xref: /linux/fs/xfs/xfs_inode.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include <linux/log2.h>
19 
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_inode.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_dir2.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_attr.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_bmap.h"
40 #include "xfs_bmap_util.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_filestream.h"
44 #include "xfs_cksum.h"
45 #include "xfs_trace.h"
46 #include "xfs_icache.h"
47 #include "xfs_symlink.h"
48 #include "xfs_trans_priv.h"
49 #include "xfs_log.h"
50 #include "xfs_bmap_btree.h"
51 
52 kmem_zone_t *xfs_inode_zone;
53 
54 /*
55  * Used in xfs_itruncate_extents().  This is the maximum number of extents
56  * freed from a file in a single transaction.
57  */
58 #define	XFS_ITRUNC_MAX_EXTENTS	2
59 
60 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
61 
62 STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
63 
64 /*
65  * helper function to extract extent size hint from inode
66  */
67 xfs_extlen_t
68 xfs_get_extsz_hint(
69 	struct xfs_inode	*ip)
70 {
71 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
72 		return ip->i_d.di_extsize;
73 	if (XFS_IS_REALTIME_INODE(ip))
74 		return ip->i_mount->m_sb.sb_rextsize;
75 	return 0;
76 }
77 
78 /*
79  * These two are wrapper routines around the xfs_ilock() routine used to
80  * centralize some grungy code.  They are used in places that wish to lock the
81  * inode solely for reading the extents.  The reason these places can't just
82  * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
83  * bringing in of the extents from disk for a file in b-tree format.  If the
84  * inode is in b-tree format, then we need to lock the inode exclusively until
85  * the extents are read in.  Locking it exclusively all the time would limit
86  * our parallelism unnecessarily, though.  What we do instead is check to see
87  * if the extents have been read in yet, and only lock the inode exclusively
88  * if they have not.
89  *
90  * The functions return a value which should be given to the corresponding
91  * xfs_iunlock() call.
92  */
93 uint
94 xfs_ilock_data_map_shared(
95 	struct xfs_inode	*ip)
96 {
97 	uint			lock_mode = XFS_ILOCK_SHARED;
98 
99 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
100 	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
101 		lock_mode = XFS_ILOCK_EXCL;
102 	xfs_ilock(ip, lock_mode);
103 	return lock_mode;
104 }
105 
106 uint
107 xfs_ilock_attr_map_shared(
108 	struct xfs_inode	*ip)
109 {
110 	uint			lock_mode = XFS_ILOCK_SHARED;
111 
112 	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
113 	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
114 		lock_mode = XFS_ILOCK_EXCL;
115 	xfs_ilock(ip, lock_mode);
116 	return lock_mode;
117 }
118 
119 /*
120  * The xfs inode contains 2 locks: a multi-reader lock called the
121  * i_iolock and a multi-reader lock called the i_lock.  This routine
122  * allows either or both of the locks to be obtained.
123  *
124  * The 2 locks should always be ordered so that the IO lock is
125  * obtained first in order to prevent deadlock.
126  *
127  * ip -- the inode being locked
128  * lock_flags -- this parameter indicates the inode's locks
129  *       to be locked.  It can be:
130  *		XFS_IOLOCK_SHARED,
131  *		XFS_IOLOCK_EXCL,
132  *		XFS_ILOCK_SHARED,
133  *		XFS_ILOCK_EXCL,
134  *		XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
135  *		XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
136  *		XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
137  *		XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
138  */
139 void
140 xfs_ilock(
141 	xfs_inode_t		*ip,
142 	uint			lock_flags)
143 {
144 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
145 
146 	/*
147 	 * You can't set both SHARED and EXCL for the same lock,
148 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
149 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
150 	 */
151 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
152 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
153 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
154 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
155 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
156 
157 	if (lock_flags & XFS_IOLOCK_EXCL)
158 		mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
159 	else if (lock_flags & XFS_IOLOCK_SHARED)
160 		mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
161 
162 	if (lock_flags & XFS_ILOCK_EXCL)
163 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
164 	else if (lock_flags & XFS_ILOCK_SHARED)
165 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
166 }
167 
168 /*
169  * This is just like xfs_ilock(), except that the caller
170  * is guaranteed not to sleep.  It returns 1 if it gets
171  * the requested locks and 0 otherwise.  If the IO lock is
172  * obtained but the inode lock cannot be, then the IO lock
173  * is dropped before returning.
174  *
175  * ip -- the inode being locked
176  * lock_flags -- this parameter indicates the inode's locks to be
177  *       to be locked.  See the comment for xfs_ilock() for a list
178  *	 of valid values.
179  */
180 int
181 xfs_ilock_nowait(
182 	xfs_inode_t		*ip,
183 	uint			lock_flags)
184 {
185 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
186 
187 	/*
188 	 * You can't set both SHARED and EXCL for the same lock,
189 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
190 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
191 	 */
192 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
193 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
194 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
195 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
196 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
197 
198 	if (lock_flags & XFS_IOLOCK_EXCL) {
199 		if (!mrtryupdate(&ip->i_iolock))
200 			goto out;
201 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
202 		if (!mrtryaccess(&ip->i_iolock))
203 			goto out;
204 	}
205 	if (lock_flags & XFS_ILOCK_EXCL) {
206 		if (!mrtryupdate(&ip->i_lock))
207 			goto out_undo_iolock;
208 	} else if (lock_flags & XFS_ILOCK_SHARED) {
209 		if (!mrtryaccess(&ip->i_lock))
210 			goto out_undo_iolock;
211 	}
212 	return 1;
213 
214  out_undo_iolock:
215 	if (lock_flags & XFS_IOLOCK_EXCL)
216 		mrunlock_excl(&ip->i_iolock);
217 	else if (lock_flags & XFS_IOLOCK_SHARED)
218 		mrunlock_shared(&ip->i_iolock);
219  out:
220 	return 0;
221 }
222 
223 /*
224  * xfs_iunlock() is used to drop the inode locks acquired with
225  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
226  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
227  * that we know which locks to drop.
228  *
229  * ip -- the inode being unlocked
230  * lock_flags -- this parameter indicates the inode's locks to be
231  *       to be unlocked.  See the comment for xfs_ilock() for a list
232  *	 of valid values for this parameter.
233  *
234  */
235 void
236 xfs_iunlock(
237 	xfs_inode_t		*ip,
238 	uint			lock_flags)
239 {
240 	/*
241 	 * You can't set both SHARED and EXCL for the same lock,
242 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
243 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
244 	 */
245 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
246 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
247 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
248 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
249 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
250 	ASSERT(lock_flags != 0);
251 
252 	if (lock_flags & XFS_IOLOCK_EXCL)
253 		mrunlock_excl(&ip->i_iolock);
254 	else if (lock_flags & XFS_IOLOCK_SHARED)
255 		mrunlock_shared(&ip->i_iolock);
256 
257 	if (lock_flags & XFS_ILOCK_EXCL)
258 		mrunlock_excl(&ip->i_lock);
259 	else if (lock_flags & XFS_ILOCK_SHARED)
260 		mrunlock_shared(&ip->i_lock);
261 
262 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
263 }
264 
265 /*
266  * give up write locks.  the i/o lock cannot be held nested
267  * if it is being demoted.
268  */
269 void
270 xfs_ilock_demote(
271 	xfs_inode_t		*ip,
272 	uint			lock_flags)
273 {
274 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
275 	ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
276 
277 	if (lock_flags & XFS_ILOCK_EXCL)
278 		mrdemote(&ip->i_lock);
279 	if (lock_flags & XFS_IOLOCK_EXCL)
280 		mrdemote(&ip->i_iolock);
281 
282 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
283 }
284 
285 #if defined(DEBUG) || defined(XFS_WARN)
286 int
287 xfs_isilocked(
288 	xfs_inode_t		*ip,
289 	uint			lock_flags)
290 {
291 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
292 		if (!(lock_flags & XFS_ILOCK_SHARED))
293 			return !!ip->i_lock.mr_writer;
294 		return rwsem_is_locked(&ip->i_lock.mr_lock);
295 	}
296 
297 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
298 		if (!(lock_flags & XFS_IOLOCK_SHARED))
299 			return !!ip->i_iolock.mr_writer;
300 		return rwsem_is_locked(&ip->i_iolock.mr_lock);
301 	}
302 
303 	ASSERT(0);
304 	return 0;
305 }
306 #endif
307 
308 #ifdef DEBUG
309 int xfs_locked_n;
310 int xfs_small_retries;
311 int xfs_middle_retries;
312 int xfs_lots_retries;
313 int xfs_lock_delays;
314 #endif
315 
316 /*
317  * Bump the subclass so xfs_lock_inodes() acquires each lock with
318  * a different value
319  */
320 static inline int
321 xfs_lock_inumorder(int lock_mode, int subclass)
322 {
323 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
324 		lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
325 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
326 		lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
327 
328 	return lock_mode;
329 }
330 
331 /*
332  * The following routine will lock n inodes in exclusive mode.
333  * We assume the caller calls us with the inodes in i_ino order.
334  *
335  * We need to detect deadlock where an inode that we lock
336  * is in the AIL and we start waiting for another inode that is locked
337  * by a thread in a long running transaction (such as truncate). This can
338  * result in deadlock since the long running trans might need to wait
339  * for the inode we just locked in order to push the tail and free space
340  * in the log.
341  */
342 void
343 xfs_lock_inodes(
344 	xfs_inode_t	**ips,
345 	int		inodes,
346 	uint		lock_mode)
347 {
348 	int		attempts = 0, i, j, try_lock;
349 	xfs_log_item_t	*lp;
350 
351 	ASSERT(ips && (inodes >= 2)); /* we need at least two */
352 
353 	try_lock = 0;
354 	i = 0;
355 
356 again:
357 	for (; i < inodes; i++) {
358 		ASSERT(ips[i]);
359 
360 		if (i && (ips[i] == ips[i-1]))	/* Already locked */
361 			continue;
362 
363 		/*
364 		 * If try_lock is not set yet, make sure all locked inodes
365 		 * are not in the AIL.
366 		 * If any are, set try_lock to be used later.
367 		 */
368 
369 		if (!try_lock) {
370 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
371 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
372 				if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
373 					try_lock++;
374 				}
375 			}
376 		}
377 
378 		/*
379 		 * If any of the previous locks we have locked is in the AIL,
380 		 * we must TRY to get the second and subsequent locks. If
381 		 * we can't get any, we must release all we have
382 		 * and try again.
383 		 */
384 
385 		if (try_lock) {
386 			/* try_lock must be 0 if i is 0. */
387 			/*
388 			 * try_lock means we have an inode locked
389 			 * that is in the AIL.
390 			 */
391 			ASSERT(i != 0);
392 			if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
393 				attempts++;
394 
395 				/*
396 				 * Unlock all previous guys and try again.
397 				 * xfs_iunlock will try to push the tail
398 				 * if the inode is in the AIL.
399 				 */
400 
401 				for(j = i - 1; j >= 0; j--) {
402 
403 					/*
404 					 * Check to see if we've already
405 					 * unlocked this one.
406 					 * Not the first one going back,
407 					 * and the inode ptr is the same.
408 					 */
409 					if ((j != (i - 1)) && ips[j] ==
410 								ips[j+1])
411 						continue;
412 
413 					xfs_iunlock(ips[j], lock_mode);
414 				}
415 
416 				if ((attempts % 5) == 0) {
417 					delay(1); /* Don't just spin the CPU */
418 #ifdef DEBUG
419 					xfs_lock_delays++;
420 #endif
421 				}
422 				i = 0;
423 				try_lock = 0;
424 				goto again;
425 			}
426 		} else {
427 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
428 		}
429 	}
430 
431 #ifdef DEBUG
432 	if (attempts) {
433 		if (attempts < 5) xfs_small_retries++;
434 		else if (attempts < 100) xfs_middle_retries++;
435 		else xfs_lots_retries++;
436 	} else {
437 		xfs_locked_n++;
438 	}
439 #endif
440 }
441 
442 /*
443  * xfs_lock_two_inodes() can only be used to lock one type of lock
444  * at a time - the iolock or the ilock, but not both at once. If
445  * we lock both at once, lockdep will report false positives saying
446  * we have violated locking orders.
447  */
448 void
449 xfs_lock_two_inodes(
450 	xfs_inode_t		*ip0,
451 	xfs_inode_t		*ip1,
452 	uint			lock_mode)
453 {
454 	xfs_inode_t		*temp;
455 	int			attempts = 0;
456 	xfs_log_item_t		*lp;
457 
458 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
459 		ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
460 	ASSERT(ip0->i_ino != ip1->i_ino);
461 
462 	if (ip0->i_ino > ip1->i_ino) {
463 		temp = ip0;
464 		ip0 = ip1;
465 		ip1 = temp;
466 	}
467 
468  again:
469 	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
470 
471 	/*
472 	 * If the first lock we have locked is in the AIL, we must TRY to get
473 	 * the second lock. If we can't get it, we must release the first one
474 	 * and try again.
475 	 */
476 	lp = (xfs_log_item_t *)ip0->i_itemp;
477 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
478 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
479 			xfs_iunlock(ip0, lock_mode);
480 			if ((++attempts % 5) == 0)
481 				delay(1); /* Don't just spin the CPU */
482 			goto again;
483 		}
484 	} else {
485 		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
486 	}
487 }
488 
489 
490 void
491 __xfs_iflock(
492 	struct xfs_inode	*ip)
493 {
494 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
495 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
496 
497 	do {
498 		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
499 		if (xfs_isiflocked(ip))
500 			io_schedule();
501 	} while (!xfs_iflock_nowait(ip));
502 
503 	finish_wait(wq, &wait.wait);
504 }
505 
506 STATIC uint
507 _xfs_dic2xflags(
508 	__uint16_t		di_flags)
509 {
510 	uint			flags = 0;
511 
512 	if (di_flags & XFS_DIFLAG_ANY) {
513 		if (di_flags & XFS_DIFLAG_REALTIME)
514 			flags |= XFS_XFLAG_REALTIME;
515 		if (di_flags & XFS_DIFLAG_PREALLOC)
516 			flags |= XFS_XFLAG_PREALLOC;
517 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
518 			flags |= XFS_XFLAG_IMMUTABLE;
519 		if (di_flags & XFS_DIFLAG_APPEND)
520 			flags |= XFS_XFLAG_APPEND;
521 		if (di_flags & XFS_DIFLAG_SYNC)
522 			flags |= XFS_XFLAG_SYNC;
523 		if (di_flags & XFS_DIFLAG_NOATIME)
524 			flags |= XFS_XFLAG_NOATIME;
525 		if (di_flags & XFS_DIFLAG_NODUMP)
526 			flags |= XFS_XFLAG_NODUMP;
527 		if (di_flags & XFS_DIFLAG_RTINHERIT)
528 			flags |= XFS_XFLAG_RTINHERIT;
529 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
530 			flags |= XFS_XFLAG_PROJINHERIT;
531 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
532 			flags |= XFS_XFLAG_NOSYMLINKS;
533 		if (di_flags & XFS_DIFLAG_EXTSIZE)
534 			flags |= XFS_XFLAG_EXTSIZE;
535 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
536 			flags |= XFS_XFLAG_EXTSZINHERIT;
537 		if (di_flags & XFS_DIFLAG_NODEFRAG)
538 			flags |= XFS_XFLAG_NODEFRAG;
539 		if (di_flags & XFS_DIFLAG_FILESTREAM)
540 			flags |= XFS_XFLAG_FILESTREAM;
541 	}
542 
543 	return flags;
544 }
545 
546 uint
547 xfs_ip2xflags(
548 	xfs_inode_t		*ip)
549 {
550 	xfs_icdinode_t		*dic = &ip->i_d;
551 
552 	return _xfs_dic2xflags(dic->di_flags) |
553 				(XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
554 }
555 
556 uint
557 xfs_dic2xflags(
558 	xfs_dinode_t		*dip)
559 {
560 	return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
561 				(XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
562 }
563 
564 /*
565  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
566  * is allowed, otherwise it has to be an exact match. If a CI match is found,
567  * ci_name->name will point to a the actual name (caller must free) or
568  * will be set to NULL if an exact match is found.
569  */
570 int
571 xfs_lookup(
572 	xfs_inode_t		*dp,
573 	struct xfs_name		*name,
574 	xfs_inode_t		**ipp,
575 	struct xfs_name		*ci_name)
576 {
577 	xfs_ino_t		inum;
578 	int			error;
579 	uint			lock_mode;
580 
581 	trace_xfs_lookup(dp, name);
582 
583 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
584 		return -EIO;
585 
586 	lock_mode = xfs_ilock_data_map_shared(dp);
587 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
588 	xfs_iunlock(dp, lock_mode);
589 
590 	if (error)
591 		goto out;
592 
593 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
594 	if (error)
595 		goto out_free_name;
596 
597 	return 0;
598 
599 out_free_name:
600 	if (ci_name)
601 		kmem_free(ci_name->name);
602 out:
603 	*ipp = NULL;
604 	return error;
605 }
606 
607 /*
608  * Allocate an inode on disk and return a copy of its in-core version.
609  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
610  * appropriately within the inode.  The uid and gid for the inode are
611  * set according to the contents of the given cred structure.
612  *
613  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
614  * has a free inode available, call xfs_iget() to obtain the in-core
615  * version of the allocated inode.  Finally, fill in the inode and
616  * log its initial contents.  In this case, ialloc_context would be
617  * set to NULL.
618  *
619  * If xfs_dialloc() does not have an available inode, it will replenish
620  * its supply by doing an allocation. Since we can only do one
621  * allocation within a transaction without deadlocks, we must commit
622  * the current transaction before returning the inode itself.
623  * In this case, therefore, we will set ialloc_context and return.
624  * The caller should then commit the current transaction, start a new
625  * transaction, and call xfs_ialloc() again to actually get the inode.
626  *
627  * To ensure that some other process does not grab the inode that
628  * was allocated during the first call to xfs_ialloc(), this routine
629  * also returns the [locked] bp pointing to the head of the freelist
630  * as ialloc_context.  The caller should hold this buffer across
631  * the commit and pass it back into this routine on the second call.
632  *
633  * If we are allocating quota inodes, we do not have a parent inode
634  * to attach to or associate with (i.e. pip == NULL) because they
635  * are not linked into the directory structure - they are attached
636  * directly to the superblock - and so have no parent.
637  */
638 int
639 xfs_ialloc(
640 	xfs_trans_t	*tp,
641 	xfs_inode_t	*pip,
642 	umode_t		mode,
643 	xfs_nlink_t	nlink,
644 	xfs_dev_t	rdev,
645 	prid_t		prid,
646 	int		okalloc,
647 	xfs_buf_t	**ialloc_context,
648 	xfs_inode_t	**ipp)
649 {
650 	struct xfs_mount *mp = tp->t_mountp;
651 	xfs_ino_t	ino;
652 	xfs_inode_t	*ip;
653 	uint		flags;
654 	int		error;
655 	struct timespec	tv;
656 
657 	/*
658 	 * Call the space management code to pick
659 	 * the on-disk inode to be allocated.
660 	 */
661 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
662 			    ialloc_context, &ino);
663 	if (error)
664 		return error;
665 	if (*ialloc_context || ino == NULLFSINO) {
666 		*ipp = NULL;
667 		return 0;
668 	}
669 	ASSERT(*ialloc_context == NULL);
670 
671 	/*
672 	 * Get the in-core inode with the lock held exclusively.
673 	 * This is because we're setting fields here we need
674 	 * to prevent others from looking at until we're done.
675 	 */
676 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
677 			 XFS_ILOCK_EXCL, &ip);
678 	if (error)
679 		return error;
680 	ASSERT(ip != NULL);
681 
682 	/*
683 	 * We always convert v1 inodes to v2 now - we only support filesystems
684 	 * with >= v2 inode capability, so there is no reason for ever leaving
685 	 * an inode in v1 format.
686 	 */
687 	if (ip->i_d.di_version == 1)
688 		ip->i_d.di_version = 2;
689 
690 	ip->i_d.di_mode = mode;
691 	ip->i_d.di_onlink = 0;
692 	ip->i_d.di_nlink = nlink;
693 	ASSERT(ip->i_d.di_nlink == nlink);
694 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
695 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
696 	xfs_set_projid(ip, prid);
697 	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
698 
699 	if (pip && XFS_INHERIT_GID(pip)) {
700 		ip->i_d.di_gid = pip->i_d.di_gid;
701 		if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
702 			ip->i_d.di_mode |= S_ISGID;
703 		}
704 	}
705 
706 	/*
707 	 * If the group ID of the new file does not match the effective group
708 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
709 	 * (and only if the irix_sgid_inherit compatibility variable is set).
710 	 */
711 	if ((irix_sgid_inherit) &&
712 	    (ip->i_d.di_mode & S_ISGID) &&
713 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
714 		ip->i_d.di_mode &= ~S_ISGID;
715 	}
716 
717 	ip->i_d.di_size = 0;
718 	ip->i_d.di_nextents = 0;
719 	ASSERT(ip->i_d.di_nblocks == 0);
720 
721 	tv = current_fs_time(mp->m_super);
722 	ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
723 	ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
724 	ip->i_d.di_atime = ip->i_d.di_mtime;
725 	ip->i_d.di_ctime = ip->i_d.di_mtime;
726 
727 	/*
728 	 * di_gen will have been taken care of in xfs_iread.
729 	 */
730 	ip->i_d.di_extsize = 0;
731 	ip->i_d.di_dmevmask = 0;
732 	ip->i_d.di_dmstate = 0;
733 	ip->i_d.di_flags = 0;
734 
735 	if (ip->i_d.di_version == 3) {
736 		ASSERT(ip->i_d.di_ino == ino);
737 		ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
738 		ip->i_d.di_crc = 0;
739 		ip->i_d.di_changecount = 1;
740 		ip->i_d.di_lsn = 0;
741 		ip->i_d.di_flags2 = 0;
742 		memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
743 		ip->i_d.di_crtime = ip->i_d.di_mtime;
744 	}
745 
746 
747 	flags = XFS_ILOG_CORE;
748 	switch (mode & S_IFMT) {
749 	case S_IFIFO:
750 	case S_IFCHR:
751 	case S_IFBLK:
752 	case S_IFSOCK:
753 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
754 		ip->i_df.if_u2.if_rdev = rdev;
755 		ip->i_df.if_flags = 0;
756 		flags |= XFS_ILOG_DEV;
757 		break;
758 	case S_IFREG:
759 	case S_IFDIR:
760 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
761 			uint	di_flags = 0;
762 
763 			if (S_ISDIR(mode)) {
764 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
765 					di_flags |= XFS_DIFLAG_RTINHERIT;
766 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
767 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
768 					ip->i_d.di_extsize = pip->i_d.di_extsize;
769 				}
770 				if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
771 					di_flags |= XFS_DIFLAG_PROJINHERIT;
772 			} else if (S_ISREG(mode)) {
773 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
774 					di_flags |= XFS_DIFLAG_REALTIME;
775 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
776 					di_flags |= XFS_DIFLAG_EXTSIZE;
777 					ip->i_d.di_extsize = pip->i_d.di_extsize;
778 				}
779 			}
780 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
781 			    xfs_inherit_noatime)
782 				di_flags |= XFS_DIFLAG_NOATIME;
783 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
784 			    xfs_inherit_nodump)
785 				di_flags |= XFS_DIFLAG_NODUMP;
786 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
787 			    xfs_inherit_sync)
788 				di_flags |= XFS_DIFLAG_SYNC;
789 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
790 			    xfs_inherit_nosymlinks)
791 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
792 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
793 			    xfs_inherit_nodefrag)
794 				di_flags |= XFS_DIFLAG_NODEFRAG;
795 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
796 				di_flags |= XFS_DIFLAG_FILESTREAM;
797 			ip->i_d.di_flags |= di_flags;
798 		}
799 		/* FALLTHROUGH */
800 	case S_IFLNK:
801 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
802 		ip->i_df.if_flags = XFS_IFEXTENTS;
803 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
804 		ip->i_df.if_u1.if_extents = NULL;
805 		break;
806 	default:
807 		ASSERT(0);
808 	}
809 	/*
810 	 * Attribute fork settings for new inode.
811 	 */
812 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
813 	ip->i_d.di_anextents = 0;
814 
815 	/*
816 	 * Log the new values stuffed into the inode.
817 	 */
818 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
819 	xfs_trans_log_inode(tp, ip, flags);
820 
821 	/* now that we have an i_mode we can setup inode ops and unlock */
822 	xfs_setup_inode(ip);
823 
824 	*ipp = ip;
825 	return 0;
826 }
827 
828 /*
829  * Allocates a new inode from disk and return a pointer to the
830  * incore copy. This routine will internally commit the current
831  * transaction and allocate a new one if the Space Manager needed
832  * to do an allocation to replenish the inode free-list.
833  *
834  * This routine is designed to be called from xfs_create and
835  * xfs_create_dir.
836  *
837  */
838 int
839 xfs_dir_ialloc(
840 	xfs_trans_t	**tpp,		/* input: current transaction;
841 					   output: may be a new transaction. */
842 	xfs_inode_t	*dp,		/* directory within whose allocate
843 					   the inode. */
844 	umode_t		mode,
845 	xfs_nlink_t	nlink,
846 	xfs_dev_t	rdev,
847 	prid_t		prid,		/* project id */
848 	int		okalloc,	/* ok to allocate new space */
849 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
850 					   locked. */
851 	int		*committed)
852 
853 {
854 	xfs_trans_t	*tp;
855 	xfs_trans_t	*ntp;
856 	xfs_inode_t	*ip;
857 	xfs_buf_t	*ialloc_context = NULL;
858 	int		code;
859 	void		*dqinfo;
860 	uint		tflags;
861 
862 	tp = *tpp;
863 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
864 
865 	/*
866 	 * xfs_ialloc will return a pointer to an incore inode if
867 	 * the Space Manager has an available inode on the free
868 	 * list. Otherwise, it will do an allocation and replenish
869 	 * the freelist.  Since we can only do one allocation per
870 	 * transaction without deadlocks, we will need to commit the
871 	 * current transaction and start a new one.  We will then
872 	 * need to call xfs_ialloc again to get the inode.
873 	 *
874 	 * If xfs_ialloc did an allocation to replenish the freelist,
875 	 * it returns the bp containing the head of the freelist as
876 	 * ialloc_context. We will hold a lock on it across the
877 	 * transaction commit so that no other process can steal
878 	 * the inode(s) that we've just allocated.
879 	 */
880 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
881 			  &ialloc_context, &ip);
882 
883 	/*
884 	 * Return an error if we were unable to allocate a new inode.
885 	 * This should only happen if we run out of space on disk or
886 	 * encounter a disk error.
887 	 */
888 	if (code) {
889 		*ipp = NULL;
890 		return code;
891 	}
892 	if (!ialloc_context && !ip) {
893 		*ipp = NULL;
894 		return -ENOSPC;
895 	}
896 
897 	/*
898 	 * If the AGI buffer is non-NULL, then we were unable to get an
899 	 * inode in one operation.  We need to commit the current
900 	 * transaction and call xfs_ialloc() again.  It is guaranteed
901 	 * to succeed the second time.
902 	 */
903 	if (ialloc_context) {
904 		struct xfs_trans_res tres;
905 
906 		/*
907 		 * Normally, xfs_trans_commit releases all the locks.
908 		 * We call bhold to hang on to the ialloc_context across
909 		 * the commit.  Holding this buffer prevents any other
910 		 * processes from doing any allocations in this
911 		 * allocation group.
912 		 */
913 		xfs_trans_bhold(tp, ialloc_context);
914 		/*
915 		 * Save the log reservation so we can use
916 		 * them in the next transaction.
917 		 */
918 		tres.tr_logres = xfs_trans_get_log_res(tp);
919 		tres.tr_logcount = xfs_trans_get_log_count(tp);
920 
921 		/*
922 		 * We want the quota changes to be associated with the next
923 		 * transaction, NOT this one. So, detach the dqinfo from this
924 		 * and attach it to the next transaction.
925 		 */
926 		dqinfo = NULL;
927 		tflags = 0;
928 		if (tp->t_dqinfo) {
929 			dqinfo = (void *)tp->t_dqinfo;
930 			tp->t_dqinfo = NULL;
931 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
932 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
933 		}
934 
935 		ntp = xfs_trans_dup(tp);
936 		code = xfs_trans_commit(tp, 0);
937 		tp = ntp;
938 		if (committed != NULL) {
939 			*committed = 1;
940 		}
941 		/*
942 		 * If we get an error during the commit processing,
943 		 * release the buffer that is still held and return
944 		 * to the caller.
945 		 */
946 		if (code) {
947 			xfs_buf_relse(ialloc_context);
948 			if (dqinfo) {
949 				tp->t_dqinfo = dqinfo;
950 				xfs_trans_free_dqinfo(tp);
951 			}
952 			*tpp = ntp;
953 			*ipp = NULL;
954 			return code;
955 		}
956 
957 		/*
958 		 * transaction commit worked ok so we can drop the extra ticket
959 		 * reference that we gained in xfs_trans_dup()
960 		 */
961 		xfs_log_ticket_put(tp->t_ticket);
962 		tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
963 		code = xfs_trans_reserve(tp, &tres, 0, 0);
964 
965 		/*
966 		 * Re-attach the quota info that we detached from prev trx.
967 		 */
968 		if (dqinfo) {
969 			tp->t_dqinfo = dqinfo;
970 			tp->t_flags |= tflags;
971 		}
972 
973 		if (code) {
974 			xfs_buf_relse(ialloc_context);
975 			*tpp = ntp;
976 			*ipp = NULL;
977 			return code;
978 		}
979 		xfs_trans_bjoin(tp, ialloc_context);
980 
981 		/*
982 		 * Call ialloc again. Since we've locked out all
983 		 * other allocations in this allocation group,
984 		 * this call should always succeed.
985 		 */
986 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
987 				  okalloc, &ialloc_context, &ip);
988 
989 		/*
990 		 * If we get an error at this point, return to the caller
991 		 * so that the current transaction can be aborted.
992 		 */
993 		if (code) {
994 			*tpp = tp;
995 			*ipp = NULL;
996 			return code;
997 		}
998 		ASSERT(!ialloc_context && ip);
999 
1000 	} else {
1001 		if (committed != NULL)
1002 			*committed = 0;
1003 	}
1004 
1005 	*ipp = ip;
1006 	*tpp = tp;
1007 
1008 	return 0;
1009 }
1010 
1011 /*
1012  * Decrement the link count on an inode & log the change.
1013  * If this causes the link count to go to zero, initiate the
1014  * logging activity required to truncate a file.
1015  */
1016 int				/* error */
1017 xfs_droplink(
1018 	xfs_trans_t *tp,
1019 	xfs_inode_t *ip)
1020 {
1021 	int	error;
1022 
1023 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1024 
1025 	ASSERT (ip->i_d.di_nlink > 0);
1026 	ip->i_d.di_nlink--;
1027 	drop_nlink(VFS_I(ip));
1028 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1029 
1030 	error = 0;
1031 	if (ip->i_d.di_nlink == 0) {
1032 		/*
1033 		 * We're dropping the last link to this file.
1034 		 * Move the on-disk inode to the AGI unlinked list.
1035 		 * From xfs_inactive() we will pull the inode from
1036 		 * the list and free it.
1037 		 */
1038 		error = xfs_iunlink(tp, ip);
1039 	}
1040 	return error;
1041 }
1042 
1043 /*
1044  * Increment the link count on an inode & log the change.
1045  */
1046 int
1047 xfs_bumplink(
1048 	xfs_trans_t *tp,
1049 	xfs_inode_t *ip)
1050 {
1051 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1052 
1053 	ASSERT(ip->i_d.di_version > 1);
1054 	ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1055 	ip->i_d.di_nlink++;
1056 	inc_nlink(VFS_I(ip));
1057 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1058 	return 0;
1059 }
1060 
1061 int
1062 xfs_create(
1063 	xfs_inode_t		*dp,
1064 	struct xfs_name		*name,
1065 	umode_t			mode,
1066 	xfs_dev_t		rdev,
1067 	xfs_inode_t		**ipp)
1068 {
1069 	int			is_dir = S_ISDIR(mode);
1070 	struct xfs_mount	*mp = dp->i_mount;
1071 	struct xfs_inode	*ip = NULL;
1072 	struct xfs_trans	*tp = NULL;
1073 	int			error;
1074 	xfs_bmap_free_t		free_list;
1075 	xfs_fsblock_t		first_block;
1076 	bool                    unlock_dp_on_error = false;
1077 	uint			cancel_flags;
1078 	int			committed;
1079 	prid_t			prid;
1080 	struct xfs_dquot	*udqp = NULL;
1081 	struct xfs_dquot	*gdqp = NULL;
1082 	struct xfs_dquot	*pdqp = NULL;
1083 	struct xfs_trans_res	*tres;
1084 	uint			resblks;
1085 
1086 	trace_xfs_create(dp, name);
1087 
1088 	if (XFS_FORCED_SHUTDOWN(mp))
1089 		return -EIO;
1090 
1091 	prid = xfs_get_initial_prid(dp);
1092 
1093 	/*
1094 	 * Make sure that we have allocated dquot(s) on disk.
1095 	 */
1096 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1097 					xfs_kgid_to_gid(current_fsgid()), prid,
1098 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1099 					&udqp, &gdqp, &pdqp);
1100 	if (error)
1101 		return error;
1102 
1103 	if (is_dir) {
1104 		rdev = 0;
1105 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1106 		tres = &M_RES(mp)->tr_mkdir;
1107 		tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1108 	} else {
1109 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1110 		tres = &M_RES(mp)->tr_create;
1111 		tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1112 	}
1113 
1114 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1115 
1116 	/*
1117 	 * Initially assume that the file does not exist and
1118 	 * reserve the resources for that case.  If that is not
1119 	 * the case we'll drop the one we have and get a more
1120 	 * appropriate transaction later.
1121 	 */
1122 	error = xfs_trans_reserve(tp, tres, resblks, 0);
1123 	if (error == -ENOSPC) {
1124 		/* flush outstanding delalloc blocks and retry */
1125 		xfs_flush_inodes(mp);
1126 		error = xfs_trans_reserve(tp, tres, resblks, 0);
1127 	}
1128 	if (error == -ENOSPC) {
1129 		/* No space at all so try a "no-allocation" reservation */
1130 		resblks = 0;
1131 		error = xfs_trans_reserve(tp, tres, 0, 0);
1132 	}
1133 	if (error) {
1134 		cancel_flags = 0;
1135 		goto out_trans_cancel;
1136 	}
1137 
1138 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1139 	unlock_dp_on_error = true;
1140 
1141 	xfs_bmap_init(&free_list, &first_block);
1142 
1143 	/*
1144 	 * Reserve disk quota and the inode.
1145 	 */
1146 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1147 						pdqp, resblks, 1, 0);
1148 	if (error)
1149 		goto out_trans_cancel;
1150 
1151 	if (!resblks) {
1152 		error = xfs_dir_canenter(tp, dp, name);
1153 		if (error)
1154 			goto out_trans_cancel;
1155 	}
1156 
1157 	/*
1158 	 * A newly created regular or special file just has one directory
1159 	 * entry pointing to them, but a directory also the "." entry
1160 	 * pointing to itself.
1161 	 */
1162 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1163 			       prid, resblks > 0, &ip, &committed);
1164 	if (error) {
1165 		if (error == -ENOSPC)
1166 			goto out_trans_cancel;
1167 		goto out_trans_abort;
1168 	}
1169 
1170 	/*
1171 	 * Now we join the directory inode to the transaction.  We do not do it
1172 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1173 	 * (and release all the locks).  An error from here on will result in
1174 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1175 	 * error path.
1176 	 */
1177 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1178 	unlock_dp_on_error = false;
1179 
1180 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1181 					&first_block, &free_list, resblks ?
1182 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1183 	if (error) {
1184 		ASSERT(error != -ENOSPC);
1185 		goto out_trans_abort;
1186 	}
1187 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1188 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1189 
1190 	if (is_dir) {
1191 		error = xfs_dir_init(tp, ip, dp);
1192 		if (error)
1193 			goto out_bmap_cancel;
1194 
1195 		error = xfs_bumplink(tp, dp);
1196 		if (error)
1197 			goto out_bmap_cancel;
1198 	}
1199 
1200 	/*
1201 	 * If this is a synchronous mount, make sure that the
1202 	 * create transaction goes to disk before returning to
1203 	 * the user.
1204 	 */
1205 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1206 		xfs_trans_set_sync(tp);
1207 
1208 	/*
1209 	 * Attach the dquot(s) to the inodes and modify them incore.
1210 	 * These ids of the inode couldn't have changed since the new
1211 	 * inode has been locked ever since it was created.
1212 	 */
1213 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1214 
1215 	error = xfs_bmap_finish(&tp, &free_list, &committed);
1216 	if (error)
1217 		goto out_bmap_cancel;
1218 
1219 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1220 	if (error)
1221 		goto out_release_inode;
1222 
1223 	xfs_qm_dqrele(udqp);
1224 	xfs_qm_dqrele(gdqp);
1225 	xfs_qm_dqrele(pdqp);
1226 
1227 	*ipp = ip;
1228 	return 0;
1229 
1230  out_bmap_cancel:
1231 	xfs_bmap_cancel(&free_list);
1232  out_trans_abort:
1233 	cancel_flags |= XFS_TRANS_ABORT;
1234  out_trans_cancel:
1235 	xfs_trans_cancel(tp, cancel_flags);
1236  out_release_inode:
1237 	/*
1238 	 * Wait until after the current transaction is aborted to
1239 	 * release the inode.  This prevents recursive transactions
1240 	 * and deadlocks from xfs_inactive.
1241 	 */
1242 	if (ip)
1243 		IRELE(ip);
1244 
1245 	xfs_qm_dqrele(udqp);
1246 	xfs_qm_dqrele(gdqp);
1247 	xfs_qm_dqrele(pdqp);
1248 
1249 	if (unlock_dp_on_error)
1250 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1251 	return error;
1252 }
1253 
1254 int
1255 xfs_create_tmpfile(
1256 	struct xfs_inode	*dp,
1257 	struct dentry		*dentry,
1258 	umode_t			mode,
1259 	struct xfs_inode	**ipp)
1260 {
1261 	struct xfs_mount	*mp = dp->i_mount;
1262 	struct xfs_inode	*ip = NULL;
1263 	struct xfs_trans	*tp = NULL;
1264 	int			error;
1265 	uint			cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1266 	prid_t                  prid;
1267 	struct xfs_dquot	*udqp = NULL;
1268 	struct xfs_dquot	*gdqp = NULL;
1269 	struct xfs_dquot	*pdqp = NULL;
1270 	struct xfs_trans_res	*tres;
1271 	uint			resblks;
1272 
1273 	if (XFS_FORCED_SHUTDOWN(mp))
1274 		return -EIO;
1275 
1276 	prid = xfs_get_initial_prid(dp);
1277 
1278 	/*
1279 	 * Make sure that we have allocated dquot(s) on disk.
1280 	 */
1281 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1282 				xfs_kgid_to_gid(current_fsgid()), prid,
1283 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1284 				&udqp, &gdqp, &pdqp);
1285 	if (error)
1286 		return error;
1287 
1288 	resblks = XFS_IALLOC_SPACE_RES(mp);
1289 	tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1290 
1291 	tres = &M_RES(mp)->tr_create_tmpfile;
1292 	error = xfs_trans_reserve(tp, tres, resblks, 0);
1293 	if (error == -ENOSPC) {
1294 		/* No space at all so try a "no-allocation" reservation */
1295 		resblks = 0;
1296 		error = xfs_trans_reserve(tp, tres, 0, 0);
1297 	}
1298 	if (error) {
1299 		cancel_flags = 0;
1300 		goto out_trans_cancel;
1301 	}
1302 
1303 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1304 						pdqp, resblks, 1, 0);
1305 	if (error)
1306 		goto out_trans_cancel;
1307 
1308 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1309 				prid, resblks > 0, &ip, NULL);
1310 	if (error) {
1311 		if (error == -ENOSPC)
1312 			goto out_trans_cancel;
1313 		goto out_trans_abort;
1314 	}
1315 
1316 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1317 		xfs_trans_set_sync(tp);
1318 
1319 	/*
1320 	 * Attach the dquot(s) to the inodes and modify them incore.
1321 	 * These ids of the inode couldn't have changed since the new
1322 	 * inode has been locked ever since it was created.
1323 	 */
1324 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1325 
1326 	ip->i_d.di_nlink--;
1327 	error = xfs_iunlink(tp, ip);
1328 	if (error)
1329 		goto out_trans_abort;
1330 
1331 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1332 	if (error)
1333 		goto out_release_inode;
1334 
1335 	xfs_qm_dqrele(udqp);
1336 	xfs_qm_dqrele(gdqp);
1337 	xfs_qm_dqrele(pdqp);
1338 
1339 	*ipp = ip;
1340 	return 0;
1341 
1342  out_trans_abort:
1343 	cancel_flags |= XFS_TRANS_ABORT;
1344  out_trans_cancel:
1345 	xfs_trans_cancel(tp, cancel_flags);
1346  out_release_inode:
1347 	/*
1348 	 * Wait until after the current transaction is aborted to
1349 	 * release the inode.  This prevents recursive transactions
1350 	 * and deadlocks from xfs_inactive.
1351 	 */
1352 	if (ip)
1353 		IRELE(ip);
1354 
1355 	xfs_qm_dqrele(udqp);
1356 	xfs_qm_dqrele(gdqp);
1357 	xfs_qm_dqrele(pdqp);
1358 
1359 	return error;
1360 }
1361 
1362 int
1363 xfs_link(
1364 	xfs_inode_t		*tdp,
1365 	xfs_inode_t		*sip,
1366 	struct xfs_name		*target_name)
1367 {
1368 	xfs_mount_t		*mp = tdp->i_mount;
1369 	xfs_trans_t		*tp;
1370 	int			error;
1371 	xfs_bmap_free_t         free_list;
1372 	xfs_fsblock_t           first_block;
1373 	int			cancel_flags;
1374 	int			committed;
1375 	int			resblks;
1376 
1377 	trace_xfs_link(tdp, target_name);
1378 
1379 	ASSERT(!S_ISDIR(sip->i_d.di_mode));
1380 
1381 	if (XFS_FORCED_SHUTDOWN(mp))
1382 		return -EIO;
1383 
1384 	error = xfs_qm_dqattach(sip, 0);
1385 	if (error)
1386 		goto std_return;
1387 
1388 	error = xfs_qm_dqattach(tdp, 0);
1389 	if (error)
1390 		goto std_return;
1391 
1392 	tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1393 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1394 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1395 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1396 	if (error == -ENOSPC) {
1397 		resblks = 0;
1398 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1399 	}
1400 	if (error) {
1401 		cancel_flags = 0;
1402 		goto error_return;
1403 	}
1404 
1405 	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1406 
1407 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1408 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1409 
1410 	/*
1411 	 * If we are using project inheritance, we only allow hard link
1412 	 * creation in our tree when the project IDs are the same; else
1413 	 * the tree quota mechanism could be circumvented.
1414 	 */
1415 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1416 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1417 		error = -EXDEV;
1418 		goto error_return;
1419 	}
1420 
1421 	if (!resblks) {
1422 		error = xfs_dir_canenter(tp, tdp, target_name);
1423 		if (error)
1424 			goto error_return;
1425 	}
1426 
1427 	xfs_bmap_init(&free_list, &first_block);
1428 
1429 	if (sip->i_d.di_nlink == 0) {
1430 		error = xfs_iunlink_remove(tp, sip);
1431 		if (error)
1432 			goto abort_return;
1433 	}
1434 
1435 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1436 					&first_block, &free_list, resblks);
1437 	if (error)
1438 		goto abort_return;
1439 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1440 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1441 
1442 	error = xfs_bumplink(tp, sip);
1443 	if (error)
1444 		goto abort_return;
1445 
1446 	/*
1447 	 * If this is a synchronous mount, make sure that the
1448 	 * link transaction goes to disk before returning to
1449 	 * the user.
1450 	 */
1451 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1452 		xfs_trans_set_sync(tp);
1453 	}
1454 
1455 	error = xfs_bmap_finish (&tp, &free_list, &committed);
1456 	if (error) {
1457 		xfs_bmap_cancel(&free_list);
1458 		goto abort_return;
1459 	}
1460 
1461 	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1462 
1463  abort_return:
1464 	cancel_flags |= XFS_TRANS_ABORT;
1465  error_return:
1466 	xfs_trans_cancel(tp, cancel_flags);
1467  std_return:
1468 	return error;
1469 }
1470 
1471 /*
1472  * Free up the underlying blocks past new_size.  The new size must be smaller
1473  * than the current size.  This routine can be used both for the attribute and
1474  * data fork, and does not modify the inode size, which is left to the caller.
1475  *
1476  * The transaction passed to this routine must have made a permanent log
1477  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1478  * given transaction and start new ones, so make sure everything involved in
1479  * the transaction is tidy before calling here.  Some transaction will be
1480  * returned to the caller to be committed.  The incoming transaction must
1481  * already include the inode, and both inode locks must be held exclusively.
1482  * The inode must also be "held" within the transaction.  On return the inode
1483  * will be "held" within the returned transaction.  This routine does NOT
1484  * require any disk space to be reserved for it within the transaction.
1485  *
1486  * If we get an error, we must return with the inode locked and linked into the
1487  * current transaction. This keeps things simple for the higher level code,
1488  * because it always knows that the inode is locked and held in the transaction
1489  * that returns to it whether errors occur or not.  We don't mark the inode
1490  * dirty on error so that transactions can be easily aborted if possible.
1491  */
1492 int
1493 xfs_itruncate_extents(
1494 	struct xfs_trans	**tpp,
1495 	struct xfs_inode	*ip,
1496 	int			whichfork,
1497 	xfs_fsize_t		new_size)
1498 {
1499 	struct xfs_mount	*mp = ip->i_mount;
1500 	struct xfs_trans	*tp = *tpp;
1501 	struct xfs_trans	*ntp;
1502 	xfs_bmap_free_t		free_list;
1503 	xfs_fsblock_t		first_block;
1504 	xfs_fileoff_t		first_unmap_block;
1505 	xfs_fileoff_t		last_block;
1506 	xfs_filblks_t		unmap_len;
1507 	int			committed;
1508 	int			error = 0;
1509 	int			done = 0;
1510 
1511 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1512 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1513 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1514 	ASSERT(new_size <= XFS_ISIZE(ip));
1515 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1516 	ASSERT(ip->i_itemp != NULL);
1517 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1518 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1519 
1520 	trace_xfs_itruncate_extents_start(ip, new_size);
1521 
1522 	/*
1523 	 * Since it is possible for space to become allocated beyond
1524 	 * the end of the file (in a crash where the space is allocated
1525 	 * but the inode size is not yet updated), simply remove any
1526 	 * blocks which show up between the new EOF and the maximum
1527 	 * possible file size.  If the first block to be removed is
1528 	 * beyond the maximum file size (ie it is the same as last_block),
1529 	 * then there is nothing to do.
1530 	 */
1531 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1532 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1533 	if (first_unmap_block == last_block)
1534 		return 0;
1535 
1536 	ASSERT(first_unmap_block < last_block);
1537 	unmap_len = last_block - first_unmap_block + 1;
1538 	while (!done) {
1539 		xfs_bmap_init(&free_list, &first_block);
1540 		error = xfs_bunmapi(tp, ip,
1541 				    first_unmap_block, unmap_len,
1542 				    xfs_bmapi_aflag(whichfork),
1543 				    XFS_ITRUNC_MAX_EXTENTS,
1544 				    &first_block, &free_list,
1545 				    &done);
1546 		if (error)
1547 			goto out_bmap_cancel;
1548 
1549 		/*
1550 		 * Duplicate the transaction that has the permanent
1551 		 * reservation and commit the old transaction.
1552 		 */
1553 		error = xfs_bmap_finish(&tp, &free_list, &committed);
1554 		if (committed)
1555 			xfs_trans_ijoin(tp, ip, 0);
1556 		if (error)
1557 			goto out_bmap_cancel;
1558 
1559 		if (committed) {
1560 			/*
1561 			 * Mark the inode dirty so it will be logged and
1562 			 * moved forward in the log as part of every commit.
1563 			 */
1564 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1565 		}
1566 
1567 		ntp = xfs_trans_dup(tp);
1568 		error = xfs_trans_commit(tp, 0);
1569 		tp = ntp;
1570 
1571 		xfs_trans_ijoin(tp, ip, 0);
1572 
1573 		if (error)
1574 			goto out;
1575 
1576 		/*
1577 		 * Transaction commit worked ok so we can drop the extra ticket
1578 		 * reference that we gained in xfs_trans_dup()
1579 		 */
1580 		xfs_log_ticket_put(tp->t_ticket);
1581 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1582 		if (error)
1583 			goto out;
1584 	}
1585 
1586 	/*
1587 	 * Always re-log the inode so that our permanent transaction can keep
1588 	 * on rolling it forward in the log.
1589 	 */
1590 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1591 
1592 	trace_xfs_itruncate_extents_end(ip, new_size);
1593 
1594 out:
1595 	*tpp = tp;
1596 	return error;
1597 out_bmap_cancel:
1598 	/*
1599 	 * If the bunmapi call encounters an error, return to the caller where
1600 	 * the transaction can be properly aborted.  We just need to make sure
1601 	 * we're not holding any resources that we were not when we came in.
1602 	 */
1603 	xfs_bmap_cancel(&free_list);
1604 	goto out;
1605 }
1606 
1607 int
1608 xfs_release(
1609 	xfs_inode_t	*ip)
1610 {
1611 	xfs_mount_t	*mp = ip->i_mount;
1612 	int		error;
1613 
1614 	if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1615 		return 0;
1616 
1617 	/* If this is a read-only mount, don't do this (would generate I/O) */
1618 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1619 		return 0;
1620 
1621 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1622 		int truncated;
1623 
1624 		/*
1625 		 * If we previously truncated this file and removed old data
1626 		 * in the process, we want to initiate "early" writeout on
1627 		 * the last close.  This is an attempt to combat the notorious
1628 		 * NULL files problem which is particularly noticeable from a
1629 		 * truncate down, buffered (re-)write (delalloc), followed by
1630 		 * a crash.  What we are effectively doing here is
1631 		 * significantly reducing the time window where we'd otherwise
1632 		 * be exposed to that problem.
1633 		 */
1634 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1635 		if (truncated) {
1636 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1637 			if (ip->i_delayed_blks > 0) {
1638 				error = filemap_flush(VFS_I(ip)->i_mapping);
1639 				if (error)
1640 					return error;
1641 			}
1642 		}
1643 	}
1644 
1645 	if (ip->i_d.di_nlink == 0)
1646 		return 0;
1647 
1648 	if (xfs_can_free_eofblocks(ip, false)) {
1649 
1650 		/*
1651 		 * If we can't get the iolock just skip truncating the blocks
1652 		 * past EOF because we could deadlock with the mmap_sem
1653 		 * otherwise.  We'll get another chance to drop them once the
1654 		 * last reference to the inode is dropped, so we'll never leak
1655 		 * blocks permanently.
1656 		 *
1657 		 * Further, check if the inode is being opened, written and
1658 		 * closed frequently and we have delayed allocation blocks
1659 		 * outstanding (e.g. streaming writes from the NFS server),
1660 		 * truncating the blocks past EOF will cause fragmentation to
1661 		 * occur.
1662 		 *
1663 		 * In this case don't do the truncation, either, but we have to
1664 		 * be careful how we detect this case. Blocks beyond EOF show
1665 		 * up as i_delayed_blks even when the inode is clean, so we
1666 		 * need to truncate them away first before checking for a dirty
1667 		 * release. Hence on the first dirty close we will still remove
1668 		 * the speculative allocation, but after that we will leave it
1669 		 * in place.
1670 		 */
1671 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1672 			return 0;
1673 
1674 		error = xfs_free_eofblocks(mp, ip, true);
1675 		if (error && error != -EAGAIN)
1676 			return error;
1677 
1678 		/* delalloc blocks after truncation means it really is dirty */
1679 		if (ip->i_delayed_blks)
1680 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1681 	}
1682 	return 0;
1683 }
1684 
1685 /*
1686  * xfs_inactive_truncate
1687  *
1688  * Called to perform a truncate when an inode becomes unlinked.
1689  */
1690 STATIC int
1691 xfs_inactive_truncate(
1692 	struct xfs_inode *ip)
1693 {
1694 	struct xfs_mount	*mp = ip->i_mount;
1695 	struct xfs_trans	*tp;
1696 	int			error;
1697 
1698 	tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1699 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1700 	if (error) {
1701 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1702 		xfs_trans_cancel(tp, 0);
1703 		return error;
1704 	}
1705 
1706 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1707 	xfs_trans_ijoin(tp, ip, 0);
1708 
1709 	/*
1710 	 * Log the inode size first to prevent stale data exposure in the event
1711 	 * of a system crash before the truncate completes. See the related
1712 	 * comment in xfs_setattr_size() for details.
1713 	 */
1714 	ip->i_d.di_size = 0;
1715 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1716 
1717 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1718 	if (error)
1719 		goto error_trans_cancel;
1720 
1721 	ASSERT(ip->i_d.di_nextents == 0);
1722 
1723 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1724 	if (error)
1725 		goto error_unlock;
1726 
1727 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1728 	return 0;
1729 
1730 error_trans_cancel:
1731 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1732 error_unlock:
1733 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1734 	return error;
1735 }
1736 
1737 /*
1738  * xfs_inactive_ifree()
1739  *
1740  * Perform the inode free when an inode is unlinked.
1741  */
1742 STATIC int
1743 xfs_inactive_ifree(
1744 	struct xfs_inode *ip)
1745 {
1746 	xfs_bmap_free_t		free_list;
1747 	xfs_fsblock_t		first_block;
1748 	int			committed;
1749 	struct xfs_mount	*mp = ip->i_mount;
1750 	struct xfs_trans	*tp;
1751 	int			error;
1752 
1753 	tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1754 
1755 	/*
1756 	 * The ifree transaction might need to allocate blocks for record
1757 	 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1758 	 * allow ifree to dip into the reserved block pool if necessary.
1759 	 *
1760 	 * Freeing large sets of inodes generally means freeing inode chunks,
1761 	 * directory and file data blocks, so this should be relatively safe.
1762 	 * Only under severe circumstances should it be possible to free enough
1763 	 * inodes to exhaust the reserve block pool via finobt expansion while
1764 	 * at the same time not creating free space in the filesystem.
1765 	 *
1766 	 * Send a warning if the reservation does happen to fail, as the inode
1767 	 * now remains allocated and sits on the unlinked list until the fs is
1768 	 * repaired.
1769 	 */
1770 	tp->t_flags |= XFS_TRANS_RESERVE;
1771 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1772 				  XFS_IFREE_SPACE_RES(mp), 0);
1773 	if (error) {
1774 		if (error == -ENOSPC) {
1775 			xfs_warn_ratelimited(mp,
1776 			"Failed to remove inode(s) from unlinked list. "
1777 			"Please free space, unmount and run xfs_repair.");
1778 		} else {
1779 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1780 		}
1781 		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1782 		return error;
1783 	}
1784 
1785 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1786 	xfs_trans_ijoin(tp, ip, 0);
1787 
1788 	xfs_bmap_init(&free_list, &first_block);
1789 	error = xfs_ifree(tp, ip, &free_list);
1790 	if (error) {
1791 		/*
1792 		 * If we fail to free the inode, shut down.  The cancel
1793 		 * might do that, we need to make sure.  Otherwise the
1794 		 * inode might be lost for a long time or forever.
1795 		 */
1796 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1797 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1798 				__func__, error);
1799 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1800 		}
1801 		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1802 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1803 		return error;
1804 	}
1805 
1806 	/*
1807 	 * Credit the quota account(s). The inode is gone.
1808 	 */
1809 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1810 
1811 	/*
1812 	 * Just ignore errors at this point.  There is nothing we can
1813 	 * do except to try to keep going. Make sure it's not a silent
1814 	 * error.
1815 	 */
1816 	error = xfs_bmap_finish(&tp,  &free_list, &committed);
1817 	if (error)
1818 		xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1819 			__func__, error);
1820 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1821 	if (error)
1822 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1823 			__func__, error);
1824 
1825 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1826 	return 0;
1827 }
1828 
1829 /*
1830  * xfs_inactive
1831  *
1832  * This is called when the vnode reference count for the vnode
1833  * goes to zero.  If the file has been unlinked, then it must
1834  * now be truncated.  Also, we clear all of the read-ahead state
1835  * kept for the inode here since the file is now closed.
1836  */
1837 void
1838 xfs_inactive(
1839 	xfs_inode_t	*ip)
1840 {
1841 	struct xfs_mount	*mp;
1842 	int			error;
1843 	int			truncate = 0;
1844 
1845 	/*
1846 	 * If the inode is already free, then there can be nothing
1847 	 * to clean up here.
1848 	 */
1849 	if (ip->i_d.di_mode == 0) {
1850 		ASSERT(ip->i_df.if_real_bytes == 0);
1851 		ASSERT(ip->i_df.if_broot_bytes == 0);
1852 		return;
1853 	}
1854 
1855 	mp = ip->i_mount;
1856 
1857 	/* If this is a read-only mount, don't do this (would generate I/O) */
1858 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1859 		return;
1860 
1861 	if (ip->i_d.di_nlink != 0) {
1862 		/*
1863 		 * force is true because we are evicting an inode from the
1864 		 * cache. Post-eof blocks must be freed, lest we end up with
1865 		 * broken free space accounting.
1866 		 */
1867 		if (xfs_can_free_eofblocks(ip, true))
1868 			xfs_free_eofblocks(mp, ip, false);
1869 
1870 		return;
1871 	}
1872 
1873 	if (S_ISREG(ip->i_d.di_mode) &&
1874 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1875 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1876 		truncate = 1;
1877 
1878 	error = xfs_qm_dqattach(ip, 0);
1879 	if (error)
1880 		return;
1881 
1882 	if (S_ISLNK(ip->i_d.di_mode))
1883 		error = xfs_inactive_symlink(ip);
1884 	else if (truncate)
1885 		error = xfs_inactive_truncate(ip);
1886 	if (error)
1887 		return;
1888 
1889 	/*
1890 	 * If there are attributes associated with the file then blow them away
1891 	 * now.  The code calls a routine that recursively deconstructs the
1892 	 * attribute fork.  We need to just commit the current transaction
1893 	 * because we can't use it for xfs_attr_inactive().
1894 	 */
1895 	if (ip->i_d.di_anextents > 0) {
1896 		ASSERT(ip->i_d.di_forkoff != 0);
1897 
1898 		error = xfs_attr_inactive(ip);
1899 		if (error)
1900 			return;
1901 	}
1902 
1903 	if (ip->i_afp)
1904 		xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1905 
1906 	ASSERT(ip->i_d.di_anextents == 0);
1907 
1908 	/*
1909 	 * Free the inode.
1910 	 */
1911 	error = xfs_inactive_ifree(ip);
1912 	if (error)
1913 		return;
1914 
1915 	/*
1916 	 * Release the dquots held by inode, if any.
1917 	 */
1918 	xfs_qm_dqdetach(ip);
1919 }
1920 
1921 /*
1922  * This is called when the inode's link count goes to 0.
1923  * We place the on-disk inode on a list in the AGI.  It
1924  * will be pulled from this list when the inode is freed.
1925  */
1926 int
1927 xfs_iunlink(
1928 	xfs_trans_t	*tp,
1929 	xfs_inode_t	*ip)
1930 {
1931 	xfs_mount_t	*mp;
1932 	xfs_agi_t	*agi;
1933 	xfs_dinode_t	*dip;
1934 	xfs_buf_t	*agibp;
1935 	xfs_buf_t	*ibp;
1936 	xfs_agino_t	agino;
1937 	short		bucket_index;
1938 	int		offset;
1939 	int		error;
1940 
1941 	ASSERT(ip->i_d.di_nlink == 0);
1942 	ASSERT(ip->i_d.di_mode != 0);
1943 
1944 	mp = tp->t_mountp;
1945 
1946 	/*
1947 	 * Get the agi buffer first.  It ensures lock ordering
1948 	 * on the list.
1949 	 */
1950 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1951 	if (error)
1952 		return error;
1953 	agi = XFS_BUF_TO_AGI(agibp);
1954 
1955 	/*
1956 	 * Get the index into the agi hash table for the
1957 	 * list this inode will go on.
1958 	 */
1959 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1960 	ASSERT(agino != 0);
1961 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1962 	ASSERT(agi->agi_unlinked[bucket_index]);
1963 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1964 
1965 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1966 		/*
1967 		 * There is already another inode in the bucket we need
1968 		 * to add ourselves to.  Add us at the front of the list.
1969 		 * Here we put the head pointer into our next pointer,
1970 		 * and then we fall through to point the head at us.
1971 		 */
1972 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1973 				       0, 0);
1974 		if (error)
1975 			return error;
1976 
1977 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1978 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1979 		offset = ip->i_imap.im_boffset +
1980 			offsetof(xfs_dinode_t, di_next_unlinked);
1981 
1982 		/* need to recalc the inode CRC if appropriate */
1983 		xfs_dinode_calc_crc(mp, dip);
1984 
1985 		xfs_trans_inode_buf(tp, ibp);
1986 		xfs_trans_log_buf(tp, ibp, offset,
1987 				  (offset + sizeof(xfs_agino_t) - 1));
1988 		xfs_inobp_check(mp, ibp);
1989 	}
1990 
1991 	/*
1992 	 * Point the bucket head pointer at the inode being inserted.
1993 	 */
1994 	ASSERT(agino != 0);
1995 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1996 	offset = offsetof(xfs_agi_t, agi_unlinked) +
1997 		(sizeof(xfs_agino_t) * bucket_index);
1998 	xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
1999 	xfs_trans_log_buf(tp, agibp, offset,
2000 			  (offset + sizeof(xfs_agino_t) - 1));
2001 	return 0;
2002 }
2003 
2004 /*
2005  * Pull the on-disk inode from the AGI unlinked list.
2006  */
2007 STATIC int
2008 xfs_iunlink_remove(
2009 	xfs_trans_t	*tp,
2010 	xfs_inode_t	*ip)
2011 {
2012 	xfs_ino_t	next_ino;
2013 	xfs_mount_t	*mp;
2014 	xfs_agi_t	*agi;
2015 	xfs_dinode_t	*dip;
2016 	xfs_buf_t	*agibp;
2017 	xfs_buf_t	*ibp;
2018 	xfs_agnumber_t	agno;
2019 	xfs_agino_t	agino;
2020 	xfs_agino_t	next_agino;
2021 	xfs_buf_t	*last_ibp;
2022 	xfs_dinode_t	*last_dip = NULL;
2023 	short		bucket_index;
2024 	int		offset, last_offset = 0;
2025 	int		error;
2026 
2027 	mp = tp->t_mountp;
2028 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2029 
2030 	/*
2031 	 * Get the agi buffer first.  It ensures lock ordering
2032 	 * on the list.
2033 	 */
2034 	error = xfs_read_agi(mp, tp, agno, &agibp);
2035 	if (error)
2036 		return error;
2037 
2038 	agi = XFS_BUF_TO_AGI(agibp);
2039 
2040 	/*
2041 	 * Get the index into the agi hash table for the
2042 	 * list this inode will go on.
2043 	 */
2044 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2045 	ASSERT(agino != 0);
2046 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2047 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2048 	ASSERT(agi->agi_unlinked[bucket_index]);
2049 
2050 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2051 		/*
2052 		 * We're at the head of the list.  Get the inode's on-disk
2053 		 * buffer to see if there is anyone after us on the list.
2054 		 * Only modify our next pointer if it is not already NULLAGINO.
2055 		 * This saves us the overhead of dealing with the buffer when
2056 		 * there is no need to change it.
2057 		 */
2058 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2059 				       0, 0);
2060 		if (error) {
2061 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2062 				__func__, error);
2063 			return error;
2064 		}
2065 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2066 		ASSERT(next_agino != 0);
2067 		if (next_agino != NULLAGINO) {
2068 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2069 			offset = ip->i_imap.im_boffset +
2070 				offsetof(xfs_dinode_t, di_next_unlinked);
2071 
2072 			/* need to recalc the inode CRC if appropriate */
2073 			xfs_dinode_calc_crc(mp, dip);
2074 
2075 			xfs_trans_inode_buf(tp, ibp);
2076 			xfs_trans_log_buf(tp, ibp, offset,
2077 					  (offset + sizeof(xfs_agino_t) - 1));
2078 			xfs_inobp_check(mp, ibp);
2079 		} else {
2080 			xfs_trans_brelse(tp, ibp);
2081 		}
2082 		/*
2083 		 * Point the bucket head pointer at the next inode.
2084 		 */
2085 		ASSERT(next_agino != 0);
2086 		ASSERT(next_agino != agino);
2087 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2088 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2089 			(sizeof(xfs_agino_t) * bucket_index);
2090 		xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
2091 		xfs_trans_log_buf(tp, agibp, offset,
2092 				  (offset + sizeof(xfs_agino_t) - 1));
2093 	} else {
2094 		/*
2095 		 * We need to search the list for the inode being freed.
2096 		 */
2097 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2098 		last_ibp = NULL;
2099 		while (next_agino != agino) {
2100 			struct xfs_imap	imap;
2101 
2102 			if (last_ibp)
2103 				xfs_trans_brelse(tp, last_ibp);
2104 
2105 			imap.im_blkno = 0;
2106 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2107 
2108 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2109 			if (error) {
2110 				xfs_warn(mp,
2111 	"%s: xfs_imap returned error %d.",
2112 					 __func__, error);
2113 				return error;
2114 			}
2115 
2116 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2117 					       &last_ibp, 0, 0);
2118 			if (error) {
2119 				xfs_warn(mp,
2120 	"%s: xfs_imap_to_bp returned error %d.",
2121 					__func__, error);
2122 				return error;
2123 			}
2124 
2125 			last_offset = imap.im_boffset;
2126 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2127 			ASSERT(next_agino != NULLAGINO);
2128 			ASSERT(next_agino != 0);
2129 		}
2130 
2131 		/*
2132 		 * Now last_ibp points to the buffer previous to us on the
2133 		 * unlinked list.  Pull us from the list.
2134 		 */
2135 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2136 				       0, 0);
2137 		if (error) {
2138 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2139 				__func__, error);
2140 			return error;
2141 		}
2142 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2143 		ASSERT(next_agino != 0);
2144 		ASSERT(next_agino != agino);
2145 		if (next_agino != NULLAGINO) {
2146 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2147 			offset = ip->i_imap.im_boffset +
2148 				offsetof(xfs_dinode_t, di_next_unlinked);
2149 
2150 			/* need to recalc the inode CRC if appropriate */
2151 			xfs_dinode_calc_crc(mp, dip);
2152 
2153 			xfs_trans_inode_buf(tp, ibp);
2154 			xfs_trans_log_buf(tp, ibp, offset,
2155 					  (offset + sizeof(xfs_agino_t) - 1));
2156 			xfs_inobp_check(mp, ibp);
2157 		} else {
2158 			xfs_trans_brelse(tp, ibp);
2159 		}
2160 		/*
2161 		 * Point the previous inode on the list to the next inode.
2162 		 */
2163 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2164 		ASSERT(next_agino != 0);
2165 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2166 
2167 		/* need to recalc the inode CRC if appropriate */
2168 		xfs_dinode_calc_crc(mp, last_dip);
2169 
2170 		xfs_trans_inode_buf(tp, last_ibp);
2171 		xfs_trans_log_buf(tp, last_ibp, offset,
2172 				  (offset + sizeof(xfs_agino_t) - 1));
2173 		xfs_inobp_check(mp, last_ibp);
2174 	}
2175 	return 0;
2176 }
2177 
2178 /*
2179  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2180  * inodes that are in memory - they all must be marked stale and attached to
2181  * the cluster buffer.
2182  */
2183 STATIC int
2184 xfs_ifree_cluster(
2185 	xfs_inode_t	*free_ip,
2186 	xfs_trans_t	*tp,
2187 	xfs_ino_t	inum)
2188 {
2189 	xfs_mount_t		*mp = free_ip->i_mount;
2190 	int			blks_per_cluster;
2191 	int			inodes_per_cluster;
2192 	int			nbufs;
2193 	int			i, j;
2194 	xfs_daddr_t		blkno;
2195 	xfs_buf_t		*bp;
2196 	xfs_inode_t		*ip;
2197 	xfs_inode_log_item_t	*iip;
2198 	xfs_log_item_t		*lip;
2199 	struct xfs_perag	*pag;
2200 
2201 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2202 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2203 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2204 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2205 
2206 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2207 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2208 					 XFS_INO_TO_AGBNO(mp, inum));
2209 
2210 		/*
2211 		 * We obtain and lock the backing buffer first in the process
2212 		 * here, as we have to ensure that any dirty inode that we
2213 		 * can't get the flush lock on is attached to the buffer.
2214 		 * If we scan the in-memory inodes first, then buffer IO can
2215 		 * complete before we get a lock on it, and hence we may fail
2216 		 * to mark all the active inodes on the buffer stale.
2217 		 */
2218 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2219 					mp->m_bsize * blks_per_cluster,
2220 					XBF_UNMAPPED);
2221 
2222 		if (!bp)
2223 			return -ENOMEM;
2224 
2225 		/*
2226 		 * This buffer may not have been correctly initialised as we
2227 		 * didn't read it from disk. That's not important because we are
2228 		 * only using to mark the buffer as stale in the log, and to
2229 		 * attach stale cached inodes on it. That means it will never be
2230 		 * dispatched for IO. If it is, we want to know about it, and we
2231 		 * want it to fail. We can acheive this by adding a write
2232 		 * verifier to the buffer.
2233 		 */
2234 		 bp->b_ops = &xfs_inode_buf_ops;
2235 
2236 		/*
2237 		 * Walk the inodes already attached to the buffer and mark them
2238 		 * stale. These will all have the flush locks held, so an
2239 		 * in-memory inode walk can't lock them. By marking them all
2240 		 * stale first, we will not attempt to lock them in the loop
2241 		 * below as the XFS_ISTALE flag will be set.
2242 		 */
2243 		lip = bp->b_fspriv;
2244 		while (lip) {
2245 			if (lip->li_type == XFS_LI_INODE) {
2246 				iip = (xfs_inode_log_item_t *)lip;
2247 				ASSERT(iip->ili_logged == 1);
2248 				lip->li_cb = xfs_istale_done;
2249 				xfs_trans_ail_copy_lsn(mp->m_ail,
2250 							&iip->ili_flush_lsn,
2251 							&iip->ili_item.li_lsn);
2252 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2253 			}
2254 			lip = lip->li_bio_list;
2255 		}
2256 
2257 
2258 		/*
2259 		 * For each inode in memory attempt to add it to the inode
2260 		 * buffer and set it up for being staled on buffer IO
2261 		 * completion.  This is safe as we've locked out tail pushing
2262 		 * and flushing by locking the buffer.
2263 		 *
2264 		 * We have already marked every inode that was part of a
2265 		 * transaction stale above, which means there is no point in
2266 		 * even trying to lock them.
2267 		 */
2268 		for (i = 0; i < inodes_per_cluster; i++) {
2269 retry:
2270 			rcu_read_lock();
2271 			ip = radix_tree_lookup(&pag->pag_ici_root,
2272 					XFS_INO_TO_AGINO(mp, (inum + i)));
2273 
2274 			/* Inode not in memory, nothing to do */
2275 			if (!ip) {
2276 				rcu_read_unlock();
2277 				continue;
2278 			}
2279 
2280 			/*
2281 			 * because this is an RCU protected lookup, we could
2282 			 * find a recently freed or even reallocated inode
2283 			 * during the lookup. We need to check under the
2284 			 * i_flags_lock for a valid inode here. Skip it if it
2285 			 * is not valid, the wrong inode or stale.
2286 			 */
2287 			spin_lock(&ip->i_flags_lock);
2288 			if (ip->i_ino != inum + i ||
2289 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2290 				spin_unlock(&ip->i_flags_lock);
2291 				rcu_read_unlock();
2292 				continue;
2293 			}
2294 			spin_unlock(&ip->i_flags_lock);
2295 
2296 			/*
2297 			 * Don't try to lock/unlock the current inode, but we
2298 			 * _cannot_ skip the other inodes that we did not find
2299 			 * in the list attached to the buffer and are not
2300 			 * already marked stale. If we can't lock it, back off
2301 			 * and retry.
2302 			 */
2303 			if (ip != free_ip &&
2304 			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2305 				rcu_read_unlock();
2306 				delay(1);
2307 				goto retry;
2308 			}
2309 			rcu_read_unlock();
2310 
2311 			xfs_iflock(ip);
2312 			xfs_iflags_set(ip, XFS_ISTALE);
2313 
2314 			/*
2315 			 * we don't need to attach clean inodes or those only
2316 			 * with unlogged changes (which we throw away, anyway).
2317 			 */
2318 			iip = ip->i_itemp;
2319 			if (!iip || xfs_inode_clean(ip)) {
2320 				ASSERT(ip != free_ip);
2321 				xfs_ifunlock(ip);
2322 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2323 				continue;
2324 			}
2325 
2326 			iip->ili_last_fields = iip->ili_fields;
2327 			iip->ili_fields = 0;
2328 			iip->ili_logged = 1;
2329 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2330 						&iip->ili_item.li_lsn);
2331 
2332 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2333 						  &iip->ili_item);
2334 
2335 			if (ip != free_ip)
2336 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2337 		}
2338 
2339 		xfs_trans_stale_inode_buf(tp, bp);
2340 		xfs_trans_binval(tp, bp);
2341 	}
2342 
2343 	xfs_perag_put(pag);
2344 	return 0;
2345 }
2346 
2347 /*
2348  * This is called to return an inode to the inode free list.
2349  * The inode should already be truncated to 0 length and have
2350  * no pages associated with it.  This routine also assumes that
2351  * the inode is already a part of the transaction.
2352  *
2353  * The on-disk copy of the inode will have been added to the list
2354  * of unlinked inodes in the AGI. We need to remove the inode from
2355  * that list atomically with respect to freeing it here.
2356  */
2357 int
2358 xfs_ifree(
2359 	xfs_trans_t	*tp,
2360 	xfs_inode_t	*ip,
2361 	xfs_bmap_free_t	*flist)
2362 {
2363 	int			error;
2364 	int			delete;
2365 	xfs_ino_t		first_ino;
2366 
2367 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2368 	ASSERT(ip->i_d.di_nlink == 0);
2369 	ASSERT(ip->i_d.di_nextents == 0);
2370 	ASSERT(ip->i_d.di_anextents == 0);
2371 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2372 	ASSERT(ip->i_d.di_nblocks == 0);
2373 
2374 	/*
2375 	 * Pull the on-disk inode from the AGI unlinked list.
2376 	 */
2377 	error = xfs_iunlink_remove(tp, ip);
2378 	if (error)
2379 		return error;
2380 
2381 	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2382 	if (error)
2383 		return error;
2384 
2385 	ip->i_d.di_mode = 0;		/* mark incore inode as free */
2386 	ip->i_d.di_flags = 0;
2387 	ip->i_d.di_dmevmask = 0;
2388 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2389 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2390 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2391 	/*
2392 	 * Bump the generation count so no one will be confused
2393 	 * by reincarnations of this inode.
2394 	 */
2395 	ip->i_d.di_gen++;
2396 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2397 
2398 	if (delete)
2399 		error = xfs_ifree_cluster(ip, tp, first_ino);
2400 
2401 	return error;
2402 }
2403 
2404 /*
2405  * This is called to unpin an inode.  The caller must have the inode locked
2406  * in at least shared mode so that the buffer cannot be subsequently pinned
2407  * once someone is waiting for it to be unpinned.
2408  */
2409 static void
2410 xfs_iunpin(
2411 	struct xfs_inode	*ip)
2412 {
2413 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2414 
2415 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2416 
2417 	/* Give the log a push to start the unpinning I/O */
2418 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2419 
2420 }
2421 
2422 static void
2423 __xfs_iunpin_wait(
2424 	struct xfs_inode	*ip)
2425 {
2426 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2427 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2428 
2429 	xfs_iunpin(ip);
2430 
2431 	do {
2432 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2433 		if (xfs_ipincount(ip))
2434 			io_schedule();
2435 	} while (xfs_ipincount(ip));
2436 	finish_wait(wq, &wait.wait);
2437 }
2438 
2439 void
2440 xfs_iunpin_wait(
2441 	struct xfs_inode	*ip)
2442 {
2443 	if (xfs_ipincount(ip))
2444 		__xfs_iunpin_wait(ip);
2445 }
2446 
2447 /*
2448  * Removing an inode from the namespace involves removing the directory entry
2449  * and dropping the link count on the inode. Removing the directory entry can
2450  * result in locking an AGF (directory blocks were freed) and removing a link
2451  * count can result in placing the inode on an unlinked list which results in
2452  * locking an AGI.
2453  *
2454  * The big problem here is that we have an ordering constraint on AGF and AGI
2455  * locking - inode allocation locks the AGI, then can allocate a new extent for
2456  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2457  * removes the inode from the unlinked list, requiring that we lock the AGI
2458  * first, and then freeing the inode can result in an inode chunk being freed
2459  * and hence freeing disk space requiring that we lock an AGF.
2460  *
2461  * Hence the ordering that is imposed by other parts of the code is AGI before
2462  * AGF. This means we cannot remove the directory entry before we drop the inode
2463  * reference count and put it on the unlinked list as this results in a lock
2464  * order of AGF then AGI, and this can deadlock against inode allocation and
2465  * freeing. Therefore we must drop the link counts before we remove the
2466  * directory entry.
2467  *
2468  * This is still safe from a transactional point of view - it is not until we
2469  * get to xfs_bmap_finish() that we have the possibility of multiple
2470  * transactions in this operation. Hence as long as we remove the directory
2471  * entry and drop the link count in the first transaction of the remove
2472  * operation, there are no transactional constraints on the ordering here.
2473  */
2474 int
2475 xfs_remove(
2476 	xfs_inode_t             *dp,
2477 	struct xfs_name		*name,
2478 	xfs_inode_t		*ip)
2479 {
2480 	xfs_mount_t		*mp = dp->i_mount;
2481 	xfs_trans_t             *tp = NULL;
2482 	int			is_dir = S_ISDIR(ip->i_d.di_mode);
2483 	int                     error = 0;
2484 	xfs_bmap_free_t         free_list;
2485 	xfs_fsblock_t           first_block;
2486 	int			cancel_flags;
2487 	int			committed;
2488 	uint			resblks;
2489 
2490 	trace_xfs_remove(dp, name);
2491 
2492 	if (XFS_FORCED_SHUTDOWN(mp))
2493 		return -EIO;
2494 
2495 	error = xfs_qm_dqattach(dp, 0);
2496 	if (error)
2497 		goto std_return;
2498 
2499 	error = xfs_qm_dqattach(ip, 0);
2500 	if (error)
2501 		goto std_return;
2502 
2503 	if (is_dir)
2504 		tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2505 	else
2506 		tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2507 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2508 
2509 	/*
2510 	 * We try to get the real space reservation first,
2511 	 * allowing for directory btree deletion(s) implying
2512 	 * possible bmap insert(s).  If we can't get the space
2513 	 * reservation then we use 0 instead, and avoid the bmap
2514 	 * btree insert(s) in the directory code by, if the bmap
2515 	 * insert tries to happen, instead trimming the LAST
2516 	 * block from the directory.
2517 	 */
2518 	resblks = XFS_REMOVE_SPACE_RES(mp);
2519 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2520 	if (error == -ENOSPC) {
2521 		resblks = 0;
2522 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2523 	}
2524 	if (error) {
2525 		ASSERT(error != -ENOSPC);
2526 		cancel_flags = 0;
2527 		goto out_trans_cancel;
2528 	}
2529 
2530 	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2531 
2532 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2533 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2534 
2535 	/*
2536 	 * If we're removing a directory perform some additional validation.
2537 	 */
2538 	cancel_flags |= XFS_TRANS_ABORT;
2539 	if (is_dir) {
2540 		ASSERT(ip->i_d.di_nlink >= 2);
2541 		if (ip->i_d.di_nlink != 2) {
2542 			error = -ENOTEMPTY;
2543 			goto out_trans_cancel;
2544 		}
2545 		if (!xfs_dir_isempty(ip)) {
2546 			error = -ENOTEMPTY;
2547 			goto out_trans_cancel;
2548 		}
2549 
2550 		/* Drop the link from ip's "..".  */
2551 		error = xfs_droplink(tp, dp);
2552 		if (error)
2553 			goto out_trans_cancel;
2554 
2555 		/* Drop the "." link from ip to self.  */
2556 		error = xfs_droplink(tp, ip);
2557 		if (error)
2558 			goto out_trans_cancel;
2559 	} else {
2560 		/*
2561 		 * When removing a non-directory we need to log the parent
2562 		 * inode here.  For a directory this is done implicitly
2563 		 * by the xfs_droplink call for the ".." entry.
2564 		 */
2565 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2566 	}
2567 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2568 
2569 	/* Drop the link from dp to ip. */
2570 	error = xfs_droplink(tp, ip);
2571 	if (error)
2572 		goto out_trans_cancel;
2573 
2574 	xfs_bmap_init(&free_list, &first_block);
2575 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2576 					&first_block, &free_list, resblks);
2577 	if (error) {
2578 		ASSERT(error != -ENOENT);
2579 		goto out_bmap_cancel;
2580 	}
2581 
2582 	/*
2583 	 * If this is a synchronous mount, make sure that the
2584 	 * remove transaction goes to disk before returning to
2585 	 * the user.
2586 	 */
2587 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2588 		xfs_trans_set_sync(tp);
2589 
2590 	error = xfs_bmap_finish(&tp, &free_list, &committed);
2591 	if (error)
2592 		goto out_bmap_cancel;
2593 
2594 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2595 	if (error)
2596 		goto std_return;
2597 
2598 	if (is_dir && xfs_inode_is_filestream(ip))
2599 		xfs_filestream_deassociate(ip);
2600 
2601 	return 0;
2602 
2603  out_bmap_cancel:
2604 	xfs_bmap_cancel(&free_list);
2605  out_trans_cancel:
2606 	xfs_trans_cancel(tp, cancel_flags);
2607  std_return:
2608 	return error;
2609 }
2610 
2611 /*
2612  * Enter all inodes for a rename transaction into a sorted array.
2613  */
2614 STATIC void
2615 xfs_sort_for_rename(
2616 	xfs_inode_t	*dp1,	/* in: old (source) directory inode */
2617 	xfs_inode_t	*dp2,	/* in: new (target) directory inode */
2618 	xfs_inode_t	*ip1,	/* in: inode of old entry */
2619 	xfs_inode_t	*ip2,	/* in: inode of new entry, if it
2620 				   already exists, NULL otherwise. */
2621 	xfs_inode_t	**i_tab,/* out: array of inode returned, sorted */
2622 	int		*num_inodes)  /* out: number of inodes in array */
2623 {
2624 	xfs_inode_t		*temp;
2625 	int			i, j;
2626 
2627 	/*
2628 	 * i_tab contains a list of pointers to inodes.  We initialize
2629 	 * the table here & we'll sort it.  We will then use it to
2630 	 * order the acquisition of the inode locks.
2631 	 *
2632 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2633 	 */
2634 	i_tab[0] = dp1;
2635 	i_tab[1] = dp2;
2636 	i_tab[2] = ip1;
2637 	if (ip2) {
2638 		*num_inodes = 4;
2639 		i_tab[3] = ip2;
2640 	} else {
2641 		*num_inodes = 3;
2642 		i_tab[3] = NULL;
2643 	}
2644 
2645 	/*
2646 	 * Sort the elements via bubble sort.  (Remember, there are at
2647 	 * most 4 elements to sort, so this is adequate.)
2648 	 */
2649 	for (i = 0; i < *num_inodes; i++) {
2650 		for (j = 1; j < *num_inodes; j++) {
2651 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2652 				temp = i_tab[j];
2653 				i_tab[j] = i_tab[j-1];
2654 				i_tab[j-1] = temp;
2655 			}
2656 		}
2657 	}
2658 }
2659 
2660 /*
2661  * xfs_cross_rename()
2662  *
2663  * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2664  */
2665 STATIC int
2666 xfs_cross_rename(
2667 	struct xfs_trans	*tp,
2668 	struct xfs_inode	*dp1,
2669 	struct xfs_name		*name1,
2670 	struct xfs_inode	*ip1,
2671 	struct xfs_inode	*dp2,
2672 	struct xfs_name		*name2,
2673 	struct xfs_inode	*ip2,
2674 	struct xfs_bmap_free	*free_list,
2675 	xfs_fsblock_t		*first_block,
2676 	int			spaceres)
2677 {
2678 	int		error = 0;
2679 	int		ip1_flags = 0;
2680 	int		ip2_flags = 0;
2681 	int		dp2_flags = 0;
2682 
2683 	/* Swap inode number for dirent in first parent */
2684 	error = xfs_dir_replace(tp, dp1, name1,
2685 				ip2->i_ino,
2686 				first_block, free_list, spaceres);
2687 	if (error)
2688 		goto out;
2689 
2690 	/* Swap inode number for dirent in second parent */
2691 	error = xfs_dir_replace(tp, dp2, name2,
2692 				ip1->i_ino,
2693 				first_block, free_list, spaceres);
2694 	if (error)
2695 		goto out;
2696 
2697 	/*
2698 	 * If we're renaming one or more directories across different parents,
2699 	 * update the respective ".." entries (and link counts) to match the new
2700 	 * parents.
2701 	 */
2702 	if (dp1 != dp2) {
2703 		dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2704 
2705 		if (S_ISDIR(ip2->i_d.di_mode)) {
2706 			error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2707 						dp1->i_ino, first_block,
2708 						free_list, spaceres);
2709 			if (error)
2710 				goto out;
2711 
2712 			/* transfer ip2 ".." reference to dp1 */
2713 			if (!S_ISDIR(ip1->i_d.di_mode)) {
2714 				error = xfs_droplink(tp, dp2);
2715 				if (error)
2716 					goto out;
2717 				error = xfs_bumplink(tp, dp1);
2718 				if (error)
2719 					goto out;
2720 			}
2721 
2722 			/*
2723 			 * Although ip1 isn't changed here, userspace needs
2724 			 * to be warned about the change, so that applications
2725 			 * relying on it (like backup ones), will properly
2726 			 * notify the change
2727 			 */
2728 			ip1_flags |= XFS_ICHGTIME_CHG;
2729 			ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2730 		}
2731 
2732 		if (S_ISDIR(ip1->i_d.di_mode)) {
2733 			error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2734 						dp2->i_ino, first_block,
2735 						free_list, spaceres);
2736 			if (error)
2737 				goto out;
2738 
2739 			/* transfer ip1 ".." reference to dp2 */
2740 			if (!S_ISDIR(ip2->i_d.di_mode)) {
2741 				error = xfs_droplink(tp, dp1);
2742 				if (error)
2743 					goto out;
2744 				error = xfs_bumplink(tp, dp2);
2745 				if (error)
2746 					goto out;
2747 			}
2748 
2749 			/*
2750 			 * Although ip2 isn't changed here, userspace needs
2751 			 * to be warned about the change, so that applications
2752 			 * relying on it (like backup ones), will properly
2753 			 * notify the change
2754 			 */
2755 			ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2756 			ip2_flags |= XFS_ICHGTIME_CHG;
2757 		}
2758 	}
2759 
2760 	if (ip1_flags) {
2761 		xfs_trans_ichgtime(tp, ip1, ip1_flags);
2762 		xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2763 	}
2764 	if (ip2_flags) {
2765 		xfs_trans_ichgtime(tp, ip2, ip2_flags);
2766 		xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2767 	}
2768 	if (dp2_flags) {
2769 		xfs_trans_ichgtime(tp, dp2, dp2_flags);
2770 		xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2771 	}
2772 	xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2773 	xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2774 out:
2775 	return error;
2776 }
2777 
2778 /*
2779  * xfs_rename
2780  */
2781 int
2782 xfs_rename(
2783 	xfs_inode_t	*src_dp,
2784 	struct xfs_name	*src_name,
2785 	xfs_inode_t	*src_ip,
2786 	xfs_inode_t	*target_dp,
2787 	struct xfs_name	*target_name,
2788 	xfs_inode_t	*target_ip,
2789 	unsigned int	flags)
2790 {
2791 	xfs_trans_t	*tp = NULL;
2792 	xfs_mount_t	*mp = src_dp->i_mount;
2793 	int		new_parent;		/* moving to a new dir */
2794 	int		src_is_directory;	/* src_name is a directory */
2795 	int		error;
2796 	xfs_bmap_free_t free_list;
2797 	xfs_fsblock_t   first_block;
2798 	int		cancel_flags;
2799 	int		committed;
2800 	xfs_inode_t	*inodes[4];
2801 	int		spaceres;
2802 	int		num_inodes;
2803 
2804 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2805 
2806 	new_parent = (src_dp != target_dp);
2807 	src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2808 
2809 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2810 				inodes, &num_inodes);
2811 
2812 	xfs_bmap_init(&free_list, &first_block);
2813 	tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2814 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2815 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2816 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2817 	if (error == -ENOSPC) {
2818 		spaceres = 0;
2819 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2820 	}
2821 	if (error) {
2822 		xfs_trans_cancel(tp, 0);
2823 		goto std_return;
2824 	}
2825 
2826 	/*
2827 	 * Attach the dquots to the inodes
2828 	 */
2829 	error = xfs_qm_vop_rename_dqattach(inodes);
2830 	if (error) {
2831 		xfs_trans_cancel(tp, cancel_flags);
2832 		goto std_return;
2833 	}
2834 
2835 	/*
2836 	 * Lock all the participating inodes. Depending upon whether
2837 	 * the target_name exists in the target directory, and
2838 	 * whether the target directory is the same as the source
2839 	 * directory, we can lock from 2 to 4 inodes.
2840 	 */
2841 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2842 
2843 	/*
2844 	 * Join all the inodes to the transaction. From this point on,
2845 	 * we can rely on either trans_commit or trans_cancel to unlock
2846 	 * them.
2847 	 */
2848 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2849 	if (new_parent)
2850 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2851 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2852 	if (target_ip)
2853 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2854 
2855 	/*
2856 	 * If we are using project inheritance, we only allow renames
2857 	 * into our tree when the project IDs are the same; else the
2858 	 * tree quota mechanism would be circumvented.
2859 	 */
2860 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2861 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2862 		error = -EXDEV;
2863 		goto error_return;
2864 	}
2865 
2866 	/*
2867 	 * Handle RENAME_EXCHANGE flags
2868 	 */
2869 	if (flags & RENAME_EXCHANGE) {
2870 		if (target_ip == NULL) {
2871 			error = -EINVAL;
2872 			goto error_return;
2873 		}
2874 		error = xfs_cross_rename(tp, src_dp, src_name, src_ip,
2875 					 target_dp, target_name, target_ip,
2876 					 &free_list, &first_block, spaceres);
2877 		if (error)
2878 			goto abort_return;
2879 		goto finish_rename;
2880 	}
2881 
2882 	/*
2883 	 * Set up the target.
2884 	 */
2885 	if (target_ip == NULL) {
2886 		/*
2887 		 * If there's no space reservation, check the entry will
2888 		 * fit before actually inserting it.
2889 		 */
2890 		if (!spaceres) {
2891 			error = xfs_dir_canenter(tp, target_dp, target_name);
2892 			if (error)
2893 				goto error_return;
2894 		}
2895 		/*
2896 		 * If target does not exist and the rename crosses
2897 		 * directories, adjust the target directory link count
2898 		 * to account for the ".." reference from the new entry.
2899 		 */
2900 		error = xfs_dir_createname(tp, target_dp, target_name,
2901 						src_ip->i_ino, &first_block,
2902 						&free_list, spaceres);
2903 		if (error == -ENOSPC)
2904 			goto error_return;
2905 		if (error)
2906 			goto abort_return;
2907 
2908 		xfs_trans_ichgtime(tp, target_dp,
2909 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2910 
2911 		if (new_parent && src_is_directory) {
2912 			error = xfs_bumplink(tp, target_dp);
2913 			if (error)
2914 				goto abort_return;
2915 		}
2916 	} else { /* target_ip != NULL */
2917 		/*
2918 		 * If target exists and it's a directory, check that both
2919 		 * target and source are directories and that target can be
2920 		 * destroyed, or that neither is a directory.
2921 		 */
2922 		if (S_ISDIR(target_ip->i_d.di_mode)) {
2923 			/*
2924 			 * Make sure target dir is empty.
2925 			 */
2926 			if (!(xfs_dir_isempty(target_ip)) ||
2927 			    (target_ip->i_d.di_nlink > 2)) {
2928 				error = -EEXIST;
2929 				goto error_return;
2930 			}
2931 		}
2932 
2933 		/*
2934 		 * Link the source inode under the target name.
2935 		 * If the source inode is a directory and we are moving
2936 		 * it across directories, its ".." entry will be
2937 		 * inconsistent until we replace that down below.
2938 		 *
2939 		 * In case there is already an entry with the same
2940 		 * name at the destination directory, remove it first.
2941 		 */
2942 		error = xfs_dir_replace(tp, target_dp, target_name,
2943 					src_ip->i_ino,
2944 					&first_block, &free_list, spaceres);
2945 		if (error)
2946 			goto abort_return;
2947 
2948 		xfs_trans_ichgtime(tp, target_dp,
2949 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2950 
2951 		/*
2952 		 * Decrement the link count on the target since the target
2953 		 * dir no longer points to it.
2954 		 */
2955 		error = xfs_droplink(tp, target_ip);
2956 		if (error)
2957 			goto abort_return;
2958 
2959 		if (src_is_directory) {
2960 			/*
2961 			 * Drop the link from the old "." entry.
2962 			 */
2963 			error = xfs_droplink(tp, target_ip);
2964 			if (error)
2965 				goto abort_return;
2966 		}
2967 	} /* target_ip != NULL */
2968 
2969 	/*
2970 	 * Remove the source.
2971 	 */
2972 	if (new_parent && src_is_directory) {
2973 		/*
2974 		 * Rewrite the ".." entry to point to the new
2975 		 * directory.
2976 		 */
2977 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2978 					target_dp->i_ino,
2979 					&first_block, &free_list, spaceres);
2980 		ASSERT(error != -EEXIST);
2981 		if (error)
2982 			goto abort_return;
2983 	}
2984 
2985 	/*
2986 	 * We always want to hit the ctime on the source inode.
2987 	 *
2988 	 * This isn't strictly required by the standards since the source
2989 	 * inode isn't really being changed, but old unix file systems did
2990 	 * it and some incremental backup programs won't work without it.
2991 	 */
2992 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2993 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2994 
2995 	/*
2996 	 * Adjust the link count on src_dp.  This is necessary when
2997 	 * renaming a directory, either within one parent when
2998 	 * the target existed, or across two parent directories.
2999 	 */
3000 	if (src_is_directory && (new_parent || target_ip != NULL)) {
3001 
3002 		/*
3003 		 * Decrement link count on src_directory since the
3004 		 * entry that's moved no longer points to it.
3005 		 */
3006 		error = xfs_droplink(tp, src_dp);
3007 		if (error)
3008 			goto abort_return;
3009 	}
3010 
3011 	error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3012 					&first_block, &free_list, spaceres);
3013 	if (error)
3014 		goto abort_return;
3015 
3016 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3017 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3018 	if (new_parent)
3019 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3020 
3021 finish_rename:
3022 	/*
3023 	 * If this is a synchronous mount, make sure that the
3024 	 * rename transaction goes to disk before returning to
3025 	 * the user.
3026 	 */
3027 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3028 		xfs_trans_set_sync(tp);
3029 	}
3030 
3031 	error = xfs_bmap_finish(&tp, &free_list, &committed);
3032 	if (error) {
3033 		xfs_bmap_cancel(&free_list);
3034 		xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3035 				 XFS_TRANS_ABORT));
3036 		goto std_return;
3037 	}
3038 
3039 	/*
3040 	 * trans_commit will unlock src_ip, target_ip & decrement
3041 	 * the vnode references.
3042 	 */
3043 	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3044 
3045  abort_return:
3046 	cancel_flags |= XFS_TRANS_ABORT;
3047  error_return:
3048 	xfs_bmap_cancel(&free_list);
3049 	xfs_trans_cancel(tp, cancel_flags);
3050  std_return:
3051 	return error;
3052 }
3053 
3054 STATIC int
3055 xfs_iflush_cluster(
3056 	xfs_inode_t	*ip,
3057 	xfs_buf_t	*bp)
3058 {
3059 	xfs_mount_t		*mp = ip->i_mount;
3060 	struct xfs_perag	*pag;
3061 	unsigned long		first_index, mask;
3062 	unsigned long		inodes_per_cluster;
3063 	int			ilist_size;
3064 	xfs_inode_t		**ilist;
3065 	xfs_inode_t		*iq;
3066 	int			nr_found;
3067 	int			clcount = 0;
3068 	int			bufwasdelwri;
3069 	int			i;
3070 
3071 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3072 
3073 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3074 	ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3075 	ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
3076 	if (!ilist)
3077 		goto out_put;
3078 
3079 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3080 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3081 	rcu_read_lock();
3082 	/* really need a gang lookup range call here */
3083 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3084 					first_index, inodes_per_cluster);
3085 	if (nr_found == 0)
3086 		goto out_free;
3087 
3088 	for (i = 0; i < nr_found; i++) {
3089 		iq = ilist[i];
3090 		if (iq == ip)
3091 			continue;
3092 
3093 		/*
3094 		 * because this is an RCU protected lookup, we could find a
3095 		 * recently freed or even reallocated inode during the lookup.
3096 		 * We need to check under the i_flags_lock for a valid inode
3097 		 * here. Skip it if it is not valid or the wrong inode.
3098 		 */
3099 		spin_lock(&ip->i_flags_lock);
3100 		if (!ip->i_ino ||
3101 		    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3102 			spin_unlock(&ip->i_flags_lock);
3103 			continue;
3104 		}
3105 		spin_unlock(&ip->i_flags_lock);
3106 
3107 		/*
3108 		 * Do an un-protected check to see if the inode is dirty and
3109 		 * is a candidate for flushing.  These checks will be repeated
3110 		 * later after the appropriate locks are acquired.
3111 		 */
3112 		if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3113 			continue;
3114 
3115 		/*
3116 		 * Try to get locks.  If any are unavailable or it is pinned,
3117 		 * then this inode cannot be flushed and is skipped.
3118 		 */
3119 
3120 		if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3121 			continue;
3122 		if (!xfs_iflock_nowait(iq)) {
3123 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
3124 			continue;
3125 		}
3126 		if (xfs_ipincount(iq)) {
3127 			xfs_ifunlock(iq);
3128 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
3129 			continue;
3130 		}
3131 
3132 		/*
3133 		 * arriving here means that this inode can be flushed.  First
3134 		 * re-check that it's dirty before flushing.
3135 		 */
3136 		if (!xfs_inode_clean(iq)) {
3137 			int	error;
3138 			error = xfs_iflush_int(iq, bp);
3139 			if (error) {
3140 				xfs_iunlock(iq, XFS_ILOCK_SHARED);
3141 				goto cluster_corrupt_out;
3142 			}
3143 			clcount++;
3144 		} else {
3145 			xfs_ifunlock(iq);
3146 		}
3147 		xfs_iunlock(iq, XFS_ILOCK_SHARED);
3148 	}
3149 
3150 	if (clcount) {
3151 		XFS_STATS_INC(xs_icluster_flushcnt);
3152 		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3153 	}
3154 
3155 out_free:
3156 	rcu_read_unlock();
3157 	kmem_free(ilist);
3158 out_put:
3159 	xfs_perag_put(pag);
3160 	return 0;
3161 
3162 
3163 cluster_corrupt_out:
3164 	/*
3165 	 * Corruption detected in the clustering loop.  Invalidate the
3166 	 * inode buffer and shut down the filesystem.
3167 	 */
3168 	rcu_read_unlock();
3169 	/*
3170 	 * Clean up the buffer.  If it was delwri, just release it --
3171 	 * brelse can handle it with no problems.  If not, shut down the
3172 	 * filesystem before releasing the buffer.
3173 	 */
3174 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3175 	if (bufwasdelwri)
3176 		xfs_buf_relse(bp);
3177 
3178 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3179 
3180 	if (!bufwasdelwri) {
3181 		/*
3182 		 * Just like incore_relse: if we have b_iodone functions,
3183 		 * mark the buffer as an error and call them.  Otherwise
3184 		 * mark it as stale and brelse.
3185 		 */
3186 		if (bp->b_iodone) {
3187 			XFS_BUF_UNDONE(bp);
3188 			xfs_buf_stale(bp);
3189 			xfs_buf_ioerror(bp, -EIO);
3190 			xfs_buf_ioend(bp);
3191 		} else {
3192 			xfs_buf_stale(bp);
3193 			xfs_buf_relse(bp);
3194 		}
3195 	}
3196 
3197 	/*
3198 	 * Unlocks the flush lock
3199 	 */
3200 	xfs_iflush_abort(iq, false);
3201 	kmem_free(ilist);
3202 	xfs_perag_put(pag);
3203 	return -EFSCORRUPTED;
3204 }
3205 
3206 /*
3207  * Flush dirty inode metadata into the backing buffer.
3208  *
3209  * The caller must have the inode lock and the inode flush lock held.  The
3210  * inode lock will still be held upon return to the caller, and the inode
3211  * flush lock will be released after the inode has reached the disk.
3212  *
3213  * The caller must write out the buffer returned in *bpp and release it.
3214  */
3215 int
3216 xfs_iflush(
3217 	struct xfs_inode	*ip,
3218 	struct xfs_buf		**bpp)
3219 {
3220 	struct xfs_mount	*mp = ip->i_mount;
3221 	struct xfs_buf		*bp;
3222 	struct xfs_dinode	*dip;
3223 	int			error;
3224 
3225 	XFS_STATS_INC(xs_iflush_count);
3226 
3227 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3228 	ASSERT(xfs_isiflocked(ip));
3229 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3230 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3231 
3232 	*bpp = NULL;
3233 
3234 	xfs_iunpin_wait(ip);
3235 
3236 	/*
3237 	 * For stale inodes we cannot rely on the backing buffer remaining
3238 	 * stale in cache for the remaining life of the stale inode and so
3239 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3240 	 * inodes below. We have to check this after ensuring the inode is
3241 	 * unpinned so that it is safe to reclaim the stale inode after the
3242 	 * flush call.
3243 	 */
3244 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3245 		xfs_ifunlock(ip);
3246 		return 0;
3247 	}
3248 
3249 	/*
3250 	 * This may have been unpinned because the filesystem is shutting
3251 	 * down forcibly. If that's the case we must not write this inode
3252 	 * to disk, because the log record didn't make it to disk.
3253 	 *
3254 	 * We also have to remove the log item from the AIL in this case,
3255 	 * as we wait for an empty AIL as part of the unmount process.
3256 	 */
3257 	if (XFS_FORCED_SHUTDOWN(mp)) {
3258 		error = -EIO;
3259 		goto abort_out;
3260 	}
3261 
3262 	/*
3263 	 * Get the buffer containing the on-disk inode.
3264 	 */
3265 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3266 			       0);
3267 	if (error || !bp) {
3268 		xfs_ifunlock(ip);
3269 		return error;
3270 	}
3271 
3272 	/*
3273 	 * First flush out the inode that xfs_iflush was called with.
3274 	 */
3275 	error = xfs_iflush_int(ip, bp);
3276 	if (error)
3277 		goto corrupt_out;
3278 
3279 	/*
3280 	 * If the buffer is pinned then push on the log now so we won't
3281 	 * get stuck waiting in the write for too long.
3282 	 */
3283 	if (xfs_buf_ispinned(bp))
3284 		xfs_log_force(mp, 0);
3285 
3286 	/*
3287 	 * inode clustering:
3288 	 * see if other inodes can be gathered into this write
3289 	 */
3290 	error = xfs_iflush_cluster(ip, bp);
3291 	if (error)
3292 		goto cluster_corrupt_out;
3293 
3294 	*bpp = bp;
3295 	return 0;
3296 
3297 corrupt_out:
3298 	xfs_buf_relse(bp);
3299 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3300 cluster_corrupt_out:
3301 	error = -EFSCORRUPTED;
3302 abort_out:
3303 	/*
3304 	 * Unlocks the flush lock
3305 	 */
3306 	xfs_iflush_abort(ip, false);
3307 	return error;
3308 }
3309 
3310 STATIC int
3311 xfs_iflush_int(
3312 	struct xfs_inode	*ip,
3313 	struct xfs_buf		*bp)
3314 {
3315 	struct xfs_inode_log_item *iip = ip->i_itemp;
3316 	struct xfs_dinode	*dip;
3317 	struct xfs_mount	*mp = ip->i_mount;
3318 
3319 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3320 	ASSERT(xfs_isiflocked(ip));
3321 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3322 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3323 	ASSERT(iip != NULL && iip->ili_fields != 0);
3324 	ASSERT(ip->i_d.di_version > 1);
3325 
3326 	/* set *dip = inode's place in the buffer */
3327 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3328 
3329 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3330 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3331 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3332 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3333 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3334 		goto corrupt_out;
3335 	}
3336 	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3337 				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3338 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3339 			"%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3340 			__func__, ip->i_ino, ip, ip->i_d.di_magic);
3341 		goto corrupt_out;
3342 	}
3343 	if (S_ISREG(ip->i_d.di_mode)) {
3344 		if (XFS_TEST_ERROR(
3345 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3346 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3347 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3348 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3349 				"%s: Bad regular inode %Lu, ptr 0x%p",
3350 				__func__, ip->i_ino, ip);
3351 			goto corrupt_out;
3352 		}
3353 	} else if (S_ISDIR(ip->i_d.di_mode)) {
3354 		if (XFS_TEST_ERROR(
3355 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3356 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3357 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3358 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3359 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3360 				"%s: Bad directory inode %Lu, ptr 0x%p",
3361 				__func__, ip->i_ino, ip);
3362 			goto corrupt_out;
3363 		}
3364 	}
3365 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3366 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3367 				XFS_RANDOM_IFLUSH_5)) {
3368 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3369 			"%s: detected corrupt incore inode %Lu, "
3370 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3371 			__func__, ip->i_ino,
3372 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3373 			ip->i_d.di_nblocks, ip);
3374 		goto corrupt_out;
3375 	}
3376 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3377 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3378 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3379 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3380 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3381 		goto corrupt_out;
3382 	}
3383 
3384 	/*
3385 	 * Inode item log recovery for v2 inodes are dependent on the
3386 	 * di_flushiter count for correct sequencing. We bump the flush
3387 	 * iteration count so we can detect flushes which postdate a log record
3388 	 * during recovery. This is redundant as we now log every change and
3389 	 * hence this can't happen but we need to still do it to ensure
3390 	 * backwards compatibility with old kernels that predate logging all
3391 	 * inode changes.
3392 	 */
3393 	if (ip->i_d.di_version < 3)
3394 		ip->i_d.di_flushiter++;
3395 
3396 	/*
3397 	 * Copy the dirty parts of the inode into the on-disk
3398 	 * inode.  We always copy out the core of the inode,
3399 	 * because if the inode is dirty at all the core must
3400 	 * be.
3401 	 */
3402 	xfs_dinode_to_disk(dip, &ip->i_d);
3403 
3404 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3405 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3406 		ip->i_d.di_flushiter = 0;
3407 
3408 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3409 	if (XFS_IFORK_Q(ip))
3410 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3411 	xfs_inobp_check(mp, bp);
3412 
3413 	/*
3414 	 * We've recorded everything logged in the inode, so we'd like to clear
3415 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3416 	 * However, we can't stop logging all this information until the data
3417 	 * we've copied into the disk buffer is written to disk.  If we did we
3418 	 * might overwrite the copy of the inode in the log with all the data
3419 	 * after re-logging only part of it, and in the face of a crash we
3420 	 * wouldn't have all the data we need to recover.
3421 	 *
3422 	 * What we do is move the bits to the ili_last_fields field.  When
3423 	 * logging the inode, these bits are moved back to the ili_fields field.
3424 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3425 	 * know that the information those bits represent is permanently on
3426 	 * disk.  As long as the flush completes before the inode is logged
3427 	 * again, then both ili_fields and ili_last_fields will be cleared.
3428 	 *
3429 	 * We can play with the ili_fields bits here, because the inode lock
3430 	 * must be held exclusively in order to set bits there and the flush
3431 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3432 	 * done routine can tell whether or not to look in the AIL.  Also, store
3433 	 * the current LSN of the inode so that we can tell whether the item has
3434 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3435 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3436 	 * atomically.
3437 	 */
3438 	iip->ili_last_fields = iip->ili_fields;
3439 	iip->ili_fields = 0;
3440 	iip->ili_logged = 1;
3441 
3442 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3443 				&iip->ili_item.li_lsn);
3444 
3445 	/*
3446 	 * Attach the function xfs_iflush_done to the inode's
3447 	 * buffer.  This will remove the inode from the AIL
3448 	 * and unlock the inode's flush lock when the inode is
3449 	 * completely written to disk.
3450 	 */
3451 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3452 
3453 	/* update the lsn in the on disk inode if required */
3454 	if (ip->i_d.di_version == 3)
3455 		dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3456 
3457 	/* generate the checksum. */
3458 	xfs_dinode_calc_crc(mp, dip);
3459 
3460 	ASSERT(bp->b_fspriv != NULL);
3461 	ASSERT(bp->b_iodone != NULL);
3462 	return 0;
3463 
3464 corrupt_out:
3465 	return -EFSCORRUPTED;
3466 }
3467