*src_ip,
2851 	struct xfs_inode	*target_dp,
2852 	struct xfs_name		*target_name,
2853 	struct xfs_inode	*target_ip,
2854 	unsigned int		flags)
2855 {
2856 	struct xfs_mount	*mp = src_dp->i_mount;
2857 	struct xfs_trans	*tp;
2858 	struct xfs_bmap_free	free_list;
2859 	xfs_fsblock_t		first_block;
2860 	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2861 	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2862 	int			num_inodes = __XFS_SORT_INODES;
2863 	bool			new_parent = (src_dp != target_dp);
2864 	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2865 	int			spaceres;
2866 	int			error;
2867 
2868 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2869 
2870 	if ((flags & RENAME_EXCHANGE) && !target_ip)
2871 		return -EINVAL;
2872 
2873 	/*
2874 	 * If we are doing a whiteout operation, allocate the whiteout inode
2875 	 * we will be placing at the target and ensure the type is set
2876 	 * appropriately.
2877 	 */
2878 	if (flags & RENAME_WHITEOUT) {
2879 		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2880 		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2881 		if (error)
2882 			return error;
2883 
2884 		/* setup target dirent info as whiteout */
2885 		src_name->type = XFS_DIR3_FT_CHRDEV;
2886 	}
2887 
2888 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2889 				inodes, &num_inodes);
2890 
2891 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2892 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2893 	if (error == -ENOSPC) {
2894 		spaceres = 0;
2895 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2896 				&tp);
2897 	}
2898 	if (error)
2899 		goto out_release_wip;
2900 
2901 	/*
2902 	 * Attach the dquots to the inodes
2903 	 */
2904 	error = xfs_qm_vop_rename_dqattach(inodes);
2905 	if (error)
2906 		goto out_trans_cancel;
2907 
2908 	/*
2909 	 * Lock all the participating inodes. Depending upon whether
2910 	 * the target_name exists in the target directory, and
2911 	 * whether the target directory is the same as the source
2912 	 * directory, we can lock from 2 to 4 inodes.
2913 	 */
2914 	if (!new_parent)
2915 		xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2916 	else
2917 		xfs_lock_two_inodes(src_dp, target_dp,
2918 				    XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2919 
2920 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2921 
2922 	/*
2923 	 * Join all the inodes to the transaction. From this point on,
2924 	 * we can rely on either trans_commit or trans_cancel to unlock
2925 	 * them.
2926 	 */
2927 	xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2928 	if (new_parent)
2929 		xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
2930 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2931 	if (target_ip)
2932 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2933 	if (wip)
2934 		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2935 
2936 	/*
2937 	 * If we are using project inheritance, we only allow renames
2938 	 * into our tree when the project IDs are the same; else the
2939 	 * tree quota mechanism would be circumvented.
2940 	 */
2941 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2942 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2943 		error = -EXDEV;
2944 		goto out_trans_cancel;
2945 	}
2946 
2947 	xfs_bmap_init(&free_list, &first_block);
2948 
2949 	/* RENAME_EXCHANGE is unique from here on. */
2950 	if (flags & RENAME_EXCHANGE)
2951 		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2952 					target_dp, target_name, target_ip,
2953 					&free_list, &first_block, spaceres);
2954 
2955 	/*
2956 	 * Set up the target.
2957 	 */
2958 	if (target_ip == NULL) {
2959 		/*
2960 		 * If there's no space reservation, check the entry will
2961 		 * fit before actually inserting it.
2962 		 */
2963 		if (!spaceres) {
2964 			error = xfs_dir_canenter(tp, target_dp, target_name);
2965 			if (error)
2966 				goto out_trans_cancel;
2967 		}
2968 		/*
2969 		 * If target does not exist and the rename crosses
2970 		 * directories, adjust the target directory link count
2971 		 * to account for the ".." reference from the new entry.
2972 		 */
2973 		error = xfs_dir_createname(tp, target_dp, target_name,
2974 						src_ip->i_ino, &first_block,
2975 						&free_list, spaceres);
2976 		if (error)
2977 			goto out_bmap_cancel;
2978 
2979 		xfs_trans_ichgtime(tp, target_dp,
2980 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2981 
2982 		if (new_parent && src_is_directory) {
2983 			error = xfs_bumplink(tp, target_dp);
2984 			if (error)
2985 				goto out_bmap_cancel;
2986 		}
2987 	} else { /* target_ip != NULL */
2988 		/*
2989 		 * If target exists and it's a directory, check that both
2990 		 * target and source are directories and that target can be
2991 		 * destroyed, or that neither is a directory.
2992 		 */
2993 		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
2994 			/*
2995 			 * Make sure target dir is empty.
2996 			 */
2997 			if (!(xfs_dir_isempty(target_ip)) ||
2998 			    (VFS_I(target_ip)->i_nlink > 2)) {
2999 				error = -EEXIST;
3000 				goto out_trans_cancel;
3001 			}
3002 		}
3003 
3004 		/*
3005 		 * Link the source inode under the target name.
3006 		 * If the source inode is a directory and we are moving
3007 		 * it across directories, its ".." entry will be
3008 		 * inconsistent until we replace that down below.
3009 		 *
3010 		 * In case there is already an entry with the same
3011 		 * name at the destination directory, remove it first.
3012 		 */
3013 		error = xfs_dir_replace(tp, target_dp, target_name,
3014 					src_ip->i_ino,
3015 					&first_block, &free_list, spaceres);
3016 		if (error)
3017 			goto out_bmap_cancel;
3018 
3019 		xfs_trans_ichgtime(tp, target_dp,
3020 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3021 
3022 		/*
3023 		 * Decrement the link count on the target since the target
3024 		 * dir no longer points to it.
3025 		 */
3026 		error = xfs_droplink(tp, target_ip);
3027 		if (error)
3028 			goto out_bmap_cancel;
3029 
3030 		if (src_is_directory) {
3031 			/*
3032 			 * Drop the link from the old "." entry.
3033 			 */
3034 			error = xfs_droplink(tp, target_ip);
3035 			if (error)
3036 				goto out_bmap_cancel;
3037 		}
3038 	} /* target_ip != NULL */
3039 
3040 	/*
3041 	 * Remove the source.
3042 	 */
3043 	if (new_parent && src_is_directory) {
3044 		/*
3045 		 * Rewrite the ".." entry to point to the new
3046 		 * directory.
3047 		 */
3048 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3049 					target_dp->i_ino,
3050 					&first_block, &free_list, spaceres);
3051 		ASSERT(error != -EEXIST);
3052 		if (error)
3053 			goto out_bmap_cancel;
3054 	}
3055 
3056 	/*
3057 	 * We always want to hit the ctime on the source inode.
3058 	 *
3059 	 * This isn't strictly required by the standards since the source
3060 	 * inode isn't really being changed, but old unix file systems did
3061 	 * it and some incremental backup programs won't work without it.
3062 	 */
3063 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3064 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3065 
3066 	/*
3067 	 * Adjust the link count on src_dp.  This is necessary when
3068 	 * renaming a directory, either within one parent when
3069 	 * the target existed, or across two parent directories.
3070 	 */
3071 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3072 
3073 		/*
3074 		 * Decrement link count on src_directory since the
3075 		 * entry that's moved no longer points to it.
3076 		 */
3077 		error = xfs_droplink(tp, src_dp);
3078 		if (error)
3079 			goto out_bmap_cancel;
3080 	}
3081 
3082 	/*
3083 	 * For whiteouts, we only need to update the source dirent with the
3084 	 * inode number of the whiteout inode rather than removing it
3085 	 * altogether.
3086 	 */
3087 	if (wip) {
3088 		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3089 					&first_block, &free_list, spaceres);
3090 	} else
3091 		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3092 					   &first_block, &free_list, spaceres);
3093 	if (error)
3094 		goto out_bmap_cancel;
3095 
3096 	/*
3097 	 * For whiteouts, we need to bump the link count on the whiteout inode.
3098 	 * This means that failures all the way up to this point leave the inode
3099 	 * on the unlinked list and so cleanup is a simple matter of dropping
3100 	 * the remaining reference to it. If we fail here after bumping the link
3101 	 * count, we're shutting down the filesystem so we'll never see the
3102 	 * intermediate state on disk.
3103 	 */
3104 	if (wip) {
3105 		ASSERT(VFS_I(wip)->i_nlink == 0);
3106 		error = xfs_bumplink(tp, wip);
3107 		if (error)
3108 			goto out_bmap_cancel;
3109 		error = xfs_iunlink_remove(tp, wip);
3110 		if (error)
3111 			goto out_bmap_cancel;
3112 		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3113 
3114 		/*
3115 		 * Now we have a real link, clear the "I'm a tmpfile" state
3116 		 * flag from the inode so it doesn't accidentally get misused in
3117 		 * future.
3118 		 */
3119 		VFS_I(wip)->i_state &= ~I_LINKABLE;
3120 	}
3121 
3122 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3123 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3124 	if (new_parent)
3125 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3126 
3127 	error = xfs_finish_rename(tp, &free_list);
3128 	if (wip)
3129 		IRELE(wip);
3130 	return error;
3131 
3132 out_bmap_cancel:
3133 	xfs_bmap_cancel(&free_list);
3134 out_trans_cancel:
3135 	xfs_trans_cancel(tp);
3136 out_release_wip:
3137 	if (wip)
3138 		IRELE(wip);
3139 	return error;
3140 }
3141 
3142 STATIC int
3143 xfs_iflush_cluster(
3144 	struct xfs_inode	*ip,
3145 	struct xfs_buf		*bp)
3146 {
3147 	struct xfs_mount	*mp = ip->i_mount;
3148 	struct xfs_perag	*pag;
3149 	unsigned long		first_index, mask;
3150 	unsigned long		inodes_per_cluster;
3151 	int			cilist_size;
3152 	struct xfs_inode	**cilist;
3153 	struct xfs_inode	*cip;
3154 	int			nr_found;
3155 	int			clcount = 0;
3156 	int			bufwasdelwri;
3157 	int			i;
3158 
3159 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3160 
3161 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3162 	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3163 	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3164 	if (!cilist)
3165 		goto out_put;
3166 
3167 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3168 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3169 	rcu_read_lock();
3170 	/* really need a gang lookup range call here */
3171 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3172 					first_index, inodes_per_cluster);
3173 	if (nr_found == 0)
3174 		goto out_free;
3175 
3176 	for (i = 0; i < nr_found; i++) {
3177 		cip = cilist[i];
3178 		if (cip == ip)
3179 			continue;
3180 
3181 		/*
3182 		 * because this is an RCU protected lookup, we could find a
3183 		 * recently freed or even reallocated inode during the lookup.
3184 		 * We need to check under the i_flags_lock for a valid inode
3185 		 * here. Skip it if it is not valid or the wrong inode.
3186 		 */
3187 		spin_lock(&cip->i_flags_lock);
3188 		if (!cip->i_ino ||
3189 		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3190 			spin_unlock(&cip->i_flags_lock);
3191 			continue;
3192 		}
3193 
3194 		/*
3195 		 * Once we fall off the end of the cluster, no point checking
3196 		 * any more inodes in the list because they will also all be
3197 		 * outside the cluster.
3198 		 */
3199 		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3200 			spin_unlock(&cip->i_flags_lock);
3201 			break;
3202 		}
3203 		spin_unlock(&cip->i_flags_lock);
3204 
3205 		/*
3206 		 * Do an un-protected check to see if the inode is dirty and
3207 		 * is a candidate for flushing.  These checks will be repeated
3208 		 * later after the appropriate locks are acquired.
3209 		 */
3210 		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3211 			continue;
3212 
3213 		/*
3214 		 * Try to get locks.  If any are unavailable or it is pinned,
3215 		 * then this inode cannot be flushed and is skipped.
3216 		 */
3217 
3218 		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3219 			continue;
3220 		if (!xfs_iflock_nowait(cip)) {
3221 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3222 			continue;
3223 		}
3224 		if (xfs_ipincount(cip)) {
3225 			xfs_ifunlock(cip);
3226 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3227 			continue;
3228 		}
3229 
3230 
3231 		/*
3232 		 * Check the inode number again, just to be certain we are not
3233 		 * racing with freeing in xfs_reclaim_inode(). See the comments
3234 		 * in that function for more information as to why the initial
3235 		 * check is not sufficient.
3236 		 */
3237 		if (!cip->i_ino) {
3238 			xfs_ifunlock(cip);
3239 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3240 			continue;
3241 		}
3242 
3243 		/*
3244 		 * arriving here means that this inode can be flushed.  First
3245 		 * re-check that it's dirty before flushing.
3246 		 */
3247 		if (!xfs_inode_clean(cip)) {
3248 			int	error;
3249 			error = xfs_iflush_int(cip, bp);
3250 			if (error) {
3251 				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3252 				goto cluster_corrupt_out;
3253 			}
3254 			clcount++;
3255 		} else {
3256 			xfs_ifunlock(cip);
3257 		}
3258 		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3259 	}
3260 
3261 	if (clcount) {
3262 		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3263 		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3264 	}
3265 
3266 out_free:
3267 	rcu_read_unlock();
3268 	kmem_free(cilist);
3269 out_put:
3270 	xfs_perag_put(pag);
3271 	return 0;
3272 
3273 
3274 cluster_corrupt_out:
3275 	/*
3276 	 * Corruption detected in the clustering loop.  Invalidate the
3277 	 * inode buffer and shut down the filesystem.
3278 	 */
3279 	rcu_read_unlock();
3280 	/*
3281 	 * Clean up the buffer.  If it was delwri, just release it --
3282 	 * brelse can handle it with no problems.  If not, shut down the
3283 	 * filesystem before releasing the buffer.
3284 	 */
3285 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3286 	if (bufwasdelwri)
3287 		xfs_buf_relse(bp);
3288 
3289 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3290 
3291 	if (!bufwasdelwri) {
3292 		/*
3293 		 * Just like incore_relse: if we have b_iodone functions,
3294 		 * mark the buffer as an error and call them.  Otherwise
3295 		 * mark it as stale and brelse.
3296 		 */
3297 		if (bp->b_iodone) {
3298 			bp->b_flags &= ~XBF_DONE;
3299 			xfs_buf_stale(bp);
3300 			xfs_buf_ioerror(bp, -EIO);
3301 			xfs_buf_ioend(bp);
3302 		} else {
3303 			xfs_buf_stale(bp);
3304 			xfs_buf_relse(bp);
3305 		}
3306 	}
3307 
3308 	/*
3309 	 * Unlocks the flush lock
3310 	 */
3311 	xfs_iflush_abort(cip, false);
3312 	kmem_free(cilist);
3313 	xfs_perag_put(pag);
3314 	return -EFSCORRUPTED;
3315 }
3316 
3317 /*
3318  * Flush dirty inode metadata into the backing buffer.
3319  *
3320  * The caller must have the inode lock and the inode flush lock held.  The
3321  * inode lock will still be held upon return to the caller, and the inode
3322  * flush lock will be released after the inode has reached the disk.
3323  *
3324  * The caller must write out the buffer returned in *bpp and release it.
3325  */
3326 int
3327 xfs_iflush(
3328 	struct xfs_inode	*ip,
3329 	struct xfs_buf		**bpp)
3330 {
3331 	struct xfs_mount	*mp = ip->i_mount;
3332 	struct xfs_buf		*bp = NULL;
3333 	struct xfs_dinode	*dip;
3334 	int			error;
3335 
3336 	XFS_STATS_INC(mp, xs_iflush_count);
3337 
3338 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3339 	ASSERT(xfs_isiflocked(ip));
3340 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3341 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3342 
3343 	*bpp = NULL;
3344 
3345 	xfs_iunpin_wait(ip);
3346 
3347 	/*
3348 	 * For stale inodes we cannot rely on the backing buffer remaining
3349 	 * stale in cache for the remaining life of the stale inode and so
3350 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3351 	 * inodes below. We have to check this after ensuring the inode is
3352 	 * unpinned so that it is safe to reclaim the stale inode after the
3353 	 * flush call.
3354 	 */
3355 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3356 		xfs_ifunlock(ip);
3357 		return 0;
3358 	}
3359 
3360 	/*
3361 	 * This may have been unpinned because the filesystem is shutting
3362 	 * down forcibly. If that's the case we must not write this inode
3363 	 * to disk, because the log record didn't make it to disk.
3364 	 *
3365 	 * We also have to remove the log item from the AIL in this case,
3366 	 * as we wait for an empty AIL as part of the unmount process.
3367 	 */
3368 	if (XFS_FORCED_SHUTDOWN(mp)) {
3369 		error = -EIO;
3370 		goto abort_out;
3371 	}
3372 
3373 	/*
3374 	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3375 	 * operation here, so we may get  an EAGAIN error. In that case, we
3376 	 * simply want to return with the inode still dirty.
3377 	 *
3378 	 * If we get any other error, we effectively have a corruption situation
3379 	 * and we cannot flush the inode, so we treat it the same as failing
3380 	 * xfs_iflush_int().
3381 	 */
3382 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3383 			       0);
3384 	if (error == -EAGAIN) {
3385 		xfs_ifunlock(ip);
3386 		return error;
3387 	}
3388 	if (error)
3389 		goto corrupt_out;
3390 
3391 	/*
3392 	 * First flush out the inode that xfs_iflush was called with.
3393 	 */
3394 	error = xfs_iflush_int(ip, bp);
3395 	if (error)
3396 		goto corrupt_out;
3397 
3398 	/*
3399 	 * If the buffer is pinned then push on the log now so we won't
3400 	 * get stuck waiting in the write for too long.
3401 	 */
3402 	if (xfs_buf_ispinned(bp))
3403 		xfs_log_force(mp, 0);
3404 
3405 	/*
3406 	 * inode clustering:
3407 	 * see if other inodes can be gathered into this write
3408 	 */
3409 	error = xfs_iflush_cluster(ip, bp);
3410 	if (error)
3411 		goto cluster_corrupt_out;
3412 
3413 	*bpp = bp;
3414 	return 0;
3415 
3416 corrupt_out:
3417 	if (bp)
3418 		xfs_buf_relse(bp);
3419 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3420 cluster_corrupt_out:
3421 	error = -EFSCORRUPTED;
3422 abort_out:
3423 	/*
3424 	 * Unlocks the flush lock
3425 	 */
3426 	xfs_iflush_abort(ip, false);
3427 	return error;
3428 }
3429 
3430 STATIC int
3431 xfs_iflush_int(
3432 	struct xfs_inode	*ip,
3433 	struct xfs_buf		*bp)
3434 {
3435 	struct xfs_inode_log_item *iip = ip->i_itemp;
3436 	struct xfs_dinode	*dip;
3437 	struct xfs_mount	*mp = ip->i_mount;
3438 
3439 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3440 	ASSERT(xfs_isiflocked(ip));
3441 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3442 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3443 	ASSERT(iip != NULL && iip->ili_fields != 0);
3444 	ASSERT(ip->i_d.di_version > 1);
3445 
3446 	/* set *dip = inode's place in the buffer */
3447 	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3448 
3449 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3450 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3451 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3452 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3453 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3454 		goto corrupt_out;
3455 	}
3456 	if (S_ISREG(VFS_I(ip)->i_mode)) {
3457 		if (XFS_TEST_ERROR(
3458 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3459 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3460 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3461 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3462 				"%s: Bad regular inode %Lu, ptr 0x%p",
3463 				__func__, ip->i_ino, ip);
3464 			goto corrupt_out;
3465 		}
3466 	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3467 		if (XFS_TEST_ERROR(
3468 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3469 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3470 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3471 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3472 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3473 				"%s: Bad directory inode %Lu, ptr 0x%p",
3474 				__func__, ip->i_ino, ip);
3475 			goto corrupt_out;
3476 		}
3477 	}
3478 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3479 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3480 				XFS_RANDOM_IFLUSH_5)) {
3481 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3482 			"%s: detected corrupt incore inode %Lu, "
3483 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3484 			__func__, ip->i_ino,
3485 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3486 			ip->i_d.di_nblocks, ip);
3487 		goto corrupt_out;
3488 	}
3489 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3490 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3491 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3492 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3493 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3494 		goto corrupt_out;
3495 	}
3496 
3497 	/*
3498 	 * Inode item log recovery for v2 inodes are dependent on the
3499 	 * di_flushiter count for correct sequencing. We bump the flush
3500 	 * iteration count so we can detect flushes which postdate a log record
3501 	 * during recovery. This is redundant as we now log every change and
3502 	 * hence this can't happen but we need to still do it to ensure
3503 	 * backwards compatibility with old kernels that predate logging all
3504 	 * inode changes.
3505 	 */
3506 	if (ip->i_d.di_version < 3)
3507 		ip->i_d.di_flushiter++;
3508 
3509 	/*
3510 	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3511 	 * copy out the core of the inode, because if the inode is dirty at all
3512 	 * the core must be.
3513 	 */
3514 	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3515 
3516 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3517 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3518 		ip->i_d.di_flushiter = 0;
3519 
3520 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3521 	if (XFS_IFORK_Q(ip))
3522 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3523 	xfs_inobp_check(mp, bp);
3524 
3525 	/*
3526 	 * We've recorded everything logged in the inode, so we'd like to clear
3527 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3528 	 * However, we can't stop logging all this information until the data
3529 	 * we've copied into the disk buffer is written to disk.  If we did we
3530 	 * might overwrite the copy of the inode in the log with all the data
3531 	 * after re-logging only part of it, and in the face of a crash we
3532 	 * wouldn't have all the data we need to recover.
3533 	 *
3534 	 * What we do is move the bits to the ili_last_fields field.  When
3535 	 * logging the inode, these bits are moved back to the ili_fields field.
3536 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3537 	 * know that the information those bits represent is permanently on
3538 	 * disk.  As long as the flush completes before the inode is logged
3539 	 * again, then both ili_fields and ili_last_fields will be cleared.
3540 	 *
3541 	 * We can play with the ili_fields bits here, because the inode lock
3542 	 * must be held exclusively in order to set bits there and the flush
3543 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3544 	 * done routine can tell whether or not to look in the AIL.  Also, store
3545 	 * the current LSN of the inode so that we can tell whether the item has
3546 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3547 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3548 	 * atomically.
3549 	 */
3550 	iip->ili_last_fields = iip->ili_fields;
3551 	iip->ili_fields = 0;
3552 	iip->ili_fsync_fields = 0;
3553 	iip->ili_logged = 1;
3554 
3555 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3556 				&iip->ili_item.li_lsn);
3557 
3558 	/*
3559 	 * Attach the function xfs_iflush_done to the inode's
3560 	 * buffer.  This will remove the inode from the AIL
3561 	 * and unlock the inode's flush lock when the inode is
3562 	 * completely written to disk.
3563 	 */
3564 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3565 
3566 	/* generate the checksum. */
3567 	xfs_dinode_calc_crc(mp, dip);
3568 
3569 	ASSERT(bp->b_fspriv != NULL);
3570 	ASSERT(bp->b_iodone != NULL);
3571 	return 0;
3572 
3573 corrupt_out:
3574 	return -EFSCORRUPTED;
3575 }
3576