xref: /openbmc/linux/fs/xfs/xfs_inode.c (revision d2168146)
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_inum.h"
27 #include "xfs_sb.h"
28 #include "xfs_ag.h"
29 #include "xfs_mount.h"
30 #include "xfs_inode.h"
31 #include "xfs_da_format.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_dir2.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_attr.h"
36 #include "xfs_trans_space.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_bmap.h"
42 #include "xfs_bmap_util.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_filestream.h"
46 #include "xfs_cksum.h"
47 #include "xfs_trace.h"
48 #include "xfs_icache.h"
49 #include "xfs_symlink.h"
50 #include "xfs_trans_priv.h"
51 #include "xfs_log.h"
52 #include "xfs_bmap_btree.h"
53 
54 kmem_zone_t *xfs_inode_zone;
55 
56 /*
57  * Used in xfs_itruncate_extents().  This is the maximum number of extents
58  * freed from a file in a single transaction.
59  */
60 #define	XFS_ITRUNC_MAX_EXTENTS	2
61 
62 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
63 
64 STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
65 
66 /*
67  * helper function to extract extent size hint from inode
68  */
69 xfs_extlen_t
70 xfs_get_extsz_hint(
71 	struct xfs_inode	*ip)
72 {
73 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
74 		return ip->i_d.di_extsize;
75 	if (XFS_IS_REALTIME_INODE(ip))
76 		return ip->i_mount->m_sb.sb_rextsize;
77 	return 0;
78 }
79 
80 /*
81  * These two are wrapper routines around the xfs_ilock() routine used to
82  * centralize some grungy code.  They are used in places that wish to lock the
83  * inode solely for reading the extents.  The reason these places can't just
84  * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
85  * bringing in of the extents from disk for a file in b-tree format.  If the
86  * inode is in b-tree format, then we need to lock the inode exclusively until
87  * the extents are read in.  Locking it exclusively all the time would limit
88  * our parallelism unnecessarily, though.  What we do instead is check to see
89  * if the extents have been read in yet, and only lock the inode exclusively
90  * if they have not.
91  *
92  * The functions return a value which should be given to the corresponding
93  * xfs_iunlock() call.
94  */
95 uint
96 xfs_ilock_data_map_shared(
97 	struct xfs_inode	*ip)
98 {
99 	uint			lock_mode = XFS_ILOCK_SHARED;
100 
101 	if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
102 	    (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
103 		lock_mode = XFS_ILOCK_EXCL;
104 	xfs_ilock(ip, lock_mode);
105 	return lock_mode;
106 }
107 
108 uint
109 xfs_ilock_attr_map_shared(
110 	struct xfs_inode	*ip)
111 {
112 	uint			lock_mode = XFS_ILOCK_SHARED;
113 
114 	if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
115 	    (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
116 		lock_mode = XFS_ILOCK_EXCL;
117 	xfs_ilock(ip, lock_mode);
118 	return lock_mode;
119 }
120 
121 /*
122  * The xfs inode contains 2 locks: a multi-reader lock called the
123  * i_iolock and a multi-reader lock called the i_lock.  This routine
124  * allows either or both of the locks to be obtained.
125  *
126  * The 2 locks should always be ordered so that the IO lock is
127  * obtained first in order to prevent deadlock.
128  *
129  * ip -- the inode being locked
130  * lock_flags -- this parameter indicates the inode's locks
131  *       to be locked.  It can be:
132  *		XFS_IOLOCK_SHARED,
133  *		XFS_IOLOCK_EXCL,
134  *		XFS_ILOCK_SHARED,
135  *		XFS_ILOCK_EXCL,
136  *		XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
137  *		XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
138  *		XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
139  *		XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
140  */
141 void
142 xfs_ilock(
143 	xfs_inode_t		*ip,
144 	uint			lock_flags)
145 {
146 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
147 
148 	/*
149 	 * You can't set both SHARED and EXCL for the same lock,
150 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
151 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
152 	 */
153 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
154 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
155 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
156 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
157 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
158 
159 	if (lock_flags & XFS_IOLOCK_EXCL)
160 		mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
161 	else if (lock_flags & XFS_IOLOCK_SHARED)
162 		mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
163 
164 	if (lock_flags & XFS_ILOCK_EXCL)
165 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
166 	else if (lock_flags & XFS_ILOCK_SHARED)
167 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
168 }
169 
170 /*
171  * This is just like xfs_ilock(), except that the caller
172  * is guaranteed not to sleep.  It returns 1 if it gets
173  * the requested locks and 0 otherwise.  If the IO lock is
174  * obtained but the inode lock cannot be, then the IO lock
175  * is dropped before returning.
176  *
177  * ip -- the inode being locked
178  * lock_flags -- this parameter indicates the inode's locks to be
179  *       to be locked.  See the comment for xfs_ilock() for a list
180  *	 of valid values.
181  */
182 int
183 xfs_ilock_nowait(
184 	xfs_inode_t		*ip,
185 	uint			lock_flags)
186 {
187 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
188 
189 	/*
190 	 * You can't set both SHARED and EXCL for the same lock,
191 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
192 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
193 	 */
194 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
195 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
196 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
197 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
198 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
199 
200 	if (lock_flags & XFS_IOLOCK_EXCL) {
201 		if (!mrtryupdate(&ip->i_iolock))
202 			goto out;
203 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
204 		if (!mrtryaccess(&ip->i_iolock))
205 			goto out;
206 	}
207 	if (lock_flags & XFS_ILOCK_EXCL) {
208 		if (!mrtryupdate(&ip->i_lock))
209 			goto out_undo_iolock;
210 	} else if (lock_flags & XFS_ILOCK_SHARED) {
211 		if (!mrtryaccess(&ip->i_lock))
212 			goto out_undo_iolock;
213 	}
214 	return 1;
215 
216  out_undo_iolock:
217 	if (lock_flags & XFS_IOLOCK_EXCL)
218 		mrunlock_excl(&ip->i_iolock);
219 	else if (lock_flags & XFS_IOLOCK_SHARED)
220 		mrunlock_shared(&ip->i_iolock);
221  out:
222 	return 0;
223 }
224 
225 /*
226  * xfs_iunlock() is used to drop the inode locks acquired with
227  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
228  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
229  * that we know which locks to drop.
230  *
231  * ip -- the inode being unlocked
232  * lock_flags -- this parameter indicates the inode's locks to be
233  *       to be unlocked.  See the comment for xfs_ilock() for a list
234  *	 of valid values for this parameter.
235  *
236  */
237 void
238 xfs_iunlock(
239 	xfs_inode_t		*ip,
240 	uint			lock_flags)
241 {
242 	/*
243 	 * You can't set both SHARED and EXCL for the same lock,
244 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
245 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
246 	 */
247 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
248 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
249 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
250 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
251 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
252 	ASSERT(lock_flags != 0);
253 
254 	if (lock_flags & XFS_IOLOCK_EXCL)
255 		mrunlock_excl(&ip->i_iolock);
256 	else if (lock_flags & XFS_IOLOCK_SHARED)
257 		mrunlock_shared(&ip->i_iolock);
258 
259 	if (lock_flags & XFS_ILOCK_EXCL)
260 		mrunlock_excl(&ip->i_lock);
261 	else if (lock_flags & XFS_ILOCK_SHARED)
262 		mrunlock_shared(&ip->i_lock);
263 
264 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
265 }
266 
267 /*
268  * give up write locks.  the i/o lock cannot be held nested
269  * if it is being demoted.
270  */
271 void
272 xfs_ilock_demote(
273 	xfs_inode_t		*ip,
274 	uint			lock_flags)
275 {
276 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
277 	ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
278 
279 	if (lock_flags & XFS_ILOCK_EXCL)
280 		mrdemote(&ip->i_lock);
281 	if (lock_flags & XFS_IOLOCK_EXCL)
282 		mrdemote(&ip->i_iolock);
283 
284 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
285 }
286 
287 #if defined(DEBUG) || defined(XFS_WARN)
288 int
289 xfs_isilocked(
290 	xfs_inode_t		*ip,
291 	uint			lock_flags)
292 {
293 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
294 		if (!(lock_flags & XFS_ILOCK_SHARED))
295 			return !!ip->i_lock.mr_writer;
296 		return rwsem_is_locked(&ip->i_lock.mr_lock);
297 	}
298 
299 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
300 		if (!(lock_flags & XFS_IOLOCK_SHARED))
301 			return !!ip->i_iolock.mr_writer;
302 		return rwsem_is_locked(&ip->i_iolock.mr_lock);
303 	}
304 
305 	ASSERT(0);
306 	return 0;
307 }
308 #endif
309 
310 #ifdef DEBUG
311 int xfs_locked_n;
312 int xfs_small_retries;
313 int xfs_middle_retries;
314 int xfs_lots_retries;
315 int xfs_lock_delays;
316 #endif
317 
318 /*
319  * Bump the subclass so xfs_lock_inodes() acquires each lock with
320  * a different value
321  */
322 static inline int
323 xfs_lock_inumorder(int lock_mode, int subclass)
324 {
325 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
326 		lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
327 	if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
328 		lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
329 
330 	return lock_mode;
331 }
332 
333 /*
334  * The following routine will lock n inodes in exclusive mode.
335  * We assume the caller calls us with the inodes in i_ino order.
336  *
337  * We need to detect deadlock where an inode that we lock
338  * is in the AIL and we start waiting for another inode that is locked
339  * by a thread in a long running transaction (such as truncate). This can
340  * result in deadlock since the long running trans might need to wait
341  * for the inode we just locked in order to push the tail and free space
342  * in the log.
343  */
344 void
345 xfs_lock_inodes(
346 	xfs_inode_t	**ips,
347 	int		inodes,
348 	uint		lock_mode)
349 {
350 	int		attempts = 0, i, j, try_lock;
351 	xfs_log_item_t	*lp;
352 
353 	ASSERT(ips && (inodes >= 2)); /* we need at least two */
354 
355 	try_lock = 0;
356 	i = 0;
357 
358 again:
359 	for (; i < inodes; i++) {
360 		ASSERT(ips[i]);
361 
362 		if (i && (ips[i] == ips[i-1]))	/* Already locked */
363 			continue;
364 
365 		/*
366 		 * If try_lock is not set yet, make sure all locked inodes
367 		 * are not in the AIL.
368 		 * If any are, set try_lock to be used later.
369 		 */
370 
371 		if (!try_lock) {
372 			for (j = (i - 1); j >= 0 && !try_lock; j--) {
373 				lp = (xfs_log_item_t *)ips[j]->i_itemp;
374 				if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
375 					try_lock++;
376 				}
377 			}
378 		}
379 
380 		/*
381 		 * If any of the previous locks we have locked is in the AIL,
382 		 * we must TRY to get the second and subsequent locks. If
383 		 * we can't get any, we must release all we have
384 		 * and try again.
385 		 */
386 
387 		if (try_lock) {
388 			/* try_lock must be 0 if i is 0. */
389 			/*
390 			 * try_lock means we have an inode locked
391 			 * that is in the AIL.
392 			 */
393 			ASSERT(i != 0);
394 			if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
395 				attempts++;
396 
397 				/*
398 				 * Unlock all previous guys and try again.
399 				 * xfs_iunlock will try to push the tail
400 				 * if the inode is in the AIL.
401 				 */
402 
403 				for(j = i - 1; j >= 0; j--) {
404 
405 					/*
406 					 * Check to see if we've already
407 					 * unlocked this one.
408 					 * Not the first one going back,
409 					 * and the inode ptr is the same.
410 					 */
411 					if ((j != (i - 1)) && ips[j] ==
412 								ips[j+1])
413 						continue;
414 
415 					xfs_iunlock(ips[j], lock_mode);
416 				}
417 
418 				if ((attempts % 5) == 0) {
419 					delay(1); /* Don't just spin the CPU */
420 #ifdef DEBUG
421 					xfs_lock_delays++;
422 #endif
423 				}
424 				i = 0;
425 				try_lock = 0;
426 				goto again;
427 			}
428 		} else {
429 			xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
430 		}
431 	}
432 
433 #ifdef DEBUG
434 	if (attempts) {
435 		if (attempts < 5) xfs_small_retries++;
436 		else if (attempts < 100) xfs_middle_retries++;
437 		else xfs_lots_retries++;
438 	} else {
439 		xfs_locked_n++;
440 	}
441 #endif
442 }
443 
444 /*
445  * xfs_lock_two_inodes() can only be used to lock one type of lock
446  * at a time - the iolock or the ilock, but not both at once. If
447  * we lock both at once, lockdep will report false positives saying
448  * we have violated locking orders.
449  */
450 void
451 xfs_lock_two_inodes(
452 	xfs_inode_t		*ip0,
453 	xfs_inode_t		*ip1,
454 	uint			lock_mode)
455 {
456 	xfs_inode_t		*temp;
457 	int			attempts = 0;
458 	xfs_log_item_t		*lp;
459 
460 	if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
461 		ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
462 	ASSERT(ip0->i_ino != ip1->i_ino);
463 
464 	if (ip0->i_ino > ip1->i_ino) {
465 		temp = ip0;
466 		ip0 = ip1;
467 		ip1 = temp;
468 	}
469 
470  again:
471 	xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
472 
473 	/*
474 	 * If the first lock we have locked is in the AIL, we must TRY to get
475 	 * the second lock. If we can't get it, we must release the first one
476 	 * and try again.
477 	 */
478 	lp = (xfs_log_item_t *)ip0->i_itemp;
479 	if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
480 		if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
481 			xfs_iunlock(ip0, lock_mode);
482 			if ((++attempts % 5) == 0)
483 				delay(1); /* Don't just spin the CPU */
484 			goto again;
485 		}
486 	} else {
487 		xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
488 	}
489 }
490 
491 
492 void
493 __xfs_iflock(
494 	struct xfs_inode	*ip)
495 {
496 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
497 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
498 
499 	do {
500 		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
501 		if (xfs_isiflocked(ip))
502 			io_schedule();
503 	} while (!xfs_iflock_nowait(ip));
504 
505 	finish_wait(wq, &wait.wait);
506 }
507 
508 STATIC uint
509 _xfs_dic2xflags(
510 	__uint16_t		di_flags)
511 {
512 	uint			flags = 0;
513 
514 	if (di_flags & XFS_DIFLAG_ANY) {
515 		if (di_flags & XFS_DIFLAG_REALTIME)
516 			flags |= XFS_XFLAG_REALTIME;
517 		if (di_flags & XFS_DIFLAG_PREALLOC)
518 			flags |= XFS_XFLAG_PREALLOC;
519 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
520 			flags |= XFS_XFLAG_IMMUTABLE;
521 		if (di_flags & XFS_DIFLAG_APPEND)
522 			flags |= XFS_XFLAG_APPEND;
523 		if (di_flags & XFS_DIFLAG_SYNC)
524 			flags |= XFS_XFLAG_SYNC;
525 		if (di_flags & XFS_DIFLAG_NOATIME)
526 			flags |= XFS_XFLAG_NOATIME;
527 		if (di_flags & XFS_DIFLAG_NODUMP)
528 			flags |= XFS_XFLAG_NODUMP;
529 		if (di_flags & XFS_DIFLAG_RTINHERIT)
530 			flags |= XFS_XFLAG_RTINHERIT;
531 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
532 			flags |= XFS_XFLAG_PROJINHERIT;
533 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
534 			flags |= XFS_XFLAG_NOSYMLINKS;
535 		if (di_flags & XFS_DIFLAG_EXTSIZE)
536 			flags |= XFS_XFLAG_EXTSIZE;
537 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
538 			flags |= XFS_XFLAG_EXTSZINHERIT;
539 		if (di_flags & XFS_DIFLAG_NODEFRAG)
540 			flags |= XFS_XFLAG_NODEFRAG;
541 		if (di_flags & XFS_DIFLAG_FILESTREAM)
542 			flags |= XFS_XFLAG_FILESTREAM;
543 	}
544 
545 	return flags;
546 }
547 
548 uint
549 xfs_ip2xflags(
550 	xfs_inode_t		*ip)
551 {
552 	xfs_icdinode_t		*dic = &ip->i_d;
553 
554 	return _xfs_dic2xflags(dic->di_flags) |
555 				(XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
556 }
557 
558 uint
559 xfs_dic2xflags(
560 	xfs_dinode_t		*dip)
561 {
562 	return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
563 				(XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
564 }
565 
566 /*
567  * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
568  * is allowed, otherwise it has to be an exact match. If a CI match is found,
569  * ci_name->name will point to a the actual name (caller must free) or
570  * will be set to NULL if an exact match is found.
571  */
572 int
573 xfs_lookup(
574 	xfs_inode_t		*dp,
575 	struct xfs_name		*name,
576 	xfs_inode_t		**ipp,
577 	struct xfs_name		*ci_name)
578 {
579 	xfs_ino_t		inum;
580 	int			error;
581 	uint			lock_mode;
582 
583 	trace_xfs_lookup(dp, name);
584 
585 	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
586 		return XFS_ERROR(EIO);
587 
588 	lock_mode = xfs_ilock_data_map_shared(dp);
589 	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
590 	xfs_iunlock(dp, lock_mode);
591 
592 	if (error)
593 		goto out;
594 
595 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
596 	if (error)
597 		goto out_free_name;
598 
599 	return 0;
600 
601 out_free_name:
602 	if (ci_name)
603 		kmem_free(ci_name->name);
604 out:
605 	*ipp = NULL;
606 	return error;
607 }
608 
609 /*
610  * Allocate an inode on disk and return a copy of its in-core version.
611  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
612  * appropriately within the inode.  The uid and gid for the inode are
613  * set according to the contents of the given cred structure.
614  *
615  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
616  * has a free inode available, call xfs_iget() to obtain the in-core
617  * version of the allocated inode.  Finally, fill in the inode and
618  * log its initial contents.  In this case, ialloc_context would be
619  * set to NULL.
620  *
621  * If xfs_dialloc() does not have an available inode, it will replenish
622  * its supply by doing an allocation. Since we can only do one
623  * allocation within a transaction without deadlocks, we must commit
624  * the current transaction before returning the inode itself.
625  * In this case, therefore, we will set ialloc_context and return.
626  * The caller should then commit the current transaction, start a new
627  * transaction, and call xfs_ialloc() again to actually get the inode.
628  *
629  * To ensure that some other process does not grab the inode that
630  * was allocated during the first call to xfs_ialloc(), this routine
631  * also returns the [locked] bp pointing to the head of the freelist
632  * as ialloc_context.  The caller should hold this buffer across
633  * the commit and pass it back into this routine on the second call.
634  *
635  * If we are allocating quota inodes, we do not have a parent inode
636  * to attach to or associate with (i.e. pip == NULL) because they
637  * are not linked into the directory structure - they are attached
638  * directly to the superblock - and so have no parent.
639  */
640 int
641 xfs_ialloc(
642 	xfs_trans_t	*tp,
643 	xfs_inode_t	*pip,
644 	umode_t		mode,
645 	xfs_nlink_t	nlink,
646 	xfs_dev_t	rdev,
647 	prid_t		prid,
648 	int		okalloc,
649 	xfs_buf_t	**ialloc_context,
650 	xfs_inode_t	**ipp)
651 {
652 	struct xfs_mount *mp = tp->t_mountp;
653 	xfs_ino_t	ino;
654 	xfs_inode_t	*ip;
655 	uint		flags;
656 	int		error;
657 	timespec_t	tv;
658 
659 	/*
660 	 * Call the space management code to pick
661 	 * the on-disk inode to be allocated.
662 	 */
663 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
664 			    ialloc_context, &ino);
665 	if (error)
666 		return error;
667 	if (*ialloc_context || ino == NULLFSINO) {
668 		*ipp = NULL;
669 		return 0;
670 	}
671 	ASSERT(*ialloc_context == NULL);
672 
673 	/*
674 	 * Get the in-core inode with the lock held exclusively.
675 	 * This is because we're setting fields here we need
676 	 * to prevent others from looking at until we're done.
677 	 */
678 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
679 			 XFS_ILOCK_EXCL, &ip);
680 	if (error)
681 		return error;
682 	ASSERT(ip != NULL);
683 
684 	/*
685 	 * We always convert v1 inodes to v2 now - we only support filesystems
686 	 * with >= v2 inode capability, so there is no reason for ever leaving
687 	 * an inode in v1 format.
688 	 */
689 	if (ip->i_d.di_version == 1)
690 		ip->i_d.di_version = 2;
691 
692 	ip->i_d.di_mode = mode;
693 	ip->i_d.di_onlink = 0;
694 	ip->i_d.di_nlink = nlink;
695 	ASSERT(ip->i_d.di_nlink == nlink);
696 	ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
697 	ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
698 	xfs_set_projid(ip, prid);
699 	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
700 
701 	if (pip && XFS_INHERIT_GID(pip)) {
702 		ip->i_d.di_gid = pip->i_d.di_gid;
703 		if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
704 			ip->i_d.di_mode |= S_ISGID;
705 		}
706 	}
707 
708 	/*
709 	 * If the group ID of the new file does not match the effective group
710 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
711 	 * (and only if the irix_sgid_inherit compatibility variable is set).
712 	 */
713 	if ((irix_sgid_inherit) &&
714 	    (ip->i_d.di_mode & S_ISGID) &&
715 	    (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
716 		ip->i_d.di_mode &= ~S_ISGID;
717 	}
718 
719 	ip->i_d.di_size = 0;
720 	ip->i_d.di_nextents = 0;
721 	ASSERT(ip->i_d.di_nblocks == 0);
722 
723 	nanotime(&tv);
724 	ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
725 	ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
726 	ip->i_d.di_atime = ip->i_d.di_mtime;
727 	ip->i_d.di_ctime = ip->i_d.di_mtime;
728 
729 	/*
730 	 * di_gen will have been taken care of in xfs_iread.
731 	 */
732 	ip->i_d.di_extsize = 0;
733 	ip->i_d.di_dmevmask = 0;
734 	ip->i_d.di_dmstate = 0;
735 	ip->i_d.di_flags = 0;
736 
737 	if (ip->i_d.di_version == 3) {
738 		ASSERT(ip->i_d.di_ino == ino);
739 		ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
740 		ip->i_d.di_crc = 0;
741 		ip->i_d.di_changecount = 1;
742 		ip->i_d.di_lsn = 0;
743 		ip->i_d.di_flags2 = 0;
744 		memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
745 		ip->i_d.di_crtime = ip->i_d.di_mtime;
746 	}
747 
748 
749 	flags = XFS_ILOG_CORE;
750 	switch (mode & S_IFMT) {
751 	case S_IFIFO:
752 	case S_IFCHR:
753 	case S_IFBLK:
754 	case S_IFSOCK:
755 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
756 		ip->i_df.if_u2.if_rdev = rdev;
757 		ip->i_df.if_flags = 0;
758 		flags |= XFS_ILOG_DEV;
759 		break;
760 	case S_IFREG:
761 	case S_IFDIR:
762 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
763 			uint	di_flags = 0;
764 
765 			if (S_ISDIR(mode)) {
766 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
767 					di_flags |= XFS_DIFLAG_RTINHERIT;
768 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
769 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
770 					ip->i_d.di_extsize = pip->i_d.di_extsize;
771 				}
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_PROJINHERIT)
793 				di_flags |= XFS_DIFLAG_PROJINHERIT;
794 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
795 			    xfs_inherit_nodefrag)
796 				di_flags |= XFS_DIFLAG_NODEFRAG;
797 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
798 				di_flags |= XFS_DIFLAG_FILESTREAM;
799 			ip->i_d.di_flags |= di_flags;
800 		}
801 		/* FALLTHROUGH */
802 	case S_IFLNK:
803 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
804 		ip->i_df.if_flags = XFS_IFEXTENTS;
805 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
806 		ip->i_df.if_u1.if_extents = NULL;
807 		break;
808 	default:
809 		ASSERT(0);
810 	}
811 	/*
812 	 * Attribute fork settings for new inode.
813 	 */
814 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
815 	ip->i_d.di_anextents = 0;
816 
817 	/*
818 	 * Log the new values stuffed into the inode.
819 	 */
820 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
821 	xfs_trans_log_inode(tp, ip, flags);
822 
823 	/* now that we have an i_mode we can setup inode ops and unlock */
824 	xfs_setup_inode(ip);
825 
826 	*ipp = ip;
827 	return 0;
828 }
829 
830 /*
831  * Allocates a new inode from disk and return a pointer to the
832  * incore copy. This routine will internally commit the current
833  * transaction and allocate a new one if the Space Manager needed
834  * to do an allocation to replenish the inode free-list.
835  *
836  * This routine is designed to be called from xfs_create and
837  * xfs_create_dir.
838  *
839  */
840 int
841 xfs_dir_ialloc(
842 	xfs_trans_t	**tpp,		/* input: current transaction;
843 					   output: may be a new transaction. */
844 	xfs_inode_t	*dp,		/* directory within whose allocate
845 					   the inode. */
846 	umode_t		mode,
847 	xfs_nlink_t	nlink,
848 	xfs_dev_t	rdev,
849 	prid_t		prid,		/* project id */
850 	int		okalloc,	/* ok to allocate new space */
851 	xfs_inode_t	**ipp,		/* pointer to inode; it will be
852 					   locked. */
853 	int		*committed)
854 
855 {
856 	xfs_trans_t	*tp;
857 	xfs_trans_t	*ntp;
858 	xfs_inode_t	*ip;
859 	xfs_buf_t	*ialloc_context = NULL;
860 	int		code;
861 	void		*dqinfo;
862 	uint		tflags;
863 
864 	tp = *tpp;
865 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
866 
867 	/*
868 	 * xfs_ialloc will return a pointer to an incore inode if
869 	 * the Space Manager has an available inode on the free
870 	 * list. Otherwise, it will do an allocation and replenish
871 	 * the freelist.  Since we can only do one allocation per
872 	 * transaction without deadlocks, we will need to commit the
873 	 * current transaction and start a new one.  We will then
874 	 * need to call xfs_ialloc again to get the inode.
875 	 *
876 	 * If xfs_ialloc did an allocation to replenish the freelist,
877 	 * it returns the bp containing the head of the freelist as
878 	 * ialloc_context. We will hold a lock on it across the
879 	 * transaction commit so that no other process can steal
880 	 * the inode(s) that we've just allocated.
881 	 */
882 	code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
883 			  &ialloc_context, &ip);
884 
885 	/*
886 	 * Return an error if we were unable to allocate a new inode.
887 	 * This should only happen if we run out of space on disk or
888 	 * encounter a disk error.
889 	 */
890 	if (code) {
891 		*ipp = NULL;
892 		return code;
893 	}
894 	if (!ialloc_context && !ip) {
895 		*ipp = NULL;
896 		return XFS_ERROR(ENOSPC);
897 	}
898 
899 	/*
900 	 * If the AGI buffer is non-NULL, then we were unable to get an
901 	 * inode in one operation.  We need to commit the current
902 	 * transaction and call xfs_ialloc() again.  It is guaranteed
903 	 * to succeed the second time.
904 	 */
905 	if (ialloc_context) {
906 		struct xfs_trans_res tres;
907 
908 		/*
909 		 * Normally, xfs_trans_commit releases all the locks.
910 		 * We call bhold to hang on to the ialloc_context across
911 		 * the commit.  Holding this buffer prevents any other
912 		 * processes from doing any allocations in this
913 		 * allocation group.
914 		 */
915 		xfs_trans_bhold(tp, ialloc_context);
916 		/*
917 		 * Save the log reservation so we can use
918 		 * them in the next transaction.
919 		 */
920 		tres.tr_logres = xfs_trans_get_log_res(tp);
921 		tres.tr_logcount = xfs_trans_get_log_count(tp);
922 
923 		/*
924 		 * We want the quota changes to be associated with the next
925 		 * transaction, NOT this one. So, detach the dqinfo from this
926 		 * and attach it to the next transaction.
927 		 */
928 		dqinfo = NULL;
929 		tflags = 0;
930 		if (tp->t_dqinfo) {
931 			dqinfo = (void *)tp->t_dqinfo;
932 			tp->t_dqinfo = NULL;
933 			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
934 			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
935 		}
936 
937 		ntp = xfs_trans_dup(tp);
938 		code = xfs_trans_commit(tp, 0);
939 		tp = ntp;
940 		if (committed != NULL) {
941 			*committed = 1;
942 		}
943 		/*
944 		 * If we get an error during the commit processing,
945 		 * release the buffer that is still held and return
946 		 * to the caller.
947 		 */
948 		if (code) {
949 			xfs_buf_relse(ialloc_context);
950 			if (dqinfo) {
951 				tp->t_dqinfo = dqinfo;
952 				xfs_trans_free_dqinfo(tp);
953 			}
954 			*tpp = ntp;
955 			*ipp = NULL;
956 			return code;
957 		}
958 
959 		/*
960 		 * transaction commit worked ok so we can drop the extra ticket
961 		 * reference that we gained in xfs_trans_dup()
962 		 */
963 		xfs_log_ticket_put(tp->t_ticket);
964 		tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
965 		code = xfs_trans_reserve(tp, &tres, 0, 0);
966 
967 		/*
968 		 * Re-attach the quota info that we detached from prev trx.
969 		 */
970 		if (dqinfo) {
971 			tp->t_dqinfo = dqinfo;
972 			tp->t_flags |= tflags;
973 		}
974 
975 		if (code) {
976 			xfs_buf_relse(ialloc_context);
977 			*tpp = ntp;
978 			*ipp = NULL;
979 			return code;
980 		}
981 		xfs_trans_bjoin(tp, ialloc_context);
982 
983 		/*
984 		 * Call ialloc again. Since we've locked out all
985 		 * other allocations in this allocation group,
986 		 * this call should always succeed.
987 		 */
988 		code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
989 				  okalloc, &ialloc_context, &ip);
990 
991 		/*
992 		 * If we get an error at this point, return to the caller
993 		 * so that the current transaction can be aborted.
994 		 */
995 		if (code) {
996 			*tpp = tp;
997 			*ipp = NULL;
998 			return code;
999 		}
1000 		ASSERT(!ialloc_context && ip);
1001 
1002 	} else {
1003 		if (committed != NULL)
1004 			*committed = 0;
1005 	}
1006 
1007 	*ipp = ip;
1008 	*tpp = tp;
1009 
1010 	return 0;
1011 }
1012 
1013 /*
1014  * Decrement the link count on an inode & log the change.
1015  * If this causes the link count to go to zero, initiate the
1016  * logging activity required to truncate a file.
1017  */
1018 int				/* error */
1019 xfs_droplink(
1020 	xfs_trans_t *tp,
1021 	xfs_inode_t *ip)
1022 {
1023 	int	error;
1024 
1025 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1026 
1027 	ASSERT (ip->i_d.di_nlink > 0);
1028 	ip->i_d.di_nlink--;
1029 	drop_nlink(VFS_I(ip));
1030 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1031 
1032 	error = 0;
1033 	if (ip->i_d.di_nlink == 0) {
1034 		/*
1035 		 * We're dropping the last link to this file.
1036 		 * Move the on-disk inode to the AGI unlinked list.
1037 		 * From xfs_inactive() we will pull the inode from
1038 		 * the list and free it.
1039 		 */
1040 		error = xfs_iunlink(tp, ip);
1041 	}
1042 	return error;
1043 }
1044 
1045 /*
1046  * Increment the link count on an inode & log the change.
1047  */
1048 int
1049 xfs_bumplink(
1050 	xfs_trans_t *tp,
1051 	xfs_inode_t *ip)
1052 {
1053 	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1054 
1055 	ASSERT(ip->i_d.di_version > 1);
1056 	ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1057 	ip->i_d.di_nlink++;
1058 	inc_nlink(VFS_I(ip));
1059 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1060 	return 0;
1061 }
1062 
1063 int
1064 xfs_create(
1065 	xfs_inode_t		*dp,
1066 	struct xfs_name		*name,
1067 	umode_t			mode,
1068 	xfs_dev_t		rdev,
1069 	xfs_inode_t		**ipp)
1070 {
1071 	int			is_dir = S_ISDIR(mode);
1072 	struct xfs_mount	*mp = dp->i_mount;
1073 	struct xfs_inode	*ip = NULL;
1074 	struct xfs_trans	*tp = NULL;
1075 	int			error;
1076 	xfs_bmap_free_t		free_list;
1077 	xfs_fsblock_t		first_block;
1078 	bool                    unlock_dp_on_error = false;
1079 	uint			cancel_flags;
1080 	int			committed;
1081 	prid_t			prid;
1082 	struct xfs_dquot	*udqp = NULL;
1083 	struct xfs_dquot	*gdqp = NULL;
1084 	struct xfs_dquot	*pdqp = NULL;
1085 	struct xfs_trans_res	tres;
1086 	uint			resblks;
1087 
1088 	trace_xfs_create(dp, name);
1089 
1090 	if (XFS_FORCED_SHUTDOWN(mp))
1091 		return XFS_ERROR(EIO);
1092 
1093 	prid = xfs_get_initial_prid(dp);
1094 
1095 	/*
1096 	 * Make sure that we have allocated dquot(s) on disk.
1097 	 */
1098 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1099 					xfs_kgid_to_gid(current_fsgid()), prid,
1100 					XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1101 					&udqp, &gdqp, &pdqp);
1102 	if (error)
1103 		return error;
1104 
1105 	if (is_dir) {
1106 		rdev = 0;
1107 		resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1108 		tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1109 		tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1110 		tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1111 	} else {
1112 		resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1113 		tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1114 		tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1115 		tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1116 	}
1117 
1118 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1119 
1120 	/*
1121 	 * Initially assume that the file does not exist and
1122 	 * reserve the resources for that case.  If that is not
1123 	 * the case we'll drop the one we have and get a more
1124 	 * appropriate transaction later.
1125 	 */
1126 	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1127 	error = xfs_trans_reserve(tp, &tres, resblks, 0);
1128 	if (error == ENOSPC) {
1129 		/* flush outstanding delalloc blocks and retry */
1130 		xfs_flush_inodes(mp);
1131 		error = xfs_trans_reserve(tp, &tres, resblks, 0);
1132 	}
1133 	if (error == ENOSPC) {
1134 		/* No space at all so try a "no-allocation" reservation */
1135 		resblks = 0;
1136 		error = xfs_trans_reserve(tp, &tres, 0, 0);
1137 	}
1138 	if (error) {
1139 		cancel_flags = 0;
1140 		goto out_trans_cancel;
1141 	}
1142 
1143 	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1144 	unlock_dp_on_error = true;
1145 
1146 	xfs_bmap_init(&free_list, &first_block);
1147 
1148 	/*
1149 	 * Reserve disk quota and the inode.
1150 	 */
1151 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1152 						pdqp, resblks, 1, 0);
1153 	if (error)
1154 		goto out_trans_cancel;
1155 
1156 	error = xfs_dir_canenter(tp, dp, name, resblks);
1157 	if (error)
1158 		goto out_trans_cancel;
1159 
1160 	/*
1161 	 * A newly created regular or special file just has one directory
1162 	 * entry pointing to them, but a directory also the "." entry
1163 	 * pointing to itself.
1164 	 */
1165 	error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1166 			       prid, resblks > 0, &ip, &committed);
1167 	if (error) {
1168 		if (error == ENOSPC)
1169 			goto out_trans_cancel;
1170 		goto out_trans_abort;
1171 	}
1172 
1173 	/*
1174 	 * Now we join the directory inode to the transaction.  We do not do it
1175 	 * earlier because xfs_dir_ialloc might commit the previous transaction
1176 	 * (and release all the locks).  An error from here on will result in
1177 	 * the transaction cancel unlocking dp so don't do it explicitly in the
1178 	 * error path.
1179 	 */
1180 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1181 	unlock_dp_on_error = false;
1182 
1183 	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1184 					&first_block, &free_list, resblks ?
1185 					resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1186 	if (error) {
1187 		ASSERT(error != ENOSPC);
1188 		goto out_trans_abort;
1189 	}
1190 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1191 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1192 
1193 	if (is_dir) {
1194 		error = xfs_dir_init(tp, ip, dp);
1195 		if (error)
1196 			goto out_bmap_cancel;
1197 
1198 		error = xfs_bumplink(tp, dp);
1199 		if (error)
1200 			goto out_bmap_cancel;
1201 	}
1202 
1203 	/*
1204 	 * If this is a synchronous mount, make sure that the
1205 	 * create transaction goes to disk before returning to
1206 	 * the user.
1207 	 */
1208 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1209 		xfs_trans_set_sync(tp);
1210 
1211 	/*
1212 	 * Attach the dquot(s) to the inodes and modify them incore.
1213 	 * These ids of the inode couldn't have changed since the new
1214 	 * inode has been locked ever since it was created.
1215 	 */
1216 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1217 
1218 	error = xfs_bmap_finish(&tp, &free_list, &committed);
1219 	if (error)
1220 		goto out_bmap_cancel;
1221 
1222 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1223 	if (error)
1224 		goto out_release_inode;
1225 
1226 	xfs_qm_dqrele(udqp);
1227 	xfs_qm_dqrele(gdqp);
1228 	xfs_qm_dqrele(pdqp);
1229 
1230 	*ipp = ip;
1231 	return 0;
1232 
1233  out_bmap_cancel:
1234 	xfs_bmap_cancel(&free_list);
1235  out_trans_abort:
1236 	cancel_flags |= XFS_TRANS_ABORT;
1237  out_trans_cancel:
1238 	xfs_trans_cancel(tp, cancel_flags);
1239  out_release_inode:
1240 	/*
1241 	 * Wait until after the current transaction is aborted to
1242 	 * release the inode.  This prevents recursive transactions
1243 	 * and deadlocks from xfs_inactive.
1244 	 */
1245 	if (ip)
1246 		IRELE(ip);
1247 
1248 	xfs_qm_dqrele(udqp);
1249 	xfs_qm_dqrele(gdqp);
1250 	xfs_qm_dqrele(pdqp);
1251 
1252 	if (unlock_dp_on_error)
1253 		xfs_iunlock(dp, XFS_ILOCK_EXCL);
1254 	return error;
1255 }
1256 
1257 int
1258 xfs_create_tmpfile(
1259 	struct xfs_inode	*dp,
1260 	struct dentry		*dentry,
1261 	umode_t			mode,
1262 	struct xfs_inode	**ipp)
1263 {
1264 	struct xfs_mount	*mp = dp->i_mount;
1265 	struct xfs_inode	*ip = NULL;
1266 	struct xfs_trans	*tp = NULL;
1267 	int			error;
1268 	uint			cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1269 	prid_t                  prid;
1270 	struct xfs_dquot	*udqp = NULL;
1271 	struct xfs_dquot	*gdqp = NULL;
1272 	struct xfs_dquot	*pdqp = NULL;
1273 	struct xfs_trans_res	*tres;
1274 	uint			resblks;
1275 
1276 	if (XFS_FORCED_SHUTDOWN(mp))
1277 		return XFS_ERROR(EIO);
1278 
1279 	prid = xfs_get_initial_prid(dp);
1280 
1281 	/*
1282 	 * Make sure that we have allocated dquot(s) on disk.
1283 	 */
1284 	error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1285 				xfs_kgid_to_gid(current_fsgid()), prid,
1286 				XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1287 				&udqp, &gdqp, &pdqp);
1288 	if (error)
1289 		return error;
1290 
1291 	resblks = XFS_IALLOC_SPACE_RES(mp);
1292 	tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1293 
1294 	tres = &M_RES(mp)->tr_create_tmpfile;
1295 	error = xfs_trans_reserve(tp, tres, resblks, 0);
1296 	if (error == ENOSPC) {
1297 		/* No space at all so try a "no-allocation" reservation */
1298 		resblks = 0;
1299 		error = xfs_trans_reserve(tp, tres, 0, 0);
1300 	}
1301 	if (error) {
1302 		cancel_flags = 0;
1303 		goto out_trans_cancel;
1304 	}
1305 
1306 	error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1307 						pdqp, resblks, 1, 0);
1308 	if (error)
1309 		goto out_trans_cancel;
1310 
1311 	error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1312 				prid, resblks > 0, &ip, NULL);
1313 	if (error) {
1314 		if (error == ENOSPC)
1315 			goto out_trans_cancel;
1316 		goto out_trans_abort;
1317 	}
1318 
1319 	if (mp->m_flags & XFS_MOUNT_WSYNC)
1320 		xfs_trans_set_sync(tp);
1321 
1322 	/*
1323 	 * Attach the dquot(s) to the inodes and modify them incore.
1324 	 * These ids of the inode couldn't have changed since the new
1325 	 * inode has been locked ever since it was created.
1326 	 */
1327 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1328 
1329 	ip->i_d.di_nlink--;
1330 	error = xfs_iunlink(tp, ip);
1331 	if (error)
1332 		goto out_trans_abort;
1333 
1334 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1335 	if (error)
1336 		goto out_release_inode;
1337 
1338 	xfs_qm_dqrele(udqp);
1339 	xfs_qm_dqrele(gdqp);
1340 	xfs_qm_dqrele(pdqp);
1341 
1342 	*ipp = ip;
1343 	return 0;
1344 
1345  out_trans_abort:
1346 	cancel_flags |= XFS_TRANS_ABORT;
1347  out_trans_cancel:
1348 	xfs_trans_cancel(tp, cancel_flags);
1349  out_release_inode:
1350 	/*
1351 	 * Wait until after the current transaction is aborted to
1352 	 * release the inode.  This prevents recursive transactions
1353 	 * and deadlocks from xfs_inactive.
1354 	 */
1355 	if (ip)
1356 		IRELE(ip);
1357 
1358 	xfs_qm_dqrele(udqp);
1359 	xfs_qm_dqrele(gdqp);
1360 	xfs_qm_dqrele(pdqp);
1361 
1362 	return error;
1363 }
1364 
1365 int
1366 xfs_link(
1367 	xfs_inode_t		*tdp,
1368 	xfs_inode_t		*sip,
1369 	struct xfs_name		*target_name)
1370 {
1371 	xfs_mount_t		*mp = tdp->i_mount;
1372 	xfs_trans_t		*tp;
1373 	int			error;
1374 	xfs_bmap_free_t         free_list;
1375 	xfs_fsblock_t           first_block;
1376 	int			cancel_flags;
1377 	int			committed;
1378 	int			resblks;
1379 
1380 	trace_xfs_link(tdp, target_name);
1381 
1382 	ASSERT(!S_ISDIR(sip->i_d.di_mode));
1383 
1384 	if (XFS_FORCED_SHUTDOWN(mp))
1385 		return XFS_ERROR(EIO);
1386 
1387 	error = xfs_qm_dqattach(sip, 0);
1388 	if (error)
1389 		goto std_return;
1390 
1391 	error = xfs_qm_dqattach(tdp, 0);
1392 	if (error)
1393 		goto std_return;
1394 
1395 	tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1396 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1397 	resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1398 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1399 	if (error == ENOSPC) {
1400 		resblks = 0;
1401 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1402 	}
1403 	if (error) {
1404 		cancel_flags = 0;
1405 		goto error_return;
1406 	}
1407 
1408 	xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1409 
1410 	xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1411 	xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1412 
1413 	/*
1414 	 * If we are using project inheritance, we only allow hard link
1415 	 * creation in our tree when the project IDs are the same; else
1416 	 * the tree quota mechanism could be circumvented.
1417 	 */
1418 	if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1419 		     (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1420 		error = XFS_ERROR(EXDEV);
1421 		goto error_return;
1422 	}
1423 
1424 	error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1425 	if (error)
1426 		goto error_return;
1427 
1428 	xfs_bmap_init(&free_list, &first_block);
1429 
1430 	if (sip->i_d.di_nlink == 0) {
1431 		error = xfs_iunlink_remove(tp, sip);
1432 		if (error)
1433 			goto abort_return;
1434 	}
1435 
1436 	error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1437 					&first_block, &free_list, resblks);
1438 	if (error)
1439 		goto abort_return;
1440 	xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1441 	xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1442 
1443 	error = xfs_bumplink(tp, sip);
1444 	if (error)
1445 		goto abort_return;
1446 
1447 	/*
1448 	 * If this is a synchronous mount, make sure that the
1449 	 * link transaction goes to disk before returning to
1450 	 * the user.
1451 	 */
1452 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1453 		xfs_trans_set_sync(tp);
1454 	}
1455 
1456 	error = xfs_bmap_finish (&tp, &free_list, &committed);
1457 	if (error) {
1458 		xfs_bmap_cancel(&free_list);
1459 		goto abort_return;
1460 	}
1461 
1462 	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1463 
1464  abort_return:
1465 	cancel_flags |= XFS_TRANS_ABORT;
1466  error_return:
1467 	xfs_trans_cancel(tp, cancel_flags);
1468  std_return:
1469 	return error;
1470 }
1471 
1472 /*
1473  * Free up the underlying blocks past new_size.  The new size must be smaller
1474  * than the current size.  This routine can be used both for the attribute and
1475  * data fork, and does not modify the inode size, which is left to the caller.
1476  *
1477  * The transaction passed to this routine must have made a permanent log
1478  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1479  * given transaction and start new ones, so make sure everything involved in
1480  * the transaction is tidy before calling here.  Some transaction will be
1481  * returned to the caller to be committed.  The incoming transaction must
1482  * already include the inode, and both inode locks must be held exclusively.
1483  * The inode must also be "held" within the transaction.  On return the inode
1484  * will be "held" within the returned transaction.  This routine does NOT
1485  * require any disk space to be reserved for it within the transaction.
1486  *
1487  * If we get an error, we must return with the inode locked and linked into the
1488  * current transaction. This keeps things simple for the higher level code,
1489  * because it always knows that the inode is locked and held in the transaction
1490  * that returns to it whether errors occur or not.  We don't mark the inode
1491  * dirty on error so that transactions can be easily aborted if possible.
1492  */
1493 int
1494 xfs_itruncate_extents(
1495 	struct xfs_trans	**tpp,
1496 	struct xfs_inode	*ip,
1497 	int			whichfork,
1498 	xfs_fsize_t		new_size)
1499 {
1500 	struct xfs_mount	*mp = ip->i_mount;
1501 	struct xfs_trans	*tp = *tpp;
1502 	struct xfs_trans	*ntp;
1503 	xfs_bmap_free_t		free_list;
1504 	xfs_fsblock_t		first_block;
1505 	xfs_fileoff_t		first_unmap_block;
1506 	xfs_fileoff_t		last_block;
1507 	xfs_filblks_t		unmap_len;
1508 	int			committed;
1509 	int			error = 0;
1510 	int			done = 0;
1511 
1512 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1513 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1514 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1515 	ASSERT(new_size <= XFS_ISIZE(ip));
1516 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1517 	ASSERT(ip->i_itemp != NULL);
1518 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1519 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1520 
1521 	trace_xfs_itruncate_extents_start(ip, new_size);
1522 
1523 	/*
1524 	 * Since it is possible for space to become allocated beyond
1525 	 * the end of the file (in a crash where the space is allocated
1526 	 * but the inode size is not yet updated), simply remove any
1527 	 * blocks which show up between the new EOF and the maximum
1528 	 * possible file size.  If the first block to be removed is
1529 	 * beyond the maximum file size (ie it is the same as last_block),
1530 	 * then there is nothing to do.
1531 	 */
1532 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1533 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1534 	if (first_unmap_block == last_block)
1535 		return 0;
1536 
1537 	ASSERT(first_unmap_block < last_block);
1538 	unmap_len = last_block - first_unmap_block + 1;
1539 	while (!done) {
1540 		xfs_bmap_init(&free_list, &first_block);
1541 		error = xfs_bunmapi(tp, ip,
1542 				    first_unmap_block, unmap_len,
1543 				    xfs_bmapi_aflag(whichfork),
1544 				    XFS_ITRUNC_MAX_EXTENTS,
1545 				    &first_block, &free_list,
1546 				    &done);
1547 		if (error)
1548 			goto out_bmap_cancel;
1549 
1550 		/*
1551 		 * Duplicate the transaction that has the permanent
1552 		 * reservation and commit the old transaction.
1553 		 */
1554 		error = xfs_bmap_finish(&tp, &free_list, &committed);
1555 		if (committed)
1556 			xfs_trans_ijoin(tp, ip, 0);
1557 		if (error)
1558 			goto out_bmap_cancel;
1559 
1560 		if (committed) {
1561 			/*
1562 			 * Mark the inode dirty so it will be logged and
1563 			 * moved forward in the log as part of every commit.
1564 			 */
1565 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1566 		}
1567 
1568 		ntp = xfs_trans_dup(tp);
1569 		error = xfs_trans_commit(tp, 0);
1570 		tp = ntp;
1571 
1572 		xfs_trans_ijoin(tp, ip, 0);
1573 
1574 		if (error)
1575 			goto out;
1576 
1577 		/*
1578 		 * Transaction commit worked ok so we can drop the extra ticket
1579 		 * reference that we gained in xfs_trans_dup()
1580 		 */
1581 		xfs_log_ticket_put(tp->t_ticket);
1582 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1583 		if (error)
1584 			goto out;
1585 	}
1586 
1587 	/*
1588 	 * Always re-log the inode so that our permanent transaction can keep
1589 	 * on rolling it forward in the log.
1590 	 */
1591 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1592 
1593 	trace_xfs_itruncate_extents_end(ip, new_size);
1594 
1595 out:
1596 	*tpp = tp;
1597 	return error;
1598 out_bmap_cancel:
1599 	/*
1600 	 * If the bunmapi call encounters an error, return to the caller where
1601 	 * the transaction can be properly aborted.  We just need to make sure
1602 	 * we're not holding any resources that we were not when we came in.
1603 	 */
1604 	xfs_bmap_cancel(&free_list);
1605 	goto out;
1606 }
1607 
1608 int
1609 xfs_release(
1610 	xfs_inode_t	*ip)
1611 {
1612 	xfs_mount_t	*mp = ip->i_mount;
1613 	int		error;
1614 
1615 	if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1616 		return 0;
1617 
1618 	/* If this is a read-only mount, don't do this (would generate I/O) */
1619 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1620 		return 0;
1621 
1622 	if (!XFS_FORCED_SHUTDOWN(mp)) {
1623 		int truncated;
1624 
1625 		/*
1626 		 * If we previously truncated this file and removed old data
1627 		 * in the process, we want to initiate "early" writeout on
1628 		 * the last close.  This is an attempt to combat the notorious
1629 		 * NULL files problem which is particularly noticeable from a
1630 		 * truncate down, buffered (re-)write (delalloc), followed by
1631 		 * a crash.  What we are effectively doing here is
1632 		 * significantly reducing the time window where we'd otherwise
1633 		 * be exposed to that problem.
1634 		 */
1635 		truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1636 		if (truncated) {
1637 			xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1638 			if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1639 				error = -filemap_flush(VFS_I(ip)->i_mapping);
1640 				if (error)
1641 					return error;
1642 			}
1643 		}
1644 	}
1645 
1646 	if (ip->i_d.di_nlink == 0)
1647 		return 0;
1648 
1649 	if (xfs_can_free_eofblocks(ip, false)) {
1650 
1651 		/*
1652 		 * If we can't get the iolock just skip truncating the blocks
1653 		 * past EOF because we could deadlock with the mmap_sem
1654 		 * otherwise.  We'll get another chance to drop them once the
1655 		 * last reference to the inode is dropped, so we'll never leak
1656 		 * blocks permanently.
1657 		 *
1658 		 * Further, check if the inode is being opened, written and
1659 		 * closed frequently and we have delayed allocation blocks
1660 		 * outstanding (e.g. streaming writes from the NFS server),
1661 		 * truncating the blocks past EOF will cause fragmentation to
1662 		 * occur.
1663 		 *
1664 		 * In this case don't do the truncation, either, but we have to
1665 		 * be careful how we detect this case. Blocks beyond EOF show
1666 		 * up as i_delayed_blks even when the inode is clean, so we
1667 		 * need to truncate them away first before checking for a dirty
1668 		 * release. Hence on the first dirty close we will still remove
1669 		 * the speculative allocation, but after that we will leave it
1670 		 * in place.
1671 		 */
1672 		if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1673 			return 0;
1674 
1675 		error = xfs_free_eofblocks(mp, ip, true);
1676 		if (error && error != EAGAIN)
1677 			return error;
1678 
1679 		/* delalloc blocks after truncation means it really is dirty */
1680 		if (ip->i_delayed_blks)
1681 			xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1682 	}
1683 	return 0;
1684 }
1685 
1686 /*
1687  * xfs_inactive_truncate
1688  *
1689  * Called to perform a truncate when an inode becomes unlinked.
1690  */
1691 STATIC int
1692 xfs_inactive_truncate(
1693 	struct xfs_inode *ip)
1694 {
1695 	struct xfs_mount	*mp = ip->i_mount;
1696 	struct xfs_trans	*tp;
1697 	int			error;
1698 
1699 	tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1700 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1701 	if (error) {
1702 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1703 		xfs_trans_cancel(tp, 0);
1704 		return error;
1705 	}
1706 
1707 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1708 	xfs_trans_ijoin(tp, ip, 0);
1709 
1710 	/*
1711 	 * Log the inode size first to prevent stale data exposure in the event
1712 	 * of a system crash before the truncate completes. See the related
1713 	 * comment in xfs_setattr_size() for details.
1714 	 */
1715 	ip->i_d.di_size = 0;
1716 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1717 
1718 	error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1719 	if (error)
1720 		goto error_trans_cancel;
1721 
1722 	ASSERT(ip->i_d.di_nextents == 0);
1723 
1724 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1725 	if (error)
1726 		goto error_unlock;
1727 
1728 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1729 	return 0;
1730 
1731 error_trans_cancel:
1732 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1733 error_unlock:
1734 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1735 	return error;
1736 }
1737 
1738 /*
1739  * xfs_inactive_ifree()
1740  *
1741  * Perform the inode free when an inode is unlinked.
1742  */
1743 STATIC int
1744 xfs_inactive_ifree(
1745 	struct xfs_inode *ip)
1746 {
1747 	xfs_bmap_free_t		free_list;
1748 	xfs_fsblock_t		first_block;
1749 	int			committed;
1750 	struct xfs_mount	*mp = ip->i_mount;
1751 	struct xfs_trans	*tp;
1752 	int			error;
1753 
1754 	tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1755 
1756 	/*
1757 	 * The ifree transaction might need to allocate blocks for record
1758 	 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1759 	 * allow ifree to dip into the reserved block pool if necessary.
1760 	 *
1761 	 * Freeing large sets of inodes generally means freeing inode chunks,
1762 	 * directory and file data blocks, so this should be relatively safe.
1763 	 * Only under severe circumstances should it be possible to free enough
1764 	 * inodes to exhaust the reserve block pool via finobt expansion while
1765 	 * at the same time not creating free space in the filesystem.
1766 	 *
1767 	 * Send a warning if the reservation does happen to fail, as the inode
1768 	 * now remains allocated and sits on the unlinked list until the fs is
1769 	 * repaired.
1770 	 */
1771 	tp->t_flags |= XFS_TRANS_RESERVE;
1772 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1773 				  XFS_IFREE_SPACE_RES(mp), 0);
1774 	if (error) {
1775 		if (error == ENOSPC) {
1776 			xfs_warn_ratelimited(mp,
1777 			"Failed to remove inode(s) from unlinked list. "
1778 			"Please free space, unmount and run xfs_repair.");
1779 		} else {
1780 			ASSERT(XFS_FORCED_SHUTDOWN(mp));
1781 		}
1782 		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1783 		return error;
1784 	}
1785 
1786 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1787 	xfs_trans_ijoin(tp, ip, 0);
1788 
1789 	xfs_bmap_init(&free_list, &first_block);
1790 	error = xfs_ifree(tp, ip, &free_list);
1791 	if (error) {
1792 		/*
1793 		 * If we fail to free the inode, shut down.  The cancel
1794 		 * might do that, we need to make sure.  Otherwise the
1795 		 * inode might be lost for a long time or forever.
1796 		 */
1797 		if (!XFS_FORCED_SHUTDOWN(mp)) {
1798 			xfs_notice(mp, "%s: xfs_ifree returned error %d",
1799 				__func__, error);
1800 			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1801 		}
1802 		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1803 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1804 		return error;
1805 	}
1806 
1807 	/*
1808 	 * Credit the quota account(s). The inode is gone.
1809 	 */
1810 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1811 
1812 	/*
1813 	 * Just ignore errors at this point.  There is nothing we can
1814 	 * do except to try to keep going. Make sure it's not a silent
1815 	 * error.
1816 	 */
1817 	error = xfs_bmap_finish(&tp,  &free_list, &committed);
1818 	if (error)
1819 		xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1820 			__func__, error);
1821 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1822 	if (error)
1823 		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1824 			__func__, error);
1825 
1826 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1827 	return 0;
1828 }
1829 
1830 /*
1831  * xfs_inactive
1832  *
1833  * This is called when the vnode reference count for the vnode
1834  * goes to zero.  If the file has been unlinked, then it must
1835  * now be truncated.  Also, we clear all of the read-ahead state
1836  * kept for the inode here since the file is now closed.
1837  */
1838 void
1839 xfs_inactive(
1840 	xfs_inode_t	*ip)
1841 {
1842 	struct xfs_mount	*mp;
1843 	int			error;
1844 	int			truncate = 0;
1845 
1846 	/*
1847 	 * If the inode is already free, then there can be nothing
1848 	 * to clean up here.
1849 	 */
1850 	if (ip->i_d.di_mode == 0) {
1851 		ASSERT(ip->i_df.if_real_bytes == 0);
1852 		ASSERT(ip->i_df.if_broot_bytes == 0);
1853 		return;
1854 	}
1855 
1856 	mp = ip->i_mount;
1857 
1858 	/* If this is a read-only mount, don't do this (would generate I/O) */
1859 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1860 		return;
1861 
1862 	if (ip->i_d.di_nlink != 0) {
1863 		/*
1864 		 * force is true because we are evicting an inode from the
1865 		 * cache. Post-eof blocks must be freed, lest we end up with
1866 		 * broken free space accounting.
1867 		 */
1868 		if (xfs_can_free_eofblocks(ip, true))
1869 			xfs_free_eofblocks(mp, ip, false);
1870 
1871 		return;
1872 	}
1873 
1874 	if (S_ISREG(ip->i_d.di_mode) &&
1875 	    (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1876 	     ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1877 		truncate = 1;
1878 
1879 	error = xfs_qm_dqattach(ip, 0);
1880 	if (error)
1881 		return;
1882 
1883 	if (S_ISLNK(ip->i_d.di_mode))
1884 		error = xfs_inactive_symlink(ip);
1885 	else if (truncate)
1886 		error = xfs_inactive_truncate(ip);
1887 	if (error)
1888 		return;
1889 
1890 	/*
1891 	 * If there are attributes associated with the file then blow them away
1892 	 * now.  The code calls a routine that recursively deconstructs the
1893 	 * attribute fork.  We need to just commit the current transaction
1894 	 * because we can't use it for xfs_attr_inactive().
1895 	 */
1896 	if (ip->i_d.di_anextents > 0) {
1897 		ASSERT(ip->i_d.di_forkoff != 0);
1898 
1899 		error = xfs_attr_inactive(ip);
1900 		if (error)
1901 			return;
1902 	}
1903 
1904 	if (ip->i_afp)
1905 		xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1906 
1907 	ASSERT(ip->i_d.di_anextents == 0);
1908 
1909 	/*
1910 	 * Free the inode.
1911 	 */
1912 	error = xfs_inactive_ifree(ip);
1913 	if (error)
1914 		return;
1915 
1916 	/*
1917 	 * Release the dquots held by inode, if any.
1918 	 */
1919 	xfs_qm_dqdetach(ip);
1920 }
1921 
1922 /*
1923  * This is called when the inode's link count goes to 0.
1924  * We place the on-disk inode on a list in the AGI.  It
1925  * will be pulled from this list when the inode is freed.
1926  */
1927 int
1928 xfs_iunlink(
1929 	xfs_trans_t	*tp,
1930 	xfs_inode_t	*ip)
1931 {
1932 	xfs_mount_t	*mp;
1933 	xfs_agi_t	*agi;
1934 	xfs_dinode_t	*dip;
1935 	xfs_buf_t	*agibp;
1936 	xfs_buf_t	*ibp;
1937 	xfs_agino_t	agino;
1938 	short		bucket_index;
1939 	int		offset;
1940 	int		error;
1941 
1942 	ASSERT(ip->i_d.di_nlink == 0);
1943 	ASSERT(ip->i_d.di_mode != 0);
1944 
1945 	mp = tp->t_mountp;
1946 
1947 	/*
1948 	 * Get the agi buffer first.  It ensures lock ordering
1949 	 * on the list.
1950 	 */
1951 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1952 	if (error)
1953 		return error;
1954 	agi = XFS_BUF_TO_AGI(agibp);
1955 
1956 	/*
1957 	 * Get the index into the agi hash table for the
1958 	 * list this inode will go on.
1959 	 */
1960 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1961 	ASSERT(agino != 0);
1962 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1963 	ASSERT(agi->agi_unlinked[bucket_index]);
1964 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1965 
1966 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1967 		/*
1968 		 * There is already another inode in the bucket we need
1969 		 * to add ourselves to.  Add us at the front of the list.
1970 		 * Here we put the head pointer into our next pointer,
1971 		 * and then we fall through to point the head at us.
1972 		 */
1973 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1974 				       0, 0);
1975 		if (error)
1976 			return error;
1977 
1978 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1979 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1980 		offset = ip->i_imap.im_boffset +
1981 			offsetof(xfs_dinode_t, di_next_unlinked);
1982 
1983 		/* need to recalc the inode CRC if appropriate */
1984 		xfs_dinode_calc_crc(mp, dip);
1985 
1986 		xfs_trans_inode_buf(tp, ibp);
1987 		xfs_trans_log_buf(tp, ibp, offset,
1988 				  (offset + sizeof(xfs_agino_t) - 1));
1989 		xfs_inobp_check(mp, ibp);
1990 	}
1991 
1992 	/*
1993 	 * Point the bucket head pointer at the inode being inserted.
1994 	 */
1995 	ASSERT(agino != 0);
1996 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1997 	offset = offsetof(xfs_agi_t, agi_unlinked) +
1998 		(sizeof(xfs_agino_t) * bucket_index);
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_log_buf(tp, agibp, offset,
2091 				  (offset + sizeof(xfs_agino_t) - 1));
2092 	} else {
2093 		/*
2094 		 * We need to search the list for the inode being freed.
2095 		 */
2096 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2097 		last_ibp = NULL;
2098 		while (next_agino != agino) {
2099 			struct xfs_imap	imap;
2100 
2101 			if (last_ibp)
2102 				xfs_trans_brelse(tp, last_ibp);
2103 
2104 			imap.im_blkno = 0;
2105 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2106 
2107 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
2108 			if (error) {
2109 				xfs_warn(mp,
2110 	"%s: xfs_imap returned error %d.",
2111 					 __func__, error);
2112 				return error;
2113 			}
2114 
2115 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2116 					       &last_ibp, 0, 0);
2117 			if (error) {
2118 				xfs_warn(mp,
2119 	"%s: xfs_imap_to_bp returned error %d.",
2120 					__func__, error);
2121 				return error;
2122 			}
2123 
2124 			last_offset = imap.im_boffset;
2125 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2126 			ASSERT(next_agino != NULLAGINO);
2127 			ASSERT(next_agino != 0);
2128 		}
2129 
2130 		/*
2131 		 * Now last_ibp points to the buffer previous to us on the
2132 		 * unlinked list.  Pull us from the list.
2133 		 */
2134 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2135 				       0, 0);
2136 		if (error) {
2137 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2138 				__func__, error);
2139 			return error;
2140 		}
2141 		next_agino = be32_to_cpu(dip->di_next_unlinked);
2142 		ASSERT(next_agino != 0);
2143 		ASSERT(next_agino != agino);
2144 		if (next_agino != NULLAGINO) {
2145 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2146 			offset = ip->i_imap.im_boffset +
2147 				offsetof(xfs_dinode_t, di_next_unlinked);
2148 
2149 			/* need to recalc the inode CRC if appropriate */
2150 			xfs_dinode_calc_crc(mp, dip);
2151 
2152 			xfs_trans_inode_buf(tp, ibp);
2153 			xfs_trans_log_buf(tp, ibp, offset,
2154 					  (offset + sizeof(xfs_agino_t) - 1));
2155 			xfs_inobp_check(mp, ibp);
2156 		} else {
2157 			xfs_trans_brelse(tp, ibp);
2158 		}
2159 		/*
2160 		 * Point the previous inode on the list to the next inode.
2161 		 */
2162 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2163 		ASSERT(next_agino != 0);
2164 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2165 
2166 		/* need to recalc the inode CRC if appropriate */
2167 		xfs_dinode_calc_crc(mp, last_dip);
2168 
2169 		xfs_trans_inode_buf(tp, last_ibp);
2170 		xfs_trans_log_buf(tp, last_ibp, offset,
2171 				  (offset + sizeof(xfs_agino_t) - 1));
2172 		xfs_inobp_check(mp, last_ibp);
2173 	}
2174 	return 0;
2175 }
2176 
2177 /*
2178  * A big issue when freeing the inode cluster is that we _cannot_ skip any
2179  * inodes that are in memory - they all must be marked stale and attached to
2180  * the cluster buffer.
2181  */
2182 STATIC int
2183 xfs_ifree_cluster(
2184 	xfs_inode_t	*free_ip,
2185 	xfs_trans_t	*tp,
2186 	xfs_ino_t	inum)
2187 {
2188 	xfs_mount_t		*mp = free_ip->i_mount;
2189 	int			blks_per_cluster;
2190 	int			inodes_per_cluster;
2191 	int			nbufs;
2192 	int			i, j;
2193 	xfs_daddr_t		blkno;
2194 	xfs_buf_t		*bp;
2195 	xfs_inode_t		*ip;
2196 	xfs_inode_log_item_t	*iip;
2197 	xfs_log_item_t		*lip;
2198 	struct xfs_perag	*pag;
2199 
2200 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2201 	blks_per_cluster = xfs_icluster_size_fsb(mp);
2202 	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2203 	nbufs = mp->m_ialloc_blks / blks_per_cluster;
2204 
2205 	for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2206 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2207 					 XFS_INO_TO_AGBNO(mp, inum));
2208 
2209 		/*
2210 		 * We obtain and lock the backing buffer first in the process
2211 		 * here, as we have to ensure that any dirty inode that we
2212 		 * can't get the flush lock on is attached to the buffer.
2213 		 * If we scan the in-memory inodes first, then buffer IO can
2214 		 * complete before we get a lock on it, and hence we may fail
2215 		 * to mark all the active inodes on the buffer stale.
2216 		 */
2217 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2218 					mp->m_bsize * blks_per_cluster,
2219 					XBF_UNMAPPED);
2220 
2221 		if (!bp)
2222 			return ENOMEM;
2223 
2224 		/*
2225 		 * This buffer may not have been correctly initialised as we
2226 		 * didn't read it from disk. That's not important because we are
2227 		 * only using to mark the buffer as stale in the log, and to
2228 		 * attach stale cached inodes on it. That means it will never be
2229 		 * dispatched for IO. If it is, we want to know about it, and we
2230 		 * want it to fail. We can acheive this by adding a write
2231 		 * verifier to the buffer.
2232 		 */
2233 		 bp->b_ops = &xfs_inode_buf_ops;
2234 
2235 		/*
2236 		 * Walk the inodes already attached to the buffer and mark them
2237 		 * stale. These will all have the flush locks held, so an
2238 		 * in-memory inode walk can't lock them. By marking them all
2239 		 * stale first, we will not attempt to lock them in the loop
2240 		 * below as the XFS_ISTALE flag will be set.
2241 		 */
2242 		lip = bp->b_fspriv;
2243 		while (lip) {
2244 			if (lip->li_type == XFS_LI_INODE) {
2245 				iip = (xfs_inode_log_item_t *)lip;
2246 				ASSERT(iip->ili_logged == 1);
2247 				lip->li_cb = xfs_istale_done;
2248 				xfs_trans_ail_copy_lsn(mp->m_ail,
2249 							&iip->ili_flush_lsn,
2250 							&iip->ili_item.li_lsn);
2251 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2252 			}
2253 			lip = lip->li_bio_list;
2254 		}
2255 
2256 
2257 		/*
2258 		 * For each inode in memory attempt to add it to the inode
2259 		 * buffer and set it up for being staled on buffer IO
2260 		 * completion.  This is safe as we've locked out tail pushing
2261 		 * and flushing by locking the buffer.
2262 		 *
2263 		 * We have already marked every inode that was part of a
2264 		 * transaction stale above, which means there is no point in
2265 		 * even trying to lock them.
2266 		 */
2267 		for (i = 0; i < inodes_per_cluster; i++) {
2268 retry:
2269 			rcu_read_lock();
2270 			ip = radix_tree_lookup(&pag->pag_ici_root,
2271 					XFS_INO_TO_AGINO(mp, (inum + i)));
2272 
2273 			/* Inode not in memory, nothing to do */
2274 			if (!ip) {
2275 				rcu_read_unlock();
2276 				continue;
2277 			}
2278 
2279 			/*
2280 			 * because this is an RCU protected lookup, we could
2281 			 * find a recently freed or even reallocated inode
2282 			 * during the lookup. We need to check under the
2283 			 * i_flags_lock for a valid inode here. Skip it if it
2284 			 * is not valid, the wrong inode or stale.
2285 			 */
2286 			spin_lock(&ip->i_flags_lock);
2287 			if (ip->i_ino != inum + i ||
2288 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
2289 				spin_unlock(&ip->i_flags_lock);
2290 				rcu_read_unlock();
2291 				continue;
2292 			}
2293 			spin_unlock(&ip->i_flags_lock);
2294 
2295 			/*
2296 			 * Don't try to lock/unlock the current inode, but we
2297 			 * _cannot_ skip the other inodes that we did not find
2298 			 * in the list attached to the buffer and are not
2299 			 * already marked stale. If we can't lock it, back off
2300 			 * and retry.
2301 			 */
2302 			if (ip != free_ip &&
2303 			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2304 				rcu_read_unlock();
2305 				delay(1);
2306 				goto retry;
2307 			}
2308 			rcu_read_unlock();
2309 
2310 			xfs_iflock(ip);
2311 			xfs_iflags_set(ip, XFS_ISTALE);
2312 
2313 			/*
2314 			 * we don't need to attach clean inodes or those only
2315 			 * with unlogged changes (which we throw away, anyway).
2316 			 */
2317 			iip = ip->i_itemp;
2318 			if (!iip || xfs_inode_clean(ip)) {
2319 				ASSERT(ip != free_ip);
2320 				xfs_ifunlock(ip);
2321 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2322 				continue;
2323 			}
2324 
2325 			iip->ili_last_fields = iip->ili_fields;
2326 			iip->ili_fields = 0;
2327 			iip->ili_logged = 1;
2328 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2329 						&iip->ili_item.li_lsn);
2330 
2331 			xfs_buf_attach_iodone(bp, xfs_istale_done,
2332 						  &iip->ili_item);
2333 
2334 			if (ip != free_ip)
2335 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2336 		}
2337 
2338 		xfs_trans_stale_inode_buf(tp, bp);
2339 		xfs_trans_binval(tp, bp);
2340 	}
2341 
2342 	xfs_perag_put(pag);
2343 	return 0;
2344 }
2345 
2346 /*
2347  * This is called to return an inode to the inode free list.
2348  * The inode should already be truncated to 0 length and have
2349  * no pages associated with it.  This routine also assumes that
2350  * the inode is already a part of the transaction.
2351  *
2352  * The on-disk copy of the inode will have been added to the list
2353  * of unlinked inodes in the AGI. We need to remove the inode from
2354  * that list atomically with respect to freeing it here.
2355  */
2356 int
2357 xfs_ifree(
2358 	xfs_trans_t	*tp,
2359 	xfs_inode_t	*ip,
2360 	xfs_bmap_free_t	*flist)
2361 {
2362 	int			error;
2363 	int			delete;
2364 	xfs_ino_t		first_ino;
2365 
2366 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2367 	ASSERT(ip->i_d.di_nlink == 0);
2368 	ASSERT(ip->i_d.di_nextents == 0);
2369 	ASSERT(ip->i_d.di_anextents == 0);
2370 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2371 	ASSERT(ip->i_d.di_nblocks == 0);
2372 
2373 	/*
2374 	 * Pull the on-disk inode from the AGI unlinked list.
2375 	 */
2376 	error = xfs_iunlink_remove(tp, ip);
2377 	if (error)
2378 		return error;
2379 
2380 	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2381 	if (error)
2382 		return error;
2383 
2384 	ip->i_d.di_mode = 0;		/* mark incore inode as free */
2385 	ip->i_d.di_flags = 0;
2386 	ip->i_d.di_dmevmask = 0;
2387 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2388 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2389 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2390 	/*
2391 	 * Bump the generation count so no one will be confused
2392 	 * by reincarnations of this inode.
2393 	 */
2394 	ip->i_d.di_gen++;
2395 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2396 
2397 	if (delete)
2398 		error = xfs_ifree_cluster(ip, tp, first_ino);
2399 
2400 	return error;
2401 }
2402 
2403 /*
2404  * This is called to unpin an inode.  The caller must have the inode locked
2405  * in at least shared mode so that the buffer cannot be subsequently pinned
2406  * once someone is waiting for it to be unpinned.
2407  */
2408 static void
2409 xfs_iunpin(
2410 	struct xfs_inode	*ip)
2411 {
2412 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2413 
2414 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2415 
2416 	/* Give the log a push to start the unpinning I/O */
2417 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2418 
2419 }
2420 
2421 static void
2422 __xfs_iunpin_wait(
2423 	struct xfs_inode	*ip)
2424 {
2425 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2426 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2427 
2428 	xfs_iunpin(ip);
2429 
2430 	do {
2431 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2432 		if (xfs_ipincount(ip))
2433 			io_schedule();
2434 	} while (xfs_ipincount(ip));
2435 	finish_wait(wq, &wait.wait);
2436 }
2437 
2438 void
2439 xfs_iunpin_wait(
2440 	struct xfs_inode	*ip)
2441 {
2442 	if (xfs_ipincount(ip))
2443 		__xfs_iunpin_wait(ip);
2444 }
2445 
2446 /*
2447  * Removing an inode from the namespace involves removing the directory entry
2448  * and dropping the link count on the inode. Removing the directory entry can
2449  * result in locking an AGF (directory blocks were freed) and removing a link
2450  * count can result in placing the inode on an unlinked list which results in
2451  * locking an AGI.
2452  *
2453  * The big problem here is that we have an ordering constraint on AGF and AGI
2454  * locking - inode allocation locks the AGI, then can allocate a new extent for
2455  * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2456  * removes the inode from the unlinked list, requiring that we lock the AGI
2457  * first, and then freeing the inode can result in an inode chunk being freed
2458  * and hence freeing disk space requiring that we lock an AGF.
2459  *
2460  * Hence the ordering that is imposed by other parts of the code is AGI before
2461  * AGF. This means we cannot remove the directory entry before we drop the inode
2462  * reference count and put it on the unlinked list as this results in a lock
2463  * order of AGF then AGI, and this can deadlock against inode allocation and
2464  * freeing. Therefore we must drop the link counts before we remove the
2465  * directory entry.
2466  *
2467  * This is still safe from a transactional point of view - it is not until we
2468  * get to xfs_bmap_finish() that we have the possibility of multiple
2469  * transactions in this operation. Hence as long as we remove the directory
2470  * entry and drop the link count in the first transaction of the remove
2471  * operation, there are no transactional constraints on the ordering here.
2472  */
2473 int
2474 xfs_remove(
2475 	xfs_inode_t             *dp,
2476 	struct xfs_name		*name,
2477 	xfs_inode_t		*ip)
2478 {
2479 	xfs_mount_t		*mp = dp->i_mount;
2480 	xfs_trans_t             *tp = NULL;
2481 	int			is_dir = S_ISDIR(ip->i_d.di_mode);
2482 	int                     error = 0;
2483 	xfs_bmap_free_t         free_list;
2484 	xfs_fsblock_t           first_block;
2485 	int			cancel_flags;
2486 	int			committed;
2487 	int			link_zero;
2488 	uint			resblks;
2489 	uint			log_count;
2490 
2491 	trace_xfs_remove(dp, name);
2492 
2493 	if (XFS_FORCED_SHUTDOWN(mp))
2494 		return XFS_ERROR(EIO);
2495 
2496 	error = xfs_qm_dqattach(dp, 0);
2497 	if (error)
2498 		goto std_return;
2499 
2500 	error = xfs_qm_dqattach(ip, 0);
2501 	if (error)
2502 		goto std_return;
2503 
2504 	if (is_dir) {
2505 		tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2506 		log_count = XFS_DEFAULT_LOG_COUNT;
2507 	} else {
2508 		tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2509 		log_count = XFS_REMOVE_LOG_COUNT;
2510 	}
2511 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2512 
2513 	/*
2514 	 * We try to get the real space reservation first,
2515 	 * allowing for directory btree deletion(s) implying
2516 	 * possible bmap insert(s).  If we can't get the space
2517 	 * reservation then we use 0 instead, and avoid the bmap
2518 	 * btree insert(s) in the directory code by, if the bmap
2519 	 * insert tries to happen, instead trimming the LAST
2520 	 * block from the directory.
2521 	 */
2522 	resblks = XFS_REMOVE_SPACE_RES(mp);
2523 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2524 	if (error == ENOSPC) {
2525 		resblks = 0;
2526 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2527 	}
2528 	if (error) {
2529 		ASSERT(error != ENOSPC);
2530 		cancel_flags = 0;
2531 		goto out_trans_cancel;
2532 	}
2533 
2534 	xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2535 
2536 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2537 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2538 
2539 	/*
2540 	 * If we're removing a directory perform some additional validation.
2541 	 */
2542 	cancel_flags |= XFS_TRANS_ABORT;
2543 	if (is_dir) {
2544 		ASSERT(ip->i_d.di_nlink >= 2);
2545 		if (ip->i_d.di_nlink != 2) {
2546 			error = XFS_ERROR(ENOTEMPTY);
2547 			goto out_trans_cancel;
2548 		}
2549 		if (!xfs_dir_isempty(ip)) {
2550 			error = XFS_ERROR(ENOTEMPTY);
2551 			goto out_trans_cancel;
2552 		}
2553 
2554 		/* Drop the link from ip's "..".  */
2555 		error = xfs_droplink(tp, dp);
2556 		if (error)
2557 			goto out_trans_cancel;
2558 
2559 		/* Drop the "." link from ip to self.  */
2560 		error = xfs_droplink(tp, ip);
2561 		if (error)
2562 			goto out_trans_cancel;
2563 	} else {
2564 		/*
2565 		 * When removing a non-directory we need to log the parent
2566 		 * inode here.  For a directory this is done implicitly
2567 		 * by the xfs_droplink call for the ".." entry.
2568 		 */
2569 		xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2570 	}
2571 	xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2572 
2573 	/* Drop the link from dp to ip. */
2574 	error = xfs_droplink(tp, ip);
2575 	if (error)
2576 		goto out_trans_cancel;
2577 
2578 	/* Determine if this is the last link while the inode is locked */
2579 	link_zero = (ip->i_d.di_nlink == 0);
2580 
2581 	xfs_bmap_init(&free_list, &first_block);
2582 	error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2583 					&first_block, &free_list, resblks);
2584 	if (error) {
2585 		ASSERT(error != ENOENT);
2586 		goto out_bmap_cancel;
2587 	}
2588 
2589 	/*
2590 	 * If this is a synchronous mount, make sure that the
2591 	 * remove transaction goes to disk before returning to
2592 	 * the user.
2593 	 */
2594 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2595 		xfs_trans_set_sync(tp);
2596 
2597 	error = xfs_bmap_finish(&tp, &free_list, &committed);
2598 	if (error)
2599 		goto out_bmap_cancel;
2600 
2601 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2602 	if (error)
2603 		goto std_return;
2604 
2605 	if (is_dir && xfs_inode_is_filestream(ip))
2606 		xfs_filestream_deassociate(ip);
2607 
2608 	return 0;
2609 
2610  out_bmap_cancel:
2611 	xfs_bmap_cancel(&free_list);
2612  out_trans_cancel:
2613 	xfs_trans_cancel(tp, cancel_flags);
2614  std_return:
2615 	return error;
2616 }
2617 
2618 /*
2619  * Enter all inodes for a rename transaction into a sorted array.
2620  */
2621 STATIC void
2622 xfs_sort_for_rename(
2623 	xfs_inode_t	*dp1,	/* in: old (source) directory inode */
2624 	xfs_inode_t	*dp2,	/* in: new (target) directory inode */
2625 	xfs_inode_t	*ip1,	/* in: inode of old entry */
2626 	xfs_inode_t	*ip2,	/* in: inode of new entry, if it
2627 				   already exists, NULL otherwise. */
2628 	xfs_inode_t	**i_tab,/* out: array of inode returned, sorted */
2629 	int		*num_inodes)  /* out: number of inodes in array */
2630 {
2631 	xfs_inode_t		*temp;
2632 	int			i, j;
2633 
2634 	/*
2635 	 * i_tab contains a list of pointers to inodes.  We initialize
2636 	 * the table here & we'll sort it.  We will then use it to
2637 	 * order the acquisition of the inode locks.
2638 	 *
2639 	 * Note that the table may contain duplicates.  e.g., dp1 == dp2.
2640 	 */
2641 	i_tab[0] = dp1;
2642 	i_tab[1] = dp2;
2643 	i_tab[2] = ip1;
2644 	if (ip2) {
2645 		*num_inodes = 4;
2646 		i_tab[3] = ip2;
2647 	} else {
2648 		*num_inodes = 3;
2649 		i_tab[3] = NULL;
2650 	}
2651 
2652 	/*
2653 	 * Sort the elements via bubble sort.  (Remember, there are at
2654 	 * most 4 elements to sort, so this is adequate.)
2655 	 */
2656 	for (i = 0; i < *num_inodes; i++) {
2657 		for (j = 1; j < *num_inodes; j++) {
2658 			if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2659 				temp = i_tab[j];
2660 				i_tab[j] = i_tab[j-1];
2661 				i_tab[j-1] = temp;
2662 			}
2663 		}
2664 	}
2665 }
2666 
2667 /*
2668  * xfs_rename
2669  */
2670 int
2671 xfs_rename(
2672 	xfs_inode_t	*src_dp,
2673 	struct xfs_name	*src_name,
2674 	xfs_inode_t	*src_ip,
2675 	xfs_inode_t	*target_dp,
2676 	struct xfs_name	*target_name,
2677 	xfs_inode_t	*target_ip)
2678 {
2679 	xfs_trans_t	*tp = NULL;
2680 	xfs_mount_t	*mp = src_dp->i_mount;
2681 	int		new_parent;		/* moving to a new dir */
2682 	int		src_is_directory;	/* src_name is a directory */
2683 	int		error;
2684 	xfs_bmap_free_t free_list;
2685 	xfs_fsblock_t   first_block;
2686 	int		cancel_flags;
2687 	int		committed;
2688 	xfs_inode_t	*inodes[4];
2689 	int		spaceres;
2690 	int		num_inodes;
2691 
2692 	trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2693 
2694 	new_parent = (src_dp != target_dp);
2695 	src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2696 
2697 	xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2698 				inodes, &num_inodes);
2699 
2700 	xfs_bmap_init(&free_list, &first_block);
2701 	tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2702 	cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2703 	spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2704 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2705 	if (error == ENOSPC) {
2706 		spaceres = 0;
2707 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2708 	}
2709 	if (error) {
2710 		xfs_trans_cancel(tp, 0);
2711 		goto std_return;
2712 	}
2713 
2714 	/*
2715 	 * Attach the dquots to the inodes
2716 	 */
2717 	error = xfs_qm_vop_rename_dqattach(inodes);
2718 	if (error) {
2719 		xfs_trans_cancel(tp, cancel_flags);
2720 		goto std_return;
2721 	}
2722 
2723 	/*
2724 	 * Lock all the participating inodes. Depending upon whether
2725 	 * the target_name exists in the target directory, and
2726 	 * whether the target directory is the same as the source
2727 	 * directory, we can lock from 2 to 4 inodes.
2728 	 */
2729 	xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2730 
2731 	/*
2732 	 * Join all the inodes to the transaction. From this point on,
2733 	 * we can rely on either trans_commit or trans_cancel to unlock
2734 	 * them.
2735 	 */
2736 	xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2737 	if (new_parent)
2738 		xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2739 	xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2740 	if (target_ip)
2741 		xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2742 
2743 	/*
2744 	 * If we are using project inheritance, we only allow renames
2745 	 * into our tree when the project IDs are the same; else the
2746 	 * tree quota mechanism would be circumvented.
2747 	 */
2748 	if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2749 		     (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2750 		error = XFS_ERROR(EXDEV);
2751 		goto error_return;
2752 	}
2753 
2754 	/*
2755 	 * Set up the target.
2756 	 */
2757 	if (target_ip == NULL) {
2758 		/*
2759 		 * If there's no space reservation, check the entry will
2760 		 * fit before actually inserting it.
2761 		 */
2762 		error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2763 		if (error)
2764 			goto error_return;
2765 		/*
2766 		 * If target does not exist and the rename crosses
2767 		 * directories, adjust the target directory link count
2768 		 * to account for the ".." reference from the new entry.
2769 		 */
2770 		error = xfs_dir_createname(tp, target_dp, target_name,
2771 						src_ip->i_ino, &first_block,
2772 						&free_list, spaceres);
2773 		if (error == ENOSPC)
2774 			goto error_return;
2775 		if (error)
2776 			goto abort_return;
2777 
2778 		xfs_trans_ichgtime(tp, target_dp,
2779 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2780 
2781 		if (new_parent && src_is_directory) {
2782 			error = xfs_bumplink(tp, target_dp);
2783 			if (error)
2784 				goto abort_return;
2785 		}
2786 	} else { /* target_ip != NULL */
2787 		/*
2788 		 * If target exists and it's a directory, check that both
2789 		 * target and source are directories and that target can be
2790 		 * destroyed, or that neither is a directory.
2791 		 */
2792 		if (S_ISDIR(target_ip->i_d.di_mode)) {
2793 			/*
2794 			 * Make sure target dir is empty.
2795 			 */
2796 			if (!(xfs_dir_isempty(target_ip)) ||
2797 			    (target_ip->i_d.di_nlink > 2)) {
2798 				error = XFS_ERROR(EEXIST);
2799 				goto error_return;
2800 			}
2801 		}
2802 
2803 		/*
2804 		 * Link the source inode under the target name.
2805 		 * If the source inode is a directory and we are moving
2806 		 * it across directories, its ".." entry will be
2807 		 * inconsistent until we replace that down below.
2808 		 *
2809 		 * In case there is already an entry with the same
2810 		 * name at the destination directory, remove it first.
2811 		 */
2812 		error = xfs_dir_replace(tp, target_dp, target_name,
2813 					src_ip->i_ino,
2814 					&first_block, &free_list, spaceres);
2815 		if (error)
2816 			goto abort_return;
2817 
2818 		xfs_trans_ichgtime(tp, target_dp,
2819 					XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2820 
2821 		/*
2822 		 * Decrement the link count on the target since the target
2823 		 * dir no longer points to it.
2824 		 */
2825 		error = xfs_droplink(tp, target_ip);
2826 		if (error)
2827 			goto abort_return;
2828 
2829 		if (src_is_directory) {
2830 			/*
2831 			 * Drop the link from the old "." entry.
2832 			 */
2833 			error = xfs_droplink(tp, target_ip);
2834 			if (error)
2835 				goto abort_return;
2836 		}
2837 	} /* target_ip != NULL */
2838 
2839 	/*
2840 	 * Remove the source.
2841 	 */
2842 	if (new_parent && src_is_directory) {
2843 		/*
2844 		 * Rewrite the ".." entry to point to the new
2845 		 * directory.
2846 		 */
2847 		error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2848 					target_dp->i_ino,
2849 					&first_block, &free_list, spaceres);
2850 		ASSERT(error != EEXIST);
2851 		if (error)
2852 			goto abort_return;
2853 	}
2854 
2855 	/*
2856 	 * We always want to hit the ctime on the source inode.
2857 	 *
2858 	 * This isn't strictly required by the standards since the source
2859 	 * inode isn't really being changed, but old unix file systems did
2860 	 * it and some incremental backup programs won't work without it.
2861 	 */
2862 	xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2863 	xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2864 
2865 	/*
2866 	 * Adjust the link count on src_dp.  This is necessary when
2867 	 * renaming a directory, either within one parent when
2868 	 * the target existed, or across two parent directories.
2869 	 */
2870 	if (src_is_directory && (new_parent || target_ip != NULL)) {
2871 
2872 		/*
2873 		 * Decrement link count on src_directory since the
2874 		 * entry that's moved no longer points to it.
2875 		 */
2876 		error = xfs_droplink(tp, src_dp);
2877 		if (error)
2878 			goto abort_return;
2879 	}
2880 
2881 	error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2882 					&first_block, &free_list, spaceres);
2883 	if (error)
2884 		goto abort_return;
2885 
2886 	xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2887 	xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2888 	if (new_parent)
2889 		xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2890 
2891 	/*
2892 	 * If this is a synchronous mount, make sure that the
2893 	 * rename transaction goes to disk before returning to
2894 	 * the user.
2895 	 */
2896 	if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2897 		xfs_trans_set_sync(tp);
2898 	}
2899 
2900 	error = xfs_bmap_finish(&tp, &free_list, &committed);
2901 	if (error) {
2902 		xfs_bmap_cancel(&free_list);
2903 		xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2904 				 XFS_TRANS_ABORT));
2905 		goto std_return;
2906 	}
2907 
2908 	/*
2909 	 * trans_commit will unlock src_ip, target_ip & decrement
2910 	 * the vnode references.
2911 	 */
2912 	return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2913 
2914  abort_return:
2915 	cancel_flags |= XFS_TRANS_ABORT;
2916  error_return:
2917 	xfs_bmap_cancel(&free_list);
2918 	xfs_trans_cancel(tp, cancel_flags);
2919  std_return:
2920 	return error;
2921 }
2922 
2923 STATIC int
2924 xfs_iflush_cluster(
2925 	xfs_inode_t	*ip,
2926 	xfs_buf_t	*bp)
2927 {
2928 	xfs_mount_t		*mp = ip->i_mount;
2929 	struct xfs_perag	*pag;
2930 	unsigned long		first_index, mask;
2931 	unsigned long		inodes_per_cluster;
2932 	int			ilist_size;
2933 	xfs_inode_t		**ilist;
2934 	xfs_inode_t		*iq;
2935 	int			nr_found;
2936 	int			clcount = 0;
2937 	int			bufwasdelwri;
2938 	int			i;
2939 
2940 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2941 
2942 	inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
2943 	ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2944 	ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2945 	if (!ilist)
2946 		goto out_put;
2947 
2948 	mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
2949 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2950 	rcu_read_lock();
2951 	/* really need a gang lookup range call here */
2952 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2953 					first_index, inodes_per_cluster);
2954 	if (nr_found == 0)
2955 		goto out_free;
2956 
2957 	for (i = 0; i < nr_found; i++) {
2958 		iq = ilist[i];
2959 		if (iq == ip)
2960 			continue;
2961 
2962 		/*
2963 		 * because this is an RCU protected lookup, we could find a
2964 		 * recently freed or even reallocated inode during the lookup.
2965 		 * We need to check under the i_flags_lock for a valid inode
2966 		 * here. Skip it if it is not valid or the wrong inode.
2967 		 */
2968 		spin_lock(&ip->i_flags_lock);
2969 		if (!ip->i_ino ||
2970 		    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2971 			spin_unlock(&ip->i_flags_lock);
2972 			continue;
2973 		}
2974 		spin_unlock(&ip->i_flags_lock);
2975 
2976 		/*
2977 		 * Do an un-protected check to see if the inode is dirty and
2978 		 * is a candidate for flushing.  These checks will be repeated
2979 		 * later after the appropriate locks are acquired.
2980 		 */
2981 		if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2982 			continue;
2983 
2984 		/*
2985 		 * Try to get locks.  If any are unavailable or it is pinned,
2986 		 * then this inode cannot be flushed and is skipped.
2987 		 */
2988 
2989 		if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2990 			continue;
2991 		if (!xfs_iflock_nowait(iq)) {
2992 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
2993 			continue;
2994 		}
2995 		if (xfs_ipincount(iq)) {
2996 			xfs_ifunlock(iq);
2997 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
2998 			continue;
2999 		}
3000 
3001 		/*
3002 		 * arriving here means that this inode can be flushed.  First
3003 		 * re-check that it's dirty before flushing.
3004 		 */
3005 		if (!xfs_inode_clean(iq)) {
3006 			int	error;
3007 			error = xfs_iflush_int(iq, bp);
3008 			if (error) {
3009 				xfs_iunlock(iq, XFS_ILOCK_SHARED);
3010 				goto cluster_corrupt_out;
3011 			}
3012 			clcount++;
3013 		} else {
3014 			xfs_ifunlock(iq);
3015 		}
3016 		xfs_iunlock(iq, XFS_ILOCK_SHARED);
3017 	}
3018 
3019 	if (clcount) {
3020 		XFS_STATS_INC(xs_icluster_flushcnt);
3021 		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3022 	}
3023 
3024 out_free:
3025 	rcu_read_unlock();
3026 	kmem_free(ilist);
3027 out_put:
3028 	xfs_perag_put(pag);
3029 	return 0;
3030 
3031 
3032 cluster_corrupt_out:
3033 	/*
3034 	 * Corruption detected in the clustering loop.  Invalidate the
3035 	 * inode buffer and shut down the filesystem.
3036 	 */
3037 	rcu_read_unlock();
3038 	/*
3039 	 * Clean up the buffer.  If it was delwri, just release it --
3040 	 * brelse can handle it with no problems.  If not, shut down the
3041 	 * filesystem before releasing the buffer.
3042 	 */
3043 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3044 	if (bufwasdelwri)
3045 		xfs_buf_relse(bp);
3046 
3047 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3048 
3049 	if (!bufwasdelwri) {
3050 		/*
3051 		 * Just like incore_relse: if we have b_iodone functions,
3052 		 * mark the buffer as an error and call them.  Otherwise
3053 		 * mark it as stale and brelse.
3054 		 */
3055 		if (bp->b_iodone) {
3056 			XFS_BUF_UNDONE(bp);
3057 			xfs_buf_stale(bp);
3058 			xfs_buf_ioerror(bp, EIO);
3059 			xfs_buf_ioend(bp, 0);
3060 		} else {
3061 			xfs_buf_stale(bp);
3062 			xfs_buf_relse(bp);
3063 		}
3064 	}
3065 
3066 	/*
3067 	 * Unlocks the flush lock
3068 	 */
3069 	xfs_iflush_abort(iq, false);
3070 	kmem_free(ilist);
3071 	xfs_perag_put(pag);
3072 	return XFS_ERROR(EFSCORRUPTED);
3073 }
3074 
3075 /*
3076  * Flush dirty inode metadata into the backing buffer.
3077  *
3078  * The caller must have the inode lock and the inode flush lock held.  The
3079  * inode lock will still be held upon return to the caller, and the inode
3080  * flush lock will be released after the inode has reached the disk.
3081  *
3082  * The caller must write out the buffer returned in *bpp and release it.
3083  */
3084 int
3085 xfs_iflush(
3086 	struct xfs_inode	*ip,
3087 	struct xfs_buf		**bpp)
3088 {
3089 	struct xfs_mount	*mp = ip->i_mount;
3090 	struct xfs_buf		*bp;
3091 	struct xfs_dinode	*dip;
3092 	int			error;
3093 
3094 	XFS_STATS_INC(xs_iflush_count);
3095 
3096 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3097 	ASSERT(xfs_isiflocked(ip));
3098 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3099 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3100 
3101 	*bpp = NULL;
3102 
3103 	xfs_iunpin_wait(ip);
3104 
3105 	/*
3106 	 * For stale inodes we cannot rely on the backing buffer remaining
3107 	 * stale in cache for the remaining life of the stale inode and so
3108 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3109 	 * inodes below. We have to check this after ensuring the inode is
3110 	 * unpinned so that it is safe to reclaim the stale inode after the
3111 	 * flush call.
3112 	 */
3113 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
3114 		xfs_ifunlock(ip);
3115 		return 0;
3116 	}
3117 
3118 	/*
3119 	 * This may have been unpinned because the filesystem is shutting
3120 	 * down forcibly. If that's the case we must not write this inode
3121 	 * to disk, because the log record didn't make it to disk.
3122 	 *
3123 	 * We also have to remove the log item from the AIL in this case,
3124 	 * as we wait for an empty AIL as part of the unmount process.
3125 	 */
3126 	if (XFS_FORCED_SHUTDOWN(mp)) {
3127 		error = XFS_ERROR(EIO);
3128 		goto abort_out;
3129 	}
3130 
3131 	/*
3132 	 * Get the buffer containing the on-disk inode.
3133 	 */
3134 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3135 			       0);
3136 	if (error || !bp) {
3137 		xfs_ifunlock(ip);
3138 		return error;
3139 	}
3140 
3141 	/*
3142 	 * First flush out the inode that xfs_iflush was called with.
3143 	 */
3144 	error = xfs_iflush_int(ip, bp);
3145 	if (error)
3146 		goto corrupt_out;
3147 
3148 	/*
3149 	 * If the buffer is pinned then push on the log now so we won't
3150 	 * get stuck waiting in the write for too long.
3151 	 */
3152 	if (xfs_buf_ispinned(bp))
3153 		xfs_log_force(mp, 0);
3154 
3155 	/*
3156 	 * inode clustering:
3157 	 * see if other inodes can be gathered into this write
3158 	 */
3159 	error = xfs_iflush_cluster(ip, bp);
3160 	if (error)
3161 		goto cluster_corrupt_out;
3162 
3163 	*bpp = bp;
3164 	return 0;
3165 
3166 corrupt_out:
3167 	xfs_buf_relse(bp);
3168 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3169 cluster_corrupt_out:
3170 	error = XFS_ERROR(EFSCORRUPTED);
3171 abort_out:
3172 	/*
3173 	 * Unlocks the flush lock
3174 	 */
3175 	xfs_iflush_abort(ip, false);
3176 	return error;
3177 }
3178 
3179 STATIC int
3180 xfs_iflush_int(
3181 	struct xfs_inode	*ip,
3182 	struct xfs_buf		*bp)
3183 {
3184 	struct xfs_inode_log_item *iip = ip->i_itemp;
3185 	struct xfs_dinode	*dip;
3186 	struct xfs_mount	*mp = ip->i_mount;
3187 
3188 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3189 	ASSERT(xfs_isiflocked(ip));
3190 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3191 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3192 	ASSERT(iip != NULL && iip->ili_fields != 0);
3193 	ASSERT(ip->i_d.di_version > 1);
3194 
3195 	/* set *dip = inode's place in the buffer */
3196 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3197 
3198 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3199 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3200 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3201 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3202 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3203 		goto corrupt_out;
3204 	}
3205 	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3206 				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3207 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3208 			"%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3209 			__func__, ip->i_ino, ip, ip->i_d.di_magic);
3210 		goto corrupt_out;
3211 	}
3212 	if (S_ISREG(ip->i_d.di_mode)) {
3213 		if (XFS_TEST_ERROR(
3214 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3215 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3216 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3217 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3218 				"%s: Bad regular inode %Lu, ptr 0x%p",
3219 				__func__, ip->i_ino, ip);
3220 			goto corrupt_out;
3221 		}
3222 	} else if (S_ISDIR(ip->i_d.di_mode)) {
3223 		if (XFS_TEST_ERROR(
3224 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3225 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3226 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3227 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3228 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3229 				"%s: Bad directory inode %Lu, ptr 0x%p",
3230 				__func__, ip->i_ino, ip);
3231 			goto corrupt_out;
3232 		}
3233 	}
3234 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3235 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3236 				XFS_RANDOM_IFLUSH_5)) {
3237 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3238 			"%s: detected corrupt incore inode %Lu, "
3239 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
3240 			__func__, ip->i_ino,
3241 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3242 			ip->i_d.di_nblocks, ip);
3243 		goto corrupt_out;
3244 	}
3245 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3246 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3247 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3248 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3249 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3250 		goto corrupt_out;
3251 	}
3252 
3253 	/*
3254 	 * Inode item log recovery for v2 inodes are dependent on the
3255 	 * di_flushiter count for correct sequencing. We bump the flush
3256 	 * iteration count so we can detect flushes which postdate a log record
3257 	 * during recovery. This is redundant as we now log every change and
3258 	 * hence this can't happen but we need to still do it to ensure
3259 	 * backwards compatibility with old kernels that predate logging all
3260 	 * inode changes.
3261 	 */
3262 	if (ip->i_d.di_version < 3)
3263 		ip->i_d.di_flushiter++;
3264 
3265 	/*
3266 	 * Copy the dirty parts of the inode into the on-disk
3267 	 * inode.  We always copy out the core of the inode,
3268 	 * because if the inode is dirty at all the core must
3269 	 * be.
3270 	 */
3271 	xfs_dinode_to_disk(dip, &ip->i_d);
3272 
3273 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3274 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3275 		ip->i_d.di_flushiter = 0;
3276 
3277 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3278 	if (XFS_IFORK_Q(ip))
3279 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3280 	xfs_inobp_check(mp, bp);
3281 
3282 	/*
3283 	 * We've recorded everything logged in the inode, so we'd like to clear
3284 	 * the ili_fields bits so we don't log and flush things unnecessarily.
3285 	 * However, we can't stop logging all this information until the data
3286 	 * we've copied into the disk buffer is written to disk.  If we did we
3287 	 * might overwrite the copy of the inode in the log with all the data
3288 	 * after re-logging only part of it, and in the face of a crash we
3289 	 * wouldn't have all the data we need to recover.
3290 	 *
3291 	 * What we do is move the bits to the ili_last_fields field.  When
3292 	 * logging the inode, these bits are moved back to the ili_fields field.
3293 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3294 	 * know that the information those bits represent is permanently on
3295 	 * disk.  As long as the flush completes before the inode is logged
3296 	 * again, then both ili_fields and ili_last_fields will be cleared.
3297 	 *
3298 	 * We can play with the ili_fields bits here, because the inode lock
3299 	 * must be held exclusively in order to set bits there and the flush
3300 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
3301 	 * done routine can tell whether or not to look in the AIL.  Also, store
3302 	 * the current LSN of the inode so that we can tell whether the item has
3303 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
3304 	 * need the AIL lock, because it is a 64 bit value that cannot be read
3305 	 * atomically.
3306 	 */
3307 	iip->ili_last_fields = iip->ili_fields;
3308 	iip->ili_fields = 0;
3309 	iip->ili_logged = 1;
3310 
3311 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3312 				&iip->ili_item.li_lsn);
3313 
3314 	/*
3315 	 * Attach the function xfs_iflush_done to the inode's
3316 	 * buffer.  This will remove the inode from the AIL
3317 	 * and unlock the inode's flush lock when the inode is
3318 	 * completely written to disk.
3319 	 */
3320 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3321 
3322 	/* update the lsn in the on disk inode if required */
3323 	if (ip->i_d.di_version == 3)
3324 		dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3325 
3326 	/* generate the checksum. */
3327 	xfs_dinode_calc_crc(mp, dip);
3328 
3329 	ASSERT(bp->b_fspriv != NULL);
3330 	ASSERT(bp->b_iodone != NULL);
3331 	return 0;
3332 
3333 corrupt_out:
3334 	return XFS_ERROR(EFSCORRUPTED);
3335 }
3336