xref: /openbmc/linux/fs/xfs/xfs_inode.c (revision 2eb0f624b709e78ec8e2f4c3412947703db99301)
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 #include <linux/iversion.h>
20 
21 #include "xfs.h"
22 #include "xfs_fs.h"
23 #include "xfs_shared.h"
24 #include "xfs_format.h"
25 #include "xfs_log_format.h"
26 #include "xfs_trans_resv.h"
27 #include "xfs_sb.h"
28 #include "xfs_mount.h"
29 #include "xfs_defer.h"
30 #include "xfs_inode.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_dir2.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_attr.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_bmap.h"
42 #include "xfs_bmap_util.h"
43 #include "xfs_errortag.h"
44 #include "xfs_error.h"
45 #include "xfs_quota.h"
46 #include "xfs_filestream.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
50 #include "xfs_symlink.h"
51 #include "xfs_trans_priv.h"
52 #include "xfs_log.h"
53 #include "xfs_bmap_btree.h"
54 #include "xfs_reflink.h"
55 #include "xfs_dir2_priv.h"
56 
57 kmem_zone_t *xfs_inode_zone;
58 
59 /*
60  * Used in xfs_itruncate_extents().  This is the maximum number of extents
61  * freed from a file in a single transaction.
62  */
63 #define	XFS_ITRUNC_MAX_EXTENTS	2
64 
65 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
66 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
67 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
68 
69 /*
70  * helper function to extract extent size hint from inode
71  */
72 xfs_extlen_t
73 xfs_get_extsz_hint(
74 	struct xfs_inode	*ip)
75 {
76 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
77 		return ip->i_d.di_extsize;
78 	if (XFS_IS_REALTIME_INODE(ip))
79 		return ip->i_mount->m_sb.sb_rextsize;
80 	return 0;
81 }
82 
83 /*
84  * Helper function to extract CoW extent size hint from inode.
85  * Between the extent size hint and the CoW extent size hint, we
86  * return the greater of the two.  If the value is zero (automatic),
87  * use the default size.
88  */
89 xfs_extlen_t
90 xfs_get_cowextsz_hint(
91 	struct xfs_inode	*ip)
92 {
93 	xfs_extlen_t		a, b;
94 
95 	a = 0;
96 	if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
97 		a = ip->i_d.di_cowextsize;
98 	b = xfs_get_extsz_hint(ip);
99 
100 	a = max(a, b);
101 	if (a == 0)
102 		return XFS_DEFAULT_COWEXTSZ_HINT;
103 	return a;
104 }
105 
106 /*
107  * These two are wrapper routines around the xfs_ilock() routine used to
108  * centralize some grungy code.  They are used in places that wish to lock the
109  * inode solely for reading the extents.  The reason these places can't just
110  * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
111  * bringing in of the extents from disk for a file in b-tree format.  If the
112  * inode is in b-tree format, then we need to lock the inode exclusively until
113  * the extents are read in.  Locking it exclusively all the time would limit
114  * our parallelism unnecessarily, though.  What we do instead is check to see
115  * if the extents have been read in yet, and only lock the inode exclusively
116  * if they have not.
117  *
118  * The functions return a value which should be given to the corresponding
119  * xfs_iunlock() call.
120  */
121 uint
122 xfs_ilock_data_map_shared(
123 	struct xfs_inode	*ip)
124 {
125 	uint			lock_mode = XFS_ILOCK_SHARED;
126 
127 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
128 	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
129 		lock_mode = XFS_ILOCK_EXCL;
130 	xfs_ilock(ip, lock_mode);
131 	return lock_mode;
132 }
133 
134 uint
135 xfs_ilock_attr_map_shared(
136 	struct xfs_inode	*ip)
137 {
138 	uint			lock_mode = XFS_ILOCK_SHARED;
139 
140 	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
141 	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
142 		lock_mode = XFS_ILOCK_EXCL;
143 	xfs_ilock(ip, lock_mode);
144 	return lock_mode;
145 }
146 
147 /*
148  * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
149  * multi-reader locks: i_mmap_lock and the i_lock.  This routine allows
150  * various combinations of the locks to be obtained.
151  *
152  * The 3 locks should always be ordered so that the IO lock is obtained first,
153  * the mmap lock second and the ilock last in order to prevent deadlock.
154  *
155  * Basic locking order:
156  *
157  * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
158  *
159  * mmap_sem locking order:
160  *
161  * i_rwsem -> page lock -> mmap_sem
162  * mmap_sem -> i_mmap_lock -> page_lock
163  *
164  * The difference in mmap_sem locking order mean that we cannot hold the
165  * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
166  * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
167  * in get_user_pages() to map the user pages into the kernel address space for
168  * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
169  * page faults already hold the mmap_sem.
170  *
171  * Hence to serialise fully against both syscall and mmap based IO, we need to
172  * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
173  * taken in places where we need to invalidate the page cache in a race
174  * free manner (e.g. truncate, hole punch and other extent manipulation
175  * functions).
176  */
177 void
178 xfs_ilock(
179 	xfs_inode_t		*ip,
180 	uint			lock_flags)
181 {
182 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
183 
184 	/*
185 	 * You can't set both SHARED and EXCL for the same lock,
186 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
187 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
188 	 */
189 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
190 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
191 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
192 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
193 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
194 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
195 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
196 
197 	if (lock_flags & XFS_IOLOCK_EXCL) {
198 		down_write_nested(&VFS_I(ip)->i_rwsem,
199 				  XFS_IOLOCK_DEP(lock_flags));
200 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
201 		down_read_nested(&VFS_I(ip)->i_rwsem,
202 				 XFS_IOLOCK_DEP(lock_flags));
203 	}
204 
205 	if (lock_flags & XFS_MMAPLOCK_EXCL)
206 		mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
207 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
208 		mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
209 
210 	if (lock_flags & XFS_ILOCK_EXCL)
211 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 	else if (lock_flags & XFS_ILOCK_SHARED)
213 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
214 }
215 
216 /*
217  * This is just like xfs_ilock(), except that the caller
218  * is guaranteed not to sleep.  It returns 1 if it gets
219  * the requested locks and 0 otherwise.  If the IO lock is
220  * obtained but the inode lock cannot be, then the IO lock
221  * is dropped before returning.
222  *
223  * ip -- the inode being locked
224  * lock_flags -- this parameter indicates the inode's locks to be
225  *       to be locked.  See the comment for xfs_ilock() for a list
226  *	 of valid values.
227  */
228 int
229 xfs_ilock_nowait(
230 	xfs_inode_t		*ip,
231 	uint			lock_flags)
232 {
233 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
234 
235 	/*
236 	 * You can't set both SHARED and EXCL for the same lock,
237 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
238 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
239 	 */
240 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
241 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
242 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
243 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
244 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
245 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
246 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
247 
248 	if (lock_flags & XFS_IOLOCK_EXCL) {
249 		if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
250 			goto out;
251 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
252 		if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
253 			goto out;
254 	}
255 
256 	if (lock_flags & XFS_MMAPLOCK_EXCL) {
257 		if (!mrtryupdate(&ip->i_mmaplock))
258 			goto out_undo_iolock;
259 	} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
260 		if (!mrtryaccess(&ip->i_mmaplock))
261 			goto out_undo_iolock;
262 	}
263 
264 	if (lock_flags & XFS_ILOCK_EXCL) {
265 		if (!mrtryupdate(&ip->i_lock))
266 			goto out_undo_mmaplock;
267 	} else if (lock_flags & XFS_ILOCK_SHARED) {
268 		if (!mrtryaccess(&ip->i_lock))
269 			goto out_undo_mmaplock;
270 	}
271 	return 1;
272 
273 out_undo_mmaplock:
274 	if (lock_flags & XFS_MMAPLOCK_EXCL)
275 		mrunlock_excl(&ip->i_mmaplock);
276 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
277 		mrunlock_shared(&ip->i_mmaplock);
278 out_undo_iolock:
279 	if (lock_flags & XFS_IOLOCK_EXCL)
280 		up_write(&VFS_I(ip)->i_rwsem);
281 	else if (lock_flags & XFS_IOLOCK_SHARED)
282 		up_read(&VFS_I(ip)->i_rwsem);
283 out:
284 	return 0;
285 }
286 
287 /*
288  * xfs_iunlock() is used to drop the inode locks acquired with
289  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
290  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
291  * that we know which locks to drop.
292  *
293  * ip -- the inode being unlocked
294  * lock_flags -- this parameter indicates the inode's locks to be
295  *       to be unlocked.  See the comment for xfs_ilock() for a list
296  *	 of valid values for this parameter.
297  *
298  */
299 void
300 xfs_iunlock(
301 	xfs_inode_t		*ip,
302 	uint			lock_flags)
303 {
304 	/*
305 	 * You can't set both SHARED and EXCL for the same lock,
306 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
307 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
308 	 */
309 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
310 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
311 	ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
312 	       (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
313 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
314 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
315 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
316 	ASSERT(lock_flags != 0);
317 
318 	if (lock_flags & XFS_IOLOCK_EXCL)
319 		up_write(&VFS_I(ip)->i_rwsem);
320 	else if (lock_flags & XFS_IOLOCK_SHARED)
321 		up_read(&VFS_I(ip)->i_rwsem);
322 
323 	if (lock_flags & XFS_MMAPLOCK_EXCL)
324 		mrunlock_excl(&ip->i_mmaplock);
325 	else if (lock_flags & XFS_MMAPLOCK_SHARED)
326 		mrunlock_shared(&ip->i_mmaplock);
327 
328 	if (lock_flags & XFS_ILOCK_EXCL)
329 		mrunlock_excl(&ip->i_lock);
330 	else if (lock_flags & XFS_ILOCK_SHARED)
331 		mrunlock_shared(&ip->i_lock);
332 
333 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
334 }
335 
336 /*
337  * give up write locks.  the i/o lock cannot be held nested
338  * if it is being demoted.
339  */
340 void
341 xfs_ilock_demote(
342 	xfs_inode_t		*ip,
343 	uint			lock_flags)
344 {
345 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
346 	ASSERT((lock_flags &
347 		~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
348 
349 	if (lock_flags & XFS_ILOCK_EXCL)
350 		mrdemote(&ip->i_lock);
351 	if (lock_flags & XFS_MMAPLOCK_EXCL)
352 		mrdemote(&ip->i_mmaplock);
353 	if (lock_flags & XFS_IOLOCK_EXCL)
354 		downgrade_write(&VFS_I(ip)->i_rwsem);
355 
356 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
357 }
358 
359 #if defined(DEBUG) || defined(XFS_WARN)
360 int
361 xfs_isilocked(
362 	xfs_inode_t		*ip,
363 	uint			lock_flags)
364 {
365 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
366 		if (!(lock_flags & XFS_ILOCK_SHARED))
367 			return !!ip->i_lock.mr_writer;
368 		return rwsem_is_locked(&ip->i_lock.mr_lock);
369 	}
370 
371 	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
372 		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
373 			return !!ip->i_mmaplock.mr_writer;
374 		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
375 	}
376 
377 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
378 		if (!(lock_flags & XFS_IOLOCK_SHARED))
379 			return !debug_locks ||
380 				lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
381 		return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
382 	}
383 
384 	ASSERT(0);
385 	return 0;
386 }
387 #endif
388 
389 /*
390  * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
391  * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
392  * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
393  * errors and warnings.
394  */
395 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
396 static bool
397 xfs_lockdep_subclass_ok(
398 	int subclass)
399 {
400 	return subclass < MAX_LOCKDEP_SUBCLASSES;
401 }
402 #else
403 #define xfs_lockdep_subclass_ok(subclass)	(true)
404 #endif
405 
406 /*
407  * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
408  * value. This can be called for any type of inode lock combination, including
409  * parent locking. Care must be taken to ensure we don't overrun the subclass
410  * storage fields in the class mask we build.
411  */
412 static inline int
413 xfs_lock_inumorder(int lock_mode, int subclass)
414 {
415 	int	class = 0;
416 
417 	ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
418 			      XFS_ILOCK_RTSUM)));
419 	ASSERT(xfs_lockdep_subclass_ok(subclass));
420 
421 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
422 		ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
423 		class += subclass << XFS_IOLOCK_SHIFT;
424 	}
425 
426 	if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
427 		ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
428 		class += subclass << XFS_MMAPLOCK_SHIFT;
429 	}
430 
431 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
432 		ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
433 		class += subclass << XFS_ILOCK_SHIFT;
434 	}
435 
436 	return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
437 }
438 
439 /*
440  * The following routine will lock n inodes in exclusive mode.  We assume the
441  * caller calls us with the inodes in i_ino order.
442  *
443  * We need to detect deadlock where an inode that we lock is in the AIL and we
444  * start waiting for another inode that is locked by a thread in a long running
445  * transaction (such as truncate). This can result in deadlock since the long
446  * running trans might need to wait for the inode we just locked in order to
447  * push the tail and free space in the log.
448  *
449  * xfs_lock_inodes() can only be used to lock one type of lock at a time -
450  * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
451  * lock more than one at a time, lockdep will report false positives saying we
452  * have violated locking orders.
453  */
454 static void
455 xfs_lock_inodes(
456 	xfs_inode_t	**ips,
457 	int		inodes,
458 	uint		lock_mode)
459 {
460 	int		attempts = 0, i, j, try_lock;
461 	xfs_log_item_t	*lp;
462 
463 	/*
464 	 * Currently supports between 2 and 5 inodes with exclusive locking.  We
465 	 * support an arbitrary depth of locking here, but absolute limits on
466 	 * inodes depend on the the type of locking and the limits placed by
467 	 * lockdep annotations in xfs_lock_inumorder.  These are all checked by
468 	 * the asserts.
469 	 */
470 	ASSERT(ips && inodes >= 2 && inodes <= 5);
471 	ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
472 			    XFS_ILOCK_EXCL));
473 	ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
474 			      XFS_ILOCK_SHARED)));
475 	ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
476 		inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
477 	ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
478 		inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
479 
480 	if (lock_mode & XFS_IOLOCK_EXCL) {
481 		ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
482 	} else if (lock_mode & XFS_MMAPLOCK_EXCL)
483 		ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
484 
485 	try_lock = 0;
486 	i = 0;
487 again:
488 	for (; i < inodes; i++) {
489 		ASSERT(ips[i]);
490 
491 		if (i && (ips[i] == ips[i - 1]))	/* Already locked */
492 			continue;
493 
494 		/*
495 		 * If try_lock is not set yet, make sure all locked inodes are
496 		 * not in the AIL.  If any are, set try_lock to be used later.
497 		 */
498 		if (!try_lock) {
499 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
500 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
501 				if (lp && (lp->li_flags & XFS_LI_IN_AIL))
502 					try_lock++;
503 			}
504 		}
505 
506 		/*
507 		 * If any of the previous locks we have locked is in the AIL,
508 		 * we must TRY to get the second and subsequent locks. If
509 		 * we can't get any, we must release all we have
510 		 * and try again.
511 		 */
512 		if (!try_lock) {
513 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
514 			continue;
515 		}
516 
517 		/* try_lock means we have an inode locked that is in the AIL. */
518 		ASSERT(i != 0);
519 		if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
520 			continue;
521 
522 		/*
523 		 * Unlock all previous guys and try again.  xfs_iunlock will try
524 		 * to push the tail if the inode is in the AIL.
525 		 */
526 		attempts++;
527 		for (j = i - 1; j >= 0; j--) {
528 			/*
529 			 * Check to see if we've already unlocked this one.  Not
530 			 * the first one going back, and the inode ptr is the
531 			 * same.
532 			 */
533 			if (j != (i - 1) && ips[j] == ips[j + 1])
534 				continue;
535 
536 			xfs_iunlock(ips[j], lock_mode);
537 		}
538 
539 		if ((attempts % 5) == 0) {
540 			delay(1); /* Don't just spin the CPU */
541 		}
542 		i = 0;
543 		try_lock = 0;
544 		goto again;
545 	}
546 }
547 
548 /*
549  * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
550  * the mmaplock or the ilock, but not more than one type at a time. If we lock
551  * more than one at a time, lockdep will report false positives saying we have
552  * violated locking orders.  The iolock must be double-locked separately since
553  * we use i_rwsem for that.  We now support taking one lock EXCL and the other
554  * SHARED.
555  */
556 void
557 xfs_lock_two_inodes(
558 	struct xfs_inode	*ip0,
559 	uint			ip0_mode,
560 	struct xfs_inode	*ip1,
561 	uint			ip1_mode)
562 {
563 	struct xfs_inode	*temp;
564 	uint			mode_temp;
565 	int			attempts = 0;
566 	xfs_log_item_t		*lp;
567 
568 	ASSERT(hweight32(ip0_mode) == 1);
569 	ASSERT(hweight32(ip1_mode) == 1);
570 	ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
571 	ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
572 	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
573 	       !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
574 	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
575 	       !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
576 	ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
577 	       !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
578 	ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
579 	       !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
580 
581 	ASSERT(ip0->i_ino != ip1->i_ino);
582 
583 	if (ip0->i_ino > ip1->i_ino) {
584 		temp = ip0;
585 		ip0 = ip1;
586 		ip1 = temp;
587 		mode_temp = ip0_mode;
588 		ip0_mode = ip1_mode;
589 		ip1_mode = mode_temp;
590 	}
591 
592  again:
593 	xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
594 
595 	/*
596 	 * If the first lock we have locked is in the AIL, we must TRY to get
597 	 * the second lock. If we can't get it, we must release the first one
598 	 * and try again.
599 	 */
600 	lp = (xfs_log_item_t *)ip0->i_itemp;
601 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
602 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
603 			xfs_iunlock(ip0, ip0_mode);
604 			if ((++attempts % 5) == 0)
605 				delay(1); /* Don't just spin the CPU */
606 			goto again;
607 		}
608 	} else {
609 		xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
610 	}
611 }
612 
613 void
614 __xfs_iflock(
615 	struct xfs_inode	*ip)
616 {
617 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
618 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
619 
620 	do {
621 		prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
622 		if (xfs_isiflocked(ip))
623 			io_schedule();
624 	} while (!xfs_iflock_nowait(ip));
625 
626 	finish_wait(wq, &wait.wq_entry);
627 }
628 
629 STATIC uint
630 _xfs_dic2xflags(
631 	uint16_t		di_flags,
632 	uint64_t		di_flags2,
633 	bool			has_attr)
634 {
635 	uint			flags = 0;
636 
637 	if (di_flags & XFS_DIFLAG_ANY) {
638 		if (di_flags & XFS_DIFLAG_REALTIME)
639 			flags |= FS_XFLAG_REALTIME;
640 		if (di_flags & XFS_DIFLAG_PREALLOC)
641 			flags |= FS_XFLAG_PREALLOC;
642 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
643 			flags |= FS_XFLAG_IMMUTABLE;
644 		if (di_flags & XFS_DIFLAG_APPEND)
645 			flags |= FS_XFLAG_APPEND;
646 		if (di_flags & XFS_DIFLAG_SYNC)
647 			flags |= FS_XFLAG_SYNC;
648 		if (di_flags & XFS_DIFLAG_NOATIME)
649 			flags |= FS_XFLAG_NOATIME;
650 		if (di_flags & XFS_DIFLAG_NODUMP)
651 			flags |= FS_XFLAG_NODUMP;
652 		if (di_flags & XFS_DIFLAG_RTINHERIT)
653 			flags |= FS_XFLAG_RTINHERIT;
654 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
655 			flags |= FS_XFLAG_PROJINHERIT;
656 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
657 			flags |= FS_XFLAG_NOSYMLINKS;
658 		if (di_flags & XFS_DIFLAG_EXTSIZE)
659 			flags |= FS_XFLAG_EXTSIZE;
660 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
661 			flags |= FS_XFLAG_EXTSZINHERIT;
662 		if (di_flags & XFS_DIFLAG_NODEFRAG)
663 			flags |= FS_XFLAG_NODEFRAG;
664 		if (di_flags & XFS_DIFLAG_FILESTREAM)
665 			flags |= FS_XFLAG_FILESTREAM;
666 	}
667 
668 	if (di_flags2 & XFS_DIFLAG2_ANY) {
669 		if (di_flags2 & XFS_DIFLAG2_DAX)
670 			flags |= FS_XFLAG_DAX;
671 		if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
672 			flags |= FS_XFLAG_COWEXTSIZE;
673 	}
674 
675 	if (has_attr)
676 		flags |= FS_XFLAG_HASATTR;
677 
678 	return flags;
679 }
680 
681 uint
682 xfs_ip2xflags(
683 	struct xfs_inode	*ip)
684 {
685 	struct xfs_icdinode	*dic = &ip->i_d;
686 
687 	return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
688 }
689 
690 /*
691  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
692  * is allowed, otherwise it has to be an exact match. If a CI match is found,
693  * ci_name->name will point to a the actual name (caller must free) or
694  * will be set to NULL if an exact match is found.
695  */
696 int
697 xfs_lookup(
698 	xfs_inode_t		*dp,
699 	struct xfs_name		*name,
700 	xfs_inode_t		**ipp,
701 	struct xfs_name		*ci_name)
702 {
703 	xfs_ino_t		inum;
704 	int			error;
705 
706 	trace_xfs_lookup(dp, name);
707 
708 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
709 		return -EIO;
710 
711 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
712 	if (error)
713 		goto out_unlock;
714 
715 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
716 	if (error)
717 		goto out_free_name;
718 
719 	return 0;
720 
721 out_free_name:
722 	if (ci_name)
723 		kmem_free(ci_name->name);
724 out_unlock:
725 	*ipp = NULL;
726 	return error;
727 }
728 
729 /*
730  * Allocate an inode on disk and return a copy of its in-core version.
731  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
732  * appropriately within the inode.  The uid and gid for the inode are
733  * set according to the contents of the given cred structure.
734  *
735  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
736  * has a free inode available, call xfs_iget() to obtain the in-core
737  * version of the allocated inode.  Finally, fill in the inode and
738  * log its initial contents.  In this case, ialloc_context would be
739  * set to NULL.
740  *
741  * If xfs_dialloc() does not have an available inode, it will replenish
742  * its supply by doing an allocation. Since we can only do one
743  * allocation within a transaction without deadlocks, we must commit
744  * the current transaction before returning the inode itself.
745  * In this case, therefore, we will set ialloc_context and return.
746  * The caller should then commit the current transaction, start a new
747  * transaction, and call xfs_ialloc() again to actually get the inode.
748  *
749  * To ensure that some other process does not grab the inode that
750  * was allocated during the first call to xfs_ialloc(), this routine
751  * also returns the [locked] bp pointing to the head of the freelist
752  * as ialloc_context.  The caller should hold this buffer across
753  * the commit and pass it back into this routine on the second call.
754  *
755  * If we are allocating quota inodes, we do not have a parent inode
756  * to attach to or associate with (i.e. pip == NULL) because they
757  * are not linked into the directory structure - they are attached
758  * directly to the superblock - and so have no parent.
759  */
760 static int
761 xfs_ialloc(
762 	xfs_trans_t	*tp,
763 	xfs_inode_t	*pip,
764 	umode_t		mode,
765 	xfs_nlink_t	nlink,
766 	dev_t		rdev,
767 	prid_t		prid,
768 	xfs_buf_t	**ialloc_context,
769 	xfs_inode_t	**ipp)
770 {
771 	struct xfs_mount *mp = tp->t_mountp;
772 	xfs_ino_t	ino;
773 	xfs_inode_t	*ip;
774 	uint		flags;
775 	int		error;
776 	struct timespec	tv;
777 	struct inode	*inode;
778 
779 	/*
780 	 * Call the space management code to pick
781 	 * the on-disk inode to be allocated.
782 	 */
783 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
784 			    ialloc_context, &ino);
785 	if (error)
786 		return error;
787 	if (*ialloc_context || ino == NULLFSINO) {
788 		*ipp = NULL;
789 		return 0;
790 	}
791 	ASSERT(*ialloc_context == NULL);
792 
793 	/*
794 	 * Get the in-core inode with the lock held exclusively.
795 	 * This is because we're setting fields here we need
796 	 * to prevent others from looking at until we're done.
797 	 */
798 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
799 			 XFS_ILOCK_EXCL, &ip);
800 	if (error)
801 		return error;
802 	ASSERT(ip != NULL);
803 	inode = VFS_I(ip);
804 
805 	/*
806 	 * We always convert v1 inodes to v2 now - we only support filesystems
807 	 * with >= v2 inode capability, so there is no reason for ever leaving
808 	 * an inode in v1 format.
809 	 */
810 	if (ip->i_d.di_version == 1)
811 		ip->i_d.di_version = 2;
812 
813 	inode->i_mode = mode;
814 	set_nlink(inode, nlink);
815 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
816 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
817 	inode->i_rdev = rdev;
818 	xfs_set_projid(ip, prid);
819 
820 	if (pip && XFS_INHERIT_GID(pip)) {
821 		ip->i_d.di_gid = pip->i_d.di_gid;
822 		if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
823 			inode->i_mode |= S_ISGID;
824 	}
825 
826 	/*
827 	 * If the group ID of the new file does not match the effective group
828 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
829 	 * (and only if the irix_sgid_inherit compatibility variable is set).
830 	 */
831 	if ((irix_sgid_inherit) &&
832 	    (inode->i_mode & S_ISGID) &&
833 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
834 		inode->i_mode &= ~S_ISGID;
835 
836 	ip->i_d.di_size = 0;
837 	ip->i_d.di_nextents = 0;
838 	ASSERT(ip->i_d.di_nblocks == 0);
839 
840 	tv = current_time(inode);
841 	inode->i_mtime = tv;
842 	inode->i_atime = tv;
843 	inode->i_ctime = tv;
844 
845 	ip->i_d.di_extsize = 0;
846 	ip->i_d.di_dmevmask = 0;
847 	ip->i_d.di_dmstate = 0;
848 	ip->i_d.di_flags = 0;
849 
850 	if (ip->i_d.di_version == 3) {
851 		inode_set_iversion(inode, 1);
852 		ip->i_d.di_flags2 = 0;
853 		ip->i_d.di_cowextsize = 0;
854 		ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec;
855 		ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec;
856 	}
857 
858 
859 	flags = XFS_ILOG_CORE;
860 	switch (mode & S_IFMT) {
861 	case S_IFIFO:
862 	case S_IFCHR:
863 	case S_IFBLK:
864 	case S_IFSOCK:
865 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
866 		ip->i_df.if_flags = 0;
867 		flags |= XFS_ILOG_DEV;
868 		break;
869 	case S_IFREG:
870 	case S_IFDIR:
871 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
872 			uint		di_flags = 0;
873 
874 			if (S_ISDIR(mode)) {
875 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
876 					di_flags |= XFS_DIFLAG_RTINHERIT;
877 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
878 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
879 					ip->i_d.di_extsize = pip->i_d.di_extsize;
880 				}
881 				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
882 					di_flags |= XFS_DIFLAG_PROJINHERIT;
883 			} else if (S_ISREG(mode)) {
884 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
885 					di_flags |= XFS_DIFLAG_REALTIME;
886 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
887 					di_flags |= XFS_DIFLAG_EXTSIZE;
888 					ip->i_d.di_extsize = pip->i_d.di_extsize;
889 				}
890 			}
891 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
892 			    xfs_inherit_noatime)
893 				di_flags |= XFS_DIFLAG_NOATIME;
894 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
895 			    xfs_inherit_nodump)
896 				di_flags |= XFS_DIFLAG_NODUMP;
897 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
898 			    xfs_inherit_sync)
899 				di_flags |= XFS_DIFLAG_SYNC;
900 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
901 			    xfs_inherit_nosymlinks)
902 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
903 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
904 			    xfs_inherit_nodefrag)
905 				di_flags |= XFS_DIFLAG_NODEFRAG;
906 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
907 				di_flags |= XFS_DIFLAG_FILESTREAM;
908 
909 			ip->i_d.di_flags |= di_flags;
910 		}
911 		if (pip &&
912 		    (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) &&
913 		    pip->i_d.di_version == 3 &&
914 		    ip->i_d.di_version == 3) {
915 			uint64_t	di_flags2 = 0;
916 
917 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
918 				di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
919 				ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
920 			}
921 			if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
922 				di_flags2 |= XFS_DIFLAG2_DAX;
923 
924 			ip->i_d.di_flags2 |= di_flags2;
925 		}
926 		/* FALLTHROUGH */
927 	case S_IFLNK:
928 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
929 		ip->i_df.if_flags = XFS_IFEXTENTS;
930 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
931 		ip->i_df.if_u1.if_root = NULL;
932 		break;
933 	default:
934 		ASSERT(0);
935 	}
936 	/*
937 	 * Attribute fork settings for new inode.
938 	 */
939 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
940 	ip->i_d.di_anextents = 0;
941 
942 	/*
943 	 * Log the new values stuffed into the inode.
944 	 */
945 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
946 	xfs_trans_log_inode(tp, ip, flags);
947 
948 	/* now that we have an i_mode we can setup the inode structure */
949 	xfs_setup_inode(ip);
950 
951 	*ipp = ip;
952 	return 0;
953 }
954 
955 /*
956  * Allocates a new inode from disk and return a pointer to the
957  * incore copy. This routine will internally commit the current
958  * transaction and allocate a new one if the Space Manager needed
959  * to do an allocation to replenish the inode free-list.
960  *
961  * This routine is designed to be called from xfs_create and
962  * xfs_create_dir.
963  *
964  */
965 int
966 xfs_dir_ialloc(
967 	xfs_trans_t	**tpp,		/* input: current transaction;
968 					   output: may be a new transaction. */
969 	xfs_inode_t	*dp,		/* directory within whose allocate
970 					   the inode. */
971 	umode_t		mode,
972 	xfs_nlink_t	nlink,
973 	dev_t		rdev,
974 	prid_t		prid,		/* project id */
975 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
976 					   locked. */
977 	int		*committed)
978 
979 {
980 	xfs_trans_t	*tp;
981 	xfs_inode_t	*ip;
982 	xfs_buf_t	*ialloc_context = NULL;
983 	int		code;
984 	void		*dqinfo;
985 	uint		tflags;
986 
987 	tp = *tpp;
988 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
989 
990 	/*
991 	 * xfs_ialloc will return a pointer to an incore inode if
992 	 * the Space Manager has an available inode on the free
993 	 * list. Otherwise, it will do an allocation and replenish
994 	 * the freelist.  Since we can only do one allocation per
995 	 * transaction without deadlocks, we will need to commit the
996 	 * current transaction and start a new one.  We will then
997 	 * need to call xfs_ialloc again to get the inode.
998 	 *
999 	 * If xfs_ialloc did an allocation to replenish the freelist,
1000 	 * it returns the bp containing the head of the freelist as
1001 	 * ialloc_context. We will hold a lock on it across the
1002 	 * transaction commit so that no other process can steal
1003 	 * the inode(s) that we've just allocated.
1004 	 */
1005 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
1006 			&ip);
1007 
1008 	/*
1009 	 * Return an error if we were unable to allocate a new inode.
1010 	 * This should only happen if we run out of space on disk or
1011 	 * encounter a disk error.
1012 	 */
1013 	if (code) {
1014 		*ipp = NULL;
1015 		return code;
1016 	}
1017 	if (!ialloc_context && !ip) {
1018 		*ipp = NULL;
1019 		return -ENOSPC;
1020 	}
1021 
1022 	/*
1023 	 * If the AGI buffer is non-NULL, then we were unable to get an
1024 	 * inode in one operation.  We need to commit the current
1025 	 * transaction and call xfs_ialloc() again.  It is guaranteed
1026 	 * to succeed the second time.
1027 	 */
1028 	if (ialloc_context) {
1029 		/*
1030 		 * Normally, xfs_trans_commit releases all the locks.
1031 		 * We call bhold to hang on to the ialloc_context across
1032 		 * the commit.  Holding this buffer prevents any other
1033 		 * processes from doing any allocations in this
1034 		 * allocation group.
1035 		 */
1036 		xfs_trans_bhold(tp, ialloc_context);
1037 
1038 		/*
1039 		 * We want the quota changes to be associated with the next
1040 		 * transaction, NOT this one. So, detach the dqinfo from this
1041 		 * and attach it to the next transaction.
1042 		 */
1043 		dqinfo = NULL;
1044 		tflags = 0;
1045 		if (tp->t_dqinfo) {
1046 			dqinfo = (void *)tp->t_dqinfo;
1047 			tp->t_dqinfo = NULL;
1048 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1049 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1050 		}
1051 
1052 		code = xfs_trans_roll(&tp);
1053 		if (committed != NULL)
1054 			*committed = 1;
1055 
1056 		/*
1057 		 * Re-attach the quota info that we detached from prev trx.
1058 		 */
1059 		if (dqinfo) {
1060 			tp->t_dqinfo = dqinfo;
1061 			tp->t_flags |= tflags;
1062 		}
1063 
1064 		if (code) {
1065 			xfs_buf_relse(ialloc_context);
1066 			*tpp = tp;
1067 			*ipp = NULL;
1068 			return code;
1069 		}
1070 		xfs_trans_bjoin(tp, ialloc_context);
1071 
1072 		/*
1073 		 * Call ialloc again. Since we've locked out all
1074 		 * other allocations in this allocation group,
1075 		 * this call should always succeed.
1076 		 */
1077 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1078 				  &ialloc_context, &ip);
1079 
1080 		/*
1081 		 * If we get an error at this point, return to the caller
1082 		 * so that the current transaction can be aborted.
1083 		 */
1084 		if (code) {
1085 			*tpp = tp;
1086 			*ipp = NULL;
1087 			return code;
1088 		}
1089 		ASSERT(!ialloc_context && ip);
1090 
1091 	} else {
1092 		if (committed != NULL)
1093 			*committed = 0;
1094 	}
1095 
1096 	*ipp = ip;
1097 	*tpp = tp;
1098 
1099 	return 0;
1100 }
1101 
1102 /*
1103  * Decrement the link count on an inode & log the change.  If this causes the
1104  * link count to go to zero, move the inode to AGI unlinked list so that it can
1105  * be freed when the last active reference goes away via xfs_inactive().
1106  */
1107 static int			/* error */
1108 xfs_droplink(
1109 	xfs_trans_t *tp,
1110 	xfs_inode_t *ip)
1111 {
1112 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1113 
1114 	drop_nlink(VFS_I(ip));
1115 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1116 
1117 	if (VFS_I(ip)->i_nlink)
1118 		return 0;
1119 
1120 	return xfs_iunlink(tp, ip);
1121 }
1122 
1123 /*
1124  * Increment the link count on an inode & log the change.
1125  */
1126 static int
1127 xfs_bumplink(
1128 	xfs_trans_t *tp,
1129 	xfs_inode_t *ip)
1130 {
1131 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1132 
1133 	ASSERT(ip->i_d.di_version > 1);
1134 	inc_nlink(VFS_I(ip));
1135 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1136 	return 0;
1137 }
1138 
1139 int
1140 xfs_create(
1141 	xfs_inode_t		*dp,
1142 	struct xfs_name		*name,
1143 	umode_t			mode,
1144 	dev_t			rdev,
1145 	xfs_inode_t		**ipp)
1146 {
1147 	int			is_dir = S_ISDIR(mode);
1148 	struct xfs_mount	*mp = dp->i_mount;
1149 	struct xfs_inode	*ip = NULL;
1150 	struct xfs_trans	*tp = NULL;
1151 	int			error;
1152 	struct xfs_defer_ops	dfops;
1153 	xfs_fsblock_t		first_block;
1154 	bool                    unlock_dp_on_error = false;
1155 	prid_t			prid;
1156 	struct xfs_dquot	*udqp = NULL;
1157 	struct xfs_dquot	*gdqp = NULL;
1158 	struct xfs_dquot	*pdqp = NULL;
1159 	struct xfs_trans_res	*tres;
1160 	uint			resblks;
1161 
1162 	trace_xfs_create(dp, name);
1163 
1164 	if (XFS_FORCED_SHUTDOWN(mp))
1165 		return -EIO;
1166 
1167 	prid = xfs_get_initial_prid(dp);
1168 
1169 	/*
1170 	 * Make sure that we have allocated dquot(s) on disk.
1171 	 */
1172 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1173 					xfs_kgid_to_gid(current_fsgid()), prid,
1174 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1175 					&udqp, &gdqp, &pdqp);
1176 	if (error)
1177 		return error;
1178 
1179 	if (is_dir) {
1180 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1181 		tres = &M_RES(mp)->tr_mkdir;
1182 	} else {
1183 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1184 		tres = &M_RES(mp)->tr_create;
1185 	}
1186 
1187 	/*
1188 	 * Initially assume that the file does not exist and
1189 	 * reserve the resources for that case.  If that is not
1190 	 * the case we'll drop the one we have and get a more
1191 	 * appropriate transaction later.
1192 	 */
1193 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1194 	if (error == -ENOSPC) {
1195 		/* flush outstanding delalloc blocks and retry */
1196 		xfs_flush_inodes(mp);
1197 		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1198 	}
1199 	if (error)
1200 		goto out_release_inode;
1201 
1202 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1203 	unlock_dp_on_error = true;
1204 
1205 	xfs_defer_init(&dfops, &first_block);
1206 
1207 	/*
1208 	 * Reserve disk quota and the inode.
1209 	 */
1210 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1211 						pdqp, resblks, 1, 0);
1212 	if (error)
1213 		goto out_trans_cancel;
1214 
1215 	/*
1216 	 * A newly created regular or special file just has one directory
1217 	 * entry pointing to them, but a directory also the "." entry
1218 	 * pointing to itself.
1219 	 */
1220 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip,
1221 			NULL);
1222 	if (error)
1223 		goto out_trans_cancel;
1224 
1225 	/*
1226 	 * Now we join the directory inode to the transaction.  We do not do it
1227 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1228 	 * (and release all the locks).  An error from here on will result in
1229 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1230 	 * error path.
1231 	 */
1232 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1233 	unlock_dp_on_error = false;
1234 
1235 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1236 					&first_block, &dfops, resblks ?
1237 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1238 	if (error) {
1239 		ASSERT(error != -ENOSPC);
1240 		goto out_trans_cancel;
1241 	}
1242 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1243 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1244 
1245 	if (is_dir) {
1246 		error = xfs_dir_init(tp, ip, dp);
1247 		if (error)
1248 			goto out_bmap_cancel;
1249 
1250 		error = xfs_bumplink(tp, dp);
1251 		if (error)
1252 			goto out_bmap_cancel;
1253 	}
1254 
1255 	/*
1256 	 * If this is a synchronous mount, make sure that the
1257 	 * create transaction goes to disk before returning to
1258 	 * the user.
1259 	 */
1260 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1261 		xfs_trans_set_sync(tp);
1262 
1263 	/*
1264 	 * Attach the dquot(s) to the inodes and modify them incore.
1265 	 * These ids of the inode couldn't have changed since the new
1266 	 * inode has been locked ever since it was created.
1267 	 */
1268 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1269 
1270 	error = xfs_defer_finish(&tp, &dfops);
1271 	if (error)
1272 		goto out_bmap_cancel;
1273 
1274 	error = xfs_trans_commit(tp);
1275 	if (error)
1276 		goto out_release_inode;
1277 
1278 	xfs_qm_dqrele(udqp);
1279 	xfs_qm_dqrele(gdqp);
1280 	xfs_qm_dqrele(pdqp);
1281 
1282 	*ipp = ip;
1283 	return 0;
1284 
1285  out_bmap_cancel:
1286 	xfs_defer_cancel(&dfops);
1287  out_trans_cancel:
1288 	xfs_trans_cancel(tp);
1289  out_release_inode:
1290 	/*
1291 	 * Wait until after the current transaction is aborted to finish the
1292 	 * setup of the inode and release the inode.  This prevents recursive
1293 	 * transactions and deadlocks from xfs_inactive.
1294 	 */
1295 	if (ip) {
1296 		xfs_finish_inode_setup(ip);
1297 		IRELE(ip);
1298 	}
1299 
1300 	xfs_qm_dqrele(udqp);
1301 	xfs_qm_dqrele(gdqp);
1302 	xfs_qm_dqrele(pdqp);
1303 
1304 	if (unlock_dp_on_error)
1305 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1306 	return error;
1307 }
1308 
1309 int
1310 xfs_create_tmpfile(
1311 	struct xfs_inode	*dp,
1312 	struct dentry		*dentry,
1313 	umode_t			mode,
1314 	struct xfs_inode	**ipp)
1315 {
1316 	struct xfs_mount	*mp = dp->i_mount;
1317 	struct xfs_inode	*ip = NULL;
1318 	struct xfs_trans	*tp = NULL;
1319 	int			error;
1320 	prid_t                  prid;
1321 	struct xfs_dquot	*udqp = NULL;
1322 	struct xfs_dquot	*gdqp = NULL;
1323 	struct xfs_dquot	*pdqp = NULL;
1324 	struct xfs_trans_res	*tres;
1325 	uint			resblks;
1326 
1327 	if (XFS_FORCED_SHUTDOWN(mp))
1328 		return -EIO;
1329 
1330 	prid = xfs_get_initial_prid(dp);
1331 
1332 	/*
1333 	 * Make sure that we have allocated dquot(s) on disk.
1334 	 */
1335 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1336 				xfs_kgid_to_gid(current_fsgid()), prid,
1337 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1338 				&udqp, &gdqp, &pdqp);
1339 	if (error)
1340 		return error;
1341 
1342 	resblks = XFS_IALLOC_SPACE_RES(mp);
1343 	tres = &M_RES(mp)->tr_create_tmpfile;
1344 
1345 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1346 	if (error)
1347 		goto out_release_inode;
1348 
1349 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1350 						pdqp, resblks, 1, 0);
1351 	if (error)
1352 		goto out_trans_cancel;
1353 
1354 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip, NULL);
1355 	if (error)
1356 		goto out_trans_cancel;
1357 
1358 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1359 		xfs_trans_set_sync(tp);
1360 
1361 	/*
1362 	 * Attach the dquot(s) to the inodes and modify them incore.
1363 	 * These ids of the inode couldn't have changed since the new
1364 	 * inode has been locked ever since it was created.
1365 	 */
1366 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1367 
1368 	error = xfs_iunlink(tp, ip);
1369 	if (error)
1370 		goto out_trans_cancel;
1371 
1372 	error = xfs_trans_commit(tp);
1373 	if (error)
1374 		goto out_release_inode;
1375 
1376 	xfs_qm_dqrele(udqp);
1377 	xfs_qm_dqrele(gdqp);
1378 	xfs_qm_dqrele(pdqp);
1379 
1380 	*ipp = ip;
1381 	return 0;
1382 
1383  out_trans_cancel:
1384 	xfs_trans_cancel(tp);
1385  out_release_inode:
1386 	/*
1387 	 * Wait until after the current transaction is aborted to finish the
1388 	 * setup of the inode and release the inode.  This prevents recursive
1389 	 * transactions and deadlocks from xfs_inactive.
1390 	 */
1391 	if (ip) {
1392 		xfs_finish_inode_setup(ip);
1393 		IRELE(ip);
1394 	}
1395 
1396 	xfs_qm_dqrele(udqp);
1397 	xfs_qm_dqrele(gdqp);
1398 	xfs_qm_dqrele(pdqp);
1399 
1400 	return error;
1401 }
1402 
1403 int
1404 xfs_link(
1405 	xfs_inode_t		*tdp,
1406 	xfs_inode_t		*sip,
1407 	struct xfs_name		*target_name)
1408 {
1409 	xfs_mount_t		*mp = tdp->i_mount;
1410 	xfs_trans_t		*tp;
1411 	int			error;
1412 	struct xfs_defer_ops	dfops;
1413 	xfs_fsblock_t           first_block;
1414 	int			resblks;
1415 
1416 	trace_xfs_link(tdp, target_name);
1417 
1418 	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1419 
1420 	if (XFS_FORCED_SHUTDOWN(mp))
1421 		return -EIO;
1422 
1423 	error = xfs_qm_dqattach(sip, 0);
1424 	if (error)
1425 		goto std_return;
1426 
1427 	error = xfs_qm_dqattach(tdp, 0);
1428 	if (error)
1429 		goto std_return;
1430 
1431 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1432 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1433 	if (error == -ENOSPC) {
1434 		resblks = 0;
1435 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1436 	}
1437 	if (error)
1438 		goto std_return;
1439 
1440 	xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1441 
1442 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1443 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1444 
1445 	/*
1446 	 * If we are using project inheritance, we only allow hard link
1447 	 * creation in our tree when the project IDs are the same; else
1448 	 * the tree quota mechanism could be circumvented.
1449 	 */
1450 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1451 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1452 		error = -EXDEV;
1453 		goto error_return;
1454 	}
1455 
1456 	if (!resblks) {
1457 		error = xfs_dir_canenter(tp, tdp, target_name);
1458 		if (error)
1459 			goto error_return;
1460 	}
1461 
1462 	xfs_defer_init(&dfops, &first_block);
1463 
1464 	/*
1465 	 * Handle initial link state of O_TMPFILE inode
1466 	 */
1467 	if (VFS_I(sip)->i_nlink == 0) {
1468 		error = xfs_iunlink_remove(tp, sip);
1469 		if (error)
1470 			goto error_return;
1471 	}
1472 
1473 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1474 					&first_block, &dfops, resblks);
1475 	if (error)
1476 		goto error_return;
1477 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1478 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1479 
1480 	error = xfs_bumplink(tp, sip);
1481 	if (error)
1482 		goto error_return;
1483 
1484 	/*
1485 	 * If this is a synchronous mount, make sure that the
1486 	 * link transaction goes to disk before returning to
1487 	 * the user.
1488 	 */
1489 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1490 		xfs_trans_set_sync(tp);
1491 
1492 	error = xfs_defer_finish(&tp, &dfops);
1493 	if (error) {
1494 		xfs_defer_cancel(&dfops);
1495 		goto error_return;
1496 	}
1497 
1498 	return xfs_trans_commit(tp);
1499 
1500  error_return:
1501 	xfs_trans_cancel(tp);
1502  std_return:
1503 	return error;
1504 }
1505 
1506 /* Clear the reflink flag and the cowblocks tag if possible. */
1507 static void
1508 xfs_itruncate_clear_reflink_flags(
1509 	struct xfs_inode	*ip)
1510 {
1511 	struct xfs_ifork	*dfork;
1512 	struct xfs_ifork	*cfork;
1513 
1514 	if (!xfs_is_reflink_inode(ip))
1515 		return;
1516 	dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1517 	cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1518 	if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1519 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1520 	if (cfork->if_bytes == 0)
1521 		xfs_inode_clear_cowblocks_tag(ip);
1522 }
1523 
1524 /*
1525  * Free up the underlying blocks past new_size.  The new size must be smaller
1526  * than the current size.  This routine can be used both for the attribute and
1527  * data fork, and does not modify the inode size, which is left to the caller.
1528  *
1529  * The transaction passed to this routine must have made a permanent log
1530  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1531  * given transaction and start new ones, so make sure everything involved in
1532  * the transaction is tidy before calling here.  Some transaction will be
1533  * returned to the caller to be committed.  The incoming transaction must
1534  * already include the inode, and both inode locks must be held exclusively.
1535  * The inode must also be "held" within the transaction.  On return the inode
1536  * will be "held" within the returned transaction.  This routine does NOT
1537  * require any disk space to be reserved for it within the transaction.
1538  *
1539  * If we get an error, we must return with the inode locked and linked into the
1540  * current transaction. This keeps things simple for the higher level code,
1541  * because it always knows that the inode is locked and held in the transaction
1542  * that returns to it whether errors occur or not.  We don't mark the inode
1543  * dirty on error so that transactions can be easily aborted if possible.
1544  */
1545 int
1546 xfs_itruncate_extents(
1547 	struct xfs_trans	**tpp,
1548 	struct xfs_inode	*ip,
1549 	int			whichfork,
1550 	xfs_fsize_t		new_size)
1551 {
1552 	struct xfs_mount	*mp = ip->i_mount;
1553 	struct xfs_trans	*tp = *tpp;
1554 	struct xfs_defer_ops	dfops;
1555 	xfs_fsblock_t		first_block;
1556 	xfs_fileoff_t		first_unmap_block;
1557 	xfs_fileoff_t		last_block;
1558 	xfs_filblks_t		unmap_len;
1559 	int			error = 0;
1560 	int			done = 0;
1561 
1562 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1563 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1564 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1565 	ASSERT(new_size <= XFS_ISIZE(ip));
1566 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1567 	ASSERT(ip->i_itemp != NULL);
1568 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1569 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1570 
1571 	trace_xfs_itruncate_extents_start(ip, new_size);
1572 
1573 	/*
1574 	 * Since it is possible for space to become allocated beyond
1575 	 * the end of the file (in a crash where the space is allocated
1576 	 * but the inode size is not yet updated), simply remove any
1577 	 * blocks which show up between the new EOF and the maximum
1578 	 * possible file size.  If the first block to be removed is
1579 	 * beyond the maximum file size (ie it is the same as last_block),
1580 	 * then there is nothing to do.
1581 	 */
1582 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1583 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1584 	if (first_unmap_block == last_block)
1585 		return 0;
1586 
1587 	ASSERT(first_unmap_block < last_block);
1588 	unmap_len = last_block - first_unmap_block + 1;
1589 	while (!done) {
1590 		xfs_defer_init(&dfops, &first_block);
1591 		error = xfs_bunmapi(tp, ip,
1592 				    first_unmap_block, unmap_len,
1593 				    xfs_bmapi_aflag(whichfork),
1594 				    XFS_ITRUNC_MAX_EXTENTS,
1595 				    &first_block, &dfops,
1596 				    &done);
1597 		if (error)
1598 			goto out_bmap_cancel;
1599 
1600 		/*
1601 		 * Duplicate the transaction that has the permanent
1602 		 * reservation and commit the old transaction.
1603 		 */
1604 		xfs_defer_ijoin(&dfops, ip);
1605 		error = xfs_defer_finish(&tp, &dfops);
1606 		if (error)
1607 			goto out_bmap_cancel;
1608 
1609 		error = xfs_trans_roll_inode(&tp, ip);
1610 		if (error)
1611 			goto out;
1612 	}
1613 
1614 	/* Remove all pending CoW reservations. */
1615 	error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block,
1616 			last_block, true);
1617 	if (error)
1618 		goto out;
1619 
1620 	xfs_itruncate_clear_reflink_flags(ip);
1621 
1622 	/*
1623 	 * Always re-log the inode so that our permanent transaction can keep
1624 	 * on rolling it forward in the log.
1625 	 */
1626 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1627 
1628 	trace_xfs_itruncate_extents_end(ip, new_size);
1629 
1630 out:
1631 	*tpp = tp;
1632 	return error;
1633 out_bmap_cancel:
1634 	/*
1635 	 * If the bunmapi call encounters an error, return to the caller where
1636 	 * the transaction can be properly aborted.  We just need to make sure
1637 	 * we're not holding any resources that we were not when we came in.
1638 	 */
1639 	xfs_defer_cancel(&dfops);
1640 	goto out;
1641 }
1642 
1643 int
1644 xfs_release(
1645 	xfs_inode_t	*ip)
1646 {
1647 	xfs_mount_t	*mp = ip->i_mount;
1648 	int		error;
1649 
1650 	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1651 		return 0;
1652 
1653 	/* If this is a read-only mount, don't do this (would generate I/O) */
1654 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1655 		return 0;
1656 
1657 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1658 		int truncated;
1659 
1660 		/*
1661 		 * If we previously truncated this file and removed old data
1662 		 * in the process, we want to initiate "early" writeout on
1663 		 * the last close.  This is an attempt to combat the notorious
1664 		 * NULL files problem which is particularly noticeable from a
1665 		 * truncate down, buffered (re-)write (delalloc), followed by
1666 		 * a crash.  What we are effectively doing here is
1667 		 * significantly reducing the time window where we'd otherwise
1668 		 * be exposed to that problem.
1669 		 */
1670 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1671 		if (truncated) {
1672 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1673 			if (ip->i_delayed_blks > 0) {
1674 				error = filemap_flush(VFS_I(ip)->i_mapping);
1675 				if (error)
1676 					return error;
1677 			}
1678 		}
1679 	}
1680 
1681 	if (VFS_I(ip)->i_nlink == 0)
1682 		return 0;
1683 
1684 	if (xfs_can_free_eofblocks(ip, false)) {
1685 
1686 		/*
1687 		 * Check if the inode is being opened, written and closed
1688 		 * frequently and we have delayed allocation blocks outstanding
1689 		 * (e.g. streaming writes from the NFS server), truncating the
1690 		 * blocks past EOF will cause fragmentation to occur.
1691 		 *
1692 		 * In this case don't do the truncation, but we have to be
1693 		 * careful how we detect this case. Blocks beyond EOF show up as
1694 		 * i_delayed_blks even when the inode is clean, so we need to
1695 		 * truncate them away first before checking for a dirty release.
1696 		 * Hence on the first dirty close we will still remove the
1697 		 * speculative allocation, but after that we will leave it in
1698 		 * place.
1699 		 */
1700 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1701 			return 0;
1702 		/*
1703 		 * If we can't get the iolock just skip truncating the blocks
1704 		 * past EOF because we could deadlock with the mmap_sem
1705 		 * otherwise. We'll get another chance to drop them once the
1706 		 * last reference to the inode is dropped, so we'll never leak
1707 		 * blocks permanently.
1708 		 */
1709 		if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1710 			error = xfs_free_eofblocks(ip);
1711 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1712 			if (error)
1713 				return error;
1714 		}
1715 
1716 		/* delalloc blocks after truncation means it really is dirty */
1717 		if (ip->i_delayed_blks)
1718 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1719 	}
1720 	return 0;
1721 }
1722 
1723 /*
1724  * xfs_inactive_truncate
1725  *
1726  * Called to perform a truncate when an inode becomes unlinked.
1727  */
1728 STATIC int
1729 xfs_inactive_truncate(
1730 	struct xfs_inode *ip)
1731 {
1732 	struct xfs_mount	*mp = ip->i_mount;
1733 	struct xfs_trans	*tp;
1734 	int			error;
1735 
1736 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1737 	if (error) {
1738 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1739 		return error;
1740 	}
1741 
1742 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1743 	xfs_trans_ijoin(tp, ip, 0);
1744 
1745 	/*
1746 	 * Log the inode size first to prevent stale data exposure in the event
1747 	 * of a system crash before the truncate completes. See the related
1748 	 * comment in xfs_vn_setattr_size() for details.
1749 	 */
1750 	ip->i_d.di_size = 0;
1751 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1752 
1753 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1754 	if (error)
1755 		goto error_trans_cancel;
1756 
1757 	ASSERT(ip->i_d.di_nextents == 0);
1758 
1759 	error = xfs_trans_commit(tp);
1760 	if (error)
1761 		goto error_unlock;
1762 
1763 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1764 	return 0;
1765 
1766 error_trans_cancel:
1767 	xfs_trans_cancel(tp);
1768 error_unlock:
1769 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1770 	return error;
1771 }
1772 
1773 /*
1774  * xfs_inactive_ifree()
1775  *
1776  * Perform the inode free when an inode is unlinked.
1777  */
1778 STATIC int
1779 xfs_inactive_ifree(
1780 	struct xfs_inode *ip)
1781 {
1782 	struct xfs_defer_ops	dfops;
1783 	xfs_fsblock_t		first_block;
1784 	struct xfs_mount	*mp = ip->i_mount;
1785 	struct xfs_trans	*tp;
1786 	int			error;
1787 
1788 	/*
1789 	 * We try to use a per-AG reservation for any block needed by the finobt
1790 	 * tree, but as the finobt feature predates the per-AG reservation
1791 	 * support a degraded file system might not have enough space for the
1792 	 * reservation at mount time.  In that case try to dip into the reserved
1793 	 * pool and pray.
1794 	 *
1795 	 * Send a warning if the reservation does happen to fail, as the inode
1796 	 * now remains allocated and sits on the unlinked list until the fs is
1797 	 * repaired.
1798 	 */
1799 	if (unlikely(mp->m_inotbt_nores)) {
1800 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1801 				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1802 				&tp);
1803 	} else {
1804 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1805 	}
1806 	if (error) {
1807 		if (error == -ENOSPC) {
1808 			xfs_warn_ratelimited(mp,
1809 			"Failed to remove inode(s) from unlinked list. "
1810 			"Please free space, unmount and run xfs_repair.");
1811 		} else {
1812 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1813 		}
1814 		return error;
1815 	}
1816 
1817 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1818 	xfs_trans_ijoin(tp, ip, 0);
1819 
1820 	xfs_defer_init(&dfops, &first_block);
1821 	error = xfs_ifree(tp, ip, &dfops);
1822 	if (error) {
1823 		/*
1824 		 * If we fail to free the inode, shut down.  The cancel
1825 		 * might do that, we need to make sure.  Otherwise the
1826 		 * inode might be lost for a long time or forever.
1827 		 */
1828 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1829 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1830 				__func__, error);
1831 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1832 		}
1833 		xfs_trans_cancel(tp);
1834 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1835 		return error;
1836 	}
1837 
1838 	/*
1839 	 * Credit the quota account(s). The inode is gone.
1840 	 */
1841 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1842 
1843 	/*
1844 	 * Just ignore errors at this point.  There is nothing we can do except
1845 	 * to try to keep going. Make sure it's not a silent error.
1846 	 */
1847 	error = xfs_defer_finish(&tp, &dfops);
1848 	if (error) {
1849 		xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1850 			__func__, error);
1851 		xfs_defer_cancel(&dfops);
1852 	}
1853 	error = xfs_trans_commit(tp);
1854 	if (error)
1855 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1856 			__func__, error);
1857 
1858 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1859 	return 0;
1860 }
1861 
1862 /*
1863  * xfs_inactive
1864  *
1865  * This is called when the vnode reference count for the vnode
1866  * goes to zero.  If the file has been unlinked, then it must
1867  * now be truncated.  Also, we clear all of the read-ahead state
1868  * kept for the inode here since the file is now closed.
1869  */
1870 void
1871 xfs_inactive(
1872 	xfs_inode_t	*ip)
1873 {
1874 	struct xfs_mount	*mp;
1875 	struct xfs_ifork	*cow_ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1876 	int			error;
1877 	int			truncate = 0;
1878 
1879 	/*
1880 	 * If the inode is already free, then there can be nothing
1881 	 * to clean up here.
1882 	 */
1883 	if (VFS_I(ip)->i_mode == 0) {
1884 		ASSERT(ip->i_df.if_real_bytes == 0);
1885 		ASSERT(ip->i_df.if_broot_bytes == 0);
1886 		return;
1887 	}
1888 
1889 	mp = ip->i_mount;
1890 	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1891 
1892 	/* If this is a read-only mount, don't do this (would generate I/O) */
1893 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1894 		return;
1895 
1896 	/* Try to clean out the cow blocks if there are any. */
1897 	if (xfs_is_reflink_inode(ip) && cow_ifp->if_bytes > 0)
1898 		xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1899 
1900 	if (VFS_I(ip)->i_nlink != 0) {
1901 		/*
1902 		 * force is true because we are evicting an inode from the
1903 		 * cache. Post-eof blocks must be freed, lest we end up with
1904 		 * broken free space accounting.
1905 		 *
1906 		 * Note: don't bother with iolock here since lockdep complains
1907 		 * about acquiring it in reclaim context. We have the only
1908 		 * reference to the inode at this point anyways.
1909 		 */
1910 		if (xfs_can_free_eofblocks(ip, true))
1911 			xfs_free_eofblocks(ip);
1912 
1913 		return;
1914 	}
1915 
1916 	if (S_ISREG(VFS_I(ip)->i_mode) &&
1917 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1918 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1919 		truncate = 1;
1920 
1921 	error = xfs_qm_dqattach(ip, 0);
1922 	if (error)
1923 		return;
1924 
1925 	if (S_ISLNK(VFS_I(ip)->i_mode))
1926 		error = xfs_inactive_symlink(ip);
1927 	else if (truncate)
1928 		error = xfs_inactive_truncate(ip);
1929 	if (error)
1930 		return;
1931 
1932 	/*
1933 	 * If there are attributes associated with the file then blow them away
1934 	 * now.  The code calls a routine that recursively deconstructs the
1935 	 * attribute fork. If also blows away the in-core attribute fork.
1936 	 */
1937 	if (XFS_IFORK_Q(ip)) {
1938 		error = xfs_attr_inactive(ip);
1939 		if (error)
1940 			return;
1941 	}
1942 
1943 	ASSERT(!ip->i_afp);
1944 	ASSERT(ip->i_d.di_anextents == 0);
1945 	ASSERT(ip->i_d.di_forkoff == 0);
1946 
1947 	/*
1948 	 * Free the inode.
1949 	 */
1950 	error = xfs_inactive_ifree(ip);
1951 	if (error)
1952 		return;
1953 
1954 	/*
1955 	 * Release the dquots held by inode, if any.
1956 	 */
1957 	xfs_qm_dqdetach(ip);
1958 }
1959 
1960 /*
1961  * This is called when the inode's link count goes to 0 or we are creating a
1962  * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1963  * set to true as the link count is dropped to zero by the VFS after we've
1964  * created the file successfully, so we have to add it to the unlinked list
1965  * while the link count is non-zero.
1966  *
1967  * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1968  * list when the inode is freed.
1969  */
1970 STATIC int
1971 xfs_iunlink(
1972 	struct xfs_trans *tp,
1973 	struct xfs_inode *ip)
1974 {
1975 	xfs_mount_t	*mp = tp->t_mountp;
1976 	xfs_agi_t	*agi;
1977 	xfs_dinode_t	*dip;
1978 	xfs_buf_t	*agibp;
1979 	xfs_buf_t	*ibp;
1980 	xfs_agino_t	agino;
1981 	short		bucket_index;
1982 	int		offset;
1983 	int		error;
1984 
1985 	ASSERT(VFS_I(ip)->i_mode != 0);
1986 
1987 	/*
1988 	 * Get the agi buffer first.  It ensures lock ordering
1989 	 * on the list.
1990 	 */
1991 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1992 	if (error)
1993 		return error;
1994 	agi = XFS_BUF_TO_AGI(agibp);
1995 
1996 	/*
1997 	 * Get the index into the agi hash table for the
1998 	 * list this inode will go on.
1999 	 */
2000 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2001 	ASSERT(agino != 0);
2002 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2003 	ASSERT(agi->agi_unlinked[bucket_index]);
2004 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2005 
2006 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2007 		/*
2008 		 * There is already another inode in the bucket we need
2009 		 * to add ourselves to.  Add us at the front of the list.
2010 		 * Here we put the head pointer into our next pointer,
2011 		 * and then we fall through to point the head at us.
2012 		 */
2013 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2014 				       0, 0);
2015 		if (error)
2016 			return error;
2017 
2018 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2019 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2020 		offset = ip->i_imap.im_boffset +
2021 			offsetof(xfs_dinode_t, di_next_unlinked);
2022 
2023 		/* need to recalc the inode CRC if appropriate */
2024 		xfs_dinode_calc_crc(mp, dip);
2025 
2026 		xfs_trans_inode_buf(tp, ibp);
2027 		xfs_trans_log_buf(tp, ibp, offset,
2028 				  (offset + sizeof(xfs_agino_t) - 1));
2029 		xfs_inobp_check(mp, ibp);
2030 	}
2031 
2032 	/*
2033 	 * Point the bucket head pointer at the inode being inserted.
2034 	 */
2035 	ASSERT(agino != 0);
2036 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2037 	offset = offsetof(xfs_agi_t, agi_unlinked) +
2038 		(sizeof(xfs_agino_t) * bucket_index);
2039 	xfs_trans_log_buf(tp, agibp, offset,
2040 			  (offset + sizeof(xfs_agino_t) - 1));
2041 	return 0;
2042 }
2043 
2044 /*
2045  * Pull the on-disk inode from the AGI unlinked list.
2046  */
2047 STATIC int
2048 xfs_iunlink_remove(
2049 	xfs_trans_t	*tp,
2050 	xfs_inode_t	*ip)
2051 {
2052 	xfs_ino_t	next_ino;
2053 	xfs_mount_t	*mp;
2054 	xfs_agi_t	*agi;
2055 	xfs_dinode_t	*dip;
2056 	xfs_buf_t	*agibp;
2057 	xfs_buf_t	*ibp;
2058 	xfs_agnumber_t	agno;
2059 	xfs_agino_t	agino;
2060 	xfs_agino_t	next_agino;
2061 	xfs_buf_t	*last_ibp;
2062 	xfs_dinode_t	*last_dip = NULL;
2063 	short		bucket_index;
2064 	int		offset, last_offset = 0;
2065 	int		error;
2066 
2067 	mp = tp->t_mountp;
2068 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2069 
2070 	/*
2071 	 * Get the agi buffer first.  It ensures lock ordering
2072 	 * on the list.
2073 	 */
2074 	error = xfs_read_agi(mp, tp, agno, &agibp);
2075 	if (error)
2076 		return error;
2077 
2078 	agi = XFS_BUF_TO_AGI(agibp);
2079 
2080 	/*
2081 	 * Get the index into the agi hash table for the
2082 	 * list this inode will go on.
2083 	 */
2084 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2085 	ASSERT(agino != 0);
2086 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2087 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2088 	ASSERT(agi->agi_unlinked[bucket_index]);
2089 
2090 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2091 		/*
2092 		 * We're at the head of the list.  Get the inode's on-disk
2093 		 * buffer to see if there is anyone after us on the list.
2094 		 * Only modify our next pointer if it is not already NULLAGINO.
2095 		 * This saves us the overhead of dealing with the buffer when
2096 		 * there is no need to change it.
2097 		 */
2098 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2099 				       0, 0);
2100 		if (error) {
2101 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2102 				__func__, error);
2103 			return error;
2104 		}
2105 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2106 		ASSERT(next_agino != 0);
2107 		if (next_agino != NULLAGINO) {
2108 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2109 			offset = ip->i_imap.im_boffset +
2110 				offsetof(xfs_dinode_t, di_next_unlinked);
2111 
2112 			/* need to recalc the inode CRC if appropriate */
2113 			xfs_dinode_calc_crc(mp, dip);
2114 
2115 			xfs_trans_inode_buf(tp, ibp);
2116 			xfs_trans_log_buf(tp, ibp, offset,
2117 					  (offset + sizeof(xfs_agino_t) - 1));
2118 			xfs_inobp_check(mp, ibp);
2119 		} else {
2120 			xfs_trans_brelse(tp, ibp);
2121 		}
2122 		/*
2123 		 * Point the bucket head pointer at the next inode.
2124 		 */
2125 		ASSERT(next_agino != 0);
2126 		ASSERT(next_agino != agino);
2127 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2128 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2129 			(sizeof(xfs_agino_t) * bucket_index);
2130 		xfs_trans_log_buf(tp, agibp, offset,
2131 				  (offset + sizeof(xfs_agino_t) - 1));
2132 	} else {
2133 		/*
2134 		 * We need to search the list for the inode being freed.
2135 		 */
2136 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2137 		last_ibp = NULL;
2138 		while (next_agino != agino) {
2139 			struct xfs_imap	imap;
2140 
2141 			if (last_ibp)
2142 				xfs_trans_brelse(tp, last_ibp);
2143 
2144 			imap.im_blkno = 0;
2145 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2146 
2147 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2148 			if (error) {
2149 				xfs_warn(mp,
2150 	"%s: xfs_imap returned error %d.",
2151 					 __func__, error);
2152 				return error;
2153 			}
2154 
2155 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2156 					       &last_ibp, 0, 0);
2157 			if (error) {
2158 				xfs_warn(mp,
2159 	"%s: xfs_imap_to_bp returned error %d.",
2160 					__func__, error);
2161 				return error;
2162 			}
2163 
2164 			last_offset = imap.im_boffset;
2165 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2166 			ASSERT(next_agino != NULLAGINO);
2167 			ASSERT(next_agino != 0);
2168 		}
2169 
2170 		/*
2171 		 * Now last_ibp points to the buffer previous to us on the
2172 		 * unlinked list.  Pull us from the list.
2173 		 */
2174 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2175 				       0, 0);
2176 		if (error) {
2177 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2178 				__func__, error);
2179 			return error;
2180 		}
2181 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2182 		ASSERT(next_agino != 0);
2183 		ASSERT(next_agino != agino);
2184 		if (next_agino != NULLAGINO) {
2185 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2186 			offset = ip->i_imap.im_boffset +
2187 				offsetof(xfs_dinode_t, di_next_unlinked);
2188 
2189 			/* need to recalc the inode CRC if appropriate */
2190 			xfs_dinode_calc_crc(mp, dip);
2191 
2192 			xfs_trans_inode_buf(tp, ibp);
2193 			xfs_trans_log_buf(tp, ibp, offset,
2194 					  (offset + sizeof(xfs_agino_t) - 1));
2195 			xfs_inobp_check(mp, ibp);
2196 		} else {
2197 			xfs_trans_brelse(tp, ibp);
2198 		}
2199 		/*
2200 		 * Point the previous inode on the list to the next inode.
2201 		 */
2202 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2203 		ASSERT(next_agino != 0);
2204 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2205 
2206 		/* need to recalc the inode CRC if appropriate */
2207 		xfs_dinode_calc_crc(mp, last_dip);
2208 
2209 		xfs_trans_inode_buf(tp, last_ibp);
2210 		xfs_trans_log_buf(tp, last_ibp, offset,
2211 				  (offset + sizeof(xfs_agino_t) - 1));
2212 		xfs_inobp_check(mp, last_ibp);
2213 	}
2214 	return 0;
2215 }
2216 
2217 /*
2218  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2219  * inodes that are in memory - they all must be marked stale and attached to
2220  * the cluster buffer.
2221  */
2222 STATIC int
2223 xfs_ifree_cluster(
2224 	xfs_inode_t		*free_ip,
2225 	xfs_trans_t		*tp,
2226 	struct xfs_icluster	*xic)
2227 {
2228 	xfs_mount_t		*mp = free_ip->i_mount;
2229 	int			blks_per_cluster;
2230 	int			inodes_per_cluster;
2231 	int			nbufs;
2232 	int			i, j;
2233 	int			ioffset;
2234 	xfs_daddr_t		blkno;
2235 	xfs_buf_t		*bp;
2236 	xfs_inode_t		*ip;
2237 	xfs_inode_log_item_t	*iip;
2238 	struct xfs_log_item	*lip;
2239 	struct xfs_perag	*pag;
2240 	xfs_ino_t		inum;
2241 
2242 	inum = xic->first_ino;
2243 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2244 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2245 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2246 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2247 
2248 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2249 		/*
2250 		 * The allocation bitmap tells us which inodes of the chunk were
2251 		 * physically allocated. Skip the cluster if an inode falls into
2252 		 * a sparse region.
2253 		 */
2254 		ioffset = inum - xic->first_ino;
2255 		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2256 			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2257 			continue;
2258 		}
2259 
2260 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2261 					 XFS_INO_TO_AGBNO(mp, inum));
2262 
2263 		/*
2264 		 * We obtain and lock the backing buffer first in the process
2265 		 * here, as we have to ensure that any dirty inode that we
2266 		 * can't get the flush lock on is attached to the buffer.
2267 		 * If we scan the in-memory inodes first, then buffer IO can
2268 		 * complete before we get a lock on it, and hence we may fail
2269 		 * to mark all the active inodes on the buffer stale.
2270 		 */
2271 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2272 					mp->m_bsize * blks_per_cluster,
2273 					XBF_UNMAPPED);
2274 
2275 		if (!bp)
2276 			return -ENOMEM;
2277 
2278 		/*
2279 		 * This buffer may not have been correctly initialised as we
2280 		 * didn't read it from disk. That's not important because we are
2281 		 * only using to mark the buffer as stale in the log, and to
2282 		 * attach stale cached inodes on it. That means it will never be
2283 		 * dispatched for IO. If it is, we want to know about it, and we
2284 		 * want it to fail. We can acheive this by adding a write
2285 		 * verifier to the buffer.
2286 		 */
2287 		 bp->b_ops = &xfs_inode_buf_ops;
2288 
2289 		/*
2290 		 * Walk the inodes already attached to the buffer and mark them
2291 		 * stale. These will all have the flush locks held, so an
2292 		 * in-memory inode walk can't lock them. By marking them all
2293 		 * stale first, we will not attempt to lock them in the loop
2294 		 * below as the XFS_ISTALE flag will be set.
2295 		 */
2296 		list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
2297 			if (lip->li_type == XFS_LI_INODE) {
2298 				iip = (xfs_inode_log_item_t *)lip;
2299 				ASSERT(iip->ili_logged == 1);
2300 				lip->li_cb = xfs_istale_done;
2301 				xfs_trans_ail_copy_lsn(mp->m_ail,
2302 							&iip->ili_flush_lsn,
2303 							&iip->ili_item.li_lsn);
2304 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2305 			}
2306 		}
2307 
2308 
2309 		/*
2310 		 * For each inode in memory attempt to add it to the inode
2311 		 * buffer and set it up for being staled on buffer IO
2312 		 * completion.  This is safe as we've locked out tail pushing
2313 		 * and flushing by locking the buffer.
2314 		 *
2315 		 * We have already marked every inode that was part of a
2316 		 * transaction stale above, which means there is no point in
2317 		 * even trying to lock them.
2318 		 */
2319 		for (i = 0; i < inodes_per_cluster; i++) {
2320 retry:
2321 			rcu_read_lock();
2322 			ip = radix_tree_lookup(&pag->pag_ici_root,
2323 					XFS_INO_TO_AGINO(mp, (inum + i)));
2324 
2325 			/* Inode not in memory, nothing to do */
2326 			if (!ip) {
2327 				rcu_read_unlock();
2328 				continue;
2329 			}
2330 
2331 			/*
2332 			 * because this is an RCU protected lookup, we could
2333 			 * find a recently freed or even reallocated inode
2334 			 * during the lookup. We need to check under the
2335 			 * i_flags_lock for a valid inode here. Skip it if it
2336 			 * is not valid, the wrong inode or stale.
2337 			 */
2338 			spin_lock(&ip->i_flags_lock);
2339 			if (ip->i_ino != inum + i ||
2340 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2341 				spin_unlock(&ip->i_flags_lock);
2342 				rcu_read_unlock();
2343 				continue;
2344 			}
2345 			spin_unlock(&ip->i_flags_lock);
2346 
2347 			/*
2348 			 * Don't try to lock/unlock the current inode, but we
2349 			 * _cannot_ skip the other inodes that we did not find
2350 			 * in the list attached to the buffer and are not
2351 			 * already marked stale. If we can't lock it, back off
2352 			 * and retry.
2353 			 */
2354 			if (ip != free_ip) {
2355 				if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2356 					rcu_read_unlock();
2357 					delay(1);
2358 					goto retry;
2359 				}
2360 
2361 				/*
2362 				 * Check the inode number again in case we're
2363 				 * racing with freeing in xfs_reclaim_inode().
2364 				 * See the comments in that function for more
2365 				 * information as to why the initial check is
2366 				 * not sufficient.
2367 				 */
2368 				if (ip->i_ino != inum + i) {
2369 					xfs_iunlock(ip, XFS_ILOCK_EXCL);
2370 					rcu_read_unlock();
2371 					continue;
2372 				}
2373 			}
2374 			rcu_read_unlock();
2375 
2376 			xfs_iflock(ip);
2377 			xfs_iflags_set(ip, XFS_ISTALE);
2378 
2379 			/*
2380 			 * we don't need to attach clean inodes or those only
2381 			 * with unlogged changes (which we throw away, anyway).
2382 			 */
2383 			iip = ip->i_itemp;
2384 			if (!iip || xfs_inode_clean(ip)) {
2385 				ASSERT(ip != free_ip);
2386 				xfs_ifunlock(ip);
2387 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2388 				continue;
2389 			}
2390 
2391 			iip->ili_last_fields = iip->ili_fields;
2392 			iip->ili_fields = 0;
2393 			iip->ili_fsync_fields = 0;
2394 			iip->ili_logged = 1;
2395 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2396 						&iip->ili_item.li_lsn);
2397 
2398 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2399 						  &iip->ili_item);
2400 
2401 			if (ip != free_ip)
2402 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2403 		}
2404 
2405 		xfs_trans_stale_inode_buf(tp, bp);
2406 		xfs_trans_binval(tp, bp);
2407 	}
2408 
2409 	xfs_perag_put(pag);
2410 	return 0;
2411 }
2412 
2413 /*
2414  * Free any local-format buffers sitting around before we reset to
2415  * extents format.
2416  */
2417 static inline void
2418 xfs_ifree_local_data(
2419 	struct xfs_inode	*ip,
2420 	int			whichfork)
2421 {
2422 	struct xfs_ifork	*ifp;
2423 
2424 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2425 		return;
2426 
2427 	ifp = XFS_IFORK_PTR(ip, whichfork);
2428 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2429 }
2430 
2431 /*
2432  * This is called to return an inode to the inode free list.
2433  * The inode should already be truncated to 0 length and have
2434  * no pages associated with it.  This routine also assumes that
2435  * the inode is already a part of the transaction.
2436  *
2437  * The on-disk copy of the inode will have been added to the list
2438  * of unlinked inodes in the AGI. We need to remove the inode from
2439  * that list atomically with respect to freeing it here.
2440  */
2441 int
2442 xfs_ifree(
2443 	xfs_trans_t	*tp,
2444 	xfs_inode_t	*ip,
2445 	struct xfs_defer_ops	*dfops)
2446 {
2447 	int			error;
2448 	struct xfs_icluster	xic = { 0 };
2449 
2450 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2451 	ASSERT(VFS_I(ip)->i_nlink == 0);
2452 	ASSERT(ip->i_d.di_nextents == 0);
2453 	ASSERT(ip->i_d.di_anextents == 0);
2454 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2455 	ASSERT(ip->i_d.di_nblocks == 0);
2456 
2457 	/*
2458 	 * Pull the on-disk inode from the AGI unlinked list.
2459 	 */
2460 	error = xfs_iunlink_remove(tp, ip);
2461 	if (error)
2462 		return error;
2463 
2464 	error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2465 	if (error)
2466 		return error;
2467 
2468 	xfs_ifree_local_data(ip, XFS_DATA_FORK);
2469 	xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2470 
2471 	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2472 	ip->i_d.di_flags = 0;
2473 	ip->i_d.di_flags2 = 0;
2474 	ip->i_d.di_dmevmask = 0;
2475 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2476 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2477 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2478 
2479 	/* Don't attempt to replay owner changes for a deleted inode */
2480 	ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER);
2481 
2482 	/*
2483 	 * Bump the generation count so no one will be confused
2484 	 * by reincarnations of this inode.
2485 	 */
2486 	VFS_I(ip)->i_generation++;
2487 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2488 
2489 	if (xic.deleted)
2490 		error = xfs_ifree_cluster(ip, tp, &xic);
2491 
2492 	return error;
2493 }
2494 
2495 /*
2496  * This is called to unpin an inode.  The caller must have the inode locked
2497  * in at least shared mode so that the buffer cannot be subsequently pinned
2498  * once someone is waiting for it to be unpinned.
2499  */
2500 static void
2501 xfs_iunpin(
2502 	struct xfs_inode	*ip)
2503 {
2504 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2505 
2506 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2507 
2508 	/* Give the log a push to start the unpinning I/O */
2509 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2510 
2511 }
2512 
2513 static void
2514 __xfs_iunpin_wait(
2515 	struct xfs_inode	*ip)
2516 {
2517 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2518 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2519 
2520 	xfs_iunpin(ip);
2521 
2522 	do {
2523 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2524 		if (xfs_ipincount(ip))
2525 			io_schedule();
2526 	} while (xfs_ipincount(ip));
2527 	finish_wait(wq, &wait.wq_entry);
2528 }
2529 
2530 void
2531 xfs_iunpin_wait(
2532 	struct xfs_inode	*ip)
2533 {
2534 	if (xfs_ipincount(ip))
2535 		__xfs_iunpin_wait(ip);
2536 }
2537 
2538 /*
2539  * Removing an inode from the namespace involves removing the directory entry
2540  * and dropping the link count on the inode. Removing the directory entry can
2541  * result in locking an AGF (directory blocks were freed) and removing a link
2542  * count can result in placing the inode on an unlinked list which results in
2543  * locking an AGI.
2544  *
2545  * The big problem here is that we have an ordering constraint on AGF and AGI
2546  * locking - inode allocation locks the AGI, then can allocate a new extent for
2547  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2548  * removes the inode from the unlinked list, requiring that we lock the AGI
2549  * first, and then freeing the inode can result in an inode chunk being freed
2550  * and hence freeing disk space requiring that we lock an AGF.
2551  *
2552  * Hence the ordering that is imposed by other parts of the code is AGI before
2553  * AGF. This means we cannot remove the directory entry before we drop the inode
2554  * reference count and put it on the unlinked list as this results in a lock
2555  * order of AGF then AGI, and this can deadlock against inode allocation and
2556  * freeing. Therefore we must drop the link counts before we remove the
2557  * directory entry.
2558  *
2559  * This is still safe from a transactional point of view - it is not until we
2560  * get to xfs_defer_finish() that we have the possibility of multiple
2561  * transactions in this operation. Hence as long as we remove the directory
2562  * entry and drop the link count in the first transaction of the remove
2563  * operation, there are no transactional constraints on the ordering here.
2564  */
2565 int
2566 xfs_remove(
2567 	xfs_inode_t             *dp,
2568 	struct xfs_name		*name,
2569 	xfs_inode_t		*ip)
2570 {
2571 	xfs_mount_t		*mp = dp->i_mount;
2572 	xfs_trans_t             *tp = NULL;
2573 	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2574 	int                     error = 0;
2575 	struct xfs_defer_ops	dfops;
2576 	xfs_fsblock_t           first_block;
2577 	uint			resblks;
2578 
2579 	trace_xfs_remove(dp, name);
2580 
2581 	if (XFS_FORCED_SHUTDOWN(mp))
2582 		return -EIO;
2583 
2584 	error = xfs_qm_dqattach(dp, 0);
2585 	if (error)
2586 		goto std_return;
2587 
2588 	error = xfs_qm_dqattach(ip, 0);
2589 	if (error)
2590 		goto std_return;
2591 
2592 	/*
2593 	 * We try to get the real space reservation first,
2594 	 * allowing for directory btree deletion(s) implying
2595 	 * possible bmap insert(s).  If we can't get the space
2596 	 * reservation then we use 0 instead, and avoid the bmap
2597 	 * btree insert(s) in the directory code by, if the bmap
2598 	 * insert tries to happen, instead trimming the LAST
2599 	 * block from the directory.
2600 	 */
2601 	resblks = XFS_REMOVE_SPACE_RES(mp);
2602 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2603 	if (error == -ENOSPC) {
2604 		resblks = 0;
2605 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2606 				&tp);
2607 	}
2608 	if (error) {
2609 		ASSERT(error != -ENOSPC);
2610 		goto std_return;
2611 	}
2612 
2613 	xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2614 
2615 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2616 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2617 
2618 	/*
2619 	 * If we're removing a directory perform some additional validation.
2620 	 */
2621 	if (is_dir) {
2622 		ASSERT(VFS_I(ip)->i_nlink >= 2);
2623 		if (VFS_I(ip)->i_nlink != 2) {
2624 			error = -ENOTEMPTY;
2625 			goto out_trans_cancel;
2626 		}
2627 		if (!xfs_dir_isempty(ip)) {
2628 			error = -ENOTEMPTY;
2629 			goto out_trans_cancel;
2630 		}
2631 
2632 		/* Drop the link from ip's "..".  */
2633 		error = xfs_droplink(tp, dp);
2634 		if (error)
2635 			goto out_trans_cancel;
2636 
2637 		/* Drop the "." link from ip to self.  */
2638 		error = xfs_droplink(tp, ip);
2639 		if (error)
2640 			goto out_trans_cancel;
2641 	} else {
2642 		/*
2643 		 * When removing a non-directory we need to log the parent
2644 		 * inode here.  For a directory this is done implicitly
2645 		 * by the xfs_droplink call for the ".." entry.
2646 		 */
2647 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2648 	}
2649 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2650 
2651 	/* Drop the link from dp to ip. */
2652 	error = xfs_droplink(tp, ip);
2653 	if (error)
2654 		goto out_trans_cancel;
2655 
2656 	xfs_defer_init(&dfops, &first_block);
2657 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2658 					&first_block, &dfops, resblks);
2659 	if (error) {
2660 		ASSERT(error != -ENOENT);
2661 		goto out_bmap_cancel;
2662 	}
2663 
2664 	/*
2665 	 * If this is a synchronous mount, make sure that the
2666 	 * remove transaction goes to disk before returning to
2667 	 * the user.
2668 	 */
2669 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2670 		xfs_trans_set_sync(tp);
2671 
2672 	error = xfs_defer_finish(&tp, &dfops);
2673 	if (error)
2674 		goto out_bmap_cancel;
2675 
2676 	error = xfs_trans_commit(tp);
2677 	if (error)
2678 		goto std_return;
2679 
2680 	if (is_dir && xfs_inode_is_filestream(ip))
2681 		xfs_filestream_deassociate(ip);
2682 
2683 	return 0;
2684 
2685  out_bmap_cancel:
2686 	xfs_defer_cancel(&dfops);
2687  out_trans_cancel:
2688 	xfs_trans_cancel(tp);
2689  std_return:
2690 	return error;
2691 }
2692 
2693 /*
2694  * Enter all inodes for a rename transaction into a sorted array.
2695  */
2696 #define __XFS_SORT_INODES	5
2697 STATIC void
2698 xfs_sort_for_rename(
2699 	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2700 	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2701 	struct xfs_inode	*ip1,	/* in: inode of old entry */
2702 	struct xfs_inode	*ip2,	/* in: inode of new entry */
2703 	struct xfs_inode	*wip,	/* in: whiteout inode */
2704 	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2705 	int			*num_inodes)  /* in/out: inodes in array */
2706 {
2707 	int			i, j;
2708 
2709 	ASSERT(*num_inodes == __XFS_SORT_INODES);
2710 	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2711 
2712 	/*
2713 	 * i_tab contains a list of pointers to inodes.  We initialize
2714 	 * the table here & we'll sort it.  We will then use it to
2715 	 * order the acquisition of the inode locks.
2716 	 *
2717 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2718 	 */
2719 	i = 0;
2720 	i_tab[i++] = dp1;
2721 	i_tab[i++] = dp2;
2722 	i_tab[i++] = ip1;
2723 	if (ip2)
2724 		i_tab[i++] = ip2;
2725 	if (wip)
2726 		i_tab[i++] = wip;
2727 	*num_inodes = i;
2728 
2729 	/*
2730 	 * Sort the elements via bubble sort.  (Remember, there are at
2731 	 * most 5 elements to sort, so this is adequate.)
2732 	 */
2733 	for (i = 0; i < *num_inodes; i++) {
2734 		for (j = 1; j < *num_inodes; j++) {
2735 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2736 				struct xfs_inode *temp = i_tab[j];
2737 				i_tab[j] = i_tab[j-1];
2738 				i_tab[j-1] = temp;
2739 			}
2740 		}
2741 	}
2742 }
2743 
2744 static int
2745 xfs_finish_rename(
2746 	struct xfs_trans	*tp,
2747 	struct xfs_defer_ops	*dfops)
2748 {
2749 	int			error;
2750 
2751 	/*
2752 	 * If this is a synchronous mount, make sure that the rename transaction
2753 	 * goes to disk before returning to the user.
2754 	 */
2755 	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2756 		xfs_trans_set_sync(tp);
2757 
2758 	error = xfs_defer_finish(&tp, dfops);
2759 	if (error) {
2760 		xfs_defer_cancel(dfops);
2761 		xfs_trans_cancel(tp);
2762 		return error;
2763 	}
2764 
2765 	return xfs_trans_commit(tp);
2766 }
2767 
2768 /*
2769  * xfs_cross_rename()
2770  *
2771  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2772  */
2773 STATIC int
2774 xfs_cross_rename(
2775 	struct xfs_trans	*tp,
2776 	struct xfs_inode	*dp1,
2777 	struct xfs_name		*name1,
2778 	struct xfs_inode	*ip1,
2779 	struct xfs_inode	*dp2,
2780 	struct xfs_name		*name2,
2781 	struct xfs_inode	*ip2,
2782 	struct xfs_defer_ops	*dfops,
2783 	xfs_fsblock_t		*first_block,
2784 	int			spaceres)
2785 {
2786 	int		error = 0;
2787 	int		ip1_flags = 0;
2788 	int		ip2_flags = 0;
2789 	int		dp2_flags = 0;
2790 
2791 	/* Swap inode number for dirent in first parent */
2792 	error = xfs_dir_replace(tp, dp1, name1,
2793 				ip2->i_ino,
2794 				first_block, dfops, spaceres);
2795 	if (error)
2796 		goto out_trans_abort;
2797 
2798 	/* Swap inode number for dirent in second parent */
2799 	error = xfs_dir_replace(tp, dp2, name2,
2800 				ip1->i_ino,
2801 				first_block, dfops, spaceres);
2802 	if (error)
2803 		goto out_trans_abort;
2804 
2805 	/*
2806 	 * If we're renaming one or more directories across different parents,
2807 	 * update the respective ".." entries (and link counts) to match the new
2808 	 * parents.
2809 	 */
2810 	if (dp1 != dp2) {
2811 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2812 
2813 		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2814 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2815 						dp1->i_ino, first_block,
2816 						dfops, spaceres);
2817 			if (error)
2818 				goto out_trans_abort;
2819 
2820 			/* transfer ip2 ".." reference to dp1 */
2821 			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2822 				error = xfs_droplink(tp, dp2);
2823 				if (error)
2824 					goto out_trans_abort;
2825 				error = xfs_bumplink(tp, dp1);
2826 				if (error)
2827 					goto out_trans_abort;
2828 			}
2829 
2830 			/*
2831 			 * Although ip1 isn't changed here, userspace needs
2832 			 * to be warned about the change, so that applications
2833 			 * relying on it (like backup ones), will properly
2834 			 * notify the change
2835 			 */
2836 			ip1_flags |= XFS_ICHGTIME_CHG;
2837 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2838 		}
2839 
2840 		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2841 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2842 						dp2->i_ino, first_block,
2843 						dfops, spaceres);
2844 			if (error)
2845 				goto out_trans_abort;
2846 
2847 			/* transfer ip1 ".." reference to dp2 */
2848 			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2849 				error = xfs_droplink(tp, dp1);
2850 				if (error)
2851 					goto out_trans_abort;
2852 				error = xfs_bumplink(tp, dp2);
2853 				if (error)
2854 					goto out_trans_abort;
2855 			}
2856 
2857 			/*
2858 			 * Although ip2 isn't changed here, userspace needs
2859 			 * to be warned about the change, so that applications
2860 			 * relying on it (like backup ones), will properly
2861 			 * notify the change
2862 			 */
2863 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2864 			ip2_flags |= XFS_ICHGTIME_CHG;
2865 		}
2866 	}
2867 
2868 	if (ip1_flags) {
2869 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2870 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2871 	}
2872 	if (ip2_flags) {
2873 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2874 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2875 	}
2876 	if (dp2_flags) {
2877 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2878 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2879 	}
2880 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2881 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2882 	return xfs_finish_rename(tp, dfops);
2883 
2884 out_trans_abort:
2885 	xfs_defer_cancel(dfops);
2886 	xfs_trans_cancel(tp);
2887 	return error;
2888 }
2889 
2890 /*
2891  * xfs_rename_alloc_whiteout()
2892  *
2893  * Return a referenced, unlinked, unlocked inode that that can be used as a
2894  * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2895  * crash between allocating the inode and linking it into the rename transaction
2896  * recovery will free the inode and we won't leak it.
2897  */
2898 static int
2899 xfs_rename_alloc_whiteout(
2900 	struct xfs_inode	*dp,
2901 	struct xfs_inode	**wip)
2902 {
2903 	struct xfs_inode	*tmpfile;
2904 	int			error;
2905 
2906 	error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2907 	if (error)
2908 		return error;
2909 
2910 	/*
2911 	 * Prepare the tmpfile inode as if it were created through the VFS.
2912 	 * Otherwise, the link increment paths will complain about nlink 0->1.
2913 	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2914 	 * and flag it as linkable.
2915 	 */
2916 	drop_nlink(VFS_I(tmpfile));
2917 	xfs_setup_iops(tmpfile);
2918 	xfs_finish_inode_setup(tmpfile);
2919 	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2920 
2921 	*wip = tmpfile;
2922 	return 0;
2923 }
2924 
2925 /*
2926  * xfs_rename
2927  */
2928 int
2929 xfs_rename(
2930 	struct xfs_inode	*src_dp,
2931 	struct xfs_name		*src_name,
2932 	struct xfs_inode	*src_ip,
2933 	struct xfs_inode	*target_dp,
2934 	struct xfs_name		*target_name,
2935 	struct xfs_inode	*target_ip,
2936 	unsigned int		flags)
2937 {
2938 	struct xfs_mount	*mp = src_dp->i_mount;
2939 	struct xfs_trans	*tp;
2940 	struct xfs_defer_ops	dfops;
2941 	xfs_fsblock_t		first_block;
2942 	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2943 	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2944 	int			num_inodes = __XFS_SORT_INODES;
2945 	bool			new_parent = (src_dp != target_dp);
2946 	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2947 	int			spaceres;
2948 	int			error;
2949 
2950 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2951 
2952 	if ((flags & RENAME_EXCHANGE) && !target_ip)
2953 		return -EINVAL;
2954 
2955 	/*
2956 	 * If we are doing a whiteout operation, allocate the whiteout inode
2957 	 * we will be placing at the target and ensure the type is set
2958 	 * appropriately.
2959 	 */
2960 	if (flags & RENAME_WHITEOUT) {
2961 		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2962 		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2963 		if (error)
2964 			return error;
2965 
2966 		/* setup target dirent info as whiteout */
2967 		src_name->type = XFS_DIR3_FT_CHRDEV;
2968 	}
2969 
2970 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2971 				inodes, &num_inodes);
2972 
2973 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2974 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2975 	if (error == -ENOSPC) {
2976 		spaceres = 0;
2977 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2978 				&tp);
2979 	}
2980 	if (error)
2981 		goto out_release_wip;
2982 
2983 	/*
2984 	 * Attach the dquots to the inodes
2985 	 */
2986 	error = xfs_qm_vop_rename_dqattach(inodes);
2987 	if (error)
2988 		goto out_trans_cancel;
2989 
2990 	/*
2991 	 * Lock all the participating inodes. Depending upon whether
2992 	 * the target_name exists in the target directory, and
2993 	 * whether the target directory is the same as the source
2994 	 * directory, we can lock from 2 to 4 inodes.
2995 	 */
2996 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2997 
2998 	/*
2999 	 * Join all the inodes to the transaction. From this point on,
3000 	 * we can rely on either trans_commit or trans_cancel to unlock
3001 	 * them.
3002 	 */
3003 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3004 	if (new_parent)
3005 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3006 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3007 	if (target_ip)
3008 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3009 	if (wip)
3010 		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3011 
3012 	/*
3013 	 * If we are using project inheritance, we only allow renames
3014 	 * into our tree when the project IDs are the same; else the
3015 	 * tree quota mechanism would be circumvented.
3016 	 */
3017 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3018 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3019 		error = -EXDEV;
3020 		goto out_trans_cancel;
3021 	}
3022 
3023 	xfs_defer_init(&dfops, &first_block);
3024 
3025 	/* RENAME_EXCHANGE is unique from here on. */
3026 	if (flags & RENAME_EXCHANGE)
3027 		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3028 					target_dp, target_name, target_ip,
3029 					&dfops, &first_block, spaceres);
3030 
3031 	/*
3032 	 * Set up the target.
3033 	 */
3034 	if (target_ip == NULL) {
3035 		/*
3036 		 * If there's no space reservation, check the entry will
3037 		 * fit before actually inserting it.
3038 		 */
3039 		if (!spaceres) {
3040 			error = xfs_dir_canenter(tp, target_dp, target_name);
3041 			if (error)
3042 				goto out_trans_cancel;
3043 		}
3044 		/*
3045 		 * If target does not exist and the rename crosses
3046 		 * directories, adjust the target directory link count
3047 		 * to account for the ".." reference from the new entry.
3048 		 */
3049 		error = xfs_dir_createname(tp, target_dp, target_name,
3050 						src_ip->i_ino, &first_block,
3051 						&dfops, spaceres);
3052 		if (error)
3053 			goto out_bmap_cancel;
3054 
3055 		xfs_trans_ichgtime(tp, target_dp,
3056 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3057 
3058 		if (new_parent && src_is_directory) {
3059 			error = xfs_bumplink(tp, target_dp);
3060 			if (error)
3061 				goto out_bmap_cancel;
3062 		}
3063 	} else { /* target_ip != NULL */
3064 		/*
3065 		 * If target exists and it's a directory, check that both
3066 		 * target and source are directories and that target can be
3067 		 * destroyed, or that neither is a directory.
3068 		 */
3069 		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3070 			/*
3071 			 * Make sure target dir is empty.
3072 			 */
3073 			if (!(xfs_dir_isempty(target_ip)) ||
3074 			    (VFS_I(target_ip)->i_nlink > 2)) {
3075 				error = -EEXIST;
3076 				goto out_trans_cancel;
3077 			}
3078 		}
3079 
3080 		/*
3081 		 * Link the source inode under the target name.
3082 		 * If the source inode is a directory and we are moving
3083 		 * it across directories, its ".." entry will be
3084 		 * inconsistent until we replace that down below.
3085 		 *
3086 		 * In case there is already an entry with the same
3087 		 * name at the destination directory, remove it first.
3088 		 */
3089 		error = xfs_dir_replace(tp, target_dp, target_name,
3090 					src_ip->i_ino,
3091 					&first_block, &dfops, spaceres);
3092 		if (error)
3093 			goto out_bmap_cancel;
3094 
3095 		xfs_trans_ichgtime(tp, target_dp,
3096 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3097 
3098 		/*
3099 		 * Decrement the link count on the target since the target
3100 		 * dir no longer points to it.
3101 		 */
3102 		error = xfs_droplink(tp, target_ip);
3103 		if (error)
3104 			goto out_bmap_cancel;
3105 
3106 		if (src_is_directory) {
3107 			/*
3108 			 * Drop the link from the old "." entry.
3109 			 */
3110 			error = xfs_droplink(tp, target_ip);
3111 			if (error)
3112 				goto out_bmap_cancel;
3113 		}
3114 	} /* target_ip != NULL */
3115 
3116 	/*
3117 	 * Remove the source.
3118 	 */
3119 	if (new_parent && src_is_directory) {
3120 		/*
3121 		 * Rewrite the ".." entry to point to the new
3122 		 * directory.
3123 		 */
3124 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3125 					target_dp->i_ino,
3126 					&first_block, &dfops, spaceres);
3127 		ASSERT(error != -EEXIST);
3128 		if (error)
3129 			goto out_bmap_cancel;
3130 	}
3131 
3132 	/*
3133 	 * We always want to hit the ctime on the source inode.
3134 	 *
3135 	 * This isn't strictly required by the standards since the source
3136 	 * inode isn't really being changed, but old unix file systems did
3137 	 * it and some incremental backup programs won't work without it.
3138 	 */
3139 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3140 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3141 
3142 	/*
3143 	 * Adjust the link count on src_dp.  This is necessary when
3144 	 * renaming a directory, either within one parent when
3145 	 * the target existed, or across two parent directories.
3146 	 */
3147 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3148 
3149 		/*
3150 		 * Decrement link count on src_directory since the
3151 		 * entry that's moved no longer points to it.
3152 		 */
3153 		error = xfs_droplink(tp, src_dp);
3154 		if (error)
3155 			goto out_bmap_cancel;
3156 	}
3157 
3158 	/*
3159 	 * For whiteouts, we only need to update the source dirent with the
3160 	 * inode number of the whiteout inode rather than removing it
3161 	 * altogether.
3162 	 */
3163 	if (wip) {
3164 		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3165 					&first_block, &dfops, spaceres);
3166 	} else
3167 		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3168 					   &first_block, &dfops, spaceres);
3169 	if (error)
3170 		goto out_bmap_cancel;
3171 
3172 	/*
3173 	 * For whiteouts, we need to bump the link count on the whiteout inode.
3174 	 * This means that failures all the way up to this point leave the inode
3175 	 * on the unlinked list and so cleanup is a simple matter of dropping
3176 	 * the remaining reference to it. If we fail here after bumping the link
3177 	 * count, we're shutting down the filesystem so we'll never see the
3178 	 * intermediate state on disk.
3179 	 */
3180 	if (wip) {
3181 		ASSERT(VFS_I(wip)->i_nlink == 0);
3182 		error = xfs_bumplink(tp, wip);
3183 		if (error)
3184 			goto out_bmap_cancel;
3185 		error = xfs_iunlink_remove(tp, wip);
3186 		if (error)
3187 			goto out_bmap_cancel;
3188 		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3189 
3190 		/*
3191 		 * Now we have a real link, clear the "I'm a tmpfile" state
3192 		 * flag from the inode so it doesn't accidentally get misused in
3193 		 * future.
3194 		 */
3195 		VFS_I(wip)->i_state &= ~I_LINKABLE;
3196 	}
3197 
3198 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3199 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3200 	if (new_parent)
3201 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3202 
3203 	error = xfs_finish_rename(tp, &dfops);
3204 	if (wip)
3205 		IRELE(wip);
3206 	return error;
3207 
3208 out_bmap_cancel:
3209 	xfs_defer_cancel(&dfops);
3210 out_trans_cancel:
3211 	xfs_trans_cancel(tp);
3212 out_release_wip:
3213 	if (wip)
3214 		IRELE(wip);
3215 	return error;
3216 }
3217 
3218 STATIC int
3219 xfs_iflush_cluster(
3220 	struct xfs_inode	*ip,
3221 	struct xfs_buf		*bp)
3222 {
3223 	struct xfs_mount	*mp = ip->i_mount;
3224 	struct xfs_perag	*pag;
3225 	unsigned long		first_index, mask;
3226 	unsigned long		inodes_per_cluster;
3227 	int			cilist_size;
3228 	struct xfs_inode	**cilist;
3229 	struct xfs_inode	*cip;
3230 	int			nr_found;
3231 	int			clcount = 0;
3232 	int			bufwasdelwri;
3233 	int			i;
3234 
3235 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3236 
3237 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3238 	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3239 	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3240 	if (!cilist)
3241 		goto out_put;
3242 
3243 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3244 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3245 	rcu_read_lock();
3246 	/* really need a gang lookup range call here */
3247 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3248 					first_index, inodes_per_cluster);
3249 	if (nr_found == 0)
3250 		goto out_free;
3251 
3252 	for (i = 0; i < nr_found; i++) {
3253 		cip = cilist[i];
3254 		if (cip == ip)
3255 			continue;
3256 
3257 		/*
3258 		 * because this is an RCU protected lookup, we could find a
3259 		 * recently freed or even reallocated inode during the lookup.
3260 		 * We need to check under the i_flags_lock for a valid inode
3261 		 * here. Skip it if it is not valid or the wrong inode.
3262 		 */
3263 		spin_lock(&cip->i_flags_lock);
3264 		if (!cip->i_ino ||
3265 		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3266 			spin_unlock(&cip->i_flags_lock);
3267 			continue;
3268 		}
3269 
3270 		/*
3271 		 * Once we fall off the end of the cluster, no point checking
3272 		 * any more inodes in the list because they will also all be
3273 		 * outside the cluster.
3274 		 */
3275 		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3276 			spin_unlock(&cip->i_flags_lock);
3277 			break;
3278 		}
3279 		spin_unlock(&cip->i_flags_lock);
3280 
3281 		/*
3282 		 * Do an un-protected check to see if the inode is dirty and
3283 		 * is a candidate for flushing.  These checks will be repeated
3284 		 * later after the appropriate locks are acquired.
3285 		 */
3286 		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3287 			continue;
3288 
3289 		/*
3290 		 * Try to get locks.  If any are unavailable or it is pinned,
3291 		 * then this inode cannot be flushed and is skipped.
3292 		 */
3293 
3294 		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3295 			continue;
3296 		if (!xfs_iflock_nowait(cip)) {
3297 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3298 			continue;
3299 		}
3300 		if (xfs_ipincount(cip)) {
3301 			xfs_ifunlock(cip);
3302 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3303 			continue;
3304 		}
3305 
3306 
3307 		/*
3308 		 * Check the inode number again, just to be certain we are not
3309 		 * racing with freeing in xfs_reclaim_inode(). See the comments
3310 		 * in that function for more information as to why the initial
3311 		 * check is not sufficient.
3312 		 */
3313 		if (!cip->i_ino) {
3314 			xfs_ifunlock(cip);
3315 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3316 			continue;
3317 		}
3318 
3319 		/*
3320 		 * arriving here means that this inode can be flushed.  First
3321 		 * re-check that it's dirty before flushing.
3322 		 */
3323 		if (!xfs_inode_clean(cip)) {
3324 			int	error;
3325 			error = xfs_iflush_int(cip, bp);
3326 			if (error) {
3327 				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3328 				goto cluster_corrupt_out;
3329 			}
3330 			clcount++;
3331 		} else {
3332 			xfs_ifunlock(cip);
3333 		}
3334 		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3335 	}
3336 
3337 	if (clcount) {
3338 		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3339 		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3340 	}
3341 
3342 out_free:
3343 	rcu_read_unlock();
3344 	kmem_free(cilist);
3345 out_put:
3346 	xfs_perag_put(pag);
3347 	return 0;
3348 
3349 
3350 cluster_corrupt_out:
3351 	/*
3352 	 * Corruption detected in the clustering loop.  Invalidate the
3353 	 * inode buffer and shut down the filesystem.
3354 	 */
3355 	rcu_read_unlock();
3356 	/*
3357 	 * Clean up the buffer.  If it was delwri, just release it --
3358 	 * brelse can handle it with no problems.  If not, shut down the
3359 	 * filesystem before releasing the buffer.
3360 	 */
3361 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3362 	if (bufwasdelwri)
3363 		xfs_buf_relse(bp);
3364 
3365 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3366 
3367 	if (!bufwasdelwri) {
3368 		/*
3369 		 * Just like incore_relse: if we have b_iodone functions,
3370 		 * mark the buffer as an error and call them.  Otherwise
3371 		 * mark it as stale and brelse.
3372 		 */
3373 		if (bp->b_iodone) {
3374 			bp->b_flags &= ~XBF_DONE;
3375 			xfs_buf_stale(bp);
3376 			xfs_buf_ioerror(bp, -EIO);
3377 			xfs_buf_ioend(bp);
3378 		} else {
3379 			xfs_buf_stale(bp);
3380 			xfs_buf_relse(bp);
3381 		}
3382 	}
3383 
3384 	/*
3385 	 * Unlocks the flush lock
3386 	 */
3387 	xfs_iflush_abort(cip, false);
3388 	kmem_free(cilist);
3389 	xfs_perag_put(pag);
3390 	return -EFSCORRUPTED;
3391 }
3392 
3393 /*
3394  * Flush dirty inode metadata into the backing buffer.
3395  *
3396  * The caller must have the inode lock and the inode flush lock held.  The
3397  * inode lock will still be held upon return to the caller, and the inode
3398  * flush lock will be released after the inode has reached the disk.
3399  *
3400  * The caller must write out the buffer returned in *bpp and release it.
3401  */
3402 int
3403 xfs_iflush(
3404 	struct xfs_inode	*ip,
3405 	struct xfs_buf		**bpp)
3406 {
3407 	struct xfs_mount	*mp = ip->i_mount;
3408 	struct xfs_buf		*bp = NULL;
3409 	struct xfs_dinode	*dip;
3410 	int			error;
3411 
3412 	XFS_STATS_INC(mp, xs_iflush_count);
3413 
3414 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3415 	ASSERT(xfs_isiflocked(ip));
3416 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3417 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3418 
3419 	*bpp = NULL;
3420 
3421 	xfs_iunpin_wait(ip);
3422 
3423 	/*
3424 	 * For stale inodes we cannot rely on the backing buffer remaining
3425 	 * stale in cache for the remaining life of the stale inode and so
3426 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3427 	 * inodes below. We have to check this after ensuring the inode is
3428 	 * unpinned so that it is safe to reclaim the stale inode after the
3429 	 * flush call.
3430 	 */
3431 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3432 		xfs_ifunlock(ip);
3433 		return 0;
3434 	}
3435 
3436 	/*
3437 	 * This may have been unpinned because the filesystem is shutting
3438 	 * down forcibly. If that's the case we must not write this inode
3439 	 * to disk, because the log record didn't make it to disk.
3440 	 *
3441 	 * We also have to remove the log item from the AIL in this case,
3442 	 * as we wait for an empty AIL as part of the unmount process.
3443 	 */
3444 	if (XFS_FORCED_SHUTDOWN(mp)) {
3445 		error = -EIO;
3446 		goto abort_out;
3447 	}
3448 
3449 	/*
3450 	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3451 	 * operation here, so we may get  an EAGAIN error. In that case, we
3452 	 * simply want to return with the inode still dirty.
3453 	 *
3454 	 * If we get any other error, we effectively have a corruption situation
3455 	 * and we cannot flush the inode, so we treat it the same as failing
3456 	 * xfs_iflush_int().
3457 	 */
3458 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3459 			       0);
3460 	if (error == -EAGAIN) {
3461 		xfs_ifunlock(ip);
3462 		return error;
3463 	}
3464 	if (error)
3465 		goto corrupt_out;
3466 
3467 	/*
3468 	 * First flush out the inode that xfs_iflush was called with.
3469 	 */
3470 	error = xfs_iflush_int(ip, bp);
3471 	if (error)
3472 		goto corrupt_out;
3473 
3474 	/*
3475 	 * If the buffer is pinned then push on the log now so we won't
3476 	 * get stuck waiting in the write for too long.
3477 	 */
3478 	if (xfs_buf_ispinned(bp))
3479 		xfs_log_force(mp, 0);
3480 
3481 	/*
3482 	 * inode clustering:
3483 	 * see if other inodes can be gathered into this write
3484 	 */
3485 	error = xfs_iflush_cluster(ip, bp);
3486 	if (error)
3487 		goto cluster_corrupt_out;
3488 
3489 	*bpp = bp;
3490 	return 0;
3491 
3492 corrupt_out:
3493 	if (bp)
3494 		xfs_buf_relse(bp);
3495 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3496 cluster_corrupt_out:
3497 	error = -EFSCORRUPTED;
3498 abort_out:
3499 	/*
3500 	 * Unlocks the flush lock
3501 	 */
3502 	xfs_iflush_abort(ip, false);
3503 	return error;
3504 }
3505 
3506 /*
3507  * If there are inline format data / attr forks attached to this inode,
3508  * make sure they're not corrupt.
3509  */
3510 bool
3511 xfs_inode_verify_forks(
3512 	struct xfs_inode	*ip)
3513 {
3514 	struct xfs_ifork	*ifp;
3515 	xfs_failaddr_t		fa;
3516 
3517 	fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops);
3518 	if (fa) {
3519 		ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3520 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
3521 				ifp->if_u1.if_data, ifp->if_bytes, fa);
3522 		return false;
3523 	}
3524 
3525 	fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops);
3526 	if (fa) {
3527 		ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
3528 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
3529 				ifp ? ifp->if_u1.if_data : NULL,
3530 				ifp ? ifp->if_bytes : 0, fa);
3531 		return false;
3532 	}
3533 	return true;
3534 }
3535 
3536 STATIC int
3537 xfs_iflush_int(
3538 	struct xfs_inode	*ip,
3539 	struct xfs_buf		*bp)
3540 {
3541 	struct xfs_inode_log_item *iip = ip->i_itemp;
3542 	struct xfs_dinode	*dip;
3543 	struct xfs_mount	*mp = ip->i_mount;
3544 
3545 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3546 	ASSERT(xfs_isiflocked(ip));
3547 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3548 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3549 	ASSERT(iip != NULL && iip->ili_fields != 0);
3550 	ASSERT(ip->i_d.di_version > 1);
3551 
3552 	/* set *dip = inode's place in the buffer */
3553 	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3554 
3555 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3556 			       mp, XFS_ERRTAG_IFLUSH_1)) {
3557 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3558 			"%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3559 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3560 		goto corrupt_out;
3561 	}
3562 	if (S_ISREG(VFS_I(ip)->i_mode)) {
3563 		if (XFS_TEST_ERROR(
3564 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3565 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3566 		    mp, XFS_ERRTAG_IFLUSH_3)) {
3567 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3568 				"%s: Bad regular inode %Lu, ptr "PTR_FMT,
3569 				__func__, ip->i_ino, ip);
3570 			goto corrupt_out;
3571 		}
3572 	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3573 		if (XFS_TEST_ERROR(
3574 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3575 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3576 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3577 		    mp, XFS_ERRTAG_IFLUSH_4)) {
3578 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3579 				"%s: Bad directory inode %Lu, ptr "PTR_FMT,
3580 				__func__, ip->i_ino, ip);
3581 			goto corrupt_out;
3582 		}
3583 	}
3584 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3585 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3586 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3587 			"%s: detected corrupt incore inode %Lu, "
3588 			"total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3589 			__func__, ip->i_ino,
3590 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3591 			ip->i_d.di_nblocks, ip);
3592 		goto corrupt_out;
3593 	}
3594 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3595 				mp, XFS_ERRTAG_IFLUSH_6)) {
3596 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3597 			"%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3598 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3599 		goto corrupt_out;
3600 	}
3601 
3602 	/*
3603 	 * Inode item log recovery for v2 inodes are dependent on the
3604 	 * di_flushiter count for correct sequencing. We bump the flush
3605 	 * iteration count so we can detect flushes which postdate a log record
3606 	 * during recovery. This is redundant as we now log every change and
3607 	 * hence this can't happen but we need to still do it to ensure
3608 	 * backwards compatibility with old kernels that predate logging all
3609 	 * inode changes.
3610 	 */
3611 	if (ip->i_d.di_version < 3)
3612 		ip->i_d.di_flushiter++;
3613 
3614 	/* Check the inline fork data before we write out. */
3615 	if (!xfs_inode_verify_forks(ip))
3616 		goto corrupt_out;
3617 
3618 	/*
3619 	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3620 	 * copy out the core of the inode, because if the inode is dirty at all
3621 	 * the core must be.
3622 	 */
3623 	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3624 
3625 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3626 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3627 		ip->i_d.di_flushiter = 0;
3628 
3629 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3630 	if (XFS_IFORK_Q(ip))
3631 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3632 	xfs_inobp_check(mp, bp);
3633 
3634 	/*
3635 	 * We've recorded everything logged in the inode, so we'd like to clear
3636 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3637 	 * However, we can't stop logging all this information until the data
3638 	 * we've copied into the disk buffer is written to disk.  If we did we
3639 	 * might overwrite the copy of the inode in the log with all the data
3640 	 * after re-logging only part of it, and in the face of a crash we
3641 	 * wouldn't have all the data we need to recover.
3642 	 *
3643 	 * What we do is move the bits to the ili_last_fields field.  When
3644 	 * logging the inode, these bits are moved back to the ili_fields field.
3645 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3646 	 * know that the information those bits represent is permanently on
3647 	 * disk.  As long as the flush completes before the inode is logged
3648 	 * again, then both ili_fields and ili_last_fields will be cleared.
3649 	 *
3650 	 * We can play with the ili_fields bits here, because the inode lock
3651 	 * must be held exclusively in order to set bits there and the flush
3652 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3653 	 * done routine can tell whether or not to look in the AIL.  Also, store
3654 	 * the current LSN of the inode so that we can tell whether the item has
3655 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3656 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3657 	 * atomically.
3658 	 */
3659 	iip->ili_last_fields = iip->ili_fields;
3660 	iip->ili_fields = 0;
3661 	iip->ili_fsync_fields = 0;
3662 	iip->ili_logged = 1;
3663 
3664 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3665 				&iip->ili_item.li_lsn);
3666 
3667 	/*
3668 	 * Attach the function xfs_iflush_done to the inode's
3669 	 * buffer.  This will remove the inode from the AIL
3670 	 * and unlock the inode's flush lock when the inode is
3671 	 * completely written to disk.
3672 	 */
3673 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3674 
3675 	/* generate the checksum. */
3676 	xfs_dinode_calc_crc(mp, dip);
3677 
3678 	ASSERT(!list_empty(&bp->b_li_list));
3679 	ASSERT(bp->b_iodone != NULL);
3680 	return 0;
3681 
3682 corrupt_out:
3683 	return -EFSCORRUPTED;
3684 }
3685