xref: /linux/fs/xfs/xfs_inode.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include <linux/log2.h>
19 
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_dir2.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_attr.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_bmap_util.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
48 #include "xfs_symlink.h"
49 #include "xfs_trans_priv.h"
50 #include "xfs_log.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_reflink.h"
53 #include "xfs_dir2_priv.h"
54 
55 kmem_zone_t *xfs_inode_zone;
56 
57 /*
58  * Used in xfs_itruncate_extents().  This is the maximum number of extents
59  * freed from a file in a single transaction.
60  */
61 #define	XFS_ITRUNC_MAX_EXTENTS	2
62 
63 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
64 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
65 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
66 
67 /*
68  * helper function to extract extent size hint from inode
69  */
70 xfs_extlen_t
71 xfs_get_extsz_hint(
72 	struct xfs_inode	*ip)
73 {
74 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
75 		return ip->i_d.di_extsize;
76 	if (XFS_IS_REALTIME_INODE(ip))
77 		return ip->i_mount->m_sb.sb_rextsize;
78 	return 0;
79 }
80 
81 /*
82  * Helper function to extract CoW extent size hint from inode.
83  * Between the extent size hint and the CoW extent size hint, we
84  * return the greater of the two.  If the value is zero (automatic),
85  * use the default size.
86  */
87 xfs_extlen_t
88 xfs_get_cowextsz_hint(
89 	struct xfs_inode	*ip)
90 {
91 	xfs_extlen_t		a, b;
92 
93 	a = 0;
94 	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
95 		a = ip->i_d.di_cowextsize;
96 	b = xfs_get_extsz_hint(ip);
97 
98 	a = max(a, b);
99 	if (a == 0)
100 		return XFS_DEFAULT_COWEXTSZ_HINT;
101 	return a;
102 }
103 
104 /*
105  * These two are wrapper routines around the xfs_ilock() routine used to
106  * centralize some grungy code.  They are used in places that wish to lock the
107  * inode solely for reading the extents.  The reason these places can't just
108  * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
109  * bringing in of the extents from disk for a file in b-tree format.  If the
110  * inode is in b-tree format, then we need to lock the inode exclusively until
111  * the extents are read in.  Locking it exclusively all the time would limit
112  * our parallelism unnecessarily, though.  What we do instead is check to see
113  * if the extents have been read in yet, and only lock the inode exclusively
114  * if they have not.
115  *
116  * The functions return a value which should be given to the corresponding
117  * xfs_iunlock() call.
118  */
119 uint
120 xfs_ilock_data_map_shared(
121 	struct xfs_inode	*ip)
122 {
123 	uint			lock_mode = XFS_ILOCK_SHARED;
124 
125 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
126 	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
127 		lock_mode = XFS_ILOCK_EXCL;
128 	xfs_ilock(ip, lock_mode);
129 	return lock_mode;
130 }
131 
132 uint
133 xfs_ilock_attr_map_shared(
134 	struct xfs_inode	*ip)
135 {
136 	uint			lock_mode = XFS_ILOCK_SHARED;
137 
138 	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
139 	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
140 		lock_mode = XFS_ILOCK_EXCL;
141 	xfs_ilock(ip, lock_mode);
142 	return lock_mode;
143 }
144 
145 /*
146  * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
147  * multi-reader locks: i_mmap_lock and the i_lock.  This routine allows
148  * various combinations of the locks to be obtained.
149  *
150  * The 3 locks should always be ordered so that the IO lock is obtained first,
151  * the mmap lock second and the ilock last in order to prevent deadlock.
152  *
153  * Basic locking order:
154  *
155  * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
156  *
157  * mmap_sem locking order:
158  *
159  * i_rwsem -> page lock -> mmap_sem
160  * mmap_sem -> i_mmap_lock -> page_lock
161  *
162  * The difference in mmap_sem locking order mean that we cannot hold the
163  * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
164  * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
165  * in get_user_pages() to map the user pages into the kernel address space for
166  * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
167  * page faults already hold the mmap_sem.
168  *
169  * Hence to serialise fully against both syscall and mmap based IO, we need to
170  * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
171  * taken in places where we need to invalidate the page cache in a race
172  * free manner (e.g. truncate, hole punch and other extent manipulation
173  * functions).
174  */
175 void
176 xfs_ilock(
177 	xfs_inode_t		*ip,
178 	uint			lock_flags)
179 {
180 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
181 
182 	/*
183 	 * You can't set both SHARED and EXCL for the same lock,
184 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
185 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
186 	 */
187 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
188 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
189 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
190 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
191 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
192 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
193 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
194 
195 	if (lock_flags & XFS_IOLOCK_EXCL) {
196 		down_write_nested(&VFS_I(ip)->i_rwsem,
197 				  XFS_IOLOCK_DEP(lock_flags));
198 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
199 		down_read_nested(&VFS_I(ip)->i_rwsem,
200 				 XFS_IOLOCK_DEP(lock_flags));
201 	}
202 
203 	if (lock_flags & XFS_MMAPLOCK_EXCL)
204 		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
205 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
206 		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
207 
208 	if (lock_flags & XFS_ILOCK_EXCL)
209 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
210 	else if (lock_flags & XFS_ILOCK_SHARED)
211 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 }
213 
214 /*
215  * This is just like xfs_ilock(), except that the caller
216  * is guaranteed not to sleep.  It returns 1 if it gets
217  * the requested locks and 0 otherwise.  If the IO lock is
218  * obtained but the inode lock cannot be, then the IO lock
219  * is dropped before returning.
220  *
221  * ip -- the inode being locked
222  * lock_flags -- this parameter indicates the inode's locks to be
223  *       to be locked.  See the comment for xfs_ilock() for a list
224  *	 of valid values.
225  */
226 int
227 xfs_ilock_nowait(
228 	xfs_inode_t		*ip,
229 	uint			lock_flags)
230 {
231 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
232 
233 	/*
234 	 * You can't set both SHARED and EXCL for the same lock,
235 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
236 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
237 	 */
238 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
239 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
240 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
241 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
242 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
243 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
244 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
245 
246 	if (lock_flags & XFS_IOLOCK_EXCL) {
247 		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
248 			goto out;
249 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
250 		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
251 			goto out;
252 	}
253 
254 	if (lock_flags & XFS_MMAPLOCK_EXCL) {
255 		if (!mrtryupdate(&ip->i_mmaplock))
256 			goto out_undo_iolock;
257 	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
258 		if (!mrtryaccess(&ip->i_mmaplock))
259 			goto out_undo_iolock;
260 	}
261 
262 	if (lock_flags & XFS_ILOCK_EXCL) {
263 		if (!mrtryupdate(&ip->i_lock))
264 			goto out_undo_mmaplock;
265 	} else if (lock_flags & XFS_ILOCK_SHARED) {
266 		if (!mrtryaccess(&ip->i_lock))
267 			goto out_undo_mmaplock;
268 	}
269 	return 1;
270 
271 out_undo_mmaplock:
272 	if (lock_flags & XFS_MMAPLOCK_EXCL)
273 		mrunlock_excl(&ip->i_mmaplock);
274 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
275 		mrunlock_shared(&ip->i_mmaplock);
276 out_undo_iolock:
277 	if (lock_flags & XFS_IOLOCK_EXCL)
278 		up_write(&VFS_I(ip)->i_rwsem);
279 	else if (lock_flags & XFS_IOLOCK_SHARED)
280 		up_read(&VFS_I(ip)->i_rwsem);
281 out:
282 	return 0;
283 }
284 
285 /*
286  * xfs_iunlock() is used to drop the inode locks acquired with
287  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
288  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
289  * that we know which locks to drop.
290  *
291  * ip -- the inode being unlocked
292  * lock_flags -- this parameter indicates the inode's locks to be
293  *       to be unlocked.  See the comment for xfs_ilock() for a list
294  *	 of valid values for this parameter.
295  *
296  */
297 void
298 xfs_iunlock(
299 	xfs_inode_t		*ip,
300 	uint			lock_flags)
301 {
302 	/*
303 	 * You can't set both SHARED and EXCL for the same lock,
304 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
305 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
306 	 */
307 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
308 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
309 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
310 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
311 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
312 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
313 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
314 	ASSERT(lock_flags != 0);
315 
316 	if (lock_flags & XFS_IOLOCK_EXCL)
317 		up_write(&VFS_I(ip)->i_rwsem);
318 	else if (lock_flags & XFS_IOLOCK_SHARED)
319 		up_read(&VFS_I(ip)->i_rwsem);
320 
321 	if (lock_flags & XFS_MMAPLOCK_EXCL)
322 		mrunlock_excl(&ip->i_mmaplock);
323 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
324 		mrunlock_shared(&ip->i_mmaplock);
325 
326 	if (lock_flags & XFS_ILOCK_EXCL)
327 		mrunlock_excl(&ip->i_lock);
328 	else if (lock_flags & XFS_ILOCK_SHARED)
329 		mrunlock_shared(&ip->i_lock);
330 
331 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
332 }
333 
334 /*
335  * give up write locks.  the i/o lock cannot be held nested
336  * if it is being demoted.
337  */
338 void
339 xfs_ilock_demote(
340 	xfs_inode_t		*ip,
341 	uint			lock_flags)
342 {
343 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
344 	ASSERT((lock_flags &
345 		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
346 
347 	if (lock_flags & XFS_ILOCK_EXCL)
348 		mrdemote(&ip->i_lock);
349 	if (lock_flags & XFS_MMAPLOCK_EXCL)
350 		mrdemote(&ip->i_mmaplock);
351 	if (lock_flags & XFS_IOLOCK_EXCL)
352 		downgrade_write(&VFS_I(ip)->i_rwsem);
353 
354 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
355 }
356 
357 #if defined(DEBUG) || defined(XFS_WARN)
358 int
359 xfs_isilocked(
360 	xfs_inode_t		*ip,
361 	uint			lock_flags)
362 {
363 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
364 		if (!(lock_flags & XFS_ILOCK_SHARED))
365 			return !!ip->i_lock.mr_writer;
366 		return rwsem_is_locked(&ip->i_lock.mr_lock);
367 	}
368 
369 	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
370 		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
371 			return !!ip->i_mmaplock.mr_writer;
372 		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
373 	}
374 
375 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
376 		if (!(lock_flags & XFS_IOLOCK_SHARED))
377 			return !debug_locks ||
378 				lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
379 		return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
380 	}
381 
382 	ASSERT(0);
383 	return 0;
384 }
385 #endif
386 
387 #ifdef DEBUG
388 int xfs_locked_n;
389 int xfs_small_retries;
390 int xfs_middle_retries;
391 int xfs_lots_retries;
392 int xfs_lock_delays;
393 #endif
394 
395 /*
396  * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
397  * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
398  * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
399  * errors and warnings.
400  */
401 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
402 static bool
403 xfs_lockdep_subclass_ok(
404 	int subclass)
405 {
406 	return subclass < MAX_LOCKDEP_SUBCLASSES;
407 }
408 #else
409 #define xfs_lockdep_subclass_ok(subclass)	(true)
410 #endif
411 
412 /*
413  * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
414  * value. This can be called for any type of inode lock combination, including
415  * parent locking. Care must be taken to ensure we don't overrun the subclass
416  * storage fields in the class mask we build.
417  */
418 static inline int
419 xfs_lock_inumorder(int lock_mode, int subclass)
420 {
421 	int	class = 0;
422 
423 	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
424 			      XFS_ILOCK_RTSUM)));
425 	ASSERT(xfs_lockdep_subclass_ok(subclass));
426 
427 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
428 		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
429 		class += subclass << XFS_IOLOCK_SHIFT;
430 	}
431 
432 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
433 		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
434 		class += subclass << XFS_MMAPLOCK_SHIFT;
435 	}
436 
437 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
438 		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
439 		class += subclass << XFS_ILOCK_SHIFT;
440 	}
441 
442 	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
443 }
444 
445 /*
446  * The following routine will lock n inodes in exclusive mode.  We assume the
447  * caller calls us with the inodes in i_ino order.
448  *
449  * We need to detect deadlock where an inode that we lock is in the AIL and we
450  * start waiting for another inode that is locked by a thread in a long running
451  * transaction (such as truncate). This can result in deadlock since the long
452  * running trans might need to wait for the inode we just locked in order to
453  * push the tail and free space in the log.
454  *
455  * xfs_lock_inodes() can only be used to lock one type of lock at a time -
456  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
457  * lock more than one at a time, lockdep will report false positives saying we
458  * have violated locking orders.
459  */
460 static void
461 xfs_lock_inodes(
462 	xfs_inode_t	**ips,
463 	int		inodes,
464 	uint		lock_mode)
465 {
466 	int		attempts = 0, i, j, try_lock;
467 	xfs_log_item_t	*lp;
468 
469 	/*
470 	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
471 	 * support an arbitrary depth of locking here, but absolute limits on
472 	 * inodes depend on the the type of locking and the limits placed by
473 	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
474 	 * the asserts.
475 	 */
476 	ASSERT(ips && inodes >= 2 && inodes <= 5);
477 	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
478 			    XFS_ILOCK_EXCL));
479 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
480 			      XFS_ILOCK_SHARED)));
481 	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
482 		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
483 	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
484 		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
485 
486 	if (lock_mode & XFS_IOLOCK_EXCL) {
487 		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
488 	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
489 		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
490 
491 	try_lock = 0;
492 	i = 0;
493 again:
494 	for (; i < inodes; i++) {
495 		ASSERT(ips[i]);
496 
497 		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
498 			continue;
499 
500 		/*
501 		 * If try_lock is not set yet, make sure all locked inodes are
502 		 * not in the AIL.  If any are, set try_lock to be used later.
503 		 */
504 		if (!try_lock) {
505 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
506 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
507 				if (lp && (lp->li_flags & XFS_LI_IN_AIL))
508 					try_lock++;
509 			}
510 		}
511 
512 		/*
513 		 * If any of the previous locks we have locked is in the AIL,
514 		 * we must TRY to get the second and subsequent locks. If
515 		 * we can't get any, we must release all we have
516 		 * and try again.
517 		 */
518 		if (!try_lock) {
519 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
520 			continue;
521 		}
522 
523 		/* try_lock means we have an inode locked that is in the AIL. */
524 		ASSERT(i != 0);
525 		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
526 			continue;
527 
528 		/*
529 		 * Unlock all previous guys and try again.  xfs_iunlock will try
530 		 * to push the tail if the inode is in the AIL.
531 		 */
532 		attempts++;
533 		for (j = i - 1; j >= 0; j--) {
534 			/*
535 			 * Check to see if we've already unlocked this one.  Not
536 			 * the first one going back, and the inode ptr is the
537 			 * same.
538 			 */
539 			if (j != (i - 1) && ips[j] == ips[j + 1])
540 				continue;
541 
542 			xfs_iunlock(ips[j], lock_mode);
543 		}
544 
545 		if ((attempts % 5) == 0) {
546 			delay(1); /* Don't just spin the CPU */
547 #ifdef DEBUG
548 			xfs_lock_delays++;
549 #endif
550 		}
551 		i = 0;
552 		try_lock = 0;
553 		goto again;
554 	}
555 
556 #ifdef DEBUG
557 	if (attempts) {
558 		if (attempts < 5) xfs_small_retries++;
559 		else if (attempts < 100) xfs_middle_retries++;
560 		else xfs_lots_retries++;
561 	} else {
562 		xfs_locked_n++;
563 	}
564 #endif
565 }
566 
567 /*
568  * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
569  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
570  * lock more than one at a time, lockdep will report false positives saying we
571  * have violated locking orders.
572  */
573 void
574 xfs_lock_two_inodes(
575 	xfs_inode_t		*ip0,
576 	xfs_inode_t		*ip1,
577 	uint			lock_mode)
578 {
579 	xfs_inode_t		*temp;
580 	int			attempts = 0;
581 	xfs_log_item_t		*lp;
582 
583 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
584 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
585 		ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
586 
587 	ASSERT(ip0->i_ino != ip1->i_ino);
588 
589 	if (ip0->i_ino > ip1->i_ino) {
590 		temp = ip0;
591 		ip0 = ip1;
592 		ip1 = temp;
593 	}
594 
595  again:
596 	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
597 
598 	/*
599 	 * If the first lock we have locked is in the AIL, we must TRY to get
600 	 * the second lock. If we can't get it, we must release the first one
601 	 * and try again.
602 	 */
603 	lp = (xfs_log_item_t *)ip0->i_itemp;
604 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
605 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
606 			xfs_iunlock(ip0, lock_mode);
607 			if ((++attempts % 5) == 0)
608 				delay(1); /* Don't just spin the CPU */
609 			goto again;
610 		}
611 	} else {
612 		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
613 	}
614 }
615 
616 
617 void
618 __xfs_iflock(
619 	struct xfs_inode	*ip)
620 {
621 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
622 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
623 
624 	do {
625 		prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
626 		if (xfs_isiflocked(ip))
627 			io_schedule();
628 	} while (!xfs_iflock_nowait(ip));
629 
630 	finish_wait(wq, &wait.wq_entry);
631 }
632 
633 STATIC uint
634 _xfs_dic2xflags(
635 	uint16_t		di_flags,
636 	uint64_t		di_flags2,
637 	bool			has_attr)
638 {
639 	uint			flags = 0;
640 
641 	if (di_flags & XFS_DIFLAG_ANY) {
642 		if (di_flags & XFS_DIFLAG_REALTIME)
643 			flags |= FS_XFLAG_REALTIME;
644 		if (di_flags & XFS_DIFLAG_PREALLOC)
645 			flags |= FS_XFLAG_PREALLOC;
646 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
647 			flags |= FS_XFLAG_IMMUTABLE;
648 		if (di_flags & XFS_DIFLAG_APPEND)
649 			flags |= FS_XFLAG_APPEND;
650 		if (di_flags & XFS_DIFLAG_SYNC)
651 			flags |= FS_XFLAG_SYNC;
652 		if (di_flags & XFS_DIFLAG_NOATIME)
653 			flags |= FS_XFLAG_NOATIME;
654 		if (di_flags & XFS_DIFLAG_NODUMP)
655 			flags |= FS_XFLAG_NODUMP;
656 		if (di_flags & XFS_DIFLAG_RTINHERIT)
657 			flags |= FS_XFLAG_RTINHERIT;
658 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
659 			flags |= FS_XFLAG_PROJINHERIT;
660 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
661 			flags |= FS_XFLAG_NOSYMLINKS;
662 		if (di_flags & XFS_DIFLAG_EXTSIZE)
663 			flags |= FS_XFLAG_EXTSIZE;
664 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
665 			flags |= FS_XFLAG_EXTSZINHERIT;
666 		if (di_flags & XFS_DIFLAG_NODEFRAG)
667 			flags |= FS_XFLAG_NODEFRAG;
668 		if (di_flags & XFS_DIFLAG_FILESTREAM)
669 			flags |= FS_XFLAG_FILESTREAM;
670 	}
671 
672 	if (di_flags2 & XFS_DIFLAG2_ANY) {
673 		if (di_flags2 & XFS_DIFLAG2_DAX)
674 			flags |= FS_XFLAG_DAX;
675 		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
676 			flags |= FS_XFLAG_COWEXTSIZE;
677 	}
678 
679 	if (has_attr)
680 		flags |= FS_XFLAG_HASATTR;
681 
682 	return flags;
683 }
684 
685 uint
686 xfs_ip2xflags(
687 	struct xfs_inode	*ip)
688 {
689 	struct xfs_icdinode	*dic = &ip->i_d;
690 
691 	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
692 }
693 
694 /*
695  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
696  * is allowed, otherwise it has to be an exact match. If a CI match is found,
697  * ci_name->name will point to a the actual name (caller must free) or
698  * will be set to NULL if an exact match is found.
699  */
700 int
701 xfs_lookup(
702 	xfs_inode_t		*dp,
703 	struct xfs_name		*name,
704 	xfs_inode_t		**ipp,
705 	struct xfs_name		*ci_name)
706 {
707 	xfs_ino_t		inum;
708 	int			error;
709 
710 	trace_xfs_lookup(dp, name);
711 
712 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
713 		return -EIO;
714 
715 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
716 	if (error)
717 		goto out_unlock;
718 
719 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
720 	if (error)
721 		goto out_free_name;
722 
723 	return 0;
724 
725 out_free_name:
726 	if (ci_name)
727 		kmem_free(ci_name->name);
728 out_unlock:
729 	*ipp = NULL;
730 	return error;
731 }
732 
733 /*
734  * Allocate an inode on disk and return a copy of its in-core version.
735  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
736  * appropriately within the inode.  The uid and gid for the inode are
737  * set according to the contents of the given cred structure.
738  *
739  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
740  * has a free inode available, call xfs_iget() to obtain the in-core
741  * version of the allocated inode.  Finally, fill in the inode and
742  * log its initial contents.  In this case, ialloc_context would be
743  * set to NULL.
744  *
745  * If xfs_dialloc() does not have an available inode, it will replenish
746  * its supply by doing an allocation. Since we can only do one
747  * allocation within a transaction without deadlocks, we must commit
748  * the current transaction before returning the inode itself.
749  * In this case, therefore, we will set ialloc_context and return.
750  * The caller should then commit the current transaction, start a new
751  * transaction, and call xfs_ialloc() again to actually get the inode.
752  *
753  * To ensure that some other process does not grab the inode that
754  * was allocated during the first call to xfs_ialloc(), this routine
755  * also returns the [locked] bp pointing to the head of the freelist
756  * as ialloc_context.  The caller should hold this buffer across
757  * the commit and pass it back into this routine on the second call.
758  *
759  * If we are allocating quota inodes, we do not have a parent inode
760  * to attach to or associate with (i.e. pip == NULL) because they
761  * are not linked into the directory structure - they are attached
762  * directly to the superblock - and so have no parent.
763  */
764 static int
765 xfs_ialloc(
766 	xfs_trans_t	*tp,
767 	xfs_inode_t	*pip,
768 	umode_t		mode,
769 	xfs_nlink_t	nlink,
770 	xfs_dev_t	rdev,
771 	prid_t		prid,
772 	int		okalloc,
773 	xfs_buf_t	**ialloc_context,
774 	xfs_inode_t	**ipp)
775 {
776 	struct xfs_mount *mp = tp->t_mountp;
777 	xfs_ino_t	ino;
778 	xfs_inode_t	*ip;
779 	uint		flags;
780 	int		error;
781 	struct timespec	tv;
782 	struct inode	*inode;
783 
784 	/*
785 	 * Call the space management code to pick
786 	 * the on-disk inode to be allocated.
787 	 */
788 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
789 			    ialloc_context, &ino);
790 	if (error)
791 		return error;
792 	if (*ialloc_context || ino == NULLFSINO) {
793 		*ipp = NULL;
794 		return 0;
795 	}
796 	ASSERT(*ialloc_context == NULL);
797 
798 	/*
799 	 * Get the in-core inode with the lock held exclusively.
800 	 * This is because we're setting fields here we need
801 	 * to prevent others from looking at until we're done.
802 	 */
803 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
804 			 XFS_ILOCK_EXCL, &ip);
805 	if (error)
806 		return error;
807 	ASSERT(ip != NULL);
808 	inode = VFS_I(ip);
809 
810 	/*
811 	 * We always convert v1 inodes to v2 now - we only support filesystems
812 	 * with >= v2 inode capability, so there is no reason for ever leaving
813 	 * an inode in v1 format.
814 	 */
815 	if (ip->i_d.di_version == 1)
816 		ip->i_d.di_version = 2;
817 
818 	inode->i_mode = mode;
819 	set_nlink(inode, nlink);
820 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
821 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
822 	xfs_set_projid(ip, prid);
823 
824 	if (pip && XFS_INHERIT_GID(pip)) {
825 		ip->i_d.di_gid = pip->i_d.di_gid;
826 		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
827 			inode->i_mode |= S_ISGID;
828 	}
829 
830 	/*
831 	 * If the group ID of the new file does not match the effective group
832 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
833 	 * (and only if the irix_sgid_inherit compatibility variable is set).
834 	 */
835 	if ((irix_sgid_inherit) &&
836 	    (inode->i_mode & S_ISGID) &&
837 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
838 		inode->i_mode &= ~S_ISGID;
839 
840 	ip->i_d.di_size = 0;
841 	ip->i_d.di_nextents = 0;
842 	ASSERT(ip->i_d.di_nblocks == 0);
843 
844 	tv = current_time(inode);
845 	inode->i_mtime = tv;
846 	inode->i_atime = tv;
847 	inode->i_ctime = tv;
848 
849 	ip->i_d.di_extsize = 0;
850 	ip->i_d.di_dmevmask = 0;
851 	ip->i_d.di_dmstate = 0;
852 	ip->i_d.di_flags = 0;
853 
854 	if (ip->i_d.di_version == 3) {
855 		inode->i_version = 1;
856 		ip->i_d.di_flags2 = 0;
857 		ip->i_d.di_cowextsize = 0;
858 		ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
859 		ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
860 	}
861 
862 
863 	flags = XFS_ILOG_CORE;
864 	switch (mode & S_IFMT) {
865 	case S_IFIFO:
866 	case S_IFCHR:
867 	case S_IFBLK:
868 	case S_IFSOCK:
869 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
870 		ip->i_df.if_u2.if_rdev = rdev;
871 		ip->i_df.if_flags = 0;
872 		flags |= XFS_ILOG_DEV;
873 		break;
874 	case S_IFREG:
875 	case S_IFDIR:
876 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
877 			uint64_t	di_flags2 = 0;
878 			uint		di_flags = 0;
879 
880 			if (S_ISDIR(mode)) {
881 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
882 					di_flags |= XFS_DIFLAG_RTINHERIT;
883 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
884 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
885 					ip->i_d.di_extsize = pip->i_d.di_extsize;
886 				}
887 				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
888 					di_flags |= XFS_DIFLAG_PROJINHERIT;
889 			} else if (S_ISREG(mode)) {
890 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
891 					di_flags |= XFS_DIFLAG_REALTIME;
892 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
893 					di_flags |= XFS_DIFLAG_EXTSIZE;
894 					ip->i_d.di_extsize = pip->i_d.di_extsize;
895 				}
896 			}
897 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
898 			    xfs_inherit_noatime)
899 				di_flags |= XFS_DIFLAG_NOATIME;
900 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
901 			    xfs_inherit_nodump)
902 				di_flags |= XFS_DIFLAG_NODUMP;
903 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
904 			    xfs_inherit_sync)
905 				di_flags |= XFS_DIFLAG_SYNC;
906 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
907 			    xfs_inherit_nosymlinks)
908 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
909 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
910 			    xfs_inherit_nodefrag)
911 				di_flags |= XFS_DIFLAG_NODEFRAG;
912 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
913 				di_flags |= XFS_DIFLAG_FILESTREAM;
914 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
915 				di_flags2 |= XFS_DIFLAG2_DAX;
916 
917 			ip->i_d.di_flags |= di_flags;
918 			ip->i_d.di_flags2 |= di_flags2;
919 		}
920 		if (pip &&
921 		    (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
922 		    pip->i_d.di_version == 3 &&
923 		    ip->i_d.di_version == 3) {
924 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
925 				ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
926 				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
927 			}
928 		}
929 		/* FALLTHROUGH */
930 	case S_IFLNK:
931 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
932 		ip->i_df.if_flags = XFS_IFEXTENTS;
933 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
934 		ip->i_df.if_u1.if_extents = NULL;
935 		break;
936 	default:
937 		ASSERT(0);
938 	}
939 	/*
940 	 * Attribute fork settings for new inode.
941 	 */
942 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
943 	ip->i_d.di_anextents = 0;
944 
945 	/*
946 	 * Log the new values stuffed into the inode.
947 	 */
948 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
949 	xfs_trans_log_inode(tp, ip, flags);
950 
951 	/* now that we have an i_mode we can setup the inode structure */
952 	xfs_setup_inode(ip);
953 
954 	*ipp = ip;
955 	return 0;
956 }
957 
958 /*
959  * Allocates a new inode from disk and return a pointer to the
960  * incore copy. This routine will internally commit the current
961  * transaction and allocate a new one if the Space Manager needed
962  * to do an allocation to replenish the inode free-list.
963  *
964  * This routine is designed to be called from xfs_create and
965  * xfs_create_dir.
966  *
967  */
968 int
969 xfs_dir_ialloc(
970 	xfs_trans_t	**tpp,		/* input: current transaction;
971 					   output: may be a new transaction. */
972 	xfs_inode_t	*dp,		/* directory within whose allocate
973 					   the inode. */
974 	umode_t		mode,
975 	xfs_nlink_t	nlink,
976 	xfs_dev_t	rdev,
977 	prid_t		prid,		/* project id */
978 	int		okalloc,	/* ok to allocate new space */
979 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
980 					   locked. */
981 	int		*committed)
982 
983 {
984 	xfs_trans_t	*tp;
985 	xfs_inode_t	*ip;
986 	xfs_buf_t	*ialloc_context = NULL;
987 	int		code;
988 	void		*dqinfo;
989 	uint		tflags;
990 
991 	tp = *tpp;
992 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
993 
994 	/*
995 	 * xfs_ialloc will return a pointer to an incore inode if
996 	 * the Space Manager has an available inode on the free
997 	 * list. Otherwise, it will do an allocation and replenish
998 	 * the freelist.  Since we can only do one allocation per
999 	 * transaction without deadlocks, we will need to commit the
1000 	 * current transaction and start a new one.  We will then
1001 	 * need to call xfs_ialloc again to get the inode.
1002 	 *
1003 	 * If xfs_ialloc did an allocation to replenish the freelist,
1004 	 * it returns the bp containing the head of the freelist as
1005 	 * ialloc_context. We will hold a lock on it across the
1006 	 * transaction commit so that no other process can steal
1007 	 * the inode(s) that we've just allocated.
1008 	 */
1009 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
1010 			  &ialloc_context, &ip);
1011 
1012 	/*
1013 	 * Return an error if we were unable to allocate a new inode.
1014 	 * This should only happen if we run out of space on disk or
1015 	 * encounter a disk error.
1016 	 */
1017 	if (code) {
1018 		*ipp = NULL;
1019 		return code;
1020 	}
1021 	if (!ialloc_context && !ip) {
1022 		*ipp = NULL;
1023 		return -ENOSPC;
1024 	}
1025 
1026 	/*
1027 	 * If the AGI buffer is non-NULL, then we were unable to get an
1028 	 * inode in one operation.  We need to commit the current
1029 	 * transaction and call xfs_ialloc() again.  It is guaranteed
1030 	 * to succeed the second time.
1031 	 */
1032 	if (ialloc_context) {
1033 		/*
1034 		 * Normally, xfs_trans_commit releases all the locks.
1035 		 * We call bhold to hang on to the ialloc_context across
1036 		 * the commit.  Holding this buffer prevents any other
1037 		 * processes from doing any allocations in this
1038 		 * allocation group.
1039 		 */
1040 		xfs_trans_bhold(tp, ialloc_context);
1041 
1042 		/*
1043 		 * We want the quota changes to be associated with the next
1044 		 * transaction, NOT this one. So, detach the dqinfo from this
1045 		 * and attach it to the next transaction.
1046 		 */
1047 		dqinfo = NULL;
1048 		tflags = 0;
1049 		if (tp->t_dqinfo) {
1050 			dqinfo = (void *)tp->t_dqinfo;
1051 			tp->t_dqinfo = NULL;
1052 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1053 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1054 		}
1055 
1056 		code = xfs_trans_roll(&tp, NULL);
1057 		if (committed != NULL)
1058 			*committed = 1;
1059 
1060 		/*
1061 		 * Re-attach the quota info that we detached from prev trx.
1062 		 */
1063 		if (dqinfo) {
1064 			tp->t_dqinfo = dqinfo;
1065 			tp->t_flags |= tflags;
1066 		}
1067 
1068 		if (code) {
1069 			xfs_buf_relse(ialloc_context);
1070 			*tpp = tp;
1071 			*ipp = NULL;
1072 			return code;
1073 		}
1074 		xfs_trans_bjoin(tp, ialloc_context);
1075 
1076 		/*
1077 		 * Call ialloc again. Since we've locked out all
1078 		 * other allocations in this allocation group,
1079 		 * this call should always succeed.
1080 		 */
1081 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1082 				  okalloc, &ialloc_context, &ip);
1083 
1084 		/*
1085 		 * If we get an error at this point, return to the caller
1086 		 * so that the current transaction can be aborted.
1087 		 */
1088 		if (code) {
1089 			*tpp = tp;
1090 			*ipp = NULL;
1091 			return code;
1092 		}
1093 		ASSERT(!ialloc_context && ip);
1094 
1095 	} else {
1096 		if (committed != NULL)
1097 			*committed = 0;
1098 	}
1099 
1100 	*ipp = ip;
1101 	*tpp = tp;
1102 
1103 	return 0;
1104 }
1105 
1106 /*
1107  * Decrement the link count on an inode & log the change.  If this causes the
1108  * link count to go to zero, move the inode to AGI unlinked list so that it can
1109  * be freed when the last active reference goes away via xfs_inactive().
1110  */
1111 static int			/* error */
1112 xfs_droplink(
1113 	xfs_trans_t *tp,
1114 	xfs_inode_t *ip)
1115 {
1116 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1117 
1118 	drop_nlink(VFS_I(ip));
1119 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1120 
1121 	if (VFS_I(ip)->i_nlink)
1122 		return 0;
1123 
1124 	return xfs_iunlink(tp, ip);
1125 }
1126 
1127 /*
1128  * Increment the link count on an inode & log the change.
1129  */
1130 static int
1131 xfs_bumplink(
1132 	xfs_trans_t *tp,
1133 	xfs_inode_t *ip)
1134 {
1135 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1136 
1137 	ASSERT(ip->i_d.di_version > 1);
1138 	inc_nlink(VFS_I(ip));
1139 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1140 	return 0;
1141 }
1142 
1143 int
1144 xfs_create(
1145 	xfs_inode_t		*dp,
1146 	struct xfs_name		*name,
1147 	umode_t			mode,
1148 	xfs_dev_t		rdev,
1149 	xfs_inode_t		**ipp)
1150 {
1151 	int			is_dir = S_ISDIR(mode);
1152 	struct xfs_mount	*mp = dp->i_mount;
1153 	struct xfs_inode	*ip = NULL;
1154 	struct xfs_trans	*tp = NULL;
1155 	int			error;
1156 	struct xfs_defer_ops	dfops;
1157 	xfs_fsblock_t		first_block;
1158 	bool                    unlock_dp_on_error = false;
1159 	prid_t			prid;
1160 	struct xfs_dquot	*udqp = NULL;
1161 	struct xfs_dquot	*gdqp = NULL;
1162 	struct xfs_dquot	*pdqp = NULL;
1163 	struct xfs_trans_res	*tres;
1164 	uint			resblks;
1165 
1166 	trace_xfs_create(dp, name);
1167 
1168 	if (XFS_FORCED_SHUTDOWN(mp))
1169 		return -EIO;
1170 
1171 	prid = xfs_get_initial_prid(dp);
1172 
1173 	/*
1174 	 * Make sure that we have allocated dquot(s) on disk.
1175 	 */
1176 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1177 					xfs_kgid_to_gid(current_fsgid()), prid,
1178 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1179 					&udqp, &gdqp, &pdqp);
1180 	if (error)
1181 		return error;
1182 
1183 	if (is_dir) {
1184 		rdev = 0;
1185 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1186 		tres = &M_RES(mp)->tr_mkdir;
1187 	} else {
1188 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1189 		tres = &M_RES(mp)->tr_create;
1190 	}
1191 
1192 	/*
1193 	 * Initially assume that the file does not exist and
1194 	 * reserve the resources for that case.  If that is not
1195 	 * the case we'll drop the one we have and get a more
1196 	 * appropriate transaction later.
1197 	 */
1198 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1199 	if (error == -ENOSPC) {
1200 		/* flush outstanding delalloc blocks and retry */
1201 		xfs_flush_inodes(mp);
1202 		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1203 	}
1204 	if (error == -ENOSPC) {
1205 		/* No space at all so try a "no-allocation" reservation */
1206 		resblks = 0;
1207 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1208 	}
1209 	if (error)
1210 		goto out_release_inode;
1211 
1212 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1213 	unlock_dp_on_error = true;
1214 
1215 	xfs_defer_init(&dfops, &first_block);
1216 
1217 	/*
1218 	 * Reserve disk quota and the inode.
1219 	 */
1220 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1221 						pdqp, resblks, 1, 0);
1222 	if (error)
1223 		goto out_trans_cancel;
1224 
1225 	if (!resblks) {
1226 		error = xfs_dir_canenter(tp, dp, name);
1227 		if (error)
1228 			goto out_trans_cancel;
1229 	}
1230 
1231 	/*
1232 	 * A newly created regular or special file just has one directory
1233 	 * entry pointing to them, but a directory also the "." entry
1234 	 * pointing to itself.
1235 	 */
1236 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1237 			       prid, resblks > 0, &ip, NULL);
1238 	if (error)
1239 		goto out_trans_cancel;
1240 
1241 	/*
1242 	 * Now we join the directory inode to the transaction.  We do not do it
1243 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1244 	 * (and release all the locks).  An error from here on will result in
1245 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1246 	 * error path.
1247 	 */
1248 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1249 	unlock_dp_on_error = false;
1250 
1251 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1252 					&first_block, &dfops, resblks ?
1253 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1254 	if (error) {
1255 		ASSERT(error != -ENOSPC);
1256 		goto out_trans_cancel;
1257 	}
1258 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1259 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1260 
1261 	if (is_dir) {
1262 		error = xfs_dir_init(tp, ip, dp);
1263 		if (error)
1264 			goto out_bmap_cancel;
1265 
1266 		error = xfs_bumplink(tp, dp);
1267 		if (error)
1268 			goto out_bmap_cancel;
1269 	}
1270 
1271 	/*
1272 	 * If this is a synchronous mount, make sure that the
1273 	 * create transaction goes to disk before returning to
1274 	 * the user.
1275 	 */
1276 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1277 		xfs_trans_set_sync(tp);
1278 
1279 	/*
1280 	 * Attach the dquot(s) to the inodes and modify them incore.
1281 	 * These ids of the inode couldn't have changed since the new
1282 	 * inode has been locked ever since it was created.
1283 	 */
1284 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1285 
1286 	error = xfs_defer_finish(&tp, &dfops, NULL);
1287 	if (error)
1288 		goto out_bmap_cancel;
1289 
1290 	error = xfs_trans_commit(tp);
1291 	if (error)
1292 		goto out_release_inode;
1293 
1294 	xfs_qm_dqrele(udqp);
1295 	xfs_qm_dqrele(gdqp);
1296 	xfs_qm_dqrele(pdqp);
1297 
1298 	*ipp = ip;
1299 	return 0;
1300 
1301  out_bmap_cancel:
1302 	xfs_defer_cancel(&dfops);
1303  out_trans_cancel:
1304 	xfs_trans_cancel(tp);
1305  out_release_inode:
1306 	/*
1307 	 * Wait until after the current transaction is aborted to finish the
1308 	 * setup of the inode and release the inode.  This prevents recursive
1309 	 * transactions and deadlocks from xfs_inactive.
1310 	 */
1311 	if (ip) {
1312 		xfs_finish_inode_setup(ip);
1313 		IRELE(ip);
1314 	}
1315 
1316 	xfs_qm_dqrele(udqp);
1317 	xfs_qm_dqrele(gdqp);
1318 	xfs_qm_dqrele(pdqp);
1319 
1320 	if (unlock_dp_on_error)
1321 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1322 	return error;
1323 }
1324 
1325 int
1326 xfs_create_tmpfile(
1327 	struct xfs_inode	*dp,
1328 	struct dentry		*dentry,
1329 	umode_t			mode,
1330 	struct xfs_inode	**ipp)
1331 {
1332 	struct xfs_mount	*mp = dp->i_mount;
1333 	struct xfs_inode	*ip = NULL;
1334 	struct xfs_trans	*tp = NULL;
1335 	int			error;
1336 	prid_t                  prid;
1337 	struct xfs_dquot	*udqp = NULL;
1338 	struct xfs_dquot	*gdqp = NULL;
1339 	struct xfs_dquot	*pdqp = NULL;
1340 	struct xfs_trans_res	*tres;
1341 	uint			resblks;
1342 
1343 	if (XFS_FORCED_SHUTDOWN(mp))
1344 		return -EIO;
1345 
1346 	prid = xfs_get_initial_prid(dp);
1347 
1348 	/*
1349 	 * Make sure that we have allocated dquot(s) on disk.
1350 	 */
1351 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1352 				xfs_kgid_to_gid(current_fsgid()), prid,
1353 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1354 				&udqp, &gdqp, &pdqp);
1355 	if (error)
1356 		return error;
1357 
1358 	resblks = XFS_IALLOC_SPACE_RES(mp);
1359 	tres = &M_RES(mp)->tr_create_tmpfile;
1360 
1361 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1362 	if (error == -ENOSPC) {
1363 		/* No space at all so try a "no-allocation" reservation */
1364 		resblks = 0;
1365 		error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
1366 	}
1367 	if (error)
1368 		goto out_release_inode;
1369 
1370 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1371 						pdqp, resblks, 1, 0);
1372 	if (error)
1373 		goto out_trans_cancel;
1374 
1375 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1376 				prid, resblks > 0, &ip, NULL);
1377 	if (error)
1378 		goto out_trans_cancel;
1379 
1380 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1381 		xfs_trans_set_sync(tp);
1382 
1383 	/*
1384 	 * Attach the dquot(s) to the inodes and modify them incore.
1385 	 * These ids of the inode couldn't have changed since the new
1386 	 * inode has been locked ever since it was created.
1387 	 */
1388 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1389 
1390 	error = xfs_iunlink(tp, ip);
1391 	if (error)
1392 		goto out_trans_cancel;
1393 
1394 	error = xfs_trans_commit(tp);
1395 	if (error)
1396 		goto out_release_inode;
1397 
1398 	xfs_qm_dqrele(udqp);
1399 	xfs_qm_dqrele(gdqp);
1400 	xfs_qm_dqrele(pdqp);
1401 
1402 	*ipp = ip;
1403 	return 0;
1404 
1405  out_trans_cancel:
1406 	xfs_trans_cancel(tp);
1407  out_release_inode:
1408 	/*
1409 	 * Wait until after the current transaction is aborted to finish the
1410 	 * setup of the inode and release the inode.  This prevents recursive
1411 	 * transactions and deadlocks from xfs_inactive.
1412 	 */
1413 	if (ip) {
1414 		xfs_finish_inode_setup(ip);
1415 		IRELE(ip);
1416 	}
1417 
1418 	xfs_qm_dqrele(udqp);
1419 	xfs_qm_dqrele(gdqp);
1420 	xfs_qm_dqrele(pdqp);
1421 
1422 	return error;
1423 }
1424 
1425 int
1426 xfs_link(
1427 	xfs_inode_t		*tdp,
1428 	xfs_inode_t		*sip,
1429 	struct xfs_name		*target_name)
1430 {
1431 	xfs_mount_t		*mp = tdp->i_mount;
1432 	xfs_trans_t		*tp;
1433 	int			error;
1434 	struct xfs_defer_ops	dfops;
1435 	xfs_fsblock_t           first_block;
1436 	int			resblks;
1437 
1438 	trace_xfs_link(tdp, target_name);
1439 
1440 	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1441 
1442 	if (XFS_FORCED_SHUTDOWN(mp))
1443 		return -EIO;
1444 
1445 	error = xfs_qm_dqattach(sip, 0);
1446 	if (error)
1447 		goto std_return;
1448 
1449 	error = xfs_qm_dqattach(tdp, 0);
1450 	if (error)
1451 		goto std_return;
1452 
1453 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1454 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1455 	if (error == -ENOSPC) {
1456 		resblks = 0;
1457 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1458 	}
1459 	if (error)
1460 		goto std_return;
1461 
1462 	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1463 
1464 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1465 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1466 
1467 	/*
1468 	 * If we are using project inheritance, we only allow hard link
1469 	 * creation in our tree when the project IDs are the same; else
1470 	 * the tree quota mechanism could be circumvented.
1471 	 */
1472 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1473 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1474 		error = -EXDEV;
1475 		goto error_return;
1476 	}
1477 
1478 	if (!resblks) {
1479 		error = xfs_dir_canenter(tp, tdp, target_name);
1480 		if (error)
1481 			goto error_return;
1482 	}
1483 
1484 	xfs_defer_init(&dfops, &first_block);
1485 
1486 	/*
1487 	 * Handle initial link state of O_TMPFILE inode
1488 	 */
1489 	if (VFS_I(sip)->i_nlink == 0) {
1490 		error = xfs_iunlink_remove(tp, sip);
1491 		if (error)
1492 			goto error_return;
1493 	}
1494 
1495 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1496 					&first_block, &dfops, resblks);
1497 	if (error)
1498 		goto error_return;
1499 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1500 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1501 
1502 	error = xfs_bumplink(tp, sip);
1503 	if (error)
1504 		goto error_return;
1505 
1506 	/*
1507 	 * If this is a synchronous mount, make sure that the
1508 	 * link transaction goes to disk before returning to
1509 	 * the user.
1510 	 */
1511 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1512 		xfs_trans_set_sync(tp);
1513 
1514 	error = xfs_defer_finish(&tp, &dfops, NULL);
1515 	if (error) {
1516 		xfs_defer_cancel(&dfops);
1517 		goto error_return;
1518 	}
1519 
1520 	return xfs_trans_commit(tp);
1521 
1522  error_return:
1523 	xfs_trans_cancel(tp);
1524  std_return:
1525 	return error;
1526 }
1527 
1528 /*
1529  * Free up the underlying blocks past new_size.  The new size must be smaller
1530  * than the current size.  This routine can be used both for the attribute and
1531  * data fork, and does not modify the inode size, which is left to the caller.
1532  *
1533  * The transaction passed to this routine must have made a permanent log
1534  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1535  * given transaction and start new ones, so make sure everything involved in
1536  * the transaction is tidy before calling here.  Some transaction will be
1537  * returned to the caller to be committed.  The incoming transaction must
1538  * already include the inode, and both inode locks must be held exclusively.
1539  * The inode must also be "held" within the transaction.  On return the inode
1540  * will be "held" within the returned transaction.  This routine does NOT
1541  * require any disk space to be reserved for it within the transaction.
1542  *
1543  * If we get an error, we must return with the inode locked and linked into the
1544  * current transaction. This keeps things simple for the higher level code,
1545  * because it always knows that the inode is locked and held in the transaction
1546  * that returns to it whether errors occur or not.  We don't mark the inode
1547  * dirty on error so that transactions can be easily aborted if possible.
1548  */
1549 int
1550 xfs_itruncate_extents(
1551 	struct xfs_trans	**tpp,
1552 	struct xfs_inode	*ip,
1553 	int			whichfork,
1554 	xfs_fsize_t		new_size)
1555 {
1556 	struct xfs_mount	*mp = ip->i_mount;
1557 	struct xfs_trans	*tp = *tpp;
1558 	struct xfs_defer_ops	dfops;
1559 	xfs_fsblock_t		first_block;
1560 	xfs_fileoff_t		first_unmap_block;
1561 	xfs_fileoff_t		last_block;
1562 	xfs_filblks_t		unmap_len;
1563 	int			error = 0;
1564 	int			done = 0;
1565 
1566 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1567 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1568 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1569 	ASSERT(new_size <= XFS_ISIZE(ip));
1570 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1571 	ASSERT(ip->i_itemp != NULL);
1572 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1573 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1574 
1575 	trace_xfs_itruncate_extents_start(ip, new_size);
1576 
1577 	/*
1578 	 * Since it is possible for space to become allocated beyond
1579 	 * the end of the file (in a crash where the space is allocated
1580 	 * but the inode size is not yet updated), simply remove any
1581 	 * blocks which show up between the new EOF and the maximum
1582 	 * possible file size.  If the first block to be removed is
1583 	 * beyond the maximum file size (ie it is the same as last_block),
1584 	 * then there is nothing to do.
1585 	 */
1586 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1587 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1588 	if (first_unmap_block == last_block)
1589 		return 0;
1590 
1591 	ASSERT(first_unmap_block < last_block);
1592 	unmap_len = last_block - first_unmap_block + 1;
1593 	while (!done) {
1594 		xfs_defer_init(&dfops, &first_block);
1595 		error = xfs_bunmapi(tp, ip,
1596 				    first_unmap_block, unmap_len,
1597 				    xfs_bmapi_aflag(whichfork),
1598 				    XFS_ITRUNC_MAX_EXTENTS,
1599 				    &first_block, &dfops,
1600 				    &done);
1601 		if (error)
1602 			goto out_bmap_cancel;
1603 
1604 		/*
1605 		 * Duplicate the transaction that has the permanent
1606 		 * reservation and commit the old transaction.
1607 		 */
1608 		error = xfs_defer_finish(&tp, &dfops, ip);
1609 		if (error)
1610 			goto out_bmap_cancel;
1611 
1612 		error = xfs_trans_roll(&tp, ip);
1613 		if (error)
1614 			goto out;
1615 	}
1616 
1617 	/* Remove all pending CoW reservations. */
1618 	error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1619 			last_block, true);
1620 	if (error)
1621 		goto out;
1622 
1623 	/*
1624 	 * Clear the reflink flag if we truncated everything.
1625 	 */
1626 	if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip)) {
1627 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1628 		xfs_inode_clear_cowblocks_tag(ip);
1629 	}
1630 
1631 	/*
1632 	 * Always re-log the inode so that our permanent transaction can keep
1633 	 * on rolling it forward in the log.
1634 	 */
1635 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1636 
1637 	trace_xfs_itruncate_extents_end(ip, new_size);
1638 
1639 out:
1640 	*tpp = tp;
1641 	return error;
1642 out_bmap_cancel:
1643 	/*
1644 	 * If the bunmapi call encounters an error, return to the caller where
1645 	 * the transaction can be properly aborted.  We just need to make sure
1646 	 * we're not holding any resources that we were not when we came in.
1647 	 */
1648 	xfs_defer_cancel(&dfops);
1649 	goto out;
1650 }
1651 
1652 int
1653 xfs_release(
1654 	xfs_inode_t	*ip)
1655 {
1656 	xfs_mount_t	*mp = ip->i_mount;
1657 	int		error;
1658 
1659 	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1660 		return 0;
1661 
1662 	/* If this is a read-only mount, don't do this (would generate I/O) */
1663 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1664 		return 0;
1665 
1666 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1667 		int truncated;
1668 
1669 		/*
1670 		 * If we previously truncated this file and removed old data
1671 		 * in the process, we want to initiate "early" writeout on
1672 		 * the last close.  This is an attempt to combat the notorious
1673 		 * NULL files problem which is particularly noticeable from a
1674 		 * truncate down, buffered (re-)write (delalloc), followed by
1675 		 * a crash.  What we are effectively doing here is
1676 		 * significantly reducing the time window where we'd otherwise
1677 		 * be exposed to that problem.
1678 		 */
1679 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1680 		if (truncated) {
1681 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1682 			if (ip->i_delayed_blks > 0) {
1683 				error = filemap_flush(VFS_I(ip)->i_mapping);
1684 				if (error)
1685 					return error;
1686 			}
1687 		}
1688 	}
1689 
1690 	if (VFS_I(ip)->i_nlink == 0)
1691 		return 0;
1692 
1693 	if (xfs_can_free_eofblocks(ip, false)) {
1694 
1695 		/*
1696 		 * Check if the inode is being opened, written and closed
1697 		 * frequently and we have delayed allocation blocks outstanding
1698 		 * (e.g. streaming writes from the NFS server), truncating the
1699 		 * blocks past EOF will cause fragmentation to occur.
1700 		 *
1701 		 * In this case don't do the truncation, but we have to be
1702 		 * careful how we detect this case. Blocks beyond EOF show up as
1703 		 * i_delayed_blks even when the inode is clean, so we need to
1704 		 * truncate them away first before checking for a dirty release.
1705 		 * Hence on the first dirty close we will still remove the
1706 		 * speculative allocation, but after that we will leave it in
1707 		 * place.
1708 		 */
1709 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1710 			return 0;
1711 		/*
1712 		 * If we can't get the iolock just skip truncating the blocks
1713 		 * past EOF because we could deadlock with the mmap_sem
1714 		 * otherwise. We'll get another chance to drop them once the
1715 		 * last reference to the inode is dropped, so we'll never leak
1716 		 * blocks permanently.
1717 		 */
1718 		if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1719 			error = xfs_free_eofblocks(ip);
1720 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1721 			if (error)
1722 				return error;
1723 		}
1724 
1725 		/* delalloc blocks after truncation means it really is dirty */
1726 		if (ip->i_delayed_blks)
1727 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1728 	}
1729 	return 0;
1730 }
1731 
1732 /*
1733  * xfs_inactive_truncate
1734  *
1735  * Called to perform a truncate when an inode becomes unlinked.
1736  */
1737 STATIC int
1738 xfs_inactive_truncate(
1739 	struct xfs_inode *ip)
1740 {
1741 	struct xfs_mount	*mp = ip->i_mount;
1742 	struct xfs_trans	*tp;
1743 	int			error;
1744 
1745 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1746 	if (error) {
1747 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1748 		return error;
1749 	}
1750 
1751 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1752 	xfs_trans_ijoin(tp, ip, 0);
1753 
1754 	/*
1755 	 * Log the inode size first to prevent stale data exposure in the event
1756 	 * of a system crash before the truncate completes. See the related
1757 	 * comment in xfs_vn_setattr_size() for details.
1758 	 */
1759 	ip->i_d.di_size = 0;
1760 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1761 
1762 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1763 	if (error)
1764 		goto error_trans_cancel;
1765 
1766 	ASSERT(ip->i_d.di_nextents == 0);
1767 
1768 	error = xfs_trans_commit(tp);
1769 	if (error)
1770 		goto error_unlock;
1771 
1772 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1773 	return 0;
1774 
1775 error_trans_cancel:
1776 	xfs_trans_cancel(tp);
1777 error_unlock:
1778 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1779 	return error;
1780 }
1781 
1782 /*
1783  * xfs_inactive_ifree()
1784  *
1785  * Perform the inode free when an inode is unlinked.
1786  */
1787 STATIC int
1788 xfs_inactive_ifree(
1789 	struct xfs_inode *ip)
1790 {
1791 	struct xfs_defer_ops	dfops;
1792 	xfs_fsblock_t		first_block;
1793 	struct xfs_mount	*mp = ip->i_mount;
1794 	struct xfs_trans	*tp;
1795 	int			error;
1796 
1797 	/*
1798 	 * We try to use a per-AG reservation for any block needed by the finobt
1799 	 * tree, but as the finobt feature predates the per-AG reservation
1800 	 * support a degraded file system might not have enough space for the
1801 	 * reservation at mount time.  In that case try to dip into the reserved
1802 	 * pool and pray.
1803 	 *
1804 	 * Send a warning if the reservation does happen to fail, as the inode
1805 	 * now remains allocated and sits on the unlinked list until the fs is
1806 	 * repaired.
1807 	 */
1808 	if (unlikely(mp->m_inotbt_nores)) {
1809 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1810 				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1811 				&tp);
1812 	} else {
1813 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1814 	}
1815 	if (error) {
1816 		if (error == -ENOSPC) {
1817 			xfs_warn_ratelimited(mp,
1818 			"Failed to remove inode(s) from unlinked list. "
1819 			"Please free space, unmount and run xfs_repair.");
1820 		} else {
1821 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1822 		}
1823 		return error;
1824 	}
1825 
1826 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1827 	xfs_trans_ijoin(tp, ip, 0);
1828 
1829 	xfs_defer_init(&dfops, &first_block);
1830 	error = xfs_ifree(tp, ip, &dfops);
1831 	if (error) {
1832 		/*
1833 		 * If we fail to free the inode, shut down.  The cancel
1834 		 * might do that, we need to make sure.  Otherwise the
1835 		 * inode might be lost for a long time or forever.
1836 		 */
1837 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1838 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1839 				__func__, error);
1840 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1841 		}
1842 		xfs_trans_cancel(tp);
1843 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1844 		return error;
1845 	}
1846 
1847 	/*
1848 	 * Credit the quota account(s). The inode is gone.
1849 	 */
1850 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1851 
1852 	/*
1853 	 * Just ignore errors at this point.  There is nothing we can do except
1854 	 * to try to keep going. Make sure it's not a silent error.
1855 	 */
1856 	error = xfs_defer_finish(&tp, &dfops, NULL);
1857 	if (error) {
1858 		xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1859 			__func__, error);
1860 		xfs_defer_cancel(&dfops);
1861 	}
1862 	error = xfs_trans_commit(tp);
1863 	if (error)
1864 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1865 			__func__, error);
1866 
1867 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1868 	return 0;
1869 }
1870 
1871 /*
1872  * xfs_inactive
1873  *
1874  * This is called when the vnode reference count for the vnode
1875  * goes to zero.  If the file has been unlinked, then it must
1876  * now be truncated.  Also, we clear all of the read-ahead state
1877  * kept for the inode here since the file is now closed.
1878  */
1879 void
1880 xfs_inactive(
1881 	xfs_inode_t	*ip)
1882 {
1883 	struct xfs_mount	*mp;
1884 	int			error;
1885 	int			truncate = 0;
1886 
1887 	/*
1888 	 * If the inode is already free, then there can be nothing
1889 	 * to clean up here.
1890 	 */
1891 	if (VFS_I(ip)->i_mode == 0) {
1892 		ASSERT(ip->i_df.if_real_bytes == 0);
1893 		ASSERT(ip->i_df.if_broot_bytes == 0);
1894 		return;
1895 	}
1896 
1897 	mp = ip->i_mount;
1898 	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1899 
1900 	/* If this is a read-only mount, don't do this (would generate I/O) */
1901 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1902 		return;
1903 
1904 	if (VFS_I(ip)->i_nlink != 0) {
1905 		/*
1906 		 * force is true because we are evicting an inode from the
1907 		 * cache. Post-eof blocks must be freed, lest we end up with
1908 		 * broken free space accounting.
1909 		 *
1910 		 * Note: don't bother with iolock here since lockdep complains
1911 		 * about acquiring it in reclaim context. We have the only
1912 		 * reference to the inode at this point anyways.
1913 		 */
1914 		if (xfs_can_free_eofblocks(ip, true))
1915 			xfs_free_eofblocks(ip);
1916 
1917 		return;
1918 	}
1919 
1920 	if (S_ISREG(VFS_I(ip)->i_mode) &&
1921 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1922 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1923 		truncate = 1;
1924 
1925 	error = xfs_qm_dqattach(ip, 0);
1926 	if (error)
1927 		return;
1928 
1929 	if (S_ISLNK(VFS_I(ip)->i_mode))
1930 		error = xfs_inactive_symlink(ip);
1931 	else if (truncate)
1932 		error = xfs_inactive_truncate(ip);
1933 	if (error)
1934 		return;
1935 
1936 	/*
1937 	 * If there are attributes associated with the file then blow them away
1938 	 * now.  The code calls a routine that recursively deconstructs the
1939 	 * attribute fork. If also blows away the in-core attribute fork.
1940 	 */
1941 	if (XFS_IFORK_Q(ip)) {
1942 		error = xfs_attr_inactive(ip);
1943 		if (error)
1944 			return;
1945 	}
1946 
1947 	ASSERT(!ip->i_afp);
1948 	ASSERT(ip->i_d.di_anextents == 0);
1949 	ASSERT(ip->i_d.di_forkoff == 0);
1950 
1951 	/*
1952 	 * Free the inode.
1953 	 */
1954 	error = xfs_inactive_ifree(ip);
1955 	if (error)
1956 		return;
1957 
1958 	/*
1959 	 * Release the dquots held by inode, if any.
1960 	 */
1961 	xfs_qm_dqdetach(ip);
1962 }
1963 
1964 /*
1965  * This is called when the inode's link count goes to 0 or we are creating a
1966  * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1967  * set to true as the link count is dropped to zero by the VFS after we've
1968  * created the file successfully, so we have to add it to the unlinked list
1969  * while the link count is non-zero.
1970  *
1971  * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1972  * list when the inode is freed.
1973  */
1974 STATIC int
1975 xfs_iunlink(
1976 	struct xfs_trans *tp,
1977 	struct xfs_inode *ip)
1978 {
1979 	xfs_mount_t	*mp = tp->t_mountp;
1980 	xfs_agi_t	*agi;
1981 	xfs_dinode_t	*dip;
1982 	xfs_buf_t	*agibp;
1983 	xfs_buf_t	*ibp;
1984 	xfs_agino_t	agino;
1985 	short		bucket_index;
1986 	int		offset;
1987 	int		error;
1988 
1989 	ASSERT(VFS_I(ip)->i_mode != 0);
1990 
1991 	/*
1992 	 * Get the agi buffer first.  It ensures lock ordering
1993 	 * on the list.
1994 	 */
1995 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1996 	if (error)
1997 		return error;
1998 	agi = XFS_BUF_TO_AGI(agibp);
1999 
2000 	/*
2001 	 * Get the index into the agi hash table for the
2002 	 * list this inode will go on.
2003 	 */
2004 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2005 	ASSERT(agino != 0);
2006 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2007 	ASSERT(agi->agi_unlinked[bucket_index]);
2008 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2009 
2010 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2011 		/*
2012 		 * There is already another inode in the bucket we need
2013 		 * to add ourselves to.  Add us at the front of the list.
2014 		 * Here we put the head pointer into our next pointer,
2015 		 * and then we fall through to point the head at us.
2016 		 */
2017 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2018 				       0, 0);
2019 		if (error)
2020 			return error;
2021 
2022 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2023 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2024 		offset = ip->i_imap.im_boffset +
2025 			offsetof(xfs_dinode_t, di_next_unlinked);
2026 
2027 		/* need to recalc the inode CRC if appropriate */
2028 		xfs_dinode_calc_crc(mp, dip);
2029 
2030 		xfs_trans_inode_buf(tp, ibp);
2031 		xfs_trans_log_buf(tp, ibp, offset,
2032 				  (offset + sizeof(xfs_agino_t) - 1));
2033 		xfs_inobp_check(mp, ibp);
2034 	}
2035 
2036 	/*
2037 	 * Point the bucket head pointer at the inode being inserted.
2038 	 */
2039 	ASSERT(agino != 0);
2040 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2041 	offset = offsetof(xfs_agi_t, agi_unlinked) +
2042 		(sizeof(xfs_agino_t) * bucket_index);
2043 	xfs_trans_log_buf(tp, agibp, offset,
2044 			  (offset + sizeof(xfs_agino_t) - 1));
2045 	return 0;
2046 }
2047 
2048 /*
2049  * Pull the on-disk inode from the AGI unlinked list.
2050  */
2051 STATIC int
2052 xfs_iunlink_remove(
2053 	xfs_trans_t	*tp,
2054 	xfs_inode_t	*ip)
2055 {
2056 	xfs_ino_t	next_ino;
2057 	xfs_mount_t	*mp;
2058 	xfs_agi_t	*agi;
2059 	xfs_dinode_t	*dip;
2060 	xfs_buf_t	*agibp;
2061 	xfs_buf_t	*ibp;
2062 	xfs_agnumber_t	agno;
2063 	xfs_agino_t	agino;
2064 	xfs_agino_t	next_agino;
2065 	xfs_buf_t	*last_ibp;
2066 	xfs_dinode_t	*last_dip = NULL;
2067 	short		bucket_index;
2068 	int		offset, last_offset = 0;
2069 	int		error;
2070 
2071 	mp = tp->t_mountp;
2072 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2073 
2074 	/*
2075 	 * Get the agi buffer first.  It ensures lock ordering
2076 	 * on the list.
2077 	 */
2078 	error = xfs_read_agi(mp, tp, agno, &agibp);
2079 	if (error)
2080 		return error;
2081 
2082 	agi = XFS_BUF_TO_AGI(agibp);
2083 
2084 	/*
2085 	 * Get the index into the agi hash table for the
2086 	 * list this inode will go on.
2087 	 */
2088 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2089 	ASSERT(agino != 0);
2090 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2091 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2092 	ASSERT(agi->agi_unlinked[bucket_index]);
2093 
2094 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2095 		/*
2096 		 * We're at the head of the list.  Get the inode's on-disk
2097 		 * buffer to see if there is anyone after us on the list.
2098 		 * Only modify our next pointer if it is not already NULLAGINO.
2099 		 * This saves us the overhead of dealing with the buffer when
2100 		 * there is no need to change it.
2101 		 */
2102 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2103 				       0, 0);
2104 		if (error) {
2105 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2106 				__func__, error);
2107 			return error;
2108 		}
2109 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2110 		ASSERT(next_agino != 0);
2111 		if (next_agino != NULLAGINO) {
2112 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2113 			offset = ip->i_imap.im_boffset +
2114 				offsetof(xfs_dinode_t, di_next_unlinked);
2115 
2116 			/* need to recalc the inode CRC if appropriate */
2117 			xfs_dinode_calc_crc(mp, dip);
2118 
2119 			xfs_trans_inode_buf(tp, ibp);
2120 			xfs_trans_log_buf(tp, ibp, offset,
2121 					  (offset + sizeof(xfs_agino_t) - 1));
2122 			xfs_inobp_check(mp, ibp);
2123 		} else {
2124 			xfs_trans_brelse(tp, ibp);
2125 		}
2126 		/*
2127 		 * Point the bucket head pointer at the next inode.
2128 		 */
2129 		ASSERT(next_agino != 0);
2130 		ASSERT(next_agino != agino);
2131 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2132 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2133 			(sizeof(xfs_agino_t) * bucket_index);
2134 		xfs_trans_log_buf(tp, agibp, offset,
2135 				  (offset + sizeof(xfs_agino_t) - 1));
2136 	} else {
2137 		/*
2138 		 * We need to search the list for the inode being freed.
2139 		 */
2140 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2141 		last_ibp = NULL;
2142 		while (next_agino != agino) {
2143 			struct xfs_imap	imap;
2144 
2145 			if (last_ibp)
2146 				xfs_trans_brelse(tp, last_ibp);
2147 
2148 			imap.im_blkno = 0;
2149 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2150 
2151 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2152 			if (error) {
2153 				xfs_warn(mp,
2154 	"%s: xfs_imap returned error %d.",
2155 					 __func__, error);
2156 				return error;
2157 			}
2158 
2159 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2160 					       &last_ibp, 0, 0);
2161 			if (error) {
2162 				xfs_warn(mp,
2163 	"%s: xfs_imap_to_bp returned error %d.",
2164 					__func__, error);
2165 				return error;
2166 			}
2167 
2168 			last_offset = imap.im_boffset;
2169 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2170 			ASSERT(next_agino != NULLAGINO);
2171 			ASSERT(next_agino != 0);
2172 		}
2173 
2174 		/*
2175 		 * Now last_ibp points to the buffer previous to us on the
2176 		 * unlinked list.  Pull us from the list.
2177 		 */
2178 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2179 				       0, 0);
2180 		if (error) {
2181 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2182 				__func__, error);
2183 			return error;
2184 		}
2185 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2186 		ASSERT(next_agino != 0);
2187 		ASSERT(next_agino != agino);
2188 		if (next_agino != NULLAGINO) {
2189 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2190 			offset = ip->i_imap.im_boffset +
2191 				offsetof(xfs_dinode_t, di_next_unlinked);
2192 
2193 			/* need to recalc the inode CRC if appropriate */
2194 			xfs_dinode_calc_crc(mp, dip);
2195 
2196 			xfs_trans_inode_buf(tp, ibp);
2197 			xfs_trans_log_buf(tp, ibp, offset,
2198 					  (offset + sizeof(xfs_agino_t) - 1));
2199 			xfs_inobp_check(mp, ibp);
2200 		} else {
2201 			xfs_trans_brelse(tp, ibp);
2202 		}
2203 		/*
2204 		 * Point the previous inode on the list to the next inode.
2205 		 */
2206 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2207 		ASSERT(next_agino != 0);
2208 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2209 
2210 		/* need to recalc the inode CRC if appropriate */
2211 		xfs_dinode_calc_crc(mp, last_dip);
2212 
2213 		xfs_trans_inode_buf(tp, last_ibp);
2214 		xfs_trans_log_buf(tp, last_ibp, offset,
2215 				  (offset + sizeof(xfs_agino_t) - 1));
2216 		xfs_inobp_check(mp, last_ibp);
2217 	}
2218 	return 0;
2219 }
2220 
2221 /*
2222  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2223  * inodes that are in memory - they all must be marked stale and attached to
2224  * the cluster buffer.
2225  */
2226 STATIC int
2227 xfs_ifree_cluster(
2228 	xfs_inode_t		*free_ip,
2229 	xfs_trans_t		*tp,
2230 	struct xfs_icluster	*xic)
2231 {
2232 	xfs_mount_t		*mp = free_ip->i_mount;
2233 	int			blks_per_cluster;
2234 	int			inodes_per_cluster;
2235 	int			nbufs;
2236 	int			i, j;
2237 	int			ioffset;
2238 	xfs_daddr_t		blkno;
2239 	xfs_buf_t		*bp;
2240 	xfs_inode_t		*ip;
2241 	xfs_inode_log_item_t	*iip;
2242 	xfs_log_item_t		*lip;
2243 	struct xfs_perag	*pag;
2244 	xfs_ino_t		inum;
2245 
2246 	inum = xic->first_ino;
2247 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2248 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2249 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2250 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2251 
2252 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2253 		/*
2254 		 * The allocation bitmap tells us which inodes of the chunk were
2255 		 * physically allocated. Skip the cluster if an inode falls into
2256 		 * a sparse region.
2257 		 */
2258 		ioffset = inum - xic->first_ino;
2259 		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2260 			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2261 			continue;
2262 		}
2263 
2264 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2265 					 XFS_INO_TO_AGBNO(mp, inum));
2266 
2267 		/*
2268 		 * We obtain and lock the backing buffer first in the process
2269 		 * here, as we have to ensure that any dirty inode that we
2270 		 * can't get the flush lock on is attached to the buffer.
2271 		 * If we scan the in-memory inodes first, then buffer IO can
2272 		 * complete before we get a lock on it, and hence we may fail
2273 		 * to mark all the active inodes on the buffer stale.
2274 		 */
2275 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2276 					mp->m_bsize * blks_per_cluster,
2277 					XBF_UNMAPPED);
2278 
2279 		if (!bp)
2280 			return -ENOMEM;
2281 
2282 		/*
2283 		 * This buffer may not have been correctly initialised as we
2284 		 * didn't read it from disk. That's not important because we are
2285 		 * only using to mark the buffer as stale in the log, and to
2286 		 * attach stale cached inodes on it. That means it will never be
2287 		 * dispatched for IO. If it is, we want to know about it, and we
2288 		 * want it to fail. We can acheive this by adding a write
2289 		 * verifier to the buffer.
2290 		 */
2291 		 bp->b_ops = &xfs_inode_buf_ops;
2292 
2293 		/*
2294 		 * Walk the inodes already attached to the buffer and mark them
2295 		 * stale. These will all have the flush locks held, so an
2296 		 * in-memory inode walk can't lock them. By marking them all
2297 		 * stale first, we will not attempt to lock them in the loop
2298 		 * below as the XFS_ISTALE flag will be set.
2299 		 */
2300 		lip = bp->b_fspriv;
2301 		while (lip) {
2302 			if (lip->li_type == XFS_LI_INODE) {
2303 				iip = (xfs_inode_log_item_t *)lip;
2304 				ASSERT(iip->ili_logged == 1);
2305 				lip->li_cb = xfs_istale_done;
2306 				xfs_trans_ail_copy_lsn(mp->m_ail,
2307 							&iip->ili_flush_lsn,
2308 							&iip->ili_item.li_lsn);
2309 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2310 			}
2311 			lip = lip->li_bio_list;
2312 		}
2313 
2314 
2315 		/*
2316 		 * For each inode in memory attempt to add it to the inode
2317 		 * buffer and set it up for being staled on buffer IO
2318 		 * completion.  This is safe as we've locked out tail pushing
2319 		 * and flushing by locking the buffer.
2320 		 *
2321 		 * We have already marked every inode that was part of a
2322 		 * transaction stale above, which means there is no point in
2323 		 * even trying to lock them.
2324 		 */
2325 		for (i = 0; i < inodes_per_cluster; i++) {
2326 retry:
2327 			rcu_read_lock();
2328 			ip = radix_tree_lookup(&pag->pag_ici_root,
2329 					XFS_INO_TO_AGINO(mp, (inum + i)));
2330 
2331 			/* Inode not in memory, nothing to do */
2332 			if (!ip) {
2333 				rcu_read_unlock();
2334 				continue;
2335 			}
2336 
2337 			/*
2338 			 * because this is an RCU protected lookup, we could
2339 			 * find a recently freed or even reallocated inode
2340 			 * during the lookup. We need to check under the
2341 			 * i_flags_lock for a valid inode here. Skip it if it
2342 			 * is not valid, the wrong inode or stale.
2343 			 */
2344 			spin_lock(&ip->i_flags_lock);
2345 			if (ip->i_ino != inum + i ||
2346 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2347 				spin_unlock(&ip->i_flags_lock);
2348 				rcu_read_unlock();
2349 				continue;
2350 			}
2351 			spin_unlock(&ip->i_flags_lock);
2352 
2353 			/*
2354 			 * Don't try to lock/unlock the current inode, but we
2355 			 * _cannot_ skip the other inodes that we did not find
2356 			 * in the list attached to the buffer and are not
2357 			 * already marked stale. If we can't lock it, back off
2358 			 * and retry.
2359 			 */
2360 			if (ip != free_ip &&
2361 			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2362 				rcu_read_unlock();
2363 				delay(1);
2364 				goto retry;
2365 			}
2366 			rcu_read_unlock();
2367 
2368 			xfs_iflock(ip);
2369 			xfs_iflags_set(ip, XFS_ISTALE);
2370 
2371 			/*
2372 			 * we don't need to attach clean inodes or those only
2373 			 * with unlogged changes (which we throw away, anyway).
2374 			 */
2375 			iip = ip->i_itemp;
2376 			if (!iip || xfs_inode_clean(ip)) {
2377 				ASSERT(ip != free_ip);
2378 				xfs_ifunlock(ip);
2379 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2380 				continue;
2381 			}
2382 
2383 			iip->ili_last_fields = iip->ili_fields;
2384 			iip->ili_fields = 0;
2385 			iip->ili_fsync_fields = 0;
2386 			iip->ili_logged = 1;
2387 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2388 						&iip->ili_item.li_lsn);
2389 
2390 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2391 						  &iip->ili_item);
2392 
2393 			if (ip != free_ip)
2394 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2395 		}
2396 
2397 		xfs_trans_stale_inode_buf(tp, bp);
2398 		xfs_trans_binval(tp, bp);
2399 	}
2400 
2401 	xfs_perag_put(pag);
2402 	return 0;
2403 }
2404 
2405 /*
2406  * This is called to return an inode to the inode free list.
2407  * The inode should already be truncated to 0 length and have
2408  * no pages associated with it.  This routine also assumes that
2409  * the inode is already a part of the transaction.
2410  *
2411  * The on-disk copy of the inode will have been added to the list
2412  * of unlinked inodes in the AGI. We need to remove the inode from
2413  * that list atomically with respect to freeing it here.
2414  */
2415 int
2416 xfs_ifree(
2417 	xfs_trans_t	*tp,
2418 	xfs_inode_t	*ip,
2419 	struct xfs_defer_ops	*dfops)
2420 {
2421 	int			error;
2422 	struct xfs_icluster	xic = { 0 };
2423 
2424 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2425 	ASSERT(VFS_I(ip)->i_nlink == 0);
2426 	ASSERT(ip->i_d.di_nextents == 0);
2427 	ASSERT(ip->i_d.di_anextents == 0);
2428 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2429 	ASSERT(ip->i_d.di_nblocks == 0);
2430 
2431 	/*
2432 	 * Pull the on-disk inode from the AGI unlinked list.
2433 	 */
2434 	error = xfs_iunlink_remove(tp, ip);
2435 	if (error)
2436 		return error;
2437 
2438 	error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2439 	if (error)
2440 		return error;
2441 
2442 	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2443 	ip->i_d.di_flags = 0;
2444 	ip->i_d.di_dmevmask = 0;
2445 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2446 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2447 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2448 	/*
2449 	 * Bump the generation count so no one will be confused
2450 	 * by reincarnations of this inode.
2451 	 */
2452 	VFS_I(ip)->i_generation++;
2453 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2454 
2455 	if (xic.deleted)
2456 		error = xfs_ifree_cluster(ip, tp, &xic);
2457 
2458 	return error;
2459 }
2460 
2461 /*
2462  * This is called to unpin an inode.  The caller must have the inode locked
2463  * in at least shared mode so that the buffer cannot be subsequently pinned
2464  * once someone is waiting for it to be unpinned.
2465  */
2466 static void
2467 xfs_iunpin(
2468 	struct xfs_inode	*ip)
2469 {
2470 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2471 
2472 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2473 
2474 	/* Give the log a push to start the unpinning I/O */
2475 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2476 
2477 }
2478 
2479 static void
2480 __xfs_iunpin_wait(
2481 	struct xfs_inode	*ip)
2482 {
2483 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2484 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2485 
2486 	xfs_iunpin(ip);
2487 
2488 	do {
2489 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2490 		if (xfs_ipincount(ip))
2491 			io_schedule();
2492 	} while (xfs_ipincount(ip));
2493 	finish_wait(wq, &wait.wq_entry);
2494 }
2495 
2496 void
2497 xfs_iunpin_wait(
2498 	struct xfs_inode	*ip)
2499 {
2500 	if (xfs_ipincount(ip))
2501 		__xfs_iunpin_wait(ip);
2502 }
2503 
2504 /*
2505  * Removing an inode from the namespace involves removing the directory entry
2506  * and dropping the link count on the inode. Removing the directory entry can
2507  * result in locking an AGF (directory blocks were freed) and removing a link
2508  * count can result in placing the inode on an unlinked list which results in
2509  * locking an AGI.
2510  *
2511  * The big problem here is that we have an ordering constraint on AGF and AGI
2512  * locking - inode allocation locks the AGI, then can allocate a new extent for
2513  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2514  * removes the inode from the unlinked list, requiring that we lock the AGI
2515  * first, and then freeing the inode can result in an inode chunk being freed
2516  * and hence freeing disk space requiring that we lock an AGF.
2517  *
2518  * Hence the ordering that is imposed by other parts of the code is AGI before
2519  * AGF. This means we cannot remove the directory entry before we drop the inode
2520  * reference count and put it on the unlinked list as this results in a lock
2521  * order of AGF then AGI, and this can deadlock against inode allocation and
2522  * freeing. Therefore we must drop the link counts before we remove the
2523  * directory entry.
2524  *
2525  * This is still safe from a transactional point of view - it is not until we
2526  * get to xfs_defer_finish() that we have the possibility of multiple
2527  * transactions in this operation. Hence as long as we remove the directory
2528  * entry and drop the link count in the first transaction of the remove
2529  * operation, there are no transactional constraints on the ordering here.
2530  */
2531 int
2532 xfs_remove(
2533 	xfs_inode_t             *dp,
2534 	struct xfs_name		*name,
2535 	xfs_inode_t		*ip)
2536 {
2537 	xfs_mount_t		*mp = dp->i_mount;
2538 	xfs_trans_t             *tp = NULL;
2539 	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2540 	int                     error = 0;
2541 	struct xfs_defer_ops	dfops;
2542 	xfs_fsblock_t           first_block;
2543 	uint			resblks;
2544 
2545 	trace_xfs_remove(dp, name);
2546 
2547 	if (XFS_FORCED_SHUTDOWN(mp))
2548 		return -EIO;
2549 
2550 	error = xfs_qm_dqattach(dp, 0);
2551 	if (error)
2552 		goto std_return;
2553 
2554 	error = xfs_qm_dqattach(ip, 0);
2555 	if (error)
2556 		goto std_return;
2557 
2558 	/*
2559 	 * We try to get the real space reservation first,
2560 	 * allowing for directory btree deletion(s) implying
2561 	 * possible bmap insert(s).  If we can't get the space
2562 	 * reservation then we use 0 instead, and avoid the bmap
2563 	 * btree insert(s) in the directory code by, if the bmap
2564 	 * insert tries to happen, instead trimming the LAST
2565 	 * block from the directory.
2566 	 */
2567 	resblks = XFS_REMOVE_SPACE_RES(mp);
2568 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2569 	if (error == -ENOSPC) {
2570 		resblks = 0;
2571 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2572 				&tp);
2573 	}
2574 	if (error) {
2575 		ASSERT(error != -ENOSPC);
2576 		goto std_return;
2577 	}
2578 
2579 	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2580 
2581 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2582 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2583 
2584 	/*
2585 	 * If we're removing a directory perform some additional validation.
2586 	 */
2587 	if (is_dir) {
2588 		ASSERT(VFS_I(ip)->i_nlink >= 2);
2589 		if (VFS_I(ip)->i_nlink != 2) {
2590 			error = -ENOTEMPTY;
2591 			goto out_trans_cancel;
2592 		}
2593 		if (!xfs_dir_isempty(ip)) {
2594 			error = -ENOTEMPTY;
2595 			goto out_trans_cancel;
2596 		}
2597 
2598 		/* Drop the link from ip's "..".  */
2599 		error = xfs_droplink(tp, dp);
2600 		if (error)
2601 			goto out_trans_cancel;
2602 
2603 		/* Drop the "." link from ip to self.  */
2604 		error = xfs_droplink(tp, ip);
2605 		if (error)
2606 			goto out_trans_cancel;
2607 	} else {
2608 		/*
2609 		 * When removing a non-directory we need to log the parent
2610 		 * inode here.  For a directory this is done implicitly
2611 		 * by the xfs_droplink call for the ".." entry.
2612 		 */
2613 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2614 	}
2615 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2616 
2617 	/* Drop the link from dp to ip. */
2618 	error = xfs_droplink(tp, ip);
2619 	if (error)
2620 		goto out_trans_cancel;
2621 
2622 	xfs_defer_init(&dfops, &first_block);
2623 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2624 					&first_block, &dfops, resblks);
2625 	if (error) {
2626 		ASSERT(error != -ENOENT);
2627 		goto out_bmap_cancel;
2628 	}
2629 
2630 	/*
2631 	 * If this is a synchronous mount, make sure that the
2632 	 * remove transaction goes to disk before returning to
2633 	 * the user.
2634 	 */
2635 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2636 		xfs_trans_set_sync(tp);
2637 
2638 	error = xfs_defer_finish(&tp, &dfops, NULL);
2639 	if (error)
2640 		goto out_bmap_cancel;
2641 
2642 	error = xfs_trans_commit(tp);
2643 	if (error)
2644 		goto std_return;
2645 
2646 	if (is_dir && xfs_inode_is_filestream(ip))
2647 		xfs_filestream_deassociate(ip);
2648 
2649 	return 0;
2650 
2651  out_bmap_cancel:
2652 	xfs_defer_cancel(&dfops);
2653  out_trans_cancel:
2654 	xfs_trans_cancel(tp);
2655  std_return:
2656 	return error;
2657 }
2658 
2659 /*
2660  * Enter all inodes for a rename transaction into a sorted array.
2661  */
2662 #define __XFS_SORT_INODES	5
2663 STATIC void
2664 xfs_sort_for_rename(
2665 	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2666 	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2667 	struct xfs_inode	*ip1,	/* in: inode of old entry */
2668 	struct xfs_inode	*ip2,	/* in: inode of new entry */
2669 	struct xfs_inode	*wip,	/* in: whiteout inode */
2670 	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2671 	int			*num_inodes)  /* in/out: inodes in array */
2672 {
2673 	int			i, j;
2674 
2675 	ASSERT(*num_inodes == __XFS_SORT_INODES);
2676 	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2677 
2678 	/*
2679 	 * i_tab contains a list of pointers to inodes.  We initialize
2680 	 * the table here & we'll sort it.  We will then use it to
2681 	 * order the acquisition of the inode locks.
2682 	 *
2683 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2684 	 */
2685 	i = 0;
2686 	i_tab[i++] = dp1;
2687 	i_tab[i++] = dp2;
2688 	i_tab[i++] = ip1;
2689 	if (ip2)
2690 		i_tab[i++] = ip2;
2691 	if (wip)
2692 		i_tab[i++] = wip;
2693 	*num_inodes = i;
2694 
2695 	/*
2696 	 * Sort the elements via bubble sort.  (Remember, there are at
2697 	 * most 5 elements to sort, so this is adequate.)
2698 	 */
2699 	for (i = 0; i < *num_inodes; i++) {
2700 		for (j = 1; j < *num_inodes; j++) {
2701 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2702 				struct xfs_inode *temp = i_tab[j];
2703 				i_tab[j] = i_tab[j-1];
2704 				i_tab[j-1] = temp;
2705 			}
2706 		}
2707 	}
2708 }
2709 
2710 static int
2711 xfs_finish_rename(
2712 	struct xfs_trans	*tp,
2713 	struct xfs_defer_ops	*dfops)
2714 {
2715 	int			error;
2716 
2717 	/*
2718 	 * If this is a synchronous mount, make sure that the rename transaction
2719 	 * goes to disk before returning to the user.
2720 	 */
2721 	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2722 		xfs_trans_set_sync(tp);
2723 
2724 	error = xfs_defer_finish(&tp, dfops, NULL);
2725 	if (error) {
2726 		xfs_defer_cancel(dfops);
2727 		xfs_trans_cancel(tp);
2728 		return error;
2729 	}
2730 
2731 	return xfs_trans_commit(tp);
2732 }
2733 
2734 /*
2735  * xfs_cross_rename()
2736  *
2737  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2738  */
2739 STATIC int
2740 xfs_cross_rename(
2741 	struct xfs_trans	*tp,
2742 	struct xfs_inode	*dp1,
2743 	struct xfs_name		*name1,
2744 	struct xfs_inode	*ip1,
2745 	struct xfs_inode	*dp2,
2746 	struct xfs_name		*name2,
2747 	struct xfs_inode	*ip2,
2748 	struct xfs_defer_ops	*dfops,
2749 	xfs_fsblock_t		*first_block,
2750 	int			spaceres)
2751 {
2752 	int		error = 0;
2753 	int		ip1_flags = 0;
2754 	int		ip2_flags = 0;
2755 	int		dp2_flags = 0;
2756 
2757 	/* Swap inode number for dirent in first parent */
2758 	error = xfs_dir_replace(tp, dp1, name1,
2759 				ip2->i_ino,
2760 				first_block, dfops, spaceres);
2761 	if (error)
2762 		goto out_trans_abort;
2763 
2764 	/* Swap inode number for dirent in second parent */
2765 	error = xfs_dir_replace(tp, dp2, name2,
2766 				ip1->i_ino,
2767 				first_block, dfops, spaceres);
2768 	if (error)
2769 		goto out_trans_abort;
2770 
2771 	/*
2772 	 * If we're renaming one or more directories across different parents,
2773 	 * update the respective ".." entries (and link counts) to match the new
2774 	 * parents.
2775 	 */
2776 	if (dp1 != dp2) {
2777 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2778 
2779 		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2780 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2781 						dp1->i_ino, first_block,
2782 						dfops, spaceres);
2783 			if (error)
2784 				goto out_trans_abort;
2785 
2786 			/* transfer ip2 ".." reference to dp1 */
2787 			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2788 				error = xfs_droplink(tp, dp2);
2789 				if (error)
2790 					goto out_trans_abort;
2791 				error = xfs_bumplink(tp, dp1);
2792 				if (error)
2793 					goto out_trans_abort;
2794 			}
2795 
2796 			/*
2797 			 * Although ip1 isn't changed here, userspace needs
2798 			 * to be warned about the change, so that applications
2799 			 * relying on it (like backup ones), will properly
2800 			 * notify the change
2801 			 */
2802 			ip1_flags |= XFS_ICHGTIME_CHG;
2803 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2804 		}
2805 
2806 		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2807 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2808 						dp2->i_ino, first_block,
2809 						dfops, spaceres);
2810 			if (error)
2811 				goto out_trans_abort;
2812 
2813 			/* transfer ip1 ".." reference to dp2 */
2814 			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2815 				error = xfs_droplink(tp, dp1);
2816 				if (error)
2817 					goto out_trans_abort;
2818 				error = xfs_bumplink(tp, dp2);
2819 				if (error)
2820 					goto out_trans_abort;
2821 			}
2822 
2823 			/*
2824 			 * Although ip2 isn't changed here, userspace needs
2825 			 * to be warned about the change, so that applications
2826 			 * relying on it (like backup ones), will properly
2827 			 * notify the change
2828 			 */
2829 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2830 			ip2_flags |= XFS_ICHGTIME_CHG;
2831 		}
2832 	}
2833 
2834 	if (ip1_flags) {
2835 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2836 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2837 	}
2838 	if (ip2_flags) {
2839 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2840 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2841 	}
2842 	if (dp2_flags) {
2843 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2844 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2845 	}
2846 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2847 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2848 	return xfs_finish_rename(tp, dfops);
2849 
2850 out_trans_abort:
2851 	xfs_defer_cancel(dfops);
2852 	xfs_trans_cancel(tp);
2853 	return error;
2854 }
2855 
2856 /*
2857  * xfs_rename_alloc_whiteout()
2858  *
2859  * Return a referenced, unlinked, unlocked inode that that can be used as a
2860  * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2861  * crash between allocating the inode and linking it into the rename transaction
2862  * recovery will free the inode and we won't leak it.
2863  */
2864 static int
2865 xfs_rename_alloc_whiteout(
2866 	struct xfs_inode	*dp,
2867 	struct xfs_inode	**wip)
2868 {
2869 	struct xfs_inode	*tmpfile;
2870 	int			error;
2871 
2872 	error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2873 	if (error)
2874 		return error;
2875 
2876 	/*
2877 	 * Prepare the tmpfile inode as if it were created through the VFS.
2878 	 * Otherwise, the link increment paths will complain about nlink 0->1.
2879 	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2880 	 * and flag it as linkable.
2881 	 */
2882 	drop_nlink(VFS_I(tmpfile));
2883 	xfs_setup_iops(tmpfile);
2884 	xfs_finish_inode_setup(tmpfile);
2885 	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2886 
2887 	*wip = tmpfile;
2888 	return 0;
2889 }
2890 
2891 /*
2892  * xfs_rename
2893  */
2894 int
2895 xfs_rename(
2896 	struct xfs_inode	*src_dp,
2897 	struct xfs_name		*src_name,
2898 	struct xfs_inode	*src_ip,
2899 	struct xfs_inode	*target_dp,
2900 	struct xfs_name		*target_name,
2901 	struct xfs_inode	*target_ip,
2902 	unsigned int		flags)
2903 {
2904 	struct xfs_mount	*mp = src_dp->i_mount;
2905 	struct xfs_trans	*tp;
2906 	struct xfs_defer_ops	dfops;
2907 	xfs_fsblock_t		first_block;
2908 	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2909 	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2910 	int			num_inodes = __XFS_SORT_INODES;
2911 	bool			new_parent = (src_dp != target_dp);
2912 	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2913 	int			spaceres;
2914 	int			error;
2915 
2916 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2917 
2918 	if ((flags & RENAME_EXCHANGE) && !target_ip)
2919 		return -EINVAL;
2920 
2921 	/*
2922 	 * If we are doing a whiteout operation, allocate the whiteout inode
2923 	 * we will be placing at the target and ensure the type is set
2924 	 * appropriately.
2925 	 */
2926 	if (flags & RENAME_WHITEOUT) {
2927 		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2928 		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2929 		if (error)
2930 			return error;
2931 
2932 		/* setup target dirent info as whiteout */
2933 		src_name->type = XFS_DIR3_FT_CHRDEV;
2934 	}
2935 
2936 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2937 				inodes, &num_inodes);
2938 
2939 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2940 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2941 	if (error == -ENOSPC) {
2942 		spaceres = 0;
2943 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2944 				&tp);
2945 	}
2946 	if (error)
2947 		goto out_release_wip;
2948 
2949 	/*
2950 	 * Attach the dquots to the inodes
2951 	 */
2952 	error = xfs_qm_vop_rename_dqattach(inodes);
2953 	if (error)
2954 		goto out_trans_cancel;
2955 
2956 	/*
2957 	 * Lock all the participating inodes. Depending upon whether
2958 	 * the target_name exists in the target directory, and
2959 	 * whether the target directory is the same as the source
2960 	 * directory, we can lock from 2 to 4 inodes.
2961 	 */
2962 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2963 
2964 	/*
2965 	 * Join all the inodes to the transaction. From this point on,
2966 	 * we can rely on either trans_commit or trans_cancel to unlock
2967 	 * them.
2968 	 */
2969 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2970 	if (new_parent)
2971 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2972 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2973 	if (target_ip)
2974 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2975 	if (wip)
2976 		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2977 
2978 	/*
2979 	 * If we are using project inheritance, we only allow renames
2980 	 * into our tree when the project IDs are the same; else the
2981 	 * tree quota mechanism would be circumvented.
2982 	 */
2983 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2984 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2985 		error = -EXDEV;
2986 		goto out_trans_cancel;
2987 	}
2988 
2989 	xfs_defer_init(&dfops, &first_block);
2990 
2991 	/* RENAME_EXCHANGE is unique from here on. */
2992 	if (flags & RENAME_EXCHANGE)
2993 		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2994 					target_dp, target_name, target_ip,
2995 					&dfops, &first_block, spaceres);
2996 
2997 	/*
2998 	 * Set up the target.
2999 	 */
3000 	if (target_ip == NULL) {
3001 		/*
3002 		 * If there's no space reservation, check the entry will
3003 		 * fit before actually inserting it.
3004 		 */
3005 		if (!spaceres) {
3006 			error = xfs_dir_canenter(tp, target_dp, target_name);
3007 			if (error)
3008 				goto out_trans_cancel;
3009 		}
3010 		/*
3011 		 * If target does not exist and the rename crosses
3012 		 * directories, adjust the target directory link count
3013 		 * to account for the ".." reference from the new entry.
3014 		 */
3015 		error = xfs_dir_createname(tp, target_dp, target_name,
3016 						src_ip->i_ino, &first_block,
3017 						&dfops, spaceres);
3018 		if (error)
3019 			goto out_bmap_cancel;
3020 
3021 		xfs_trans_ichgtime(tp, target_dp,
3022 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3023 
3024 		if (new_parent && src_is_directory) {
3025 			error = xfs_bumplink(tp, target_dp);
3026 			if (error)
3027 				goto out_bmap_cancel;
3028 		}
3029 	} else { /* target_ip != NULL */
3030 		/*
3031 		 * If target exists and it's a directory, check that both
3032 		 * target and source are directories and that target can be
3033 		 * destroyed, or that neither is a directory.
3034 		 */
3035 		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3036 			/*
3037 			 * Make sure target dir is empty.
3038 			 */
3039 			if (!(xfs_dir_isempty(target_ip)) ||
3040 			    (VFS_I(target_ip)->i_nlink > 2)) {
3041 				error = -EEXIST;
3042 				goto out_trans_cancel;
3043 			}
3044 		}
3045 
3046 		/*
3047 		 * Link the source inode under the target name.
3048 		 * If the source inode is a directory and we are moving
3049 		 * it across directories, its ".." entry will be
3050 		 * inconsistent until we replace that down below.
3051 		 *
3052 		 * In case there is already an entry with the same
3053 		 * name at the destination directory, remove it first.
3054 		 */
3055 		error = xfs_dir_replace(tp, target_dp, target_name,
3056 					src_ip->i_ino,
3057 					&first_block, &dfops, spaceres);
3058 		if (error)
3059 			goto out_bmap_cancel;
3060 
3061 		xfs_trans_ichgtime(tp, target_dp,
3062 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3063 
3064 		/*
3065 		 * Decrement the link count on the target since the target
3066 		 * dir no longer points to it.
3067 		 */
3068 		error = xfs_droplink(tp, target_ip);
3069 		if (error)
3070 			goto out_bmap_cancel;
3071 
3072 		if (src_is_directory) {
3073 			/*
3074 			 * Drop the link from the old "." entry.
3075 			 */
3076 			error = xfs_droplink(tp, target_ip);
3077 			if (error)
3078 				goto out_bmap_cancel;
3079 		}
3080 	} /* target_ip != NULL */
3081 
3082 	/*
3083 	 * Remove the source.
3084 	 */
3085 	if (new_parent && src_is_directory) {
3086 		/*
3087 		 * Rewrite the ".." entry to point to the new
3088 		 * directory.
3089 		 */
3090 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3091 					target_dp->i_ino,
3092 					&first_block, &dfops, spaceres);
3093 		ASSERT(error != -EEXIST);
3094 		if (error)
3095 			goto out_bmap_cancel;
3096 	}
3097 
3098 	/*
3099 	 * We always want to hit the ctime on the source inode.
3100 	 *
3101 	 * This isn't strictly required by the standards since the source
3102 	 * inode isn't really being changed, but old unix file systems did
3103 	 * it and some incremental backup programs won't work without it.
3104 	 */
3105 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3106 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3107 
3108 	/*
3109 	 * Adjust the link count on src_dp.  This is necessary when
3110 	 * renaming a directory, either within one parent when
3111 	 * the target existed, or across two parent directories.
3112 	 */
3113 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3114 
3115 		/*
3116 		 * Decrement link count on src_directory since the
3117 		 * entry that's moved no longer points to it.
3118 		 */
3119 		error = xfs_droplink(tp, src_dp);
3120 		if (error)
3121 			goto out_bmap_cancel;
3122 	}
3123 
3124 	/*
3125 	 * For whiteouts, we only need to update the source dirent with the
3126 	 * inode number of the whiteout inode rather than removing it
3127 	 * altogether.
3128 	 */
3129 	if (wip) {
3130 		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3131 					&first_block, &dfops, spaceres);
3132 	} else
3133 		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3134 					   &first_block, &dfops, spaceres);
3135 	if (error)
3136 		goto out_bmap_cancel;
3137 
3138 	/*
3139 	 * For whiteouts, we need to bump the link count on the whiteout inode.
3140 	 * This means that failures all the way up to this point leave the inode
3141 	 * on the unlinked list and so cleanup is a simple matter of dropping
3142 	 * the remaining reference to it. If we fail here after bumping the link
3143 	 * count, we're shutting down the filesystem so we'll never see the
3144 	 * intermediate state on disk.
3145 	 */
3146 	if (wip) {
3147 		ASSERT(VFS_I(wip)->i_nlink == 0);
3148 		error = xfs_bumplink(tp, wip);
3149 		if (error)
3150 			goto out_bmap_cancel;
3151 		error = xfs_iunlink_remove(tp, wip);
3152 		if (error)
3153 			goto out_bmap_cancel;
3154 		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3155 
3156 		/*
3157 		 * Now we have a real link, clear the "I'm a tmpfile" state
3158 		 * flag from the inode so it doesn't accidentally get misused in
3159 		 * future.
3160 		 */
3161 		VFS_I(wip)->i_state &= ~I_LINKABLE;
3162 	}
3163 
3164 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3165 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3166 	if (new_parent)
3167 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3168 
3169 	error = xfs_finish_rename(tp, &dfops);
3170 	if (wip)
3171 		IRELE(wip);
3172 	return error;
3173 
3174 out_bmap_cancel:
3175 	xfs_defer_cancel(&dfops);
3176 out_trans_cancel:
3177 	xfs_trans_cancel(tp);
3178 out_release_wip:
3179 	if (wip)
3180 		IRELE(wip);
3181 	return error;
3182 }
3183 
3184 STATIC int
3185 xfs_iflush_cluster(
3186 	struct xfs_inode	*ip,
3187 	struct xfs_buf		*bp)
3188 {
3189 	struct xfs_mount	*mp = ip->i_mount;
3190 	struct xfs_perag	*pag;
3191 	unsigned long		first_index, mask;
3192 	unsigned long		inodes_per_cluster;
3193 	int			cilist_size;
3194 	struct xfs_inode	**cilist;
3195 	struct xfs_inode	*cip;
3196 	int			nr_found;
3197 	int			clcount = 0;
3198 	int			bufwasdelwri;
3199 	int			i;
3200 
3201 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3202 
3203 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3204 	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3205 	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3206 	if (!cilist)
3207 		goto out_put;
3208 
3209 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3210 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3211 	rcu_read_lock();
3212 	/* really need a gang lookup range call here */
3213 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3214 					first_index, inodes_per_cluster);
3215 	if (nr_found == 0)
3216 		goto out_free;
3217 
3218 	for (i = 0; i < nr_found; i++) {
3219 		cip = cilist[i];
3220 		if (cip == ip)
3221 			continue;
3222 
3223 		/*
3224 		 * because this is an RCU protected lookup, we could find a
3225 		 * recently freed or even reallocated inode during the lookup.
3226 		 * We need to check under the i_flags_lock for a valid inode
3227 		 * here. Skip it if it is not valid or the wrong inode.
3228 		 */
3229 		spin_lock(&cip->i_flags_lock);
3230 		if (!cip->i_ino ||
3231 		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3232 			spin_unlock(&cip->i_flags_lock);
3233 			continue;
3234 		}
3235 
3236 		/*
3237 		 * Once we fall off the end of the cluster, no point checking
3238 		 * any more inodes in the list because they will also all be
3239 		 * outside the cluster.
3240 		 */
3241 		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3242 			spin_unlock(&cip->i_flags_lock);
3243 			break;
3244 		}
3245 		spin_unlock(&cip->i_flags_lock);
3246 
3247 		/*
3248 		 * Do an un-protected check to see if the inode is dirty and
3249 		 * is a candidate for flushing.  These checks will be repeated
3250 		 * later after the appropriate locks are acquired.
3251 		 */
3252 		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3253 			continue;
3254 
3255 		/*
3256 		 * Try to get locks.  If any are unavailable or it is pinned,
3257 		 * then this inode cannot be flushed and is skipped.
3258 		 */
3259 
3260 		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3261 			continue;
3262 		if (!xfs_iflock_nowait(cip)) {
3263 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3264 			continue;
3265 		}
3266 		if (xfs_ipincount(cip)) {
3267 			xfs_ifunlock(cip);
3268 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3269 			continue;
3270 		}
3271 
3272 
3273 		/*
3274 		 * Check the inode number again, just to be certain we are not
3275 		 * racing with freeing in xfs_reclaim_inode(). See the comments
3276 		 * in that function for more information as to why the initial
3277 		 * check is not sufficient.
3278 		 */
3279 		if (!cip->i_ino) {
3280 			xfs_ifunlock(cip);
3281 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3282 			continue;
3283 		}
3284 
3285 		/*
3286 		 * arriving here means that this inode can be flushed.  First
3287 		 * re-check that it's dirty before flushing.
3288 		 */
3289 		if (!xfs_inode_clean(cip)) {
3290 			int	error;
3291 			error = xfs_iflush_int(cip, bp);
3292 			if (error) {
3293 				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3294 				goto cluster_corrupt_out;
3295 			}
3296 			clcount++;
3297 		} else {
3298 			xfs_ifunlock(cip);
3299 		}
3300 		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3301 	}
3302 
3303 	if (clcount) {
3304 		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3305 		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3306 	}
3307 
3308 out_free:
3309 	rcu_read_unlock();
3310 	kmem_free(cilist);
3311 out_put:
3312 	xfs_perag_put(pag);
3313 	return 0;
3314 
3315 
3316 cluster_corrupt_out:
3317 	/*
3318 	 * Corruption detected in the clustering loop.  Invalidate the
3319 	 * inode buffer and shut down the filesystem.
3320 	 */
3321 	rcu_read_unlock();
3322 	/*
3323 	 * Clean up the buffer.  If it was delwri, just release it --
3324 	 * brelse can handle it with no problems.  If not, shut down the
3325 	 * filesystem before releasing the buffer.
3326 	 */
3327 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3328 	if (bufwasdelwri)
3329 		xfs_buf_relse(bp);
3330 
3331 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3332 
3333 	if (!bufwasdelwri) {
3334 		/*
3335 		 * Just like incore_relse: if we have b_iodone functions,
3336 		 * mark the buffer as an error and call them.  Otherwise
3337 		 * mark it as stale and brelse.
3338 		 */
3339 		if (bp->b_iodone) {
3340 			bp->b_flags &= ~XBF_DONE;
3341 			xfs_buf_stale(bp);
3342 			xfs_buf_ioerror(bp, -EIO);
3343 			xfs_buf_ioend(bp);
3344 		} else {
3345 			xfs_buf_stale(bp);
3346 			xfs_buf_relse(bp);
3347 		}
3348 	}
3349 
3350 	/*
3351 	 * Unlocks the flush lock
3352 	 */
3353 	xfs_iflush_abort(cip, false);
3354 	kmem_free(cilist);
3355 	xfs_perag_put(pag);
3356 	return -EFSCORRUPTED;
3357 }
3358 
3359 /*
3360  * Flush dirty inode metadata into the backing buffer.
3361  *
3362  * The caller must have the inode lock and the inode flush lock held.  The
3363  * inode lock will still be held upon return to the caller, and the inode
3364  * flush lock will be released after the inode has reached the disk.
3365  *
3366  * The caller must write out the buffer returned in *bpp and release it.
3367  */
3368 int
3369 xfs_iflush(
3370 	struct xfs_inode	*ip,
3371 	struct xfs_buf		**bpp)
3372 {
3373 	struct xfs_mount	*mp = ip->i_mount;
3374 	struct xfs_buf		*bp = NULL;
3375 	struct xfs_dinode	*dip;
3376 	int			error;
3377 
3378 	XFS_STATS_INC(mp, xs_iflush_count);
3379 
3380 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3381 	ASSERT(xfs_isiflocked(ip));
3382 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3383 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3384 
3385 	*bpp = NULL;
3386 
3387 	xfs_iunpin_wait(ip);
3388 
3389 	/*
3390 	 * For stale inodes we cannot rely on the backing buffer remaining
3391 	 * stale in cache for the remaining life of the stale inode and so
3392 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3393 	 * inodes below. We have to check this after ensuring the inode is
3394 	 * unpinned so that it is safe to reclaim the stale inode after the
3395 	 * flush call.
3396 	 */
3397 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3398 		xfs_ifunlock(ip);
3399 		return 0;
3400 	}
3401 
3402 	/*
3403 	 * This may have been unpinned because the filesystem is shutting
3404 	 * down forcibly. If that's the case we must not write this inode
3405 	 * to disk, because the log record didn't make it to disk.
3406 	 *
3407 	 * We also have to remove the log item from the AIL in this case,
3408 	 * as we wait for an empty AIL as part of the unmount process.
3409 	 */
3410 	if (XFS_FORCED_SHUTDOWN(mp)) {
3411 		error = -EIO;
3412 		goto abort_out;
3413 	}
3414 
3415 	/*
3416 	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3417 	 * operation here, so we may get  an EAGAIN error. In that case, we
3418 	 * simply want to return with the inode still dirty.
3419 	 *
3420 	 * If we get any other error, we effectively have a corruption situation
3421 	 * and we cannot flush the inode, so we treat it the same as failing
3422 	 * xfs_iflush_int().
3423 	 */
3424 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3425 			       0);
3426 	if (error == -EAGAIN) {
3427 		xfs_ifunlock(ip);
3428 		return error;
3429 	}
3430 	if (error)
3431 		goto corrupt_out;
3432 
3433 	/*
3434 	 * First flush out the inode that xfs_iflush was called with.
3435 	 */
3436 	error = xfs_iflush_int(ip, bp);
3437 	if (error)
3438 		goto corrupt_out;
3439 
3440 	/*
3441 	 * If the buffer is pinned then push on the log now so we won't
3442 	 * get stuck waiting in the write for too long.
3443 	 */
3444 	if (xfs_buf_ispinned(bp))
3445 		xfs_log_force(mp, 0);
3446 
3447 	/*
3448 	 * inode clustering:
3449 	 * see if other inodes can be gathered into this write
3450 	 */
3451 	error = xfs_iflush_cluster(ip, bp);
3452 	if (error)
3453 		goto cluster_corrupt_out;
3454 
3455 	*bpp = bp;
3456 	return 0;
3457 
3458 corrupt_out:
3459 	if (bp)
3460 		xfs_buf_relse(bp);
3461 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3462 cluster_corrupt_out:
3463 	error = -EFSCORRUPTED;
3464 abort_out:
3465 	/*
3466 	 * Unlocks the flush lock
3467 	 */
3468 	xfs_iflush_abort(ip, false);
3469 	return error;
3470 }
3471 
3472 STATIC int
3473 xfs_iflush_int(
3474 	struct xfs_inode	*ip,
3475 	struct xfs_buf		*bp)
3476 {
3477 	struct xfs_inode_log_item *iip = ip->i_itemp;
3478 	struct xfs_dinode	*dip;
3479 	struct xfs_mount	*mp = ip->i_mount;
3480 
3481 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3482 	ASSERT(xfs_isiflocked(ip));
3483 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3484 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3485 	ASSERT(iip != NULL && iip->ili_fields != 0);
3486 	ASSERT(ip->i_d.di_version > 1);
3487 
3488 	/* set *dip = inode's place in the buffer */
3489 	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3490 
3491 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3492 			       mp, XFS_ERRTAG_IFLUSH_1)) {
3493 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3494 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3495 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3496 		goto corrupt_out;
3497 	}
3498 	if (S_ISREG(VFS_I(ip)->i_mode)) {
3499 		if (XFS_TEST_ERROR(
3500 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3501 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3502 		    mp, XFS_ERRTAG_IFLUSH_3)) {
3503 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3504 				"%s: Bad regular inode %Lu, ptr 0x%p",
3505 				__func__, ip->i_ino, ip);
3506 			goto corrupt_out;
3507 		}
3508 	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3509 		if (XFS_TEST_ERROR(
3510 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3511 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3512 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3513 		    mp, XFS_ERRTAG_IFLUSH_4)) {
3514 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3515 				"%s: Bad directory inode %Lu, ptr 0x%p",
3516 				__func__, ip->i_ino, ip);
3517 			goto corrupt_out;
3518 		}
3519 	}
3520 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3521 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3522 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3523 			"%s: detected corrupt incore inode %Lu, "
3524 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3525 			__func__, ip->i_ino,
3526 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3527 			ip->i_d.di_nblocks, ip);
3528 		goto corrupt_out;
3529 	}
3530 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3531 				mp, XFS_ERRTAG_IFLUSH_6)) {
3532 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3533 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3534 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3535 		goto corrupt_out;
3536 	}
3537 
3538 	/*
3539 	 * Inode item log recovery for v2 inodes are dependent on the
3540 	 * di_flushiter count for correct sequencing. We bump the flush
3541 	 * iteration count so we can detect flushes which postdate a log record
3542 	 * during recovery. This is redundant as we now log every change and
3543 	 * hence this can't happen but we need to still do it to ensure
3544 	 * backwards compatibility with old kernels that predate logging all
3545 	 * inode changes.
3546 	 */
3547 	if (ip->i_d.di_version < 3)
3548 		ip->i_d.di_flushiter++;
3549 
3550 	/* Check the inline directory data. */
3551 	if (S_ISDIR(VFS_I(ip)->i_mode) &&
3552 	    ip->i_d.di_format == XFS_DINODE_FMT_LOCAL &&
3553 	    xfs_dir2_sf_verify(ip))
3554 		goto corrupt_out;
3555 
3556 	/*
3557 	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3558 	 * copy out the core of the inode, because if the inode is dirty at all
3559 	 * the core must be.
3560 	 */
3561 	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3562 
3563 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3564 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3565 		ip->i_d.di_flushiter = 0;
3566 
3567 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3568 	if (XFS_IFORK_Q(ip))
3569 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3570 	xfs_inobp_check(mp, bp);
3571 
3572 	/*
3573 	 * We've recorded everything logged in the inode, so we'd like to clear
3574 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3575 	 * However, we can't stop logging all this information until the data
3576 	 * we've copied into the disk buffer is written to disk.  If we did we
3577 	 * might overwrite the copy of the inode in the log with all the data
3578 	 * after re-logging only part of it, and in the face of a crash we
3579 	 * wouldn't have all the data we need to recover.
3580 	 *
3581 	 * What we do is move the bits to the ili_last_fields field.  When
3582 	 * logging the inode, these bits are moved back to the ili_fields field.
3583 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3584 	 * know that the information those bits represent is permanently on
3585 	 * disk.  As long as the flush completes before the inode is logged
3586 	 * again, then both ili_fields and ili_last_fields will be cleared.
3587 	 *
3588 	 * We can play with the ili_fields bits here, because the inode lock
3589 	 * must be held exclusively in order to set bits there and the flush
3590 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3591 	 * done routine can tell whether or not to look in the AIL.  Also, store
3592 	 * the current LSN of the inode so that we can tell whether the item has
3593 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3594 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3595 	 * atomically.
3596 	 */
3597 	iip->ili_last_fields = iip->ili_fields;
3598 	iip->ili_fields = 0;
3599 	iip->ili_fsync_fields = 0;
3600 	iip->ili_logged = 1;
3601 
3602 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3603 				&iip->ili_item.li_lsn);
3604 
3605 	/*
3606 	 * Attach the function xfs_iflush_done to the inode's
3607 	 * buffer.  This will remove the inode from the AIL
3608 	 * and unlock the inode's flush lock when the inode is
3609 	 * completely written to disk.
3610 	 */
3611 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3612 
3613 	/* generate the checksum. */
3614 	xfs_dinode_calc_crc(mp, dip);
3615 
3616 	ASSERT(bp->b_fspriv != NULL);
3617 	ASSERT(bp->b_iodone != NULL);
3618 	return 0;
3619 
3620 corrupt_out:
3621 	return -EFSCORRUPTED;
3622 }
3623