xref: /openbmc/linux/fs/xfs/xfs_inode.c (revision d2ba09c1)
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 {
978 	xfs_trans_t	*tp;
979 	xfs_inode_t	*ip;
980 	xfs_buf_t	*ialloc_context = NULL;
981 	int		code;
982 	void		*dqinfo;
983 	uint		tflags;
984 
985 	tp = *tpp;
986 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
987 
988 	/*
989 	 * xfs_ialloc will return a pointer to an incore inode if
990 	 * the Space Manager has an available inode on the free
991 	 * list. Otherwise, it will do an allocation and replenish
992 	 * the freelist.  Since we can only do one allocation per
993 	 * transaction without deadlocks, we will need to commit the
994 	 * current transaction and start a new one.  We will then
995 	 * need to call xfs_ialloc again to get the inode.
996 	 *
997 	 * If xfs_ialloc did an allocation to replenish the freelist,
998 	 * it returns the bp containing the head of the freelist as
999 	 * ialloc_context. We will hold a lock on it across the
1000 	 * transaction commit so that no other process can steal
1001 	 * the inode(s) that we've just allocated.
1002 	 */
1003 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
1004 			&ip);
1005 
1006 	/*
1007 	 * Return an error if we were unable to allocate a new inode.
1008 	 * This should only happen if we run out of space on disk or
1009 	 * encounter a disk error.
1010 	 */
1011 	if (code) {
1012 		*ipp = NULL;
1013 		return code;
1014 	}
1015 	if (!ialloc_context && !ip) {
1016 		*ipp = NULL;
1017 		return -ENOSPC;
1018 	}
1019 
1020 	/*
1021 	 * If the AGI buffer is non-NULL, then we were unable to get an
1022 	 * inode in one operation.  We need to commit the current
1023 	 * transaction and call xfs_ialloc() again.  It is guaranteed
1024 	 * to succeed the second time.
1025 	 */
1026 	if (ialloc_context) {
1027 		/*
1028 		 * Normally, xfs_trans_commit releases all the locks.
1029 		 * We call bhold to hang on to the ialloc_context across
1030 		 * the commit.  Holding this buffer prevents any other
1031 		 * processes from doing any allocations in this
1032 		 * allocation group.
1033 		 */
1034 		xfs_trans_bhold(tp, ialloc_context);
1035 
1036 		/*
1037 		 * We want the quota changes to be associated with the next
1038 		 * transaction, NOT this one. So, detach the dqinfo from this
1039 		 * and attach it to the next transaction.
1040 		 */
1041 		dqinfo = NULL;
1042 		tflags = 0;
1043 		if (tp->t_dqinfo) {
1044 			dqinfo = (void *)tp->t_dqinfo;
1045 			tp->t_dqinfo = NULL;
1046 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1047 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1048 		}
1049 
1050 		code = xfs_trans_roll(&tp);
1051 
1052 		/*
1053 		 * Re-attach the quota info that we detached from prev trx.
1054 		 */
1055 		if (dqinfo) {
1056 			tp->t_dqinfo = dqinfo;
1057 			tp->t_flags |= tflags;
1058 		}
1059 
1060 		if (code) {
1061 			xfs_buf_relse(ialloc_context);
1062 			*tpp = tp;
1063 			*ipp = NULL;
1064 			return code;
1065 		}
1066 		xfs_trans_bjoin(tp, ialloc_context);
1067 
1068 		/*
1069 		 * Call ialloc again. Since we've locked out all
1070 		 * other allocations in this allocation group,
1071 		 * this call should always succeed.
1072 		 */
1073 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1074 				  &ialloc_context, &ip);
1075 
1076 		/*
1077 		 * If we get an error at this point, return to the caller
1078 		 * so that the current transaction can be aborted.
1079 		 */
1080 		if (code) {
1081 			*tpp = tp;
1082 			*ipp = NULL;
1083 			return code;
1084 		}
1085 		ASSERT(!ialloc_context && ip);
1086 
1087 	}
1088 
1089 	*ipp = ip;
1090 	*tpp = tp;
1091 
1092 	return 0;
1093 }
1094 
1095 /*
1096  * Decrement the link count on an inode & log the change.  If this causes the
1097  * link count to go to zero, move the inode to AGI unlinked list so that it can
1098  * be freed when the last active reference goes away via xfs_inactive().
1099  */
1100 static int			/* error */
1101 xfs_droplink(
1102 	xfs_trans_t *tp,
1103 	xfs_inode_t *ip)
1104 {
1105 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1106 
1107 	drop_nlink(VFS_I(ip));
1108 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1109 
1110 	if (VFS_I(ip)->i_nlink)
1111 		return 0;
1112 
1113 	return xfs_iunlink(tp, ip);
1114 }
1115 
1116 /*
1117  * Increment the link count on an inode & log the change.
1118  */
1119 static int
1120 xfs_bumplink(
1121 	xfs_trans_t *tp,
1122 	xfs_inode_t *ip)
1123 {
1124 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1125 
1126 	ASSERT(ip->i_d.di_version > 1);
1127 	inc_nlink(VFS_I(ip));
1128 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1129 	return 0;
1130 }
1131 
1132 int
1133 xfs_create(
1134 	xfs_inode_t		*dp,
1135 	struct xfs_name		*name,
1136 	umode_t			mode,
1137 	dev_t			rdev,
1138 	xfs_inode_t		**ipp)
1139 {
1140 	int			is_dir = S_ISDIR(mode);
1141 	struct xfs_mount	*mp = dp->i_mount;
1142 	struct xfs_inode	*ip = NULL;
1143 	struct xfs_trans	*tp = NULL;
1144 	int			error;
1145 	struct xfs_defer_ops	dfops;
1146 	xfs_fsblock_t		first_block;
1147 	bool                    unlock_dp_on_error = false;
1148 	prid_t			prid;
1149 	struct xfs_dquot	*udqp = NULL;
1150 	struct xfs_dquot	*gdqp = NULL;
1151 	struct xfs_dquot	*pdqp = NULL;
1152 	struct xfs_trans_res	*tres;
1153 	uint			resblks;
1154 
1155 	trace_xfs_create(dp, name);
1156 
1157 	if (XFS_FORCED_SHUTDOWN(mp))
1158 		return -EIO;
1159 
1160 	prid = xfs_get_initial_prid(dp);
1161 
1162 	/*
1163 	 * Make sure that we have allocated dquot(s) on disk.
1164 	 */
1165 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1166 					xfs_kgid_to_gid(current_fsgid()), prid,
1167 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1168 					&udqp, &gdqp, &pdqp);
1169 	if (error)
1170 		return error;
1171 
1172 	if (is_dir) {
1173 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1174 		tres = &M_RES(mp)->tr_mkdir;
1175 	} else {
1176 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1177 		tres = &M_RES(mp)->tr_create;
1178 	}
1179 
1180 	/*
1181 	 * Initially assume that the file does not exist and
1182 	 * reserve the resources for that case.  If that is not
1183 	 * the case we'll drop the one we have and get a more
1184 	 * appropriate transaction later.
1185 	 */
1186 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1187 	if (error == -ENOSPC) {
1188 		/* flush outstanding delalloc blocks and retry */
1189 		xfs_flush_inodes(mp);
1190 		error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1191 	}
1192 	if (error)
1193 		goto out_release_inode;
1194 
1195 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1196 	unlock_dp_on_error = true;
1197 
1198 	xfs_defer_init(&dfops, &first_block);
1199 
1200 	/*
1201 	 * Reserve disk quota and the inode.
1202 	 */
1203 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1204 						pdqp, resblks, 1, 0);
1205 	if (error)
1206 		goto out_trans_cancel;
1207 
1208 	/*
1209 	 * A newly created regular or special file just has one directory
1210 	 * entry pointing to them, but a directory also the "." entry
1211 	 * pointing to itself.
1212 	 */
1213 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
1214 	if (error)
1215 		goto out_trans_cancel;
1216 
1217 	/*
1218 	 * Now we join the directory inode to the transaction.  We do not do it
1219 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1220 	 * (and release all the locks).  An error from here on will result in
1221 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1222 	 * error path.
1223 	 */
1224 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1225 	unlock_dp_on_error = false;
1226 
1227 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1228 					&first_block, &dfops, resblks ?
1229 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1230 	if (error) {
1231 		ASSERT(error != -ENOSPC);
1232 		goto out_trans_cancel;
1233 	}
1234 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1235 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1236 
1237 	if (is_dir) {
1238 		error = xfs_dir_init(tp, ip, dp);
1239 		if (error)
1240 			goto out_bmap_cancel;
1241 
1242 		error = xfs_bumplink(tp, dp);
1243 		if (error)
1244 			goto out_bmap_cancel;
1245 	}
1246 
1247 	/*
1248 	 * If this is a synchronous mount, make sure that the
1249 	 * create transaction goes to disk before returning to
1250 	 * the user.
1251 	 */
1252 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1253 		xfs_trans_set_sync(tp);
1254 
1255 	/*
1256 	 * Attach the dquot(s) to the inodes and modify them incore.
1257 	 * These ids of the inode couldn't have changed since the new
1258 	 * inode has been locked ever since it was created.
1259 	 */
1260 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1261 
1262 	error = xfs_defer_finish(&tp, &dfops);
1263 	if (error)
1264 		goto out_bmap_cancel;
1265 
1266 	error = xfs_trans_commit(tp);
1267 	if (error)
1268 		goto out_release_inode;
1269 
1270 	xfs_qm_dqrele(udqp);
1271 	xfs_qm_dqrele(gdqp);
1272 	xfs_qm_dqrele(pdqp);
1273 
1274 	*ipp = ip;
1275 	return 0;
1276 
1277  out_bmap_cancel:
1278 	xfs_defer_cancel(&dfops);
1279  out_trans_cancel:
1280 	xfs_trans_cancel(tp);
1281  out_release_inode:
1282 	/*
1283 	 * Wait until after the current transaction is aborted to finish the
1284 	 * setup of the inode and release the inode.  This prevents recursive
1285 	 * transactions and deadlocks from xfs_inactive.
1286 	 */
1287 	if (ip) {
1288 		xfs_finish_inode_setup(ip);
1289 		IRELE(ip);
1290 	}
1291 
1292 	xfs_qm_dqrele(udqp);
1293 	xfs_qm_dqrele(gdqp);
1294 	xfs_qm_dqrele(pdqp);
1295 
1296 	if (unlock_dp_on_error)
1297 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1298 	return error;
1299 }
1300 
1301 int
1302 xfs_create_tmpfile(
1303 	struct xfs_inode	*dp,
1304 	umode_t			mode,
1305 	struct xfs_inode	**ipp)
1306 {
1307 	struct xfs_mount	*mp = dp->i_mount;
1308 	struct xfs_inode	*ip = NULL;
1309 	struct xfs_trans	*tp = NULL;
1310 	int			error;
1311 	prid_t                  prid;
1312 	struct xfs_dquot	*udqp = NULL;
1313 	struct xfs_dquot	*gdqp = NULL;
1314 	struct xfs_dquot	*pdqp = NULL;
1315 	struct xfs_trans_res	*tres;
1316 	uint			resblks;
1317 
1318 	if (XFS_FORCED_SHUTDOWN(mp))
1319 		return -EIO;
1320 
1321 	prid = xfs_get_initial_prid(dp);
1322 
1323 	/*
1324 	 * Make sure that we have allocated dquot(s) on disk.
1325 	 */
1326 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1327 				xfs_kgid_to_gid(current_fsgid()), prid,
1328 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1329 				&udqp, &gdqp, &pdqp);
1330 	if (error)
1331 		return error;
1332 
1333 	resblks = XFS_IALLOC_SPACE_RES(mp);
1334 	tres = &M_RES(mp)->tr_create_tmpfile;
1335 
1336 	error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1337 	if (error)
1338 		goto out_release_inode;
1339 
1340 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1341 						pdqp, resblks, 1, 0);
1342 	if (error)
1343 		goto out_trans_cancel;
1344 
1345 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip);
1346 	if (error)
1347 		goto out_trans_cancel;
1348 
1349 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1350 		xfs_trans_set_sync(tp);
1351 
1352 	/*
1353 	 * Attach the dquot(s) to the inodes and modify them incore.
1354 	 * These ids of the inode couldn't have changed since the new
1355 	 * inode has been locked ever since it was created.
1356 	 */
1357 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1358 
1359 	error = xfs_iunlink(tp, ip);
1360 	if (error)
1361 		goto out_trans_cancel;
1362 
1363 	error = xfs_trans_commit(tp);
1364 	if (error)
1365 		goto out_release_inode;
1366 
1367 	xfs_qm_dqrele(udqp);
1368 	xfs_qm_dqrele(gdqp);
1369 	xfs_qm_dqrele(pdqp);
1370 
1371 	*ipp = ip;
1372 	return 0;
1373 
1374  out_trans_cancel:
1375 	xfs_trans_cancel(tp);
1376  out_release_inode:
1377 	/*
1378 	 * Wait until after the current transaction is aborted to finish the
1379 	 * setup of the inode and release the inode.  This prevents recursive
1380 	 * transactions and deadlocks from xfs_inactive.
1381 	 */
1382 	if (ip) {
1383 		xfs_finish_inode_setup(ip);
1384 		IRELE(ip);
1385 	}
1386 
1387 	xfs_qm_dqrele(udqp);
1388 	xfs_qm_dqrele(gdqp);
1389 	xfs_qm_dqrele(pdqp);
1390 
1391 	return error;
1392 }
1393 
1394 int
1395 xfs_link(
1396 	xfs_inode_t		*tdp,
1397 	xfs_inode_t		*sip,
1398 	struct xfs_name		*target_name)
1399 {
1400 	xfs_mount_t		*mp = tdp->i_mount;
1401 	xfs_trans_t		*tp;
1402 	int			error;
1403 	struct xfs_defer_ops	dfops;
1404 	xfs_fsblock_t           first_block;
1405 	int			resblks;
1406 
1407 	trace_xfs_link(tdp, target_name);
1408 
1409 	ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1410 
1411 	if (XFS_FORCED_SHUTDOWN(mp))
1412 		return -EIO;
1413 
1414 	error = xfs_qm_dqattach(sip, 0);
1415 	if (error)
1416 		goto std_return;
1417 
1418 	error = xfs_qm_dqattach(tdp, 0);
1419 	if (error)
1420 		goto std_return;
1421 
1422 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1423 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1424 	if (error == -ENOSPC) {
1425 		resblks = 0;
1426 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1427 	}
1428 	if (error)
1429 		goto std_return;
1430 
1431 	xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1432 
1433 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1434 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1435 
1436 	/*
1437 	 * If we are using project inheritance, we only allow hard link
1438 	 * creation in our tree when the project IDs are the same; else
1439 	 * the tree quota mechanism could be circumvented.
1440 	 */
1441 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1442 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1443 		error = -EXDEV;
1444 		goto error_return;
1445 	}
1446 
1447 	if (!resblks) {
1448 		error = xfs_dir_canenter(tp, tdp, target_name);
1449 		if (error)
1450 			goto error_return;
1451 	}
1452 
1453 	xfs_defer_init(&dfops, &first_block);
1454 
1455 	/*
1456 	 * Handle initial link state of O_TMPFILE inode
1457 	 */
1458 	if (VFS_I(sip)->i_nlink == 0) {
1459 		error = xfs_iunlink_remove(tp, sip);
1460 		if (error)
1461 			goto error_return;
1462 	}
1463 
1464 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1465 					&first_block, &dfops, resblks);
1466 	if (error)
1467 		goto error_return;
1468 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1469 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1470 
1471 	error = xfs_bumplink(tp, sip);
1472 	if (error)
1473 		goto error_return;
1474 
1475 	/*
1476 	 * If this is a synchronous mount, make sure that the
1477 	 * link transaction goes to disk before returning to
1478 	 * the user.
1479 	 */
1480 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1481 		xfs_trans_set_sync(tp);
1482 
1483 	error = xfs_defer_finish(&tp, &dfops);
1484 	if (error) {
1485 		xfs_defer_cancel(&dfops);
1486 		goto error_return;
1487 	}
1488 
1489 	return xfs_trans_commit(tp);
1490 
1491  error_return:
1492 	xfs_trans_cancel(tp);
1493  std_return:
1494 	return error;
1495 }
1496 
1497 /* Clear the reflink flag and the cowblocks tag if possible. */
1498 static void
1499 xfs_itruncate_clear_reflink_flags(
1500 	struct xfs_inode	*ip)
1501 {
1502 	struct xfs_ifork	*dfork;
1503 	struct xfs_ifork	*cfork;
1504 
1505 	if (!xfs_is_reflink_inode(ip))
1506 		return;
1507 	dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1508 	cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1509 	if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1510 		ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1511 	if (cfork->if_bytes == 0)
1512 		xfs_inode_clear_cowblocks_tag(ip);
1513 }
1514 
1515 /*
1516  * Free up the underlying blocks past new_size.  The new size must be smaller
1517  * than the current size.  This routine can be used both for the attribute and
1518  * data fork, and does not modify the inode size, which is left to the caller.
1519  *
1520  * The transaction passed to this routine must have made a permanent log
1521  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1522  * given transaction and start new ones, so make sure everything involved in
1523  * the transaction is tidy before calling here.  Some transaction will be
1524  * returned to the caller to be committed.  The incoming transaction must
1525  * already include the inode, and both inode locks must be held exclusively.
1526  * The inode must also be "held" within the transaction.  On return the inode
1527  * will be "held" within the returned transaction.  This routine does NOT
1528  * require any disk space to be reserved for it within the transaction.
1529  *
1530  * If we get an error, we must return with the inode locked and linked into the
1531  * current transaction. This keeps things simple for the higher level code,
1532  * because it always knows that the inode is locked and held in the transaction
1533  * that returns to it whether errors occur or not.  We don't mark the inode
1534  * dirty on error so that transactions can be easily aborted if possible.
1535  */
1536 int
1537 xfs_itruncate_extents(
1538 	struct xfs_trans	**tpp,
1539 	struct xfs_inode	*ip,
1540 	int			whichfork,
1541 	xfs_fsize_t		new_size)
1542 {
1543 	struct xfs_mount	*mp = ip->i_mount;
1544 	struct xfs_trans	*tp = *tpp;
1545 	struct xfs_defer_ops	dfops;
1546 	xfs_fsblock_t		first_block;
1547 	xfs_fileoff_t		first_unmap_block;
1548 	xfs_fileoff_t		last_block;
1549 	xfs_filblks_t		unmap_len;
1550 	int			error = 0;
1551 	int			done = 0;
1552 
1553 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1554 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1555 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1556 	ASSERT(new_size <= XFS_ISIZE(ip));
1557 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1558 	ASSERT(ip->i_itemp != NULL);
1559 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1560 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1561 
1562 	trace_xfs_itruncate_extents_start(ip, new_size);
1563 
1564 	/*
1565 	 * Since it is possible for space to become allocated beyond
1566 	 * the end of the file (in a crash where the space is allocated
1567 	 * but the inode size is not yet updated), simply remove any
1568 	 * blocks which show up between the new EOF and the maximum
1569 	 * possible file size.  If the first block to be removed is
1570 	 * beyond the maximum file size (ie it is the same as last_block),
1571 	 * then there is nothing to do.
1572 	 */
1573 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1574 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1575 	if (first_unmap_block == last_block)
1576 		return 0;
1577 
1578 	ASSERT(first_unmap_block < last_block);
1579 	unmap_len = last_block - first_unmap_block + 1;
1580 	while (!done) {
1581 		xfs_defer_init(&dfops, &first_block);
1582 		error = xfs_bunmapi(tp, ip,
1583 				    first_unmap_block, unmap_len,
1584 				    xfs_bmapi_aflag(whichfork),
1585 				    XFS_ITRUNC_MAX_EXTENTS,
1586 				    &first_block, &dfops,
1587 				    &done);
1588 		if (error)
1589 			goto out_bmap_cancel;
1590 
1591 		/*
1592 		 * Duplicate the transaction that has the permanent
1593 		 * reservation and commit the old transaction.
1594 		 */
1595 		xfs_defer_ijoin(&dfops, ip);
1596 		error = xfs_defer_finish(&tp, &dfops);
1597 		if (error)
1598 			goto out_bmap_cancel;
1599 
1600 		error = xfs_trans_roll_inode(&tp, ip);
1601 		if (error)
1602 			goto out;
1603 	}
1604 
1605 	if (whichfork == XFS_DATA_FORK) {
1606 		/* Remove all pending CoW reservations. */
1607 		error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1608 				first_unmap_block, last_block, true);
1609 		if (error)
1610 			goto out;
1611 
1612 		xfs_itruncate_clear_reflink_flags(ip);
1613 	}
1614 
1615 	/*
1616 	 * Always re-log the inode so that our permanent transaction can keep
1617 	 * on rolling it forward in the log.
1618 	 */
1619 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1620 
1621 	trace_xfs_itruncate_extents_end(ip, new_size);
1622 
1623 out:
1624 	*tpp = tp;
1625 	return error;
1626 out_bmap_cancel:
1627 	/*
1628 	 * If the bunmapi call encounters an error, return to the caller where
1629 	 * the transaction can be properly aborted.  We just need to make sure
1630 	 * we're not holding any resources that we were not when we came in.
1631 	 */
1632 	xfs_defer_cancel(&dfops);
1633 	goto out;
1634 }
1635 
1636 int
1637 xfs_release(
1638 	xfs_inode_t	*ip)
1639 {
1640 	xfs_mount_t	*mp = ip->i_mount;
1641 	int		error;
1642 
1643 	if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1644 		return 0;
1645 
1646 	/* If this is a read-only mount, don't do this (would generate I/O) */
1647 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1648 		return 0;
1649 
1650 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1651 		int truncated;
1652 
1653 		/*
1654 		 * If we previously truncated this file and removed old data
1655 		 * in the process, we want to initiate "early" writeout on
1656 		 * the last close.  This is an attempt to combat the notorious
1657 		 * NULL files problem which is particularly noticeable from a
1658 		 * truncate down, buffered (re-)write (delalloc), followed by
1659 		 * a crash.  What we are effectively doing here is
1660 		 * significantly reducing the time window where we'd otherwise
1661 		 * be exposed to that problem.
1662 		 */
1663 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1664 		if (truncated) {
1665 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1666 			if (ip->i_delayed_blks > 0) {
1667 				error = filemap_flush(VFS_I(ip)->i_mapping);
1668 				if (error)
1669 					return error;
1670 			}
1671 		}
1672 	}
1673 
1674 	if (VFS_I(ip)->i_nlink == 0)
1675 		return 0;
1676 
1677 	if (xfs_can_free_eofblocks(ip, false)) {
1678 
1679 		/*
1680 		 * Check if the inode is being opened, written and closed
1681 		 * frequently and we have delayed allocation blocks outstanding
1682 		 * (e.g. streaming writes from the NFS server), truncating the
1683 		 * blocks past EOF will cause fragmentation to occur.
1684 		 *
1685 		 * In this case don't do the truncation, but we have to be
1686 		 * careful how we detect this case. Blocks beyond EOF show up as
1687 		 * i_delayed_blks even when the inode is clean, so we need to
1688 		 * truncate them away first before checking for a dirty release.
1689 		 * Hence on the first dirty close we will still remove the
1690 		 * speculative allocation, but after that we will leave it in
1691 		 * place.
1692 		 */
1693 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1694 			return 0;
1695 		/*
1696 		 * If we can't get the iolock just skip truncating the blocks
1697 		 * past EOF because we could deadlock with the mmap_sem
1698 		 * otherwise. We'll get another chance to drop them once the
1699 		 * last reference to the inode is dropped, so we'll never leak
1700 		 * blocks permanently.
1701 		 */
1702 		if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1703 			error = xfs_free_eofblocks(ip);
1704 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1705 			if (error)
1706 				return error;
1707 		}
1708 
1709 		/* delalloc blocks after truncation means it really is dirty */
1710 		if (ip->i_delayed_blks)
1711 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1712 	}
1713 	return 0;
1714 }
1715 
1716 /*
1717  * xfs_inactive_truncate
1718  *
1719  * Called to perform a truncate when an inode becomes unlinked.
1720  */
1721 STATIC int
1722 xfs_inactive_truncate(
1723 	struct xfs_inode *ip)
1724 {
1725 	struct xfs_mount	*mp = ip->i_mount;
1726 	struct xfs_trans	*tp;
1727 	int			error;
1728 
1729 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1730 	if (error) {
1731 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1732 		return error;
1733 	}
1734 
1735 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1736 	xfs_trans_ijoin(tp, ip, 0);
1737 
1738 	/*
1739 	 * Log the inode size first to prevent stale data exposure in the event
1740 	 * of a system crash before the truncate completes. See the related
1741 	 * comment in xfs_vn_setattr_size() for details.
1742 	 */
1743 	ip->i_d.di_size = 0;
1744 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1745 
1746 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1747 	if (error)
1748 		goto error_trans_cancel;
1749 
1750 	ASSERT(ip->i_d.di_nextents == 0);
1751 
1752 	error = xfs_trans_commit(tp);
1753 	if (error)
1754 		goto error_unlock;
1755 
1756 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1757 	return 0;
1758 
1759 error_trans_cancel:
1760 	xfs_trans_cancel(tp);
1761 error_unlock:
1762 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1763 	return error;
1764 }
1765 
1766 /*
1767  * xfs_inactive_ifree()
1768  *
1769  * Perform the inode free when an inode is unlinked.
1770  */
1771 STATIC int
1772 xfs_inactive_ifree(
1773 	struct xfs_inode *ip)
1774 {
1775 	struct xfs_defer_ops	dfops;
1776 	xfs_fsblock_t		first_block;
1777 	struct xfs_mount	*mp = ip->i_mount;
1778 	struct xfs_trans	*tp;
1779 	int			error;
1780 
1781 	/*
1782 	 * We try to use a per-AG reservation for any block needed by the finobt
1783 	 * tree, but as the finobt feature predates the per-AG reservation
1784 	 * support a degraded file system might not have enough space for the
1785 	 * reservation at mount time.  In that case try to dip into the reserved
1786 	 * pool and pray.
1787 	 *
1788 	 * Send a warning if the reservation does happen to fail, as the inode
1789 	 * now remains allocated and sits on the unlinked list until the fs is
1790 	 * repaired.
1791 	 */
1792 	if (unlikely(mp->m_inotbt_nores)) {
1793 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1794 				XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1795 				&tp);
1796 	} else {
1797 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1798 	}
1799 	if (error) {
1800 		if (error == -ENOSPC) {
1801 			xfs_warn_ratelimited(mp,
1802 			"Failed to remove inode(s) from unlinked list. "
1803 			"Please free space, unmount and run xfs_repair.");
1804 		} else {
1805 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1806 		}
1807 		return error;
1808 	}
1809 
1810 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1811 	xfs_trans_ijoin(tp, ip, 0);
1812 
1813 	xfs_defer_init(&dfops, &first_block);
1814 	error = xfs_ifree(tp, ip, &dfops);
1815 	if (error) {
1816 		/*
1817 		 * If we fail to free the inode, shut down.  The cancel
1818 		 * might do that, we need to make sure.  Otherwise the
1819 		 * inode might be lost for a long time or forever.
1820 		 */
1821 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1822 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1823 				__func__, error);
1824 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1825 		}
1826 		xfs_trans_cancel(tp);
1827 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1828 		return error;
1829 	}
1830 
1831 	/*
1832 	 * Credit the quota account(s). The inode is gone.
1833 	 */
1834 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1835 
1836 	/*
1837 	 * Just ignore errors at this point.  There is nothing we can do except
1838 	 * to try to keep going. Make sure it's not a silent error.
1839 	 */
1840 	error = xfs_defer_finish(&tp, &dfops);
1841 	if (error) {
1842 		xfs_notice(mp, "%s: xfs_defer_finish returned error %d",
1843 			__func__, error);
1844 		xfs_defer_cancel(&dfops);
1845 	}
1846 	error = xfs_trans_commit(tp);
1847 	if (error)
1848 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1849 			__func__, error);
1850 
1851 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1852 	return 0;
1853 }
1854 
1855 /*
1856  * xfs_inactive
1857  *
1858  * This is called when the vnode reference count for the vnode
1859  * goes to zero.  If the file has been unlinked, then it must
1860  * now be truncated.  Also, we clear all of the read-ahead state
1861  * kept for the inode here since the file is now closed.
1862  */
1863 void
1864 xfs_inactive(
1865 	xfs_inode_t	*ip)
1866 {
1867 	struct xfs_mount	*mp;
1868 	struct xfs_ifork	*cow_ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1869 	int			error;
1870 	int			truncate = 0;
1871 
1872 	/*
1873 	 * If the inode is already free, then there can be nothing
1874 	 * to clean up here.
1875 	 */
1876 	if (VFS_I(ip)->i_mode == 0) {
1877 		ASSERT(ip->i_df.if_real_bytes == 0);
1878 		ASSERT(ip->i_df.if_broot_bytes == 0);
1879 		return;
1880 	}
1881 
1882 	mp = ip->i_mount;
1883 	ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1884 
1885 	/* If this is a read-only mount, don't do this (would generate I/O) */
1886 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1887 		return;
1888 
1889 	/* Try to clean out the cow blocks if there are any. */
1890 	if (xfs_is_reflink_inode(ip) && cow_ifp->if_bytes > 0)
1891 		xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1892 
1893 	if (VFS_I(ip)->i_nlink != 0) {
1894 		/*
1895 		 * force is true because we are evicting an inode from the
1896 		 * cache. Post-eof blocks must be freed, lest we end up with
1897 		 * broken free space accounting.
1898 		 *
1899 		 * Note: don't bother with iolock here since lockdep complains
1900 		 * about acquiring it in reclaim context. We have the only
1901 		 * reference to the inode at this point anyways.
1902 		 */
1903 		if (xfs_can_free_eofblocks(ip, true))
1904 			xfs_free_eofblocks(ip);
1905 
1906 		return;
1907 	}
1908 
1909 	if (S_ISREG(VFS_I(ip)->i_mode) &&
1910 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1911 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1912 		truncate = 1;
1913 
1914 	error = xfs_qm_dqattach(ip, 0);
1915 	if (error)
1916 		return;
1917 
1918 	if (S_ISLNK(VFS_I(ip)->i_mode))
1919 		error = xfs_inactive_symlink(ip);
1920 	else if (truncate)
1921 		error = xfs_inactive_truncate(ip);
1922 	if (error)
1923 		return;
1924 
1925 	/*
1926 	 * If there are attributes associated with the file then blow them away
1927 	 * now.  The code calls a routine that recursively deconstructs the
1928 	 * attribute fork. If also blows away the in-core attribute fork.
1929 	 */
1930 	if (XFS_IFORK_Q(ip)) {
1931 		error = xfs_attr_inactive(ip);
1932 		if (error)
1933 			return;
1934 	}
1935 
1936 	ASSERT(!ip->i_afp);
1937 	ASSERT(ip->i_d.di_anextents == 0);
1938 	ASSERT(ip->i_d.di_forkoff == 0);
1939 
1940 	/*
1941 	 * Free the inode.
1942 	 */
1943 	error = xfs_inactive_ifree(ip);
1944 	if (error)
1945 		return;
1946 
1947 	/*
1948 	 * Release the dquots held by inode, if any.
1949 	 */
1950 	xfs_qm_dqdetach(ip);
1951 }
1952 
1953 /*
1954  * This is called when the inode's link count goes to 0 or we are creating a
1955  * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1956  * set to true as the link count is dropped to zero by the VFS after we've
1957  * created the file successfully, so we have to add it to the unlinked list
1958  * while the link count is non-zero.
1959  *
1960  * We place the on-disk inode on a list in the AGI.  It will be pulled from this
1961  * list when the inode is freed.
1962  */
1963 STATIC int
1964 xfs_iunlink(
1965 	struct xfs_trans *tp,
1966 	struct xfs_inode *ip)
1967 {
1968 	xfs_mount_t	*mp = tp->t_mountp;
1969 	xfs_agi_t	*agi;
1970 	xfs_dinode_t	*dip;
1971 	xfs_buf_t	*agibp;
1972 	xfs_buf_t	*ibp;
1973 	xfs_agino_t	agino;
1974 	short		bucket_index;
1975 	int		offset;
1976 	int		error;
1977 
1978 	ASSERT(VFS_I(ip)->i_mode != 0);
1979 
1980 	/*
1981 	 * Get the agi buffer first.  It ensures lock ordering
1982 	 * on the list.
1983 	 */
1984 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1985 	if (error)
1986 		return error;
1987 	agi = XFS_BUF_TO_AGI(agibp);
1988 
1989 	/*
1990 	 * Get the index into the agi hash table for the
1991 	 * list this inode will go on.
1992 	 */
1993 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1994 	ASSERT(agino != 0);
1995 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1996 	ASSERT(agi->agi_unlinked[bucket_index]);
1997 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1998 
1999 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2000 		/*
2001 		 * There is already another inode in the bucket we need
2002 		 * to add ourselves to.  Add us at the front of the list.
2003 		 * Here we put the head pointer into our next pointer,
2004 		 * and then we fall through to point the head at us.
2005 		 */
2006 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2007 				       0, 0);
2008 		if (error)
2009 			return error;
2010 
2011 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2012 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2013 		offset = ip->i_imap.im_boffset +
2014 			offsetof(xfs_dinode_t, di_next_unlinked);
2015 
2016 		/* need to recalc the inode CRC if appropriate */
2017 		xfs_dinode_calc_crc(mp, dip);
2018 
2019 		xfs_trans_inode_buf(tp, ibp);
2020 		xfs_trans_log_buf(tp, ibp, offset,
2021 				  (offset + sizeof(xfs_agino_t) - 1));
2022 		xfs_inobp_check(mp, ibp);
2023 	}
2024 
2025 	/*
2026 	 * Point the bucket head pointer at the inode being inserted.
2027 	 */
2028 	ASSERT(agino != 0);
2029 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2030 	offset = offsetof(xfs_agi_t, agi_unlinked) +
2031 		(sizeof(xfs_agino_t) * bucket_index);
2032 	xfs_trans_log_buf(tp, agibp, offset,
2033 			  (offset + sizeof(xfs_agino_t) - 1));
2034 	return 0;
2035 }
2036 
2037 /*
2038  * Pull the on-disk inode from the AGI unlinked list.
2039  */
2040 STATIC int
2041 xfs_iunlink_remove(
2042 	xfs_trans_t	*tp,
2043 	xfs_inode_t	*ip)
2044 {
2045 	xfs_ino_t	next_ino;
2046 	xfs_mount_t	*mp;
2047 	xfs_agi_t	*agi;
2048 	xfs_dinode_t	*dip;
2049 	xfs_buf_t	*agibp;
2050 	xfs_buf_t	*ibp;
2051 	xfs_agnumber_t	agno;
2052 	xfs_agino_t	agino;
2053 	xfs_agino_t	next_agino;
2054 	xfs_buf_t	*last_ibp;
2055 	xfs_dinode_t	*last_dip = NULL;
2056 	short		bucket_index;
2057 	int		offset, last_offset = 0;
2058 	int		error;
2059 
2060 	mp = tp->t_mountp;
2061 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2062 
2063 	/*
2064 	 * Get the agi buffer first.  It ensures lock ordering
2065 	 * on the list.
2066 	 */
2067 	error = xfs_read_agi(mp, tp, agno, &agibp);
2068 	if (error)
2069 		return error;
2070 
2071 	agi = XFS_BUF_TO_AGI(agibp);
2072 
2073 	/*
2074 	 * Get the index into the agi hash table for the
2075 	 * list this inode will go on.
2076 	 */
2077 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2078 	ASSERT(agino != 0);
2079 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2080 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2081 	ASSERT(agi->agi_unlinked[bucket_index]);
2082 
2083 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2084 		/*
2085 		 * We're at the head of the list.  Get the inode's on-disk
2086 		 * buffer to see if there is anyone after us on the list.
2087 		 * Only modify our next pointer if it is not already NULLAGINO.
2088 		 * This saves us the overhead of dealing with the buffer when
2089 		 * there is no need to change it.
2090 		 */
2091 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2092 				       0, 0);
2093 		if (error) {
2094 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2095 				__func__, error);
2096 			return error;
2097 		}
2098 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2099 		ASSERT(next_agino != 0);
2100 		if (next_agino != NULLAGINO) {
2101 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2102 			offset = ip->i_imap.im_boffset +
2103 				offsetof(xfs_dinode_t, di_next_unlinked);
2104 
2105 			/* need to recalc the inode CRC if appropriate */
2106 			xfs_dinode_calc_crc(mp, dip);
2107 
2108 			xfs_trans_inode_buf(tp, ibp);
2109 			xfs_trans_log_buf(tp, ibp, offset,
2110 					  (offset + sizeof(xfs_agino_t) - 1));
2111 			xfs_inobp_check(mp, ibp);
2112 		} else {
2113 			xfs_trans_brelse(tp, ibp);
2114 		}
2115 		/*
2116 		 * Point the bucket head pointer at the next inode.
2117 		 */
2118 		ASSERT(next_agino != 0);
2119 		ASSERT(next_agino != agino);
2120 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2121 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2122 			(sizeof(xfs_agino_t) * bucket_index);
2123 		xfs_trans_log_buf(tp, agibp, offset,
2124 				  (offset + sizeof(xfs_agino_t) - 1));
2125 	} else {
2126 		/*
2127 		 * We need to search the list for the inode being freed.
2128 		 */
2129 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2130 		last_ibp = NULL;
2131 		while (next_agino != agino) {
2132 			struct xfs_imap	imap;
2133 
2134 			if (last_ibp)
2135 				xfs_trans_brelse(tp, last_ibp);
2136 
2137 			imap.im_blkno = 0;
2138 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2139 
2140 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2141 			if (error) {
2142 				xfs_warn(mp,
2143 	"%s: xfs_imap returned error %d.",
2144 					 __func__, error);
2145 				return error;
2146 			}
2147 
2148 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2149 					       &last_ibp, 0, 0);
2150 			if (error) {
2151 				xfs_warn(mp,
2152 	"%s: xfs_imap_to_bp returned error %d.",
2153 					__func__, error);
2154 				return error;
2155 			}
2156 
2157 			last_offset = imap.im_boffset;
2158 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2159 			ASSERT(next_agino != NULLAGINO);
2160 			ASSERT(next_agino != 0);
2161 		}
2162 
2163 		/*
2164 		 * Now last_ibp points to the buffer previous to us on the
2165 		 * unlinked list.  Pull us from the list.
2166 		 */
2167 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2168 				       0, 0);
2169 		if (error) {
2170 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2171 				__func__, error);
2172 			return error;
2173 		}
2174 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2175 		ASSERT(next_agino != 0);
2176 		ASSERT(next_agino != agino);
2177 		if (next_agino != NULLAGINO) {
2178 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2179 			offset = ip->i_imap.im_boffset +
2180 				offsetof(xfs_dinode_t, di_next_unlinked);
2181 
2182 			/* need to recalc the inode CRC if appropriate */
2183 			xfs_dinode_calc_crc(mp, dip);
2184 
2185 			xfs_trans_inode_buf(tp, ibp);
2186 			xfs_trans_log_buf(tp, ibp, offset,
2187 					  (offset + sizeof(xfs_agino_t) - 1));
2188 			xfs_inobp_check(mp, ibp);
2189 		} else {
2190 			xfs_trans_brelse(tp, ibp);
2191 		}
2192 		/*
2193 		 * Point the previous inode on the list to the next inode.
2194 		 */
2195 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2196 		ASSERT(next_agino != 0);
2197 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2198 
2199 		/* need to recalc the inode CRC if appropriate */
2200 		xfs_dinode_calc_crc(mp, last_dip);
2201 
2202 		xfs_trans_inode_buf(tp, last_ibp);
2203 		xfs_trans_log_buf(tp, last_ibp, offset,
2204 				  (offset + sizeof(xfs_agino_t) - 1));
2205 		xfs_inobp_check(mp, last_ibp);
2206 	}
2207 	return 0;
2208 }
2209 
2210 /*
2211  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2212  * inodes that are in memory - they all must be marked stale and attached to
2213  * the cluster buffer.
2214  */
2215 STATIC int
2216 xfs_ifree_cluster(
2217 	xfs_inode_t		*free_ip,
2218 	xfs_trans_t		*tp,
2219 	struct xfs_icluster	*xic)
2220 {
2221 	xfs_mount_t		*mp = free_ip->i_mount;
2222 	int			blks_per_cluster;
2223 	int			inodes_per_cluster;
2224 	int			nbufs;
2225 	int			i, j;
2226 	int			ioffset;
2227 	xfs_daddr_t		blkno;
2228 	xfs_buf_t		*bp;
2229 	xfs_inode_t		*ip;
2230 	xfs_inode_log_item_t	*iip;
2231 	struct xfs_log_item	*lip;
2232 	struct xfs_perag	*pag;
2233 	xfs_ino_t		inum;
2234 
2235 	inum = xic->first_ino;
2236 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2237 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2238 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2239 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2240 
2241 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2242 		/*
2243 		 * The allocation bitmap tells us which inodes of the chunk were
2244 		 * physically allocated. Skip the cluster if an inode falls into
2245 		 * a sparse region.
2246 		 */
2247 		ioffset = inum - xic->first_ino;
2248 		if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2249 			ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
2250 			continue;
2251 		}
2252 
2253 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2254 					 XFS_INO_TO_AGBNO(mp, inum));
2255 
2256 		/*
2257 		 * We obtain and lock the backing buffer first in the process
2258 		 * here, as we have to ensure that any dirty inode that we
2259 		 * can't get the flush lock on is attached to the buffer.
2260 		 * If we scan the in-memory inodes first, then buffer IO can
2261 		 * complete before we get a lock on it, and hence we may fail
2262 		 * to mark all the active inodes on the buffer stale.
2263 		 */
2264 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2265 					mp->m_bsize * blks_per_cluster,
2266 					XBF_UNMAPPED);
2267 
2268 		if (!bp)
2269 			return -ENOMEM;
2270 
2271 		/*
2272 		 * This buffer may not have been correctly initialised as we
2273 		 * didn't read it from disk. That's not important because we are
2274 		 * only using to mark the buffer as stale in the log, and to
2275 		 * attach stale cached inodes on it. That means it will never be
2276 		 * dispatched for IO. If it is, we want to know about it, and we
2277 		 * want it to fail. We can acheive this by adding a write
2278 		 * verifier to the buffer.
2279 		 */
2280 		 bp->b_ops = &xfs_inode_buf_ops;
2281 
2282 		/*
2283 		 * Walk the inodes already attached to the buffer and mark them
2284 		 * stale. These will all have the flush locks held, so an
2285 		 * in-memory inode walk can't lock them. By marking them all
2286 		 * stale first, we will not attempt to lock them in the loop
2287 		 * below as the XFS_ISTALE flag will be set.
2288 		 */
2289 		list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
2290 			if (lip->li_type == XFS_LI_INODE) {
2291 				iip = (xfs_inode_log_item_t *)lip;
2292 				ASSERT(iip->ili_logged == 1);
2293 				lip->li_cb = xfs_istale_done;
2294 				xfs_trans_ail_copy_lsn(mp->m_ail,
2295 							&iip->ili_flush_lsn,
2296 							&iip->ili_item.li_lsn);
2297 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2298 			}
2299 		}
2300 
2301 
2302 		/*
2303 		 * For each inode in memory attempt to add it to the inode
2304 		 * buffer and set it up for being staled on buffer IO
2305 		 * completion.  This is safe as we've locked out tail pushing
2306 		 * and flushing by locking the buffer.
2307 		 *
2308 		 * We have already marked every inode that was part of a
2309 		 * transaction stale above, which means there is no point in
2310 		 * even trying to lock them.
2311 		 */
2312 		for (i = 0; i < inodes_per_cluster; i++) {
2313 retry:
2314 			rcu_read_lock();
2315 			ip = radix_tree_lookup(&pag->pag_ici_root,
2316 					XFS_INO_TO_AGINO(mp, (inum + i)));
2317 
2318 			/* Inode not in memory, nothing to do */
2319 			if (!ip) {
2320 				rcu_read_unlock();
2321 				continue;
2322 			}
2323 
2324 			/*
2325 			 * because this is an RCU protected lookup, we could
2326 			 * find a recently freed or even reallocated inode
2327 			 * during the lookup. We need to check under the
2328 			 * i_flags_lock for a valid inode here. Skip it if it
2329 			 * is not valid, the wrong inode or stale.
2330 			 */
2331 			spin_lock(&ip->i_flags_lock);
2332 			if (ip->i_ino != inum + i ||
2333 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2334 				spin_unlock(&ip->i_flags_lock);
2335 				rcu_read_unlock();
2336 				continue;
2337 			}
2338 			spin_unlock(&ip->i_flags_lock);
2339 
2340 			/*
2341 			 * Don't try to lock/unlock the current inode, but we
2342 			 * _cannot_ skip the other inodes that we did not find
2343 			 * in the list attached to the buffer and are not
2344 			 * already marked stale. If we can't lock it, back off
2345 			 * and retry.
2346 			 */
2347 			if (ip != free_ip) {
2348 				if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2349 					rcu_read_unlock();
2350 					delay(1);
2351 					goto retry;
2352 				}
2353 
2354 				/*
2355 				 * Check the inode number again in case we're
2356 				 * racing with freeing in xfs_reclaim_inode().
2357 				 * See the comments in that function for more
2358 				 * information as to why the initial check is
2359 				 * not sufficient.
2360 				 */
2361 				if (ip->i_ino != inum + i) {
2362 					xfs_iunlock(ip, XFS_ILOCK_EXCL);
2363 					rcu_read_unlock();
2364 					continue;
2365 				}
2366 			}
2367 			rcu_read_unlock();
2368 
2369 			xfs_iflock(ip);
2370 			xfs_iflags_set(ip, XFS_ISTALE);
2371 
2372 			/*
2373 			 * we don't need to attach clean inodes or those only
2374 			 * with unlogged changes (which we throw away, anyway).
2375 			 */
2376 			iip = ip->i_itemp;
2377 			if (!iip || xfs_inode_clean(ip)) {
2378 				ASSERT(ip != free_ip);
2379 				xfs_ifunlock(ip);
2380 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2381 				continue;
2382 			}
2383 
2384 			iip->ili_last_fields = iip->ili_fields;
2385 			iip->ili_fields = 0;
2386 			iip->ili_fsync_fields = 0;
2387 			iip->ili_logged = 1;
2388 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2389 						&iip->ili_item.li_lsn);
2390 
2391 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2392 						  &iip->ili_item);
2393 
2394 			if (ip != free_ip)
2395 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2396 		}
2397 
2398 		xfs_trans_stale_inode_buf(tp, bp);
2399 		xfs_trans_binval(tp, bp);
2400 	}
2401 
2402 	xfs_perag_put(pag);
2403 	return 0;
2404 }
2405 
2406 /*
2407  * Free any local-format buffers sitting around before we reset to
2408  * extents format.
2409  */
2410 static inline void
2411 xfs_ifree_local_data(
2412 	struct xfs_inode	*ip,
2413 	int			whichfork)
2414 {
2415 	struct xfs_ifork	*ifp;
2416 
2417 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
2418 		return;
2419 
2420 	ifp = XFS_IFORK_PTR(ip, whichfork);
2421 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
2422 }
2423 
2424 /*
2425  * This is called to return an inode to the inode free list.
2426  * The inode should already be truncated to 0 length and have
2427  * no pages associated with it.  This routine also assumes that
2428  * the inode is already a part of the transaction.
2429  *
2430  * The on-disk copy of the inode will have been added to the list
2431  * of unlinked inodes in the AGI. We need to remove the inode from
2432  * that list atomically with respect to freeing it here.
2433  */
2434 int
2435 xfs_ifree(
2436 	xfs_trans_t	*tp,
2437 	xfs_inode_t	*ip,
2438 	struct xfs_defer_ops	*dfops)
2439 {
2440 	int			error;
2441 	struct xfs_icluster	xic = { 0 };
2442 
2443 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2444 	ASSERT(VFS_I(ip)->i_nlink == 0);
2445 	ASSERT(ip->i_d.di_nextents == 0);
2446 	ASSERT(ip->i_d.di_anextents == 0);
2447 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2448 	ASSERT(ip->i_d.di_nblocks == 0);
2449 
2450 	/*
2451 	 * Pull the on-disk inode from the AGI unlinked list.
2452 	 */
2453 	error = xfs_iunlink_remove(tp, ip);
2454 	if (error)
2455 		return error;
2456 
2457 	error = xfs_difree(tp, ip->i_ino, dfops, &xic);
2458 	if (error)
2459 		return error;
2460 
2461 	xfs_ifree_local_data(ip, XFS_DATA_FORK);
2462 	xfs_ifree_local_data(ip, XFS_ATTR_FORK);
2463 
2464 	VFS_I(ip)->i_mode = 0;		/* mark incore inode as free */
2465 	ip->i_d.di_flags = 0;
2466 	ip->i_d.di_flags2 = 0;
2467 	ip->i_d.di_dmevmask = 0;
2468 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2469 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2470 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2471 
2472 	/* Don't attempt to replay owner changes for a deleted inode */
2473 	ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER);
2474 
2475 	/*
2476 	 * Bump the generation count so no one will be confused
2477 	 * by reincarnations of this inode.
2478 	 */
2479 	VFS_I(ip)->i_generation++;
2480 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2481 
2482 	if (xic.deleted)
2483 		error = xfs_ifree_cluster(ip, tp, &xic);
2484 
2485 	return error;
2486 }
2487 
2488 /*
2489  * This is called to unpin an inode.  The caller must have the inode locked
2490  * in at least shared mode so that the buffer cannot be subsequently pinned
2491  * once someone is waiting for it to be unpinned.
2492  */
2493 static void
2494 xfs_iunpin(
2495 	struct xfs_inode	*ip)
2496 {
2497 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2498 
2499 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2500 
2501 	/* Give the log a push to start the unpinning I/O */
2502 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2503 
2504 }
2505 
2506 static void
2507 __xfs_iunpin_wait(
2508 	struct xfs_inode	*ip)
2509 {
2510 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2511 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2512 
2513 	xfs_iunpin(ip);
2514 
2515 	do {
2516 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2517 		if (xfs_ipincount(ip))
2518 			io_schedule();
2519 	} while (xfs_ipincount(ip));
2520 	finish_wait(wq, &wait.wq_entry);
2521 }
2522 
2523 void
2524 xfs_iunpin_wait(
2525 	struct xfs_inode	*ip)
2526 {
2527 	if (xfs_ipincount(ip))
2528 		__xfs_iunpin_wait(ip);
2529 }
2530 
2531 /*
2532  * Removing an inode from the namespace involves removing the directory entry
2533  * and dropping the link count on the inode. Removing the directory entry can
2534  * result in locking an AGF (directory blocks were freed) and removing a link
2535  * count can result in placing the inode on an unlinked list which results in
2536  * locking an AGI.
2537  *
2538  * The big problem here is that we have an ordering constraint on AGF and AGI
2539  * locking - inode allocation locks the AGI, then can allocate a new extent for
2540  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2541  * removes the inode from the unlinked list, requiring that we lock the AGI
2542  * first, and then freeing the inode can result in an inode chunk being freed
2543  * and hence freeing disk space requiring that we lock an AGF.
2544  *
2545  * Hence the ordering that is imposed by other parts of the code is AGI before
2546  * AGF. This means we cannot remove the directory entry before we drop the inode
2547  * reference count and put it on the unlinked list as this results in a lock
2548  * order of AGF then AGI, and this can deadlock against inode allocation and
2549  * freeing. Therefore we must drop the link counts before we remove the
2550  * directory entry.
2551  *
2552  * This is still safe from a transactional point of view - it is not until we
2553  * get to xfs_defer_finish() that we have the possibility of multiple
2554  * transactions in this operation. Hence as long as we remove the directory
2555  * entry and drop the link count in the first transaction of the remove
2556  * operation, there are no transactional constraints on the ordering here.
2557  */
2558 int
2559 xfs_remove(
2560 	xfs_inode_t             *dp,
2561 	struct xfs_name		*name,
2562 	xfs_inode_t		*ip)
2563 {
2564 	xfs_mount_t		*mp = dp->i_mount;
2565 	xfs_trans_t             *tp = NULL;
2566 	int			is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2567 	int                     error = 0;
2568 	struct xfs_defer_ops	dfops;
2569 	xfs_fsblock_t           first_block;
2570 	uint			resblks;
2571 
2572 	trace_xfs_remove(dp, name);
2573 
2574 	if (XFS_FORCED_SHUTDOWN(mp))
2575 		return -EIO;
2576 
2577 	error = xfs_qm_dqattach(dp, 0);
2578 	if (error)
2579 		goto std_return;
2580 
2581 	error = xfs_qm_dqattach(ip, 0);
2582 	if (error)
2583 		goto std_return;
2584 
2585 	/*
2586 	 * We try to get the real space reservation first,
2587 	 * allowing for directory btree deletion(s) implying
2588 	 * possible bmap insert(s).  If we can't get the space
2589 	 * reservation then we use 0 instead, and avoid the bmap
2590 	 * btree insert(s) in the directory code by, if the bmap
2591 	 * insert tries to happen, instead trimming the LAST
2592 	 * block from the directory.
2593 	 */
2594 	resblks = XFS_REMOVE_SPACE_RES(mp);
2595 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2596 	if (error == -ENOSPC) {
2597 		resblks = 0;
2598 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2599 				&tp);
2600 	}
2601 	if (error) {
2602 		ASSERT(error != -ENOSPC);
2603 		goto std_return;
2604 	}
2605 
2606 	xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2607 
2608 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2609 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2610 
2611 	/*
2612 	 * If we're removing a directory perform some additional validation.
2613 	 */
2614 	if (is_dir) {
2615 		ASSERT(VFS_I(ip)->i_nlink >= 2);
2616 		if (VFS_I(ip)->i_nlink != 2) {
2617 			error = -ENOTEMPTY;
2618 			goto out_trans_cancel;
2619 		}
2620 		if (!xfs_dir_isempty(ip)) {
2621 			error = -ENOTEMPTY;
2622 			goto out_trans_cancel;
2623 		}
2624 
2625 		/* Drop the link from ip's "..".  */
2626 		error = xfs_droplink(tp, dp);
2627 		if (error)
2628 			goto out_trans_cancel;
2629 
2630 		/* Drop the "." link from ip to self.  */
2631 		error = xfs_droplink(tp, ip);
2632 		if (error)
2633 			goto out_trans_cancel;
2634 	} else {
2635 		/*
2636 		 * When removing a non-directory we need to log the parent
2637 		 * inode here.  For a directory this is done implicitly
2638 		 * by the xfs_droplink call for the ".." entry.
2639 		 */
2640 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2641 	}
2642 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2643 
2644 	/* Drop the link from dp to ip. */
2645 	error = xfs_droplink(tp, ip);
2646 	if (error)
2647 		goto out_trans_cancel;
2648 
2649 	xfs_defer_init(&dfops, &first_block);
2650 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2651 					&first_block, &dfops, resblks);
2652 	if (error) {
2653 		ASSERT(error != -ENOENT);
2654 		goto out_bmap_cancel;
2655 	}
2656 
2657 	/*
2658 	 * If this is a synchronous mount, make sure that the
2659 	 * remove transaction goes to disk before returning to
2660 	 * the user.
2661 	 */
2662 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2663 		xfs_trans_set_sync(tp);
2664 
2665 	error = xfs_defer_finish(&tp, &dfops);
2666 	if (error)
2667 		goto out_bmap_cancel;
2668 
2669 	error = xfs_trans_commit(tp);
2670 	if (error)
2671 		goto std_return;
2672 
2673 	if (is_dir && xfs_inode_is_filestream(ip))
2674 		xfs_filestream_deassociate(ip);
2675 
2676 	return 0;
2677 
2678  out_bmap_cancel:
2679 	xfs_defer_cancel(&dfops);
2680  out_trans_cancel:
2681 	xfs_trans_cancel(tp);
2682  std_return:
2683 	return error;
2684 }
2685 
2686 /*
2687  * Enter all inodes for a rename transaction into a sorted array.
2688  */
2689 #define __XFS_SORT_INODES	5
2690 STATIC void
2691 xfs_sort_for_rename(
2692 	struct xfs_inode	*dp1,	/* in: old (source) directory inode */
2693 	struct xfs_inode	*dp2,	/* in: new (target) directory inode */
2694 	struct xfs_inode	*ip1,	/* in: inode of old entry */
2695 	struct xfs_inode	*ip2,	/* in: inode of new entry */
2696 	struct xfs_inode	*wip,	/* in: whiteout inode */
2697 	struct xfs_inode	**i_tab,/* out: sorted array of inodes */
2698 	int			*num_inodes)  /* in/out: inodes in array */
2699 {
2700 	int			i, j;
2701 
2702 	ASSERT(*num_inodes == __XFS_SORT_INODES);
2703 	memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2704 
2705 	/*
2706 	 * i_tab contains a list of pointers to inodes.  We initialize
2707 	 * the table here & we'll sort it.  We will then use it to
2708 	 * order the acquisition of the inode locks.
2709 	 *
2710 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2711 	 */
2712 	i = 0;
2713 	i_tab[i++] = dp1;
2714 	i_tab[i++] = dp2;
2715 	i_tab[i++] = ip1;
2716 	if (ip2)
2717 		i_tab[i++] = ip2;
2718 	if (wip)
2719 		i_tab[i++] = wip;
2720 	*num_inodes = i;
2721 
2722 	/*
2723 	 * Sort the elements via bubble sort.  (Remember, there are at
2724 	 * most 5 elements to sort, so this is adequate.)
2725 	 */
2726 	for (i = 0; i < *num_inodes; i++) {
2727 		for (j = 1; j < *num_inodes; j++) {
2728 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2729 				struct xfs_inode *temp = i_tab[j];
2730 				i_tab[j] = i_tab[j-1];
2731 				i_tab[j-1] = temp;
2732 			}
2733 		}
2734 	}
2735 }
2736 
2737 static int
2738 xfs_finish_rename(
2739 	struct xfs_trans	*tp,
2740 	struct xfs_defer_ops	*dfops)
2741 {
2742 	int			error;
2743 
2744 	/*
2745 	 * If this is a synchronous mount, make sure that the rename transaction
2746 	 * goes to disk before returning to the user.
2747 	 */
2748 	if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2749 		xfs_trans_set_sync(tp);
2750 
2751 	error = xfs_defer_finish(&tp, dfops);
2752 	if (error) {
2753 		xfs_defer_cancel(dfops);
2754 		xfs_trans_cancel(tp);
2755 		return error;
2756 	}
2757 
2758 	return xfs_trans_commit(tp);
2759 }
2760 
2761 /*
2762  * xfs_cross_rename()
2763  *
2764  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2765  */
2766 STATIC int
2767 xfs_cross_rename(
2768 	struct xfs_trans	*tp,
2769 	struct xfs_inode	*dp1,
2770 	struct xfs_name		*name1,
2771 	struct xfs_inode	*ip1,
2772 	struct xfs_inode	*dp2,
2773 	struct xfs_name		*name2,
2774 	struct xfs_inode	*ip2,
2775 	struct xfs_defer_ops	*dfops,
2776 	xfs_fsblock_t		*first_block,
2777 	int			spaceres)
2778 {
2779 	int		error = 0;
2780 	int		ip1_flags = 0;
2781 	int		ip2_flags = 0;
2782 	int		dp2_flags = 0;
2783 
2784 	/* Swap inode number for dirent in first parent */
2785 	error = xfs_dir_replace(tp, dp1, name1,
2786 				ip2->i_ino,
2787 				first_block, dfops, spaceres);
2788 	if (error)
2789 		goto out_trans_abort;
2790 
2791 	/* Swap inode number for dirent in second parent */
2792 	error = xfs_dir_replace(tp, dp2, name2,
2793 				ip1->i_ino,
2794 				first_block, dfops, spaceres);
2795 	if (error)
2796 		goto out_trans_abort;
2797 
2798 	/*
2799 	 * If we're renaming one or more directories across different parents,
2800 	 * update the respective ".." entries (and link counts) to match the new
2801 	 * parents.
2802 	 */
2803 	if (dp1 != dp2) {
2804 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2805 
2806 		if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2807 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2808 						dp1->i_ino, first_block,
2809 						dfops, spaceres);
2810 			if (error)
2811 				goto out_trans_abort;
2812 
2813 			/* transfer ip2 ".." reference to dp1 */
2814 			if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2815 				error = xfs_droplink(tp, dp2);
2816 				if (error)
2817 					goto out_trans_abort;
2818 				error = xfs_bumplink(tp, dp1);
2819 				if (error)
2820 					goto out_trans_abort;
2821 			}
2822 
2823 			/*
2824 			 * Although ip1 isn't changed here, userspace needs
2825 			 * to be warned about the change, so that applications
2826 			 * relying on it (like backup ones), will properly
2827 			 * notify the change
2828 			 */
2829 			ip1_flags |= XFS_ICHGTIME_CHG;
2830 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2831 		}
2832 
2833 		if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2834 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2835 						dp2->i_ino, first_block,
2836 						dfops, spaceres);
2837 			if (error)
2838 				goto out_trans_abort;
2839 
2840 			/* transfer ip1 ".." reference to dp2 */
2841 			if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2842 				error = xfs_droplink(tp, dp1);
2843 				if (error)
2844 					goto out_trans_abort;
2845 				error = xfs_bumplink(tp, dp2);
2846 				if (error)
2847 					goto out_trans_abort;
2848 			}
2849 
2850 			/*
2851 			 * Although ip2 isn't changed here, userspace needs
2852 			 * to be warned about the change, so that applications
2853 			 * relying on it (like backup ones), will properly
2854 			 * notify the change
2855 			 */
2856 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2857 			ip2_flags |= XFS_ICHGTIME_CHG;
2858 		}
2859 	}
2860 
2861 	if (ip1_flags) {
2862 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2863 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2864 	}
2865 	if (ip2_flags) {
2866 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2867 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2868 	}
2869 	if (dp2_flags) {
2870 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2871 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2872 	}
2873 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2874 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2875 	return xfs_finish_rename(tp, dfops);
2876 
2877 out_trans_abort:
2878 	xfs_defer_cancel(dfops);
2879 	xfs_trans_cancel(tp);
2880 	return error;
2881 }
2882 
2883 /*
2884  * xfs_rename_alloc_whiteout()
2885  *
2886  * Return a referenced, unlinked, unlocked inode that that can be used as a
2887  * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2888  * crash between allocating the inode and linking it into the rename transaction
2889  * recovery will free the inode and we won't leak it.
2890  */
2891 static int
2892 xfs_rename_alloc_whiteout(
2893 	struct xfs_inode	*dp,
2894 	struct xfs_inode	**wip)
2895 {
2896 	struct xfs_inode	*tmpfile;
2897 	int			error;
2898 
2899 	error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2900 	if (error)
2901 		return error;
2902 
2903 	/*
2904 	 * Prepare the tmpfile inode as if it were created through the VFS.
2905 	 * Otherwise, the link increment paths will complain about nlink 0->1.
2906 	 * Drop the link count as done by d_tmpfile(), complete the inode setup
2907 	 * and flag it as linkable.
2908 	 */
2909 	drop_nlink(VFS_I(tmpfile));
2910 	xfs_setup_iops(tmpfile);
2911 	xfs_finish_inode_setup(tmpfile);
2912 	VFS_I(tmpfile)->i_state |= I_LINKABLE;
2913 
2914 	*wip = tmpfile;
2915 	return 0;
2916 }
2917 
2918 /*
2919  * xfs_rename
2920  */
2921 int
2922 xfs_rename(
2923 	struct xfs_inode	*src_dp,
2924 	struct xfs_name		*src_name,
2925 	struct xfs_inode	*src_ip,
2926 	struct xfs_inode	*target_dp,
2927 	struct xfs_name		*target_name,
2928 	struct xfs_inode	*target_ip,
2929 	unsigned int		flags)
2930 {
2931 	struct xfs_mount	*mp = src_dp->i_mount;
2932 	struct xfs_trans	*tp;
2933 	struct xfs_defer_ops	dfops;
2934 	xfs_fsblock_t		first_block;
2935 	struct xfs_inode	*wip = NULL;		/* whiteout inode */
2936 	struct xfs_inode	*inodes[__XFS_SORT_INODES];
2937 	int			num_inodes = __XFS_SORT_INODES;
2938 	bool			new_parent = (src_dp != target_dp);
2939 	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2940 	int			spaceres;
2941 	int			error;
2942 
2943 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2944 
2945 	if ((flags & RENAME_EXCHANGE) && !target_ip)
2946 		return -EINVAL;
2947 
2948 	/*
2949 	 * If we are doing a whiteout operation, allocate the whiteout inode
2950 	 * we will be placing at the target and ensure the type is set
2951 	 * appropriately.
2952 	 */
2953 	if (flags & RENAME_WHITEOUT) {
2954 		ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2955 		error = xfs_rename_alloc_whiteout(target_dp, &wip);
2956 		if (error)
2957 			return error;
2958 
2959 		/* setup target dirent info as whiteout */
2960 		src_name->type = XFS_DIR3_FT_CHRDEV;
2961 	}
2962 
2963 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2964 				inodes, &num_inodes);
2965 
2966 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2967 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2968 	if (error == -ENOSPC) {
2969 		spaceres = 0;
2970 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2971 				&tp);
2972 	}
2973 	if (error)
2974 		goto out_release_wip;
2975 
2976 	/*
2977 	 * Attach the dquots to the inodes
2978 	 */
2979 	error = xfs_qm_vop_rename_dqattach(inodes);
2980 	if (error)
2981 		goto out_trans_cancel;
2982 
2983 	/*
2984 	 * Lock all the participating inodes. Depending upon whether
2985 	 * the target_name exists in the target directory, and
2986 	 * whether the target directory is the same as the source
2987 	 * directory, we can lock from 2 to 4 inodes.
2988 	 */
2989 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2990 
2991 	/*
2992 	 * Join all the inodes to the transaction. From this point on,
2993 	 * we can rely on either trans_commit or trans_cancel to unlock
2994 	 * them.
2995 	 */
2996 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2997 	if (new_parent)
2998 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2999 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3000 	if (target_ip)
3001 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3002 	if (wip)
3003 		xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3004 
3005 	/*
3006 	 * If we are using project inheritance, we only allow renames
3007 	 * into our tree when the project IDs are the same; else the
3008 	 * tree quota mechanism would be circumvented.
3009 	 */
3010 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3011 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
3012 		error = -EXDEV;
3013 		goto out_trans_cancel;
3014 	}
3015 
3016 	xfs_defer_init(&dfops, &first_block);
3017 
3018 	/* RENAME_EXCHANGE is unique from here on. */
3019 	if (flags & RENAME_EXCHANGE)
3020 		return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3021 					target_dp, target_name, target_ip,
3022 					&dfops, &first_block, spaceres);
3023 
3024 	/*
3025 	 * Set up the target.
3026 	 */
3027 	if (target_ip == NULL) {
3028 		/*
3029 		 * If there's no space reservation, check the entry will
3030 		 * fit before actually inserting it.
3031 		 */
3032 		if (!spaceres) {
3033 			error = xfs_dir_canenter(tp, target_dp, target_name);
3034 			if (error)
3035 				goto out_trans_cancel;
3036 		}
3037 		/*
3038 		 * If target does not exist and the rename crosses
3039 		 * directories, adjust the target directory link count
3040 		 * to account for the ".." reference from the new entry.
3041 		 */
3042 		error = xfs_dir_createname(tp, target_dp, target_name,
3043 						src_ip->i_ino, &first_block,
3044 						&dfops, spaceres);
3045 		if (error)
3046 			goto out_bmap_cancel;
3047 
3048 		xfs_trans_ichgtime(tp, target_dp,
3049 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3050 
3051 		if (new_parent && src_is_directory) {
3052 			error = xfs_bumplink(tp, target_dp);
3053 			if (error)
3054 				goto out_bmap_cancel;
3055 		}
3056 	} else { /* target_ip != NULL */
3057 		/*
3058 		 * If target exists and it's a directory, check that both
3059 		 * target and source are directories and that target can be
3060 		 * destroyed, or that neither is a directory.
3061 		 */
3062 		if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
3063 			/*
3064 			 * Make sure target dir is empty.
3065 			 */
3066 			if (!(xfs_dir_isempty(target_ip)) ||
3067 			    (VFS_I(target_ip)->i_nlink > 2)) {
3068 				error = -EEXIST;
3069 				goto out_trans_cancel;
3070 			}
3071 		}
3072 
3073 		/*
3074 		 * Link the source inode under the target name.
3075 		 * If the source inode is a directory and we are moving
3076 		 * it across directories, its ".." entry will be
3077 		 * inconsistent until we replace that down below.
3078 		 *
3079 		 * In case there is already an entry with the same
3080 		 * name at the destination directory, remove it first.
3081 		 */
3082 		error = xfs_dir_replace(tp, target_dp, target_name,
3083 					src_ip->i_ino,
3084 					&first_block, &dfops, spaceres);
3085 		if (error)
3086 			goto out_bmap_cancel;
3087 
3088 		xfs_trans_ichgtime(tp, target_dp,
3089 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3090 
3091 		/*
3092 		 * Decrement the link count on the target since the target
3093 		 * dir no longer points to it.
3094 		 */
3095 		error = xfs_droplink(tp, target_ip);
3096 		if (error)
3097 			goto out_bmap_cancel;
3098 
3099 		if (src_is_directory) {
3100 			/*
3101 			 * Drop the link from the old "." entry.
3102 			 */
3103 			error = xfs_droplink(tp, target_ip);
3104 			if (error)
3105 				goto out_bmap_cancel;
3106 		}
3107 	} /* target_ip != NULL */
3108 
3109 	/*
3110 	 * Remove the source.
3111 	 */
3112 	if (new_parent && src_is_directory) {
3113 		/*
3114 		 * Rewrite the ".." entry to point to the new
3115 		 * directory.
3116 		 */
3117 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3118 					target_dp->i_ino,
3119 					&first_block, &dfops, spaceres);
3120 		ASSERT(error != -EEXIST);
3121 		if (error)
3122 			goto out_bmap_cancel;
3123 	}
3124 
3125 	/*
3126 	 * We always want to hit the ctime on the source inode.
3127 	 *
3128 	 * This isn't strictly required by the standards since the source
3129 	 * inode isn't really being changed, but old unix file systems did
3130 	 * it and some incremental backup programs won't work without it.
3131 	 */
3132 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3133 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3134 
3135 	/*
3136 	 * Adjust the link count on src_dp.  This is necessary when
3137 	 * renaming a directory, either within one parent when
3138 	 * the target existed, or across two parent directories.
3139 	 */
3140 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3141 
3142 		/*
3143 		 * Decrement link count on src_directory since the
3144 		 * entry that's moved no longer points to it.
3145 		 */
3146 		error = xfs_droplink(tp, src_dp);
3147 		if (error)
3148 			goto out_bmap_cancel;
3149 	}
3150 
3151 	/*
3152 	 * For whiteouts, we only need to update the source dirent with the
3153 	 * inode number of the whiteout inode rather than removing it
3154 	 * altogether.
3155 	 */
3156 	if (wip) {
3157 		error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3158 					&first_block, &dfops, spaceres);
3159 	} else
3160 		error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3161 					   &first_block, &dfops, spaceres);
3162 	if (error)
3163 		goto out_bmap_cancel;
3164 
3165 	/*
3166 	 * For whiteouts, we need to bump the link count on the whiteout inode.
3167 	 * This means that failures all the way up to this point leave the inode
3168 	 * on the unlinked list and so cleanup is a simple matter of dropping
3169 	 * the remaining reference to it. If we fail here after bumping the link
3170 	 * count, we're shutting down the filesystem so we'll never see the
3171 	 * intermediate state on disk.
3172 	 */
3173 	if (wip) {
3174 		ASSERT(VFS_I(wip)->i_nlink == 0);
3175 		error = xfs_bumplink(tp, wip);
3176 		if (error)
3177 			goto out_bmap_cancel;
3178 		error = xfs_iunlink_remove(tp, wip);
3179 		if (error)
3180 			goto out_bmap_cancel;
3181 		xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3182 
3183 		/*
3184 		 * Now we have a real link, clear the "I'm a tmpfile" state
3185 		 * flag from the inode so it doesn't accidentally get misused in
3186 		 * future.
3187 		 */
3188 		VFS_I(wip)->i_state &= ~I_LINKABLE;
3189 	}
3190 
3191 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3192 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3193 	if (new_parent)
3194 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3195 
3196 	error = xfs_finish_rename(tp, &dfops);
3197 	if (wip)
3198 		IRELE(wip);
3199 	return error;
3200 
3201 out_bmap_cancel:
3202 	xfs_defer_cancel(&dfops);
3203 out_trans_cancel:
3204 	xfs_trans_cancel(tp);
3205 out_release_wip:
3206 	if (wip)
3207 		IRELE(wip);
3208 	return error;
3209 }
3210 
3211 STATIC int
3212 xfs_iflush_cluster(
3213 	struct xfs_inode	*ip,
3214 	struct xfs_buf		*bp)
3215 {
3216 	struct xfs_mount	*mp = ip->i_mount;
3217 	struct xfs_perag	*pag;
3218 	unsigned long		first_index, mask;
3219 	unsigned long		inodes_per_cluster;
3220 	int			cilist_size;
3221 	struct xfs_inode	**cilist;
3222 	struct xfs_inode	*cip;
3223 	int			nr_found;
3224 	int			clcount = 0;
3225 	int			bufwasdelwri;
3226 	int			i;
3227 
3228 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3229 
3230 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3231 	cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3232 	cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS);
3233 	if (!cilist)
3234 		goto out_put;
3235 
3236 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3237 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3238 	rcu_read_lock();
3239 	/* really need a gang lookup range call here */
3240 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist,
3241 					first_index, inodes_per_cluster);
3242 	if (nr_found == 0)
3243 		goto out_free;
3244 
3245 	for (i = 0; i < nr_found; i++) {
3246 		cip = cilist[i];
3247 		if (cip == ip)
3248 			continue;
3249 
3250 		/*
3251 		 * because this is an RCU protected lookup, we could find a
3252 		 * recently freed or even reallocated inode during the lookup.
3253 		 * We need to check under the i_flags_lock for a valid inode
3254 		 * here. Skip it if it is not valid or the wrong inode.
3255 		 */
3256 		spin_lock(&cip->i_flags_lock);
3257 		if (!cip->i_ino ||
3258 		    __xfs_iflags_test(cip, XFS_ISTALE)) {
3259 			spin_unlock(&cip->i_flags_lock);
3260 			continue;
3261 		}
3262 
3263 		/*
3264 		 * Once we fall off the end of the cluster, no point checking
3265 		 * any more inodes in the list because they will also all be
3266 		 * outside the cluster.
3267 		 */
3268 		if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) {
3269 			spin_unlock(&cip->i_flags_lock);
3270 			break;
3271 		}
3272 		spin_unlock(&cip->i_flags_lock);
3273 
3274 		/*
3275 		 * Do an un-protected check to see if the inode is dirty and
3276 		 * is a candidate for flushing.  These checks will be repeated
3277 		 * later after the appropriate locks are acquired.
3278 		 */
3279 		if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0)
3280 			continue;
3281 
3282 		/*
3283 		 * Try to get locks.  If any are unavailable or it is pinned,
3284 		 * then this inode cannot be flushed and is skipped.
3285 		 */
3286 
3287 		if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED))
3288 			continue;
3289 		if (!xfs_iflock_nowait(cip)) {
3290 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3291 			continue;
3292 		}
3293 		if (xfs_ipincount(cip)) {
3294 			xfs_ifunlock(cip);
3295 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3296 			continue;
3297 		}
3298 
3299 
3300 		/*
3301 		 * Check the inode number again, just to be certain we are not
3302 		 * racing with freeing in xfs_reclaim_inode(). See the comments
3303 		 * in that function for more information as to why the initial
3304 		 * check is not sufficient.
3305 		 */
3306 		if (!cip->i_ino) {
3307 			xfs_ifunlock(cip);
3308 			xfs_iunlock(cip, XFS_ILOCK_SHARED);
3309 			continue;
3310 		}
3311 
3312 		/*
3313 		 * arriving here means that this inode can be flushed.  First
3314 		 * re-check that it's dirty before flushing.
3315 		 */
3316 		if (!xfs_inode_clean(cip)) {
3317 			int	error;
3318 			error = xfs_iflush_int(cip, bp);
3319 			if (error) {
3320 				xfs_iunlock(cip, XFS_ILOCK_SHARED);
3321 				goto cluster_corrupt_out;
3322 			}
3323 			clcount++;
3324 		} else {
3325 			xfs_ifunlock(cip);
3326 		}
3327 		xfs_iunlock(cip, XFS_ILOCK_SHARED);
3328 	}
3329 
3330 	if (clcount) {
3331 		XFS_STATS_INC(mp, xs_icluster_flushcnt);
3332 		XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3333 	}
3334 
3335 out_free:
3336 	rcu_read_unlock();
3337 	kmem_free(cilist);
3338 out_put:
3339 	xfs_perag_put(pag);
3340 	return 0;
3341 
3342 
3343 cluster_corrupt_out:
3344 	/*
3345 	 * Corruption detected in the clustering loop.  Invalidate the
3346 	 * inode buffer and shut down the filesystem.
3347 	 */
3348 	rcu_read_unlock();
3349 	/*
3350 	 * Clean up the buffer.  If it was delwri, just release it --
3351 	 * brelse can handle it with no problems.  If not, shut down the
3352 	 * filesystem before releasing the buffer.
3353 	 */
3354 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3355 	if (bufwasdelwri)
3356 		xfs_buf_relse(bp);
3357 
3358 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3359 
3360 	if (!bufwasdelwri) {
3361 		/*
3362 		 * Just like incore_relse: if we have b_iodone functions,
3363 		 * mark the buffer as an error and call them.  Otherwise
3364 		 * mark it as stale and brelse.
3365 		 */
3366 		if (bp->b_iodone) {
3367 			bp->b_flags &= ~XBF_DONE;
3368 			xfs_buf_stale(bp);
3369 			xfs_buf_ioerror(bp, -EIO);
3370 			xfs_buf_ioend(bp);
3371 		} else {
3372 			xfs_buf_stale(bp);
3373 			xfs_buf_relse(bp);
3374 		}
3375 	}
3376 
3377 	/*
3378 	 * Unlocks the flush lock
3379 	 */
3380 	xfs_iflush_abort(cip, false);
3381 	kmem_free(cilist);
3382 	xfs_perag_put(pag);
3383 	return -EFSCORRUPTED;
3384 }
3385 
3386 /*
3387  * Flush dirty inode metadata into the backing buffer.
3388  *
3389  * The caller must have the inode lock and the inode flush lock held.  The
3390  * inode lock will still be held upon return to the caller, and the inode
3391  * flush lock will be released after the inode has reached the disk.
3392  *
3393  * The caller must write out the buffer returned in *bpp and release it.
3394  */
3395 int
3396 xfs_iflush(
3397 	struct xfs_inode	*ip,
3398 	struct xfs_buf		**bpp)
3399 {
3400 	struct xfs_mount	*mp = ip->i_mount;
3401 	struct xfs_buf		*bp = NULL;
3402 	struct xfs_dinode	*dip;
3403 	int			error;
3404 
3405 	XFS_STATS_INC(mp, xs_iflush_count);
3406 
3407 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3408 	ASSERT(xfs_isiflocked(ip));
3409 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3410 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3411 
3412 	*bpp = NULL;
3413 
3414 	xfs_iunpin_wait(ip);
3415 
3416 	/*
3417 	 * For stale inodes we cannot rely on the backing buffer remaining
3418 	 * stale in cache for the remaining life of the stale inode and so
3419 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3420 	 * inodes below. We have to check this after ensuring the inode is
3421 	 * unpinned so that it is safe to reclaim the stale inode after the
3422 	 * flush call.
3423 	 */
3424 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3425 		xfs_ifunlock(ip);
3426 		return 0;
3427 	}
3428 
3429 	/*
3430 	 * This may have been unpinned because the filesystem is shutting
3431 	 * down forcibly. If that's the case we must not write this inode
3432 	 * to disk, because the log record didn't make it to disk.
3433 	 *
3434 	 * We also have to remove the log item from the AIL in this case,
3435 	 * as we wait for an empty AIL as part of the unmount process.
3436 	 */
3437 	if (XFS_FORCED_SHUTDOWN(mp)) {
3438 		error = -EIO;
3439 		goto abort_out;
3440 	}
3441 
3442 	/*
3443 	 * Get the buffer containing the on-disk inode. We are doing a try-lock
3444 	 * operation here, so we may get  an EAGAIN error. In that case, we
3445 	 * simply want to return with the inode still dirty.
3446 	 *
3447 	 * If we get any other error, we effectively have a corruption situation
3448 	 * and we cannot flush the inode, so we treat it the same as failing
3449 	 * xfs_iflush_int().
3450 	 */
3451 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3452 			       0);
3453 	if (error == -EAGAIN) {
3454 		xfs_ifunlock(ip);
3455 		return error;
3456 	}
3457 	if (error)
3458 		goto corrupt_out;
3459 
3460 	/*
3461 	 * First flush out the inode that xfs_iflush was called with.
3462 	 */
3463 	error = xfs_iflush_int(ip, bp);
3464 	if (error)
3465 		goto corrupt_out;
3466 
3467 	/*
3468 	 * If the buffer is pinned then push on the log now so we won't
3469 	 * get stuck waiting in the write for too long.
3470 	 */
3471 	if (xfs_buf_ispinned(bp))
3472 		xfs_log_force(mp, 0);
3473 
3474 	/*
3475 	 * inode clustering:
3476 	 * see if other inodes can be gathered into this write
3477 	 */
3478 	error = xfs_iflush_cluster(ip, bp);
3479 	if (error)
3480 		goto cluster_corrupt_out;
3481 
3482 	*bpp = bp;
3483 	return 0;
3484 
3485 corrupt_out:
3486 	if (bp)
3487 		xfs_buf_relse(bp);
3488 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3489 cluster_corrupt_out:
3490 	error = -EFSCORRUPTED;
3491 abort_out:
3492 	/*
3493 	 * Unlocks the flush lock
3494 	 */
3495 	xfs_iflush_abort(ip, false);
3496 	return error;
3497 }
3498 
3499 /*
3500  * If there are inline format data / attr forks attached to this inode,
3501  * make sure they're not corrupt.
3502  */
3503 bool
3504 xfs_inode_verify_forks(
3505 	struct xfs_inode	*ip)
3506 {
3507 	struct xfs_ifork	*ifp;
3508 	xfs_failaddr_t		fa;
3509 
3510 	fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops);
3511 	if (fa) {
3512 		ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
3513 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork",
3514 				ifp->if_u1.if_data, ifp->if_bytes, fa);
3515 		return false;
3516 	}
3517 
3518 	fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops);
3519 	if (fa) {
3520 		ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
3521 		xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork",
3522 				ifp ? ifp->if_u1.if_data : NULL,
3523 				ifp ? ifp->if_bytes : 0, fa);
3524 		return false;
3525 	}
3526 	return true;
3527 }
3528 
3529 STATIC int
3530 xfs_iflush_int(
3531 	struct xfs_inode	*ip,
3532 	struct xfs_buf		*bp)
3533 {
3534 	struct xfs_inode_log_item *iip = ip->i_itemp;
3535 	struct xfs_dinode	*dip;
3536 	struct xfs_mount	*mp = ip->i_mount;
3537 
3538 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3539 	ASSERT(xfs_isiflocked(ip));
3540 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3541 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3542 	ASSERT(iip != NULL && iip->ili_fields != 0);
3543 	ASSERT(ip->i_d.di_version > 1);
3544 
3545 	/* set *dip = inode's place in the buffer */
3546 	dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3547 
3548 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3549 			       mp, XFS_ERRTAG_IFLUSH_1)) {
3550 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3551 			"%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3552 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3553 		goto corrupt_out;
3554 	}
3555 	if (S_ISREG(VFS_I(ip)->i_mode)) {
3556 		if (XFS_TEST_ERROR(
3557 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3558 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3559 		    mp, XFS_ERRTAG_IFLUSH_3)) {
3560 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3561 				"%s: Bad regular inode %Lu, ptr "PTR_FMT,
3562 				__func__, ip->i_ino, ip);
3563 			goto corrupt_out;
3564 		}
3565 	} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3566 		if (XFS_TEST_ERROR(
3567 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3568 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3569 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3570 		    mp, XFS_ERRTAG_IFLUSH_4)) {
3571 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3572 				"%s: Bad directory inode %Lu, ptr "PTR_FMT,
3573 				__func__, ip->i_ino, ip);
3574 			goto corrupt_out;
3575 		}
3576 	}
3577 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3578 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3579 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3580 			"%s: detected corrupt incore inode %Lu, "
3581 			"total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3582 			__func__, ip->i_ino,
3583 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3584 			ip->i_d.di_nblocks, ip);
3585 		goto corrupt_out;
3586 	}
3587 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3588 				mp, XFS_ERRTAG_IFLUSH_6)) {
3589 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3590 			"%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3591 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3592 		goto corrupt_out;
3593 	}
3594 
3595 	/*
3596 	 * Inode item log recovery for v2 inodes are dependent on the
3597 	 * di_flushiter count for correct sequencing. We bump the flush
3598 	 * iteration count so we can detect flushes which postdate a log record
3599 	 * during recovery. This is redundant as we now log every change and
3600 	 * hence this can't happen but we need to still do it to ensure
3601 	 * backwards compatibility with old kernels that predate logging all
3602 	 * inode changes.
3603 	 */
3604 	if (ip->i_d.di_version < 3)
3605 		ip->i_d.di_flushiter++;
3606 
3607 	/* Check the inline fork data before we write out. */
3608 	if (!xfs_inode_verify_forks(ip))
3609 		goto corrupt_out;
3610 
3611 	/*
3612 	 * Copy the dirty parts of the inode into the on-disk inode.  We always
3613 	 * copy out the core of the inode, because if the inode is dirty at all
3614 	 * the core must be.
3615 	 */
3616 	xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3617 
3618 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3619 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3620 		ip->i_d.di_flushiter = 0;
3621 
3622 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3623 	if (XFS_IFORK_Q(ip))
3624 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3625 	xfs_inobp_check(mp, bp);
3626 
3627 	/*
3628 	 * We've recorded everything logged in the inode, so we'd like to clear
3629 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3630 	 * However, we can't stop logging all this information until the data
3631 	 * we've copied into the disk buffer is written to disk.  If we did we
3632 	 * might overwrite the copy of the inode in the log with all the data
3633 	 * after re-logging only part of it, and in the face of a crash we
3634 	 * wouldn't have all the data we need to recover.
3635 	 *
3636 	 * What we do is move the bits to the ili_last_fields field.  When
3637 	 * logging the inode, these bits are moved back to the ili_fields field.
3638 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3639 	 * know that the information those bits represent is permanently on
3640 	 * disk.  As long as the flush completes before the inode is logged
3641 	 * again, then both ili_fields and ili_last_fields will be cleared.
3642 	 *
3643 	 * We can play with the ili_fields bits here, because the inode lock
3644 	 * must be held exclusively in order to set bits there and the flush
3645 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3646 	 * done routine can tell whether or not to look in the AIL.  Also, store
3647 	 * the current LSN of the inode so that we can tell whether the item has
3648 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3649 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3650 	 * atomically.
3651 	 */
3652 	iip->ili_last_fields = iip->ili_fields;
3653 	iip->ili_fields = 0;
3654 	iip->ili_fsync_fields = 0;
3655 	iip->ili_logged = 1;
3656 
3657 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3658 				&iip->ili_item.li_lsn);
3659 
3660 	/*
3661 	 * Attach the function xfs_iflush_done to the inode's
3662 	 * buffer.  This will remove the inode from the AIL
3663 	 * and unlock the inode's flush lock when the inode is
3664 	 * completely written to disk.
3665 	 */
3666 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3667 
3668 	/* generate the checksum. */
3669 	xfs_dinode_calc_crc(mp, dip);
3670 
3671 	ASSERT(!list_empty(&bp->b_li_list));
3672 	ASSERT(bp->b_iodone != NULL);
3673 	return 0;
3674 
3675 corrupt_out:
3676 	return -EFSCORRUPTED;
3677 }
3678