xref: /openbmc/linux/fs/xfs/xfs_inode.c (revision f35e839a)
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_types.h"
23 #include "xfs_log.h"
24 #include "xfs_inum.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
27 #include "xfs_sb.h"
28 #include "xfs_ag.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_btree.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_bmap.h"
42 #include "xfs_error.h"
43 #include "xfs_utils.h"
44 #include "xfs_quota.h"
45 #include "xfs_filestream.h"
46 #include "xfs_vnodeops.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
50 
51 kmem_zone_t *xfs_ifork_zone;
52 kmem_zone_t *xfs_inode_zone;
53 
54 /*
55  * Used in xfs_itruncate_extents().  This is the maximum number of extents
56  * freed from a file in a single transaction.
57  */
58 #define	XFS_ITRUNC_MAX_EXTENTS	2
59 
60 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
61 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
62 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
63 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
64 
65 /*
66  * helper function to extract extent size hint from inode
67  */
68 xfs_extlen_t
69 xfs_get_extsz_hint(
70 	struct xfs_inode	*ip)
71 {
72 	if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
73 		return ip->i_d.di_extsize;
74 	if (XFS_IS_REALTIME_INODE(ip))
75 		return ip->i_mount->m_sb.sb_rextsize;
76 	return 0;
77 }
78 
79 /*
80  * This is a wrapper routine around the xfs_ilock() routine used to centralize
81  * some grungy code.  It is used in places that wish to lock the inode solely
82  * for reading the extents.  The reason these places can't just call
83  * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
84  * extents from disk for a file in b-tree format.  If the inode is in b-tree
85  * format, then we need to lock the inode exclusively until the extents are read
86  * in.  Locking it exclusively all the time would limit our parallelism
87  * unnecessarily, though.  What we do instead is check to see if the extents
88  * have been read in yet, and only lock the inode exclusively if they have not.
89  *
90  * The function returns a value which should be given to the corresponding
91  * xfs_iunlock_map_shared().  This value is the mode in which the lock was
92  * actually taken.
93  */
94 uint
95 xfs_ilock_map_shared(
96 	xfs_inode_t	*ip)
97 {
98 	uint	lock_mode;
99 
100 	if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
101 	    ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
102 		lock_mode = XFS_ILOCK_EXCL;
103 	} else {
104 		lock_mode = XFS_ILOCK_SHARED;
105 	}
106 
107 	xfs_ilock(ip, lock_mode);
108 
109 	return lock_mode;
110 }
111 
112 /*
113  * This is simply the unlock routine to go with xfs_ilock_map_shared().
114  * All it does is call xfs_iunlock() with the given lock_mode.
115  */
116 void
117 xfs_iunlock_map_shared(
118 	xfs_inode_t	*ip,
119 	unsigned int	lock_mode)
120 {
121 	xfs_iunlock(ip, lock_mode);
122 }
123 
124 /*
125  * The xfs inode contains 2 locks: a multi-reader lock called the
126  * i_iolock and a multi-reader lock called the i_lock.  This routine
127  * allows either or both of the locks to be obtained.
128  *
129  * The 2 locks should always be ordered so that the IO lock is
130  * obtained first in order to prevent deadlock.
131  *
132  * ip -- the inode being locked
133  * lock_flags -- this parameter indicates the inode's locks
134  *       to be locked.  It can be:
135  *		XFS_IOLOCK_SHARED,
136  *		XFS_IOLOCK_EXCL,
137  *		XFS_ILOCK_SHARED,
138  *		XFS_ILOCK_EXCL,
139  *		XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
140  *		XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
141  *		XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
142  *		XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
143  */
144 void
145 xfs_ilock(
146 	xfs_inode_t		*ip,
147 	uint			lock_flags)
148 {
149 	trace_xfs_ilock(ip, lock_flags, _RET_IP_);
150 
151 	/*
152 	 * You can't set both SHARED and EXCL for the same lock,
153 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
154 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
155 	 */
156 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
157 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
158 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
159 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
160 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
161 
162 	if (lock_flags & XFS_IOLOCK_EXCL)
163 		mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
164 	else if (lock_flags & XFS_IOLOCK_SHARED)
165 		mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
166 
167 	if (lock_flags & XFS_ILOCK_EXCL)
168 		mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
169 	else if (lock_flags & XFS_ILOCK_SHARED)
170 		mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
171 }
172 
173 /*
174  * This is just like xfs_ilock(), except that the caller
175  * is guaranteed not to sleep.  It returns 1 if it gets
176  * the requested locks and 0 otherwise.  If the IO lock is
177  * obtained but the inode lock cannot be, then the IO lock
178  * is dropped before returning.
179  *
180  * ip -- the inode being locked
181  * lock_flags -- this parameter indicates the inode's locks to be
182  *       to be locked.  See the comment for xfs_ilock() for a list
183  *	 of valid values.
184  */
185 int
186 xfs_ilock_nowait(
187 	xfs_inode_t		*ip,
188 	uint			lock_flags)
189 {
190 	trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
191 
192 	/*
193 	 * You can't set both SHARED and EXCL for the same lock,
194 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
195 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
196 	 */
197 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
198 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
199 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
200 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
201 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
202 
203 	if (lock_flags & XFS_IOLOCK_EXCL) {
204 		if (!mrtryupdate(&ip->i_iolock))
205 			goto out;
206 	} else if (lock_flags & XFS_IOLOCK_SHARED) {
207 		if (!mrtryaccess(&ip->i_iolock))
208 			goto out;
209 	}
210 	if (lock_flags & XFS_ILOCK_EXCL) {
211 		if (!mrtryupdate(&ip->i_lock))
212 			goto out_undo_iolock;
213 	} else if (lock_flags & XFS_ILOCK_SHARED) {
214 		if (!mrtryaccess(&ip->i_lock))
215 			goto out_undo_iolock;
216 	}
217 	return 1;
218 
219  out_undo_iolock:
220 	if (lock_flags & XFS_IOLOCK_EXCL)
221 		mrunlock_excl(&ip->i_iolock);
222 	else if (lock_flags & XFS_IOLOCK_SHARED)
223 		mrunlock_shared(&ip->i_iolock);
224  out:
225 	return 0;
226 }
227 
228 /*
229  * xfs_iunlock() is used to drop the inode locks acquired with
230  * xfs_ilock() and xfs_ilock_nowait().  The caller must pass
231  * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
232  * that we know which locks to drop.
233  *
234  * ip -- the inode being unlocked
235  * lock_flags -- this parameter indicates the inode's locks to be
236  *       to be unlocked.  See the comment for xfs_ilock() for a list
237  *	 of valid values for this parameter.
238  *
239  */
240 void
241 xfs_iunlock(
242 	xfs_inode_t		*ip,
243 	uint			lock_flags)
244 {
245 	/*
246 	 * You can't set both SHARED and EXCL for the same lock,
247 	 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
248 	 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
249 	 */
250 	ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
251 	       (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
252 	ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
253 	       (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
254 	ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
255 	ASSERT(lock_flags != 0);
256 
257 	if (lock_flags & XFS_IOLOCK_EXCL)
258 		mrunlock_excl(&ip->i_iolock);
259 	else if (lock_flags & XFS_IOLOCK_SHARED)
260 		mrunlock_shared(&ip->i_iolock);
261 
262 	if (lock_flags & XFS_ILOCK_EXCL)
263 		mrunlock_excl(&ip->i_lock);
264 	else if (lock_flags & XFS_ILOCK_SHARED)
265 		mrunlock_shared(&ip->i_lock);
266 
267 	trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
268 }
269 
270 /*
271  * give up write locks.  the i/o lock cannot be held nested
272  * if it is being demoted.
273  */
274 void
275 xfs_ilock_demote(
276 	xfs_inode_t		*ip,
277 	uint			lock_flags)
278 {
279 	ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
280 	ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
281 
282 	if (lock_flags & XFS_ILOCK_EXCL)
283 		mrdemote(&ip->i_lock);
284 	if (lock_flags & XFS_IOLOCK_EXCL)
285 		mrdemote(&ip->i_iolock);
286 
287 	trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
288 }
289 
290 #if defined(DEBUG) || defined(XFS_WARN)
291 int
292 xfs_isilocked(
293 	xfs_inode_t		*ip,
294 	uint			lock_flags)
295 {
296 	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
297 		if (!(lock_flags & XFS_ILOCK_SHARED))
298 			return !!ip->i_lock.mr_writer;
299 		return rwsem_is_locked(&ip->i_lock.mr_lock);
300 	}
301 
302 	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
303 		if (!(lock_flags & XFS_IOLOCK_SHARED))
304 			return !!ip->i_iolock.mr_writer;
305 		return rwsem_is_locked(&ip->i_iolock.mr_lock);
306 	}
307 
308 	ASSERT(0);
309 	return 0;
310 }
311 #endif
312 
313 void
314 __xfs_iflock(
315 	struct xfs_inode	*ip)
316 {
317 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
318 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
319 
320 	do {
321 		prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
322 		if (xfs_isiflocked(ip))
323 			io_schedule();
324 	} while (!xfs_iflock_nowait(ip));
325 
326 	finish_wait(wq, &wait.wait);
327 }
328 
329 #ifdef DEBUG
330 /*
331  * Make sure that the extents in the given memory buffer
332  * are valid.
333  */
334 STATIC void
335 xfs_validate_extents(
336 	xfs_ifork_t		*ifp,
337 	int			nrecs,
338 	xfs_exntfmt_t		fmt)
339 {
340 	xfs_bmbt_irec_t		irec;
341 	xfs_bmbt_rec_host_t	rec;
342 	int			i;
343 
344 	for (i = 0; i < nrecs; i++) {
345 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
346 		rec.l0 = get_unaligned(&ep->l0);
347 		rec.l1 = get_unaligned(&ep->l1);
348 		xfs_bmbt_get_all(&rec, &irec);
349 		if (fmt == XFS_EXTFMT_NOSTATE)
350 			ASSERT(irec.br_state == XFS_EXT_NORM);
351 	}
352 }
353 #else /* DEBUG */
354 #define xfs_validate_extents(ifp, nrecs, fmt)
355 #endif /* DEBUG */
356 
357 /*
358  * Check that none of the inode's in the buffer have a next
359  * unlinked field of 0.
360  */
361 #if defined(DEBUG)
362 void
363 xfs_inobp_check(
364 	xfs_mount_t	*mp,
365 	xfs_buf_t	*bp)
366 {
367 	int		i;
368 	int		j;
369 	xfs_dinode_t	*dip;
370 
371 	j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
372 
373 	for (i = 0; i < j; i++) {
374 		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
375 					i * mp->m_sb.sb_inodesize);
376 		if (!dip->di_next_unlinked)  {
377 			xfs_alert(mp,
378 	"Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
379 				bp);
380 			ASSERT(dip->di_next_unlinked);
381 		}
382 	}
383 }
384 #endif
385 
386 static void
387 xfs_inode_buf_verify(
388 	struct xfs_buf	*bp)
389 {
390 	struct xfs_mount *mp = bp->b_target->bt_mount;
391 	int		i;
392 	int		ni;
393 
394 	/*
395 	 * Validate the magic number and version of every inode in the buffer
396 	 */
397 	ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
398 	for (i = 0; i < ni; i++) {
399 		int		di_ok;
400 		xfs_dinode_t	*dip;
401 
402 		dip = (struct xfs_dinode *)xfs_buf_offset(bp,
403 					(i << mp->m_sb.sb_inodelog));
404 		di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
405 			    XFS_DINODE_GOOD_VERSION(dip->di_version);
406 		if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
407 						XFS_ERRTAG_ITOBP_INOTOBP,
408 						XFS_RANDOM_ITOBP_INOTOBP))) {
409 			xfs_buf_ioerror(bp, EFSCORRUPTED);
410 			XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH,
411 					     mp, dip);
412 #ifdef DEBUG
413 			xfs_emerg(mp,
414 				"bad inode magic/vsn daddr %lld #%d (magic=%x)",
415 				(unsigned long long)bp->b_bn, i,
416 				be16_to_cpu(dip->di_magic));
417 			ASSERT(0);
418 #endif
419 		}
420 	}
421 	xfs_inobp_check(mp, bp);
422 }
423 
424 
425 static void
426 xfs_inode_buf_read_verify(
427 	struct xfs_buf	*bp)
428 {
429 	xfs_inode_buf_verify(bp);
430 }
431 
432 static void
433 xfs_inode_buf_write_verify(
434 	struct xfs_buf	*bp)
435 {
436 	xfs_inode_buf_verify(bp);
437 }
438 
439 const struct xfs_buf_ops xfs_inode_buf_ops = {
440 	.verify_read = xfs_inode_buf_read_verify,
441 	.verify_write = xfs_inode_buf_write_verify,
442 };
443 
444 
445 /*
446  * This routine is called to map an inode to the buffer containing the on-disk
447  * version of the inode.  It returns a pointer to the buffer containing the
448  * on-disk inode in the bpp parameter, and in the dipp parameter it returns a
449  * pointer to the on-disk inode within that buffer.
450  *
451  * If a non-zero error is returned, then the contents of bpp and dipp are
452  * undefined.
453  */
454 int
455 xfs_imap_to_bp(
456 	struct xfs_mount	*mp,
457 	struct xfs_trans	*tp,
458 	struct xfs_imap		*imap,
459 	struct xfs_dinode       **dipp,
460 	struct xfs_buf		**bpp,
461 	uint			buf_flags,
462 	uint			iget_flags)
463 {
464 	struct xfs_buf		*bp;
465 	int			error;
466 
467 	buf_flags |= XBF_UNMAPPED;
468 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
469 				   (int)imap->im_len, buf_flags, &bp,
470 				   &xfs_inode_buf_ops);
471 	if (error) {
472 		if (error == EAGAIN) {
473 			ASSERT(buf_flags & XBF_TRYLOCK);
474 			return error;
475 		}
476 
477 		if (error == EFSCORRUPTED &&
478 		    (iget_flags & XFS_IGET_UNTRUSTED))
479 			return XFS_ERROR(EINVAL);
480 
481 		xfs_warn(mp, "%s: xfs_trans_read_buf() returned error %d.",
482 			__func__, error);
483 		return error;
484 	}
485 
486 	*bpp = bp;
487 	*dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset);
488 	return 0;
489 }
490 
491 /*
492  * Move inode type and inode format specific information from the
493  * on-disk inode to the in-core inode.  For fifos, devs, and sockets
494  * this means set if_rdev to the proper value.  For files, directories,
495  * and symlinks this means to bring in the in-line data or extent
496  * pointers.  For a file in B-tree format, only the root is immediately
497  * brought in-core.  The rest will be in-lined in if_extents when it
498  * is first referenced (see xfs_iread_extents()).
499  */
500 STATIC int
501 xfs_iformat(
502 	xfs_inode_t		*ip,
503 	xfs_dinode_t		*dip)
504 {
505 	xfs_attr_shortform_t	*atp;
506 	int			size;
507 	int			error = 0;
508 	xfs_fsize_t             di_size;
509 
510 	if (unlikely(be32_to_cpu(dip->di_nextents) +
511 		     be16_to_cpu(dip->di_anextents) >
512 		     be64_to_cpu(dip->di_nblocks))) {
513 		xfs_warn(ip->i_mount,
514 			"corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
515 			(unsigned long long)ip->i_ino,
516 			(int)(be32_to_cpu(dip->di_nextents) +
517 			      be16_to_cpu(dip->di_anextents)),
518 			(unsigned long long)
519 				be64_to_cpu(dip->di_nblocks));
520 		XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
521 				     ip->i_mount, dip);
522 		return XFS_ERROR(EFSCORRUPTED);
523 	}
524 
525 	if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
526 		xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
527 			(unsigned long long)ip->i_ino,
528 			dip->di_forkoff);
529 		XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
530 				     ip->i_mount, dip);
531 		return XFS_ERROR(EFSCORRUPTED);
532 	}
533 
534 	if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
535 		     !ip->i_mount->m_rtdev_targp)) {
536 		xfs_warn(ip->i_mount,
537 			"corrupt dinode %Lu, has realtime flag set.",
538 			ip->i_ino);
539 		XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
540 				     XFS_ERRLEVEL_LOW, ip->i_mount, dip);
541 		return XFS_ERROR(EFSCORRUPTED);
542 	}
543 
544 	switch (ip->i_d.di_mode & S_IFMT) {
545 	case S_IFIFO:
546 	case S_IFCHR:
547 	case S_IFBLK:
548 	case S_IFSOCK:
549 		if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
550 			XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
551 					      ip->i_mount, dip);
552 			return XFS_ERROR(EFSCORRUPTED);
553 		}
554 		ip->i_d.di_size = 0;
555 		ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
556 		break;
557 
558 	case S_IFREG:
559 	case S_IFLNK:
560 	case S_IFDIR:
561 		switch (dip->di_format) {
562 		case XFS_DINODE_FMT_LOCAL:
563 			/*
564 			 * no local regular files yet
565 			 */
566 			if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
567 				xfs_warn(ip->i_mount,
568 			"corrupt inode %Lu (local format for regular file).",
569 					(unsigned long long) ip->i_ino);
570 				XFS_CORRUPTION_ERROR("xfs_iformat(4)",
571 						     XFS_ERRLEVEL_LOW,
572 						     ip->i_mount, dip);
573 				return XFS_ERROR(EFSCORRUPTED);
574 			}
575 
576 			di_size = be64_to_cpu(dip->di_size);
577 			if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
578 				xfs_warn(ip->i_mount,
579 			"corrupt inode %Lu (bad size %Ld for local inode).",
580 					(unsigned long long) ip->i_ino,
581 					(long long) di_size);
582 				XFS_CORRUPTION_ERROR("xfs_iformat(5)",
583 						     XFS_ERRLEVEL_LOW,
584 						     ip->i_mount, dip);
585 				return XFS_ERROR(EFSCORRUPTED);
586 			}
587 
588 			size = (int)di_size;
589 			error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
590 			break;
591 		case XFS_DINODE_FMT_EXTENTS:
592 			error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
593 			break;
594 		case XFS_DINODE_FMT_BTREE:
595 			error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
596 			break;
597 		default:
598 			XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
599 					 ip->i_mount);
600 			return XFS_ERROR(EFSCORRUPTED);
601 		}
602 		break;
603 
604 	default:
605 		XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
606 		return XFS_ERROR(EFSCORRUPTED);
607 	}
608 	if (error) {
609 		return error;
610 	}
611 	if (!XFS_DFORK_Q(dip))
612 		return 0;
613 
614 	ASSERT(ip->i_afp == NULL);
615 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
616 
617 	switch (dip->di_aformat) {
618 	case XFS_DINODE_FMT_LOCAL:
619 		atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
620 		size = be16_to_cpu(atp->hdr.totsize);
621 
622 		if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
623 			xfs_warn(ip->i_mount,
624 				"corrupt inode %Lu (bad attr fork size %Ld).",
625 				(unsigned long long) ip->i_ino,
626 				(long long) size);
627 			XFS_CORRUPTION_ERROR("xfs_iformat(8)",
628 					     XFS_ERRLEVEL_LOW,
629 					     ip->i_mount, dip);
630 			return XFS_ERROR(EFSCORRUPTED);
631 		}
632 
633 		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
634 		break;
635 	case XFS_DINODE_FMT_EXTENTS:
636 		error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
637 		break;
638 	case XFS_DINODE_FMT_BTREE:
639 		error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
640 		break;
641 	default:
642 		error = XFS_ERROR(EFSCORRUPTED);
643 		break;
644 	}
645 	if (error) {
646 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
647 		ip->i_afp = NULL;
648 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
649 	}
650 	return error;
651 }
652 
653 /*
654  * The file is in-lined in the on-disk inode.
655  * If it fits into if_inline_data, then copy
656  * it there, otherwise allocate a buffer for it
657  * and copy the data there.  Either way, set
658  * if_data to point at the data.
659  * If we allocate a buffer for the data, make
660  * sure that its size is a multiple of 4 and
661  * record the real size in i_real_bytes.
662  */
663 STATIC int
664 xfs_iformat_local(
665 	xfs_inode_t	*ip,
666 	xfs_dinode_t	*dip,
667 	int		whichfork,
668 	int		size)
669 {
670 	xfs_ifork_t	*ifp;
671 	int		real_size;
672 
673 	/*
674 	 * If the size is unreasonable, then something
675 	 * is wrong and we just bail out rather than crash in
676 	 * kmem_alloc() or memcpy() below.
677 	 */
678 	if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
679 		xfs_warn(ip->i_mount,
680 	"corrupt inode %Lu (bad size %d for local fork, size = %d).",
681 			(unsigned long long) ip->i_ino, size,
682 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
683 		XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
684 				     ip->i_mount, dip);
685 		return XFS_ERROR(EFSCORRUPTED);
686 	}
687 	ifp = XFS_IFORK_PTR(ip, whichfork);
688 	real_size = 0;
689 	if (size == 0)
690 		ifp->if_u1.if_data = NULL;
691 	else if (size <= sizeof(ifp->if_u2.if_inline_data))
692 		ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
693 	else {
694 		real_size = roundup(size, 4);
695 		ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
696 	}
697 	ifp->if_bytes = size;
698 	ifp->if_real_bytes = real_size;
699 	if (size)
700 		memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
701 	ifp->if_flags &= ~XFS_IFEXTENTS;
702 	ifp->if_flags |= XFS_IFINLINE;
703 	return 0;
704 }
705 
706 /*
707  * The file consists of a set of extents all
708  * of which fit into the on-disk inode.
709  * If there are few enough extents to fit into
710  * the if_inline_ext, then copy them there.
711  * Otherwise allocate a buffer for them and copy
712  * them into it.  Either way, set if_extents
713  * to point at the extents.
714  */
715 STATIC int
716 xfs_iformat_extents(
717 	xfs_inode_t	*ip,
718 	xfs_dinode_t	*dip,
719 	int		whichfork)
720 {
721 	xfs_bmbt_rec_t	*dp;
722 	xfs_ifork_t	*ifp;
723 	int		nex;
724 	int		size;
725 	int		i;
726 
727 	ifp = XFS_IFORK_PTR(ip, whichfork);
728 	nex = XFS_DFORK_NEXTENTS(dip, whichfork);
729 	size = nex * (uint)sizeof(xfs_bmbt_rec_t);
730 
731 	/*
732 	 * If the number of extents is unreasonable, then something
733 	 * is wrong and we just bail out rather than crash in
734 	 * kmem_alloc() or memcpy() below.
735 	 */
736 	if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
737 		xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
738 			(unsigned long long) ip->i_ino, nex);
739 		XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
740 				     ip->i_mount, dip);
741 		return XFS_ERROR(EFSCORRUPTED);
742 	}
743 
744 	ifp->if_real_bytes = 0;
745 	if (nex == 0)
746 		ifp->if_u1.if_extents = NULL;
747 	else if (nex <= XFS_INLINE_EXTS)
748 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
749 	else
750 		xfs_iext_add(ifp, 0, nex);
751 
752 	ifp->if_bytes = size;
753 	if (size) {
754 		dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
755 		xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
756 		for (i = 0; i < nex; i++, dp++) {
757 			xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
758 			ep->l0 = get_unaligned_be64(&dp->l0);
759 			ep->l1 = get_unaligned_be64(&dp->l1);
760 		}
761 		XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
762 		if (whichfork != XFS_DATA_FORK ||
763 			XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
764 				if (unlikely(xfs_check_nostate_extents(
765 				    ifp, 0, nex))) {
766 					XFS_ERROR_REPORT("xfs_iformat_extents(2)",
767 							 XFS_ERRLEVEL_LOW,
768 							 ip->i_mount);
769 					return XFS_ERROR(EFSCORRUPTED);
770 				}
771 	}
772 	ifp->if_flags |= XFS_IFEXTENTS;
773 	return 0;
774 }
775 
776 /*
777  * The file has too many extents to fit into
778  * the inode, so they are in B-tree format.
779  * Allocate a buffer for the root of the B-tree
780  * and copy the root into it.  The i_extents
781  * field will remain NULL until all of the
782  * extents are read in (when they are needed).
783  */
784 STATIC int
785 xfs_iformat_btree(
786 	xfs_inode_t		*ip,
787 	xfs_dinode_t		*dip,
788 	int			whichfork)
789 {
790 	struct xfs_mount	*mp = ip->i_mount;
791 	xfs_bmdr_block_t	*dfp;
792 	xfs_ifork_t		*ifp;
793 	/* REFERENCED */
794 	int			nrecs;
795 	int			size;
796 
797 	ifp = XFS_IFORK_PTR(ip, whichfork);
798 	dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
799 	size = XFS_BMAP_BROOT_SPACE(mp, dfp);
800 	nrecs = be16_to_cpu(dfp->bb_numrecs);
801 
802 	/*
803 	 * blow out if -- fork has less extents than can fit in
804 	 * fork (fork shouldn't be a btree format), root btree
805 	 * block has more records than can fit into the fork,
806 	 * or the number of extents is greater than the number of
807 	 * blocks.
808 	 */
809 	if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
810 					XFS_IFORK_MAXEXT(ip, whichfork) ||
811 		     XFS_BMDR_SPACE_CALC(nrecs) >
812 					XFS_DFORK_SIZE(dip, mp, whichfork) ||
813 		     XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
814 		xfs_warn(mp, "corrupt inode %Lu (btree).",
815 					(unsigned long long) ip->i_ino);
816 		XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
817 					 mp, dip);
818 		return XFS_ERROR(EFSCORRUPTED);
819 	}
820 
821 	ifp->if_broot_bytes = size;
822 	ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
823 	ASSERT(ifp->if_broot != NULL);
824 	/*
825 	 * Copy and convert from the on-disk structure
826 	 * to the in-memory structure.
827 	 */
828 	xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
829 			 ifp->if_broot, size);
830 	ifp->if_flags &= ~XFS_IFEXTENTS;
831 	ifp->if_flags |= XFS_IFBROOT;
832 
833 	return 0;
834 }
835 
836 STATIC void
837 xfs_dinode_from_disk(
838 	xfs_icdinode_t		*to,
839 	xfs_dinode_t		*from)
840 {
841 	to->di_magic = be16_to_cpu(from->di_magic);
842 	to->di_mode = be16_to_cpu(from->di_mode);
843 	to->di_version = from ->di_version;
844 	to->di_format = from->di_format;
845 	to->di_onlink = be16_to_cpu(from->di_onlink);
846 	to->di_uid = be32_to_cpu(from->di_uid);
847 	to->di_gid = be32_to_cpu(from->di_gid);
848 	to->di_nlink = be32_to_cpu(from->di_nlink);
849 	to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
850 	to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
851 	memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
852 	to->di_flushiter = be16_to_cpu(from->di_flushiter);
853 	to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
854 	to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
855 	to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
856 	to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
857 	to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
858 	to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
859 	to->di_size = be64_to_cpu(from->di_size);
860 	to->di_nblocks = be64_to_cpu(from->di_nblocks);
861 	to->di_extsize = be32_to_cpu(from->di_extsize);
862 	to->di_nextents = be32_to_cpu(from->di_nextents);
863 	to->di_anextents = be16_to_cpu(from->di_anextents);
864 	to->di_forkoff = from->di_forkoff;
865 	to->di_aformat	= from->di_aformat;
866 	to->di_dmevmask	= be32_to_cpu(from->di_dmevmask);
867 	to->di_dmstate	= be16_to_cpu(from->di_dmstate);
868 	to->di_flags	= be16_to_cpu(from->di_flags);
869 	to->di_gen	= be32_to_cpu(from->di_gen);
870 
871 	if (to->di_version == 3) {
872 		to->di_changecount = be64_to_cpu(from->di_changecount);
873 		to->di_crtime.t_sec = be32_to_cpu(from->di_crtime.t_sec);
874 		to->di_crtime.t_nsec = be32_to_cpu(from->di_crtime.t_nsec);
875 		to->di_flags2 = be64_to_cpu(from->di_flags2);
876 		to->di_ino = be64_to_cpu(from->di_ino);
877 		to->di_lsn = be64_to_cpu(from->di_lsn);
878 		memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2));
879 		uuid_copy(&to->di_uuid, &from->di_uuid);
880 	}
881 }
882 
883 void
884 xfs_dinode_to_disk(
885 	xfs_dinode_t		*to,
886 	xfs_icdinode_t		*from)
887 {
888 	to->di_magic = cpu_to_be16(from->di_magic);
889 	to->di_mode = cpu_to_be16(from->di_mode);
890 	to->di_version = from ->di_version;
891 	to->di_format = from->di_format;
892 	to->di_onlink = cpu_to_be16(from->di_onlink);
893 	to->di_uid = cpu_to_be32(from->di_uid);
894 	to->di_gid = cpu_to_be32(from->di_gid);
895 	to->di_nlink = cpu_to_be32(from->di_nlink);
896 	to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
897 	to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
898 	memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
899 	to->di_flushiter = cpu_to_be16(from->di_flushiter);
900 	to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
901 	to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
902 	to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
903 	to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
904 	to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
905 	to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
906 	to->di_size = cpu_to_be64(from->di_size);
907 	to->di_nblocks = cpu_to_be64(from->di_nblocks);
908 	to->di_extsize = cpu_to_be32(from->di_extsize);
909 	to->di_nextents = cpu_to_be32(from->di_nextents);
910 	to->di_anextents = cpu_to_be16(from->di_anextents);
911 	to->di_forkoff = from->di_forkoff;
912 	to->di_aformat = from->di_aformat;
913 	to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
914 	to->di_dmstate = cpu_to_be16(from->di_dmstate);
915 	to->di_flags = cpu_to_be16(from->di_flags);
916 	to->di_gen = cpu_to_be32(from->di_gen);
917 
918 	if (from->di_version == 3) {
919 		to->di_changecount = cpu_to_be64(from->di_changecount);
920 		to->di_crtime.t_sec = cpu_to_be32(from->di_crtime.t_sec);
921 		to->di_crtime.t_nsec = cpu_to_be32(from->di_crtime.t_nsec);
922 		to->di_flags2 = cpu_to_be64(from->di_flags2);
923 		to->di_ino = cpu_to_be64(from->di_ino);
924 		to->di_lsn = cpu_to_be64(from->di_lsn);
925 		memcpy(to->di_pad2, from->di_pad2, sizeof(to->di_pad2));
926 		uuid_copy(&to->di_uuid, &from->di_uuid);
927 	}
928 }
929 
930 STATIC uint
931 _xfs_dic2xflags(
932 	__uint16_t		di_flags)
933 {
934 	uint			flags = 0;
935 
936 	if (di_flags & XFS_DIFLAG_ANY) {
937 		if (di_flags & XFS_DIFLAG_REALTIME)
938 			flags |= XFS_XFLAG_REALTIME;
939 		if (di_flags & XFS_DIFLAG_PREALLOC)
940 			flags |= XFS_XFLAG_PREALLOC;
941 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
942 			flags |= XFS_XFLAG_IMMUTABLE;
943 		if (di_flags & XFS_DIFLAG_APPEND)
944 			flags |= XFS_XFLAG_APPEND;
945 		if (di_flags & XFS_DIFLAG_SYNC)
946 			flags |= XFS_XFLAG_SYNC;
947 		if (di_flags & XFS_DIFLAG_NOATIME)
948 			flags |= XFS_XFLAG_NOATIME;
949 		if (di_flags & XFS_DIFLAG_NODUMP)
950 			flags |= XFS_XFLAG_NODUMP;
951 		if (di_flags & XFS_DIFLAG_RTINHERIT)
952 			flags |= XFS_XFLAG_RTINHERIT;
953 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
954 			flags |= XFS_XFLAG_PROJINHERIT;
955 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
956 			flags |= XFS_XFLAG_NOSYMLINKS;
957 		if (di_flags & XFS_DIFLAG_EXTSIZE)
958 			flags |= XFS_XFLAG_EXTSIZE;
959 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
960 			flags |= XFS_XFLAG_EXTSZINHERIT;
961 		if (di_flags & XFS_DIFLAG_NODEFRAG)
962 			flags |= XFS_XFLAG_NODEFRAG;
963 		if (di_flags & XFS_DIFLAG_FILESTREAM)
964 			flags |= XFS_XFLAG_FILESTREAM;
965 	}
966 
967 	return flags;
968 }
969 
970 uint
971 xfs_ip2xflags(
972 	xfs_inode_t		*ip)
973 {
974 	xfs_icdinode_t		*dic = &ip->i_d;
975 
976 	return _xfs_dic2xflags(dic->di_flags) |
977 				(XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
978 }
979 
980 uint
981 xfs_dic2xflags(
982 	xfs_dinode_t		*dip)
983 {
984 	return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
985 				(XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
986 }
987 
988 static bool
989 xfs_dinode_verify(
990 	struct xfs_mount	*mp,
991 	struct xfs_inode	*ip,
992 	struct xfs_dinode	*dip)
993 {
994 	if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
995 		return false;
996 
997 	/* only version 3 or greater inodes are extensively verified here */
998 	if (dip->di_version < 3)
999 		return true;
1000 
1001 	if (!xfs_sb_version_hascrc(&mp->m_sb))
1002 		return false;
1003 	if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
1004 			      offsetof(struct xfs_dinode, di_crc)))
1005 		return false;
1006 	if (be64_to_cpu(dip->di_ino) != ip->i_ino)
1007 		return false;
1008 	if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_uuid))
1009 		return false;
1010 	return true;
1011 }
1012 
1013 void
1014 xfs_dinode_calc_crc(
1015 	struct xfs_mount	*mp,
1016 	struct xfs_dinode	*dip)
1017 {
1018 	__uint32_t		crc;
1019 
1020 	if (dip->di_version < 3)
1021 		return;
1022 
1023 	ASSERT(xfs_sb_version_hascrc(&mp->m_sb));
1024 	crc = xfs_start_cksum((char *)dip, mp->m_sb.sb_inodesize,
1025 			      offsetof(struct xfs_dinode, di_crc));
1026 	dip->di_crc = xfs_end_cksum(crc);
1027 }
1028 
1029 /*
1030  * Read the disk inode attributes into the in-core inode structure.
1031  */
1032 int
1033 xfs_iread(
1034 	xfs_mount_t	*mp,
1035 	xfs_trans_t	*tp,
1036 	xfs_inode_t	*ip,
1037 	uint		iget_flags)
1038 {
1039 	xfs_buf_t	*bp;
1040 	xfs_dinode_t	*dip;
1041 	int		error;
1042 
1043 	/*
1044 	 * Fill in the location information in the in-core inode.
1045 	 */
1046 	error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
1047 	if (error)
1048 		return error;
1049 
1050 	/*
1051 	 * Get pointers to the on-disk inode and the buffer containing it.
1052 	 */
1053 	error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags);
1054 	if (error)
1055 		return error;
1056 
1057 	/* even unallocated inodes are verified */
1058 	if (!xfs_dinode_verify(mp, ip, dip)) {
1059 		xfs_alert(mp, "%s: validation failed for inode %lld failed",
1060 				__func__, ip->i_ino);
1061 
1062 		XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, dip);
1063 		error = XFS_ERROR(EFSCORRUPTED);
1064 		goto out_brelse;
1065 	}
1066 
1067 	/*
1068 	 * If the on-disk inode is already linked to a directory
1069 	 * entry, copy all of the inode into the in-core inode.
1070 	 * xfs_iformat() handles copying in the inode format
1071 	 * specific information.
1072 	 * Otherwise, just get the truly permanent information.
1073 	 */
1074 	if (dip->di_mode) {
1075 		xfs_dinode_from_disk(&ip->i_d, dip);
1076 		error = xfs_iformat(ip, dip);
1077 		if (error)  {
1078 #ifdef DEBUG
1079 			xfs_alert(mp, "%s: xfs_iformat() returned error %d",
1080 				__func__, error);
1081 #endif /* DEBUG */
1082 			goto out_brelse;
1083 		}
1084 	} else {
1085 		/*
1086 		 * Partial initialisation of the in-core inode. Just the bits
1087 		 * that xfs_ialloc won't overwrite or relies on being correct.
1088 		 */
1089 		ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
1090 		ip->i_d.di_version = dip->di_version;
1091 		ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
1092 		ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
1093 
1094 		if (dip->di_version == 3) {
1095 			ip->i_d.di_ino = be64_to_cpu(dip->di_ino);
1096 			uuid_copy(&ip->i_d.di_uuid, &dip->di_uuid);
1097 		}
1098 
1099 		/*
1100 		 * Make sure to pull in the mode here as well in
1101 		 * case the inode is released without being used.
1102 		 * This ensures that xfs_inactive() will see that
1103 		 * the inode is already free and not try to mess
1104 		 * with the uninitialized part of it.
1105 		 */
1106 		ip->i_d.di_mode = 0;
1107 	}
1108 
1109 	/*
1110 	 * The inode format changed when we moved the link count and
1111 	 * made it 32 bits long.  If this is an old format inode,
1112 	 * convert it in memory to look like a new one.  If it gets
1113 	 * flushed to disk we will convert back before flushing or
1114 	 * logging it.  We zero out the new projid field and the old link
1115 	 * count field.  We'll handle clearing the pad field (the remains
1116 	 * of the old uuid field) when we actually convert the inode to
1117 	 * the new format. We don't change the version number so that we
1118 	 * can distinguish this from a real new format inode.
1119 	 */
1120 	if (ip->i_d.di_version == 1) {
1121 		ip->i_d.di_nlink = ip->i_d.di_onlink;
1122 		ip->i_d.di_onlink = 0;
1123 		xfs_set_projid(ip, 0);
1124 	}
1125 
1126 	ip->i_delayed_blks = 0;
1127 
1128 	/*
1129 	 * Mark the buffer containing the inode as something to keep
1130 	 * around for a while.  This helps to keep recently accessed
1131 	 * meta-data in-core longer.
1132 	 */
1133 	xfs_buf_set_ref(bp, XFS_INO_REF);
1134 
1135 	/*
1136 	 * Use xfs_trans_brelse() to release the buffer containing the
1137 	 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1138 	 * in xfs_imap_to_bp() above.  If tp is NULL, this is just a normal
1139 	 * brelse().  If we're within a transaction, then xfs_trans_brelse()
1140 	 * will only release the buffer if it is not dirty within the
1141 	 * transaction.  It will be OK to release the buffer in this case,
1142 	 * because inodes on disk are never destroyed and we will be
1143 	 * locking the new in-core inode before putting it in the hash
1144 	 * table where other processes can find it.  Thus we don't have
1145 	 * to worry about the inode being changed just because we released
1146 	 * the buffer.
1147 	 */
1148  out_brelse:
1149 	xfs_trans_brelse(tp, bp);
1150 	return error;
1151 }
1152 
1153 /*
1154  * Read in extents from a btree-format inode.
1155  * Allocate and fill in if_extents.  Real work is done in xfs_bmap.c.
1156  */
1157 int
1158 xfs_iread_extents(
1159 	xfs_trans_t	*tp,
1160 	xfs_inode_t	*ip,
1161 	int		whichfork)
1162 {
1163 	int		error;
1164 	xfs_ifork_t	*ifp;
1165 	xfs_extnum_t	nextents;
1166 
1167 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1168 		XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1169 				 ip->i_mount);
1170 		return XFS_ERROR(EFSCORRUPTED);
1171 	}
1172 	nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1173 	ifp = XFS_IFORK_PTR(ip, whichfork);
1174 
1175 	/*
1176 	 * We know that the size is valid (it's checked in iformat_btree)
1177 	 */
1178 	ifp->if_bytes = ifp->if_real_bytes = 0;
1179 	ifp->if_flags |= XFS_IFEXTENTS;
1180 	xfs_iext_add(ifp, 0, nextents);
1181 	error = xfs_bmap_read_extents(tp, ip, whichfork);
1182 	if (error) {
1183 		xfs_iext_destroy(ifp);
1184 		ifp->if_flags &= ~XFS_IFEXTENTS;
1185 		return error;
1186 	}
1187 	xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
1188 	return 0;
1189 }
1190 
1191 /*
1192  * Allocate an inode on disk and return a copy of its in-core version.
1193  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
1194  * appropriately within the inode.  The uid and gid for the inode are
1195  * set according to the contents of the given cred structure.
1196  *
1197  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1198  * has a free inode available, call xfs_iget() to obtain the in-core
1199  * version of the allocated inode.  Finally, fill in the inode and
1200  * log its initial contents.  In this case, ialloc_context would be
1201  * set to NULL.
1202  *
1203  * If xfs_dialloc() does not have an available inode, it will replenish
1204  * its supply by doing an allocation. Since we can only do one
1205  * allocation within a transaction without deadlocks, we must commit
1206  * the current transaction before returning the inode itself.
1207  * In this case, therefore, we will set ialloc_context and return.
1208  * The caller should then commit the current transaction, start a new
1209  * transaction, and call xfs_ialloc() again to actually get the inode.
1210  *
1211  * To ensure that some other process does not grab the inode that
1212  * was allocated during the first call to xfs_ialloc(), this routine
1213  * also returns the [locked] bp pointing to the head of the freelist
1214  * as ialloc_context.  The caller should hold this buffer across
1215  * the commit and pass it back into this routine on the second call.
1216  *
1217  * If we are allocating quota inodes, we do not have a parent inode
1218  * to attach to or associate with (i.e. pip == NULL) because they
1219  * are not linked into the directory structure - they are attached
1220  * directly to the superblock - and so have no parent.
1221  */
1222 int
1223 xfs_ialloc(
1224 	xfs_trans_t	*tp,
1225 	xfs_inode_t	*pip,
1226 	umode_t		mode,
1227 	xfs_nlink_t	nlink,
1228 	xfs_dev_t	rdev,
1229 	prid_t		prid,
1230 	int		okalloc,
1231 	xfs_buf_t	**ialloc_context,
1232 	xfs_inode_t	**ipp)
1233 {
1234 	struct xfs_mount *mp = tp->t_mountp;
1235 	xfs_ino_t	ino;
1236 	xfs_inode_t	*ip;
1237 	uint		flags;
1238 	int		error;
1239 	timespec_t	tv;
1240 	int		filestreams = 0;
1241 
1242 	/*
1243 	 * Call the space management code to pick
1244 	 * the on-disk inode to be allocated.
1245 	 */
1246 	error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1247 			    ialloc_context, &ino);
1248 	if (error)
1249 		return error;
1250 	if (*ialloc_context || ino == NULLFSINO) {
1251 		*ipp = NULL;
1252 		return 0;
1253 	}
1254 	ASSERT(*ialloc_context == NULL);
1255 
1256 	/*
1257 	 * Get the in-core inode with the lock held exclusively.
1258 	 * This is because we're setting fields here we need
1259 	 * to prevent others from looking at until we're done.
1260 	 */
1261 	error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
1262 			 XFS_ILOCK_EXCL, &ip);
1263 	if (error)
1264 		return error;
1265 	ASSERT(ip != NULL);
1266 
1267 	ip->i_d.di_mode = mode;
1268 	ip->i_d.di_onlink = 0;
1269 	ip->i_d.di_nlink = nlink;
1270 	ASSERT(ip->i_d.di_nlink == nlink);
1271 	ip->i_d.di_uid = current_fsuid();
1272 	ip->i_d.di_gid = current_fsgid();
1273 	xfs_set_projid(ip, prid);
1274 	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1275 
1276 	/*
1277 	 * If the superblock version is up to where we support new format
1278 	 * inodes and this is currently an old format inode, then change
1279 	 * the inode version number now.  This way we only do the conversion
1280 	 * here rather than here and in the flush/logging code.
1281 	 */
1282 	if (xfs_sb_version_hasnlink(&mp->m_sb) &&
1283 	    ip->i_d.di_version == 1) {
1284 		ip->i_d.di_version = 2;
1285 		/*
1286 		 * We've already zeroed the old link count, the projid field,
1287 		 * and the pad field.
1288 		 */
1289 	}
1290 
1291 	/*
1292 	 * Project ids won't be stored on disk if we are using a version 1 inode.
1293 	 */
1294 	if ((prid != 0) && (ip->i_d.di_version == 1))
1295 		xfs_bump_ino_vers2(tp, ip);
1296 
1297 	if (pip && XFS_INHERIT_GID(pip)) {
1298 		ip->i_d.di_gid = pip->i_d.di_gid;
1299 		if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
1300 			ip->i_d.di_mode |= S_ISGID;
1301 		}
1302 	}
1303 
1304 	/*
1305 	 * If the group ID of the new file does not match the effective group
1306 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1307 	 * (and only if the irix_sgid_inherit compatibility variable is set).
1308 	 */
1309 	if ((irix_sgid_inherit) &&
1310 	    (ip->i_d.di_mode & S_ISGID) &&
1311 	    (!in_group_p((gid_t)ip->i_d.di_gid))) {
1312 		ip->i_d.di_mode &= ~S_ISGID;
1313 	}
1314 
1315 	ip->i_d.di_size = 0;
1316 	ip->i_d.di_nextents = 0;
1317 	ASSERT(ip->i_d.di_nblocks == 0);
1318 
1319 	nanotime(&tv);
1320 	ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1321 	ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1322 	ip->i_d.di_atime = ip->i_d.di_mtime;
1323 	ip->i_d.di_ctime = ip->i_d.di_mtime;
1324 
1325 	/*
1326 	 * di_gen will have been taken care of in xfs_iread.
1327 	 */
1328 	ip->i_d.di_extsize = 0;
1329 	ip->i_d.di_dmevmask = 0;
1330 	ip->i_d.di_dmstate = 0;
1331 	ip->i_d.di_flags = 0;
1332 
1333 	if (ip->i_d.di_version == 3) {
1334 		ASSERT(ip->i_d.di_ino == ino);
1335 		ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
1336 		ip->i_d.di_crc = 0;
1337 		ip->i_d.di_changecount = 1;
1338 		ip->i_d.di_lsn = 0;
1339 		ip->i_d.di_flags2 = 0;
1340 		memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
1341 		ip->i_d.di_crtime = ip->i_d.di_mtime;
1342 	}
1343 
1344 
1345 	flags = XFS_ILOG_CORE;
1346 	switch (mode & S_IFMT) {
1347 	case S_IFIFO:
1348 	case S_IFCHR:
1349 	case S_IFBLK:
1350 	case S_IFSOCK:
1351 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1352 		ip->i_df.if_u2.if_rdev = rdev;
1353 		ip->i_df.if_flags = 0;
1354 		flags |= XFS_ILOG_DEV;
1355 		break;
1356 	case S_IFREG:
1357 		/*
1358 		 * we can't set up filestreams until after the VFS inode
1359 		 * is set up properly.
1360 		 */
1361 		if (pip && xfs_inode_is_filestream(pip))
1362 			filestreams = 1;
1363 		/* fall through */
1364 	case S_IFDIR:
1365 		if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1366 			uint	di_flags = 0;
1367 
1368 			if (S_ISDIR(mode)) {
1369 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1370 					di_flags |= XFS_DIFLAG_RTINHERIT;
1371 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1372 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1373 					ip->i_d.di_extsize = pip->i_d.di_extsize;
1374 				}
1375 			} else if (S_ISREG(mode)) {
1376 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1377 					di_flags |= XFS_DIFLAG_REALTIME;
1378 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1379 					di_flags |= XFS_DIFLAG_EXTSIZE;
1380 					ip->i_d.di_extsize = pip->i_d.di_extsize;
1381 				}
1382 			}
1383 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1384 			    xfs_inherit_noatime)
1385 				di_flags |= XFS_DIFLAG_NOATIME;
1386 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1387 			    xfs_inherit_nodump)
1388 				di_flags |= XFS_DIFLAG_NODUMP;
1389 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1390 			    xfs_inherit_sync)
1391 				di_flags |= XFS_DIFLAG_SYNC;
1392 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1393 			    xfs_inherit_nosymlinks)
1394 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
1395 			if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1396 				di_flags |= XFS_DIFLAG_PROJINHERIT;
1397 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1398 			    xfs_inherit_nodefrag)
1399 				di_flags |= XFS_DIFLAG_NODEFRAG;
1400 			if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1401 				di_flags |= XFS_DIFLAG_FILESTREAM;
1402 			ip->i_d.di_flags |= di_flags;
1403 		}
1404 		/* FALLTHROUGH */
1405 	case S_IFLNK:
1406 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1407 		ip->i_df.if_flags = XFS_IFEXTENTS;
1408 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1409 		ip->i_df.if_u1.if_extents = NULL;
1410 		break;
1411 	default:
1412 		ASSERT(0);
1413 	}
1414 	/*
1415 	 * Attribute fork settings for new inode.
1416 	 */
1417 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1418 	ip->i_d.di_anextents = 0;
1419 
1420 	/*
1421 	 * Log the new values stuffed into the inode.
1422 	 */
1423 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1424 	xfs_trans_log_inode(tp, ip, flags);
1425 
1426 	/* now that we have an i_mode we can setup inode ops and unlock */
1427 	xfs_setup_inode(ip);
1428 
1429 	/* now we have set up the vfs inode we can associate the filestream */
1430 	if (filestreams) {
1431 		error = xfs_filestream_associate(pip, ip);
1432 		if (error < 0)
1433 			return -error;
1434 		if (!error)
1435 			xfs_iflags_set(ip, XFS_IFILESTREAM);
1436 	}
1437 
1438 	*ipp = ip;
1439 	return 0;
1440 }
1441 
1442 /*
1443  * Free up the underlying blocks past new_size.  The new size must be smaller
1444  * than the current size.  This routine can be used both for the attribute and
1445  * data fork, and does not modify the inode size, which is left to the caller.
1446  *
1447  * The transaction passed to this routine must have made a permanent log
1448  * reservation of at least XFS_ITRUNCATE_LOG_RES.  This routine may commit the
1449  * given transaction and start new ones, so make sure everything involved in
1450  * the transaction is tidy before calling here.  Some transaction will be
1451  * returned to the caller to be committed.  The incoming transaction must
1452  * already include the inode, and both inode locks must be held exclusively.
1453  * The inode must also be "held" within the transaction.  On return the inode
1454  * will be "held" within the returned transaction.  This routine does NOT
1455  * require any disk space to be reserved for it within the transaction.
1456  *
1457  * If we get an error, we must return with the inode locked and linked into the
1458  * current transaction. This keeps things simple for the higher level code,
1459  * because it always knows that the inode is locked and held in the transaction
1460  * that returns to it whether errors occur or not.  We don't mark the inode
1461  * dirty on error so that transactions can be easily aborted if possible.
1462  */
1463 int
1464 xfs_itruncate_extents(
1465 	struct xfs_trans	**tpp,
1466 	struct xfs_inode	*ip,
1467 	int			whichfork,
1468 	xfs_fsize_t		new_size)
1469 {
1470 	struct xfs_mount	*mp = ip->i_mount;
1471 	struct xfs_trans	*tp = *tpp;
1472 	struct xfs_trans	*ntp;
1473 	xfs_bmap_free_t		free_list;
1474 	xfs_fsblock_t		first_block;
1475 	xfs_fileoff_t		first_unmap_block;
1476 	xfs_fileoff_t		last_block;
1477 	xfs_filblks_t		unmap_len;
1478 	int			committed;
1479 	int			error = 0;
1480 	int			done = 0;
1481 
1482 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1483 	ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1484 	       xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1485 	ASSERT(new_size <= XFS_ISIZE(ip));
1486 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1487 	ASSERT(ip->i_itemp != NULL);
1488 	ASSERT(ip->i_itemp->ili_lock_flags == 0);
1489 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1490 
1491 	trace_xfs_itruncate_extents_start(ip, new_size);
1492 
1493 	/*
1494 	 * Since it is possible for space to become allocated beyond
1495 	 * the end of the file (in a crash where the space is allocated
1496 	 * but the inode size is not yet updated), simply remove any
1497 	 * blocks which show up between the new EOF and the maximum
1498 	 * possible file size.  If the first block to be removed is
1499 	 * beyond the maximum file size (ie it is the same as last_block),
1500 	 * then there is nothing to do.
1501 	 */
1502 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1503 	last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1504 	if (first_unmap_block == last_block)
1505 		return 0;
1506 
1507 	ASSERT(first_unmap_block < last_block);
1508 	unmap_len = last_block - first_unmap_block + 1;
1509 	while (!done) {
1510 		xfs_bmap_init(&free_list, &first_block);
1511 		error = xfs_bunmapi(tp, ip,
1512 				    first_unmap_block, unmap_len,
1513 				    xfs_bmapi_aflag(whichfork),
1514 				    XFS_ITRUNC_MAX_EXTENTS,
1515 				    &first_block, &free_list,
1516 				    &done);
1517 		if (error)
1518 			goto out_bmap_cancel;
1519 
1520 		/*
1521 		 * Duplicate the transaction that has the permanent
1522 		 * reservation and commit the old transaction.
1523 		 */
1524 		error = xfs_bmap_finish(&tp, &free_list, &committed);
1525 		if (committed)
1526 			xfs_trans_ijoin(tp, ip, 0);
1527 		if (error)
1528 			goto out_bmap_cancel;
1529 
1530 		if (committed) {
1531 			/*
1532 			 * Mark the inode dirty so it will be logged and
1533 			 * moved forward in the log as part of every commit.
1534 			 */
1535 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1536 		}
1537 
1538 		ntp = xfs_trans_dup(tp);
1539 		error = xfs_trans_commit(tp, 0);
1540 		tp = ntp;
1541 
1542 		xfs_trans_ijoin(tp, ip, 0);
1543 
1544 		if (error)
1545 			goto out;
1546 
1547 		/*
1548 		 * Transaction commit worked ok so we can drop the extra ticket
1549 		 * reference that we gained in xfs_trans_dup()
1550 		 */
1551 		xfs_log_ticket_put(tp->t_ticket);
1552 		error = xfs_trans_reserve(tp, 0,
1553 					XFS_ITRUNCATE_LOG_RES(mp), 0,
1554 					XFS_TRANS_PERM_LOG_RES,
1555 					XFS_ITRUNCATE_LOG_COUNT);
1556 		if (error)
1557 			goto out;
1558 	}
1559 
1560 	/*
1561 	 * Always re-log the inode so that our permanent transaction can keep
1562 	 * on rolling it forward in the log.
1563 	 */
1564 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1565 
1566 	trace_xfs_itruncate_extents_end(ip, new_size);
1567 
1568 out:
1569 	*tpp = tp;
1570 	return error;
1571 out_bmap_cancel:
1572 	/*
1573 	 * If the bunmapi call encounters an error, return to the caller where
1574 	 * the transaction can be properly aborted.  We just need to make sure
1575 	 * we're not holding any resources that we were not when we came in.
1576 	 */
1577 	xfs_bmap_cancel(&free_list);
1578 	goto out;
1579 }
1580 
1581 /*
1582  * This is called when the inode's link count goes to 0.
1583  * We place the on-disk inode on a list in the AGI.  It
1584  * will be pulled from this list when the inode is freed.
1585  */
1586 int
1587 xfs_iunlink(
1588 	xfs_trans_t	*tp,
1589 	xfs_inode_t	*ip)
1590 {
1591 	xfs_mount_t	*mp;
1592 	xfs_agi_t	*agi;
1593 	xfs_dinode_t	*dip;
1594 	xfs_buf_t	*agibp;
1595 	xfs_buf_t	*ibp;
1596 	xfs_agino_t	agino;
1597 	short		bucket_index;
1598 	int		offset;
1599 	int		error;
1600 
1601 	ASSERT(ip->i_d.di_nlink == 0);
1602 	ASSERT(ip->i_d.di_mode != 0);
1603 
1604 	mp = tp->t_mountp;
1605 
1606 	/*
1607 	 * Get the agi buffer first.  It ensures lock ordering
1608 	 * on the list.
1609 	 */
1610 	error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1611 	if (error)
1612 		return error;
1613 	agi = XFS_BUF_TO_AGI(agibp);
1614 
1615 	/*
1616 	 * Get the index into the agi hash table for the
1617 	 * list this inode will go on.
1618 	 */
1619 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1620 	ASSERT(agino != 0);
1621 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1622 	ASSERT(agi->agi_unlinked[bucket_index]);
1623 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1624 
1625 	if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1626 		/*
1627 		 * There is already another inode in the bucket we need
1628 		 * to add ourselves to.  Add us at the front of the list.
1629 		 * Here we put the head pointer into our next pointer,
1630 		 * and then we fall through to point the head at us.
1631 		 */
1632 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1633 				       0, 0);
1634 		if (error)
1635 			return error;
1636 
1637 		ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1638 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1639 		offset = ip->i_imap.im_boffset +
1640 			offsetof(xfs_dinode_t, di_next_unlinked);
1641 		xfs_trans_inode_buf(tp, ibp);
1642 		xfs_trans_log_buf(tp, ibp, offset,
1643 				  (offset + sizeof(xfs_agino_t) - 1));
1644 		xfs_inobp_check(mp, ibp);
1645 	}
1646 
1647 	/*
1648 	 * Point the bucket head pointer at the inode being inserted.
1649 	 */
1650 	ASSERT(agino != 0);
1651 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1652 	offset = offsetof(xfs_agi_t, agi_unlinked) +
1653 		(sizeof(xfs_agino_t) * bucket_index);
1654 	xfs_trans_log_buf(tp, agibp, offset,
1655 			  (offset + sizeof(xfs_agino_t) - 1));
1656 	return 0;
1657 }
1658 
1659 /*
1660  * Pull the on-disk inode from the AGI unlinked list.
1661  */
1662 STATIC int
1663 xfs_iunlink_remove(
1664 	xfs_trans_t	*tp,
1665 	xfs_inode_t	*ip)
1666 {
1667 	xfs_ino_t	next_ino;
1668 	xfs_mount_t	*mp;
1669 	xfs_agi_t	*agi;
1670 	xfs_dinode_t	*dip;
1671 	xfs_buf_t	*agibp;
1672 	xfs_buf_t	*ibp;
1673 	xfs_agnumber_t	agno;
1674 	xfs_agino_t	agino;
1675 	xfs_agino_t	next_agino;
1676 	xfs_buf_t	*last_ibp;
1677 	xfs_dinode_t	*last_dip = NULL;
1678 	short		bucket_index;
1679 	int		offset, last_offset = 0;
1680 	int		error;
1681 
1682 	mp = tp->t_mountp;
1683 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1684 
1685 	/*
1686 	 * Get the agi buffer first.  It ensures lock ordering
1687 	 * on the list.
1688 	 */
1689 	error = xfs_read_agi(mp, tp, agno, &agibp);
1690 	if (error)
1691 		return error;
1692 
1693 	agi = XFS_BUF_TO_AGI(agibp);
1694 
1695 	/*
1696 	 * Get the index into the agi hash table for the
1697 	 * list this inode will go on.
1698 	 */
1699 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1700 	ASSERT(agino != 0);
1701 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1702 	ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1703 	ASSERT(agi->agi_unlinked[bucket_index]);
1704 
1705 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1706 		/*
1707 		 * We're at the head of the list.  Get the inode's on-disk
1708 		 * buffer to see if there is anyone after us on the list.
1709 		 * Only modify our next pointer if it is not already NULLAGINO.
1710 		 * This saves us the overhead of dealing with the buffer when
1711 		 * there is no need to change it.
1712 		 */
1713 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1714 				       0, 0);
1715 		if (error) {
1716 			xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
1717 				__func__, error);
1718 			return error;
1719 		}
1720 		next_agino = be32_to_cpu(dip->di_next_unlinked);
1721 		ASSERT(next_agino != 0);
1722 		if (next_agino != NULLAGINO) {
1723 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1724 			offset = ip->i_imap.im_boffset +
1725 				offsetof(xfs_dinode_t, di_next_unlinked);
1726 			xfs_trans_inode_buf(tp, ibp);
1727 			xfs_trans_log_buf(tp, ibp, offset,
1728 					  (offset + sizeof(xfs_agino_t) - 1));
1729 			xfs_inobp_check(mp, ibp);
1730 		} else {
1731 			xfs_trans_brelse(tp, ibp);
1732 		}
1733 		/*
1734 		 * Point the bucket head pointer at the next inode.
1735 		 */
1736 		ASSERT(next_agino != 0);
1737 		ASSERT(next_agino != agino);
1738 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1739 		offset = offsetof(xfs_agi_t, agi_unlinked) +
1740 			(sizeof(xfs_agino_t) * bucket_index);
1741 		xfs_trans_log_buf(tp, agibp, offset,
1742 				  (offset + sizeof(xfs_agino_t) - 1));
1743 	} else {
1744 		/*
1745 		 * We need to search the list for the inode being freed.
1746 		 */
1747 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1748 		last_ibp = NULL;
1749 		while (next_agino != agino) {
1750 			struct xfs_imap	imap;
1751 
1752 			if (last_ibp)
1753 				xfs_trans_brelse(tp, last_ibp);
1754 
1755 			imap.im_blkno = 0;
1756 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1757 
1758 			error = xfs_imap(mp, tp, next_ino, &imap, 0);
1759 			if (error) {
1760 				xfs_warn(mp,
1761 	"%s: xfs_imap returned error %d.",
1762 					 __func__, error);
1763 				return error;
1764 			}
1765 
1766 			error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
1767 					       &last_ibp, 0, 0);
1768 			if (error) {
1769 				xfs_warn(mp,
1770 	"%s: xfs_imap_to_bp returned error %d.",
1771 					__func__, error);
1772 				return error;
1773 			}
1774 
1775 			last_offset = imap.im_boffset;
1776 			next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1777 			ASSERT(next_agino != NULLAGINO);
1778 			ASSERT(next_agino != 0);
1779 		}
1780 
1781 		/*
1782 		 * Now last_ibp points to the buffer previous to us on the
1783 		 * unlinked list.  Pull us from the list.
1784 		 */
1785 		error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1786 				       0, 0);
1787 		if (error) {
1788 			xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
1789 				__func__, error);
1790 			return error;
1791 		}
1792 		next_agino = be32_to_cpu(dip->di_next_unlinked);
1793 		ASSERT(next_agino != 0);
1794 		ASSERT(next_agino != agino);
1795 		if (next_agino != NULLAGINO) {
1796 			dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1797 			offset = ip->i_imap.im_boffset +
1798 				offsetof(xfs_dinode_t, di_next_unlinked);
1799 			xfs_trans_inode_buf(tp, ibp);
1800 			xfs_trans_log_buf(tp, ibp, offset,
1801 					  (offset + sizeof(xfs_agino_t) - 1));
1802 			xfs_inobp_check(mp, ibp);
1803 		} else {
1804 			xfs_trans_brelse(tp, ibp);
1805 		}
1806 		/*
1807 		 * Point the previous inode on the list to the next inode.
1808 		 */
1809 		last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1810 		ASSERT(next_agino != 0);
1811 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1812 		xfs_trans_inode_buf(tp, last_ibp);
1813 		xfs_trans_log_buf(tp, last_ibp, offset,
1814 				  (offset + sizeof(xfs_agino_t) - 1));
1815 		xfs_inobp_check(mp, last_ibp);
1816 	}
1817 	return 0;
1818 }
1819 
1820 /*
1821  * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1822  * inodes that are in memory - they all must be marked stale and attached to
1823  * the cluster buffer.
1824  */
1825 STATIC int
1826 xfs_ifree_cluster(
1827 	xfs_inode_t	*free_ip,
1828 	xfs_trans_t	*tp,
1829 	xfs_ino_t	inum)
1830 {
1831 	xfs_mount_t		*mp = free_ip->i_mount;
1832 	int			blks_per_cluster;
1833 	int			nbufs;
1834 	int			ninodes;
1835 	int			i, j;
1836 	xfs_daddr_t		blkno;
1837 	xfs_buf_t		*bp;
1838 	xfs_inode_t		*ip;
1839 	xfs_inode_log_item_t	*iip;
1840 	xfs_log_item_t		*lip;
1841 	struct xfs_perag	*pag;
1842 
1843 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1844 	if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1845 		blks_per_cluster = 1;
1846 		ninodes = mp->m_sb.sb_inopblock;
1847 		nbufs = XFS_IALLOC_BLOCKS(mp);
1848 	} else {
1849 		blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1850 					mp->m_sb.sb_blocksize;
1851 		ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1852 		nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1853 	}
1854 
1855 	for (j = 0; j < nbufs; j++, inum += ninodes) {
1856 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1857 					 XFS_INO_TO_AGBNO(mp, inum));
1858 
1859 		/*
1860 		 * We obtain and lock the backing buffer first in the process
1861 		 * here, as we have to ensure that any dirty inode that we
1862 		 * can't get the flush lock on is attached to the buffer.
1863 		 * If we scan the in-memory inodes first, then buffer IO can
1864 		 * complete before we get a lock on it, and hence we may fail
1865 		 * to mark all the active inodes on the buffer stale.
1866 		 */
1867 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1868 					mp->m_bsize * blks_per_cluster,
1869 					XBF_UNMAPPED);
1870 
1871 		if (!bp)
1872 			return ENOMEM;
1873 
1874 		/*
1875 		 * This buffer may not have been correctly initialised as we
1876 		 * didn't read it from disk. That's not important because we are
1877 		 * only using to mark the buffer as stale in the log, and to
1878 		 * attach stale cached inodes on it. That means it will never be
1879 		 * dispatched for IO. If it is, we want to know about it, and we
1880 		 * want it to fail. We can acheive this by adding a write
1881 		 * verifier to the buffer.
1882 		 */
1883 		 bp->b_ops = &xfs_inode_buf_ops;
1884 
1885 		/*
1886 		 * Walk the inodes already attached to the buffer and mark them
1887 		 * stale. These will all have the flush locks held, so an
1888 		 * in-memory inode walk can't lock them. By marking them all
1889 		 * stale first, we will not attempt to lock them in the loop
1890 		 * below as the XFS_ISTALE flag will be set.
1891 		 */
1892 		lip = bp->b_fspriv;
1893 		while (lip) {
1894 			if (lip->li_type == XFS_LI_INODE) {
1895 				iip = (xfs_inode_log_item_t *)lip;
1896 				ASSERT(iip->ili_logged == 1);
1897 				lip->li_cb = xfs_istale_done;
1898 				xfs_trans_ail_copy_lsn(mp->m_ail,
1899 							&iip->ili_flush_lsn,
1900 							&iip->ili_item.li_lsn);
1901 				xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1902 			}
1903 			lip = lip->li_bio_list;
1904 		}
1905 
1906 
1907 		/*
1908 		 * For each inode in memory attempt to add it to the inode
1909 		 * buffer and set it up for being staled on buffer IO
1910 		 * completion.  This is safe as we've locked out tail pushing
1911 		 * and flushing by locking the buffer.
1912 		 *
1913 		 * We have already marked every inode that was part of a
1914 		 * transaction stale above, which means there is no point in
1915 		 * even trying to lock them.
1916 		 */
1917 		for (i = 0; i < ninodes; i++) {
1918 retry:
1919 			rcu_read_lock();
1920 			ip = radix_tree_lookup(&pag->pag_ici_root,
1921 					XFS_INO_TO_AGINO(mp, (inum + i)));
1922 
1923 			/* Inode not in memory, nothing to do */
1924 			if (!ip) {
1925 				rcu_read_unlock();
1926 				continue;
1927 			}
1928 
1929 			/*
1930 			 * because this is an RCU protected lookup, we could
1931 			 * find a recently freed or even reallocated inode
1932 			 * during the lookup. We need to check under the
1933 			 * i_flags_lock for a valid inode here. Skip it if it
1934 			 * is not valid, the wrong inode or stale.
1935 			 */
1936 			spin_lock(&ip->i_flags_lock);
1937 			if (ip->i_ino != inum + i ||
1938 			    __xfs_iflags_test(ip, XFS_ISTALE)) {
1939 				spin_unlock(&ip->i_flags_lock);
1940 				rcu_read_unlock();
1941 				continue;
1942 			}
1943 			spin_unlock(&ip->i_flags_lock);
1944 
1945 			/*
1946 			 * Don't try to lock/unlock the current inode, but we
1947 			 * _cannot_ skip the other inodes that we did not find
1948 			 * in the list attached to the buffer and are not
1949 			 * already marked stale. If we can't lock it, back off
1950 			 * and retry.
1951 			 */
1952 			if (ip != free_ip &&
1953 			    !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1954 				rcu_read_unlock();
1955 				delay(1);
1956 				goto retry;
1957 			}
1958 			rcu_read_unlock();
1959 
1960 			xfs_iflock(ip);
1961 			xfs_iflags_set(ip, XFS_ISTALE);
1962 
1963 			/*
1964 			 * we don't need to attach clean inodes or those only
1965 			 * with unlogged changes (which we throw away, anyway).
1966 			 */
1967 			iip = ip->i_itemp;
1968 			if (!iip || xfs_inode_clean(ip)) {
1969 				ASSERT(ip != free_ip);
1970 				xfs_ifunlock(ip);
1971 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
1972 				continue;
1973 			}
1974 
1975 			iip->ili_last_fields = iip->ili_fields;
1976 			iip->ili_fields = 0;
1977 			iip->ili_logged = 1;
1978 			xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1979 						&iip->ili_item.li_lsn);
1980 
1981 			xfs_buf_attach_iodone(bp, xfs_istale_done,
1982 						  &iip->ili_item);
1983 
1984 			if (ip != free_ip)
1985 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
1986 		}
1987 
1988 		xfs_trans_stale_inode_buf(tp, bp);
1989 		xfs_trans_binval(tp, bp);
1990 	}
1991 
1992 	xfs_perag_put(pag);
1993 	return 0;
1994 }
1995 
1996 /*
1997  * This is called to return an inode to the inode free list.
1998  * The inode should already be truncated to 0 length and have
1999  * no pages associated with it.  This routine also assumes that
2000  * the inode is already a part of the transaction.
2001  *
2002  * The on-disk copy of the inode will have been added to the list
2003  * of unlinked inodes in the AGI. We need to remove the inode from
2004  * that list atomically with respect to freeing it here.
2005  */
2006 int
2007 xfs_ifree(
2008 	xfs_trans_t	*tp,
2009 	xfs_inode_t	*ip,
2010 	xfs_bmap_free_t	*flist)
2011 {
2012 	int			error;
2013 	int			delete;
2014 	xfs_ino_t		first_ino;
2015 	xfs_dinode_t    	*dip;
2016 	xfs_buf_t       	*ibp;
2017 
2018 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2019 	ASSERT(ip->i_d.di_nlink == 0);
2020 	ASSERT(ip->i_d.di_nextents == 0);
2021 	ASSERT(ip->i_d.di_anextents == 0);
2022 	ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2023 	ASSERT(ip->i_d.di_nblocks == 0);
2024 
2025 	/*
2026 	 * Pull the on-disk inode from the AGI unlinked list.
2027 	 */
2028 	error = xfs_iunlink_remove(tp, ip);
2029 	if (error != 0) {
2030 		return error;
2031 	}
2032 
2033 	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2034 	if (error != 0) {
2035 		return error;
2036 	}
2037 	ip->i_d.di_mode = 0;		/* mark incore inode as free */
2038 	ip->i_d.di_flags = 0;
2039 	ip->i_d.di_dmevmask = 0;
2040 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2041 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2042 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2043 	/*
2044 	 * Bump the generation count so no one will be confused
2045 	 * by reincarnations of this inode.
2046 	 */
2047 	ip->i_d.di_gen++;
2048 
2049 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2050 
2051 	error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp,
2052 			       0, 0);
2053 	if (error)
2054 		return error;
2055 
2056         /*
2057 	* Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2058 	* from picking up this inode when it is reclaimed (its incore state
2059 	* initialzed but not flushed to disk yet). The in-core di_mode is
2060 	* already cleared  and a corresponding transaction logged.
2061 	* The hack here just synchronizes the in-core to on-disk
2062 	* di_mode value in advance before the actual inode sync to disk.
2063 	* This is OK because the inode is already unlinked and would never
2064 	* change its di_mode again for this inode generation.
2065 	* This is a temporary hack that would require a proper fix
2066 	* in the future.
2067 	*/
2068 	dip->di_mode = 0;
2069 
2070 	if (delete) {
2071 		error = xfs_ifree_cluster(ip, tp, first_ino);
2072 	}
2073 
2074 	return error;
2075 }
2076 
2077 /*
2078  * Reallocate the space for if_broot based on the number of records
2079  * being added or deleted as indicated in rec_diff.  Move the records
2080  * and pointers in if_broot to fit the new size.  When shrinking this
2081  * will eliminate holes between the records and pointers created by
2082  * the caller.  When growing this will create holes to be filled in
2083  * by the caller.
2084  *
2085  * The caller must not request to add more records than would fit in
2086  * the on-disk inode root.  If the if_broot is currently NULL, then
2087  * if we adding records one will be allocated.  The caller must also
2088  * not request that the number of records go below zero, although
2089  * it can go to zero.
2090  *
2091  * ip -- the inode whose if_broot area is changing
2092  * ext_diff -- the change in the number of records, positive or negative,
2093  *	 requested for the if_broot array.
2094  */
2095 void
2096 xfs_iroot_realloc(
2097 	xfs_inode_t		*ip,
2098 	int			rec_diff,
2099 	int			whichfork)
2100 {
2101 	struct xfs_mount	*mp = ip->i_mount;
2102 	int			cur_max;
2103 	xfs_ifork_t		*ifp;
2104 	struct xfs_btree_block	*new_broot;
2105 	int			new_max;
2106 	size_t			new_size;
2107 	char			*np;
2108 	char			*op;
2109 
2110 	/*
2111 	 * Handle the degenerate case quietly.
2112 	 */
2113 	if (rec_diff == 0) {
2114 		return;
2115 	}
2116 
2117 	ifp = XFS_IFORK_PTR(ip, whichfork);
2118 	if (rec_diff > 0) {
2119 		/*
2120 		 * If there wasn't any memory allocated before, just
2121 		 * allocate it now and get out.
2122 		 */
2123 		if (ifp->if_broot_bytes == 0) {
2124 			new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
2125 			ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2126 			ifp->if_broot_bytes = (int)new_size;
2127 			return;
2128 		}
2129 
2130 		/*
2131 		 * If there is already an existing if_broot, then we need
2132 		 * to realloc() it and shift the pointers to their new
2133 		 * location.  The records don't change location because
2134 		 * they are kept butted up against the btree block header.
2135 		 */
2136 		cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2137 		new_max = cur_max + rec_diff;
2138 		new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
2139 		ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2140 				XFS_BMAP_BROOT_SPACE_CALC(mp, cur_max),
2141 				KM_SLEEP | KM_NOFS);
2142 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2143 						     ifp->if_broot_bytes);
2144 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2145 						     (int)new_size);
2146 		ifp->if_broot_bytes = (int)new_size;
2147 		ASSERT(ifp->if_broot_bytes <=
2148 			XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ(ip));
2149 		memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2150 		return;
2151 	}
2152 
2153 	/*
2154 	 * rec_diff is less than 0.  In this case, we are shrinking the
2155 	 * if_broot buffer.  It must already exist.  If we go to zero
2156 	 * records, just get rid of the root and clear the status bit.
2157 	 */
2158 	ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2159 	cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2160 	new_max = cur_max + rec_diff;
2161 	ASSERT(new_max >= 0);
2162 	if (new_max > 0)
2163 		new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
2164 	else
2165 		new_size = 0;
2166 	if (new_size > 0) {
2167 		new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2168 		/*
2169 		 * First copy over the btree block header.
2170 		 */
2171 		memcpy(new_broot, ifp->if_broot,
2172 			XFS_BMBT_BLOCK_LEN(ip->i_mount));
2173 	} else {
2174 		new_broot = NULL;
2175 		ifp->if_flags &= ~XFS_IFBROOT;
2176 	}
2177 
2178 	/*
2179 	 * Only copy the records and pointers if there are any.
2180 	 */
2181 	if (new_max > 0) {
2182 		/*
2183 		 * First copy the records.
2184 		 */
2185 		op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2186 		np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2187 		memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2188 
2189 		/*
2190 		 * Then copy the pointers.
2191 		 */
2192 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2193 						     ifp->if_broot_bytes);
2194 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2195 						     (int)new_size);
2196 		memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2197 	}
2198 	kmem_free(ifp->if_broot);
2199 	ifp->if_broot = new_broot;
2200 	ifp->if_broot_bytes = (int)new_size;
2201 	ASSERT(ifp->if_broot_bytes <=
2202 		XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ(ip));
2203 	return;
2204 }
2205 
2206 
2207 /*
2208  * This is called when the amount of space needed for if_data
2209  * is increased or decreased.  The change in size is indicated by
2210  * the number of bytes that need to be added or deleted in the
2211  * byte_diff parameter.
2212  *
2213  * If the amount of space needed has decreased below the size of the
2214  * inline buffer, then switch to using the inline buffer.  Otherwise,
2215  * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2216  * to what is needed.
2217  *
2218  * ip -- the inode whose if_data area is changing
2219  * byte_diff -- the change in the number of bytes, positive or negative,
2220  *	 requested for the if_data array.
2221  */
2222 void
2223 xfs_idata_realloc(
2224 	xfs_inode_t	*ip,
2225 	int		byte_diff,
2226 	int		whichfork)
2227 {
2228 	xfs_ifork_t	*ifp;
2229 	int		new_size;
2230 	int		real_size;
2231 
2232 	if (byte_diff == 0) {
2233 		return;
2234 	}
2235 
2236 	ifp = XFS_IFORK_PTR(ip, whichfork);
2237 	new_size = (int)ifp->if_bytes + byte_diff;
2238 	ASSERT(new_size >= 0);
2239 
2240 	if (new_size == 0) {
2241 		if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2242 			kmem_free(ifp->if_u1.if_data);
2243 		}
2244 		ifp->if_u1.if_data = NULL;
2245 		real_size = 0;
2246 	} else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2247 		/*
2248 		 * If the valid extents/data can fit in if_inline_ext/data,
2249 		 * copy them from the malloc'd vector and free it.
2250 		 */
2251 		if (ifp->if_u1.if_data == NULL) {
2252 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2253 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2254 			ASSERT(ifp->if_real_bytes != 0);
2255 			memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2256 			      new_size);
2257 			kmem_free(ifp->if_u1.if_data);
2258 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2259 		}
2260 		real_size = 0;
2261 	} else {
2262 		/*
2263 		 * Stuck with malloc/realloc.
2264 		 * For inline data, the underlying buffer must be
2265 		 * a multiple of 4 bytes in size so that it can be
2266 		 * logged and stay on word boundaries.  We enforce
2267 		 * that here.
2268 		 */
2269 		real_size = roundup(new_size, 4);
2270 		if (ifp->if_u1.if_data == NULL) {
2271 			ASSERT(ifp->if_real_bytes == 0);
2272 			ifp->if_u1.if_data = kmem_alloc(real_size,
2273 							KM_SLEEP | KM_NOFS);
2274 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2275 			/*
2276 			 * Only do the realloc if the underlying size
2277 			 * is really changing.
2278 			 */
2279 			if (ifp->if_real_bytes != real_size) {
2280 				ifp->if_u1.if_data =
2281 					kmem_realloc(ifp->if_u1.if_data,
2282 							real_size,
2283 							ifp->if_real_bytes,
2284 							KM_SLEEP | KM_NOFS);
2285 			}
2286 		} else {
2287 			ASSERT(ifp->if_real_bytes == 0);
2288 			ifp->if_u1.if_data = kmem_alloc(real_size,
2289 							KM_SLEEP | KM_NOFS);
2290 			memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2291 				ifp->if_bytes);
2292 		}
2293 	}
2294 	ifp->if_real_bytes = real_size;
2295 	ifp->if_bytes = new_size;
2296 	ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2297 }
2298 
2299 void
2300 xfs_idestroy_fork(
2301 	xfs_inode_t	*ip,
2302 	int		whichfork)
2303 {
2304 	xfs_ifork_t	*ifp;
2305 
2306 	ifp = XFS_IFORK_PTR(ip, whichfork);
2307 	if (ifp->if_broot != NULL) {
2308 		kmem_free(ifp->if_broot);
2309 		ifp->if_broot = NULL;
2310 	}
2311 
2312 	/*
2313 	 * If the format is local, then we can't have an extents
2314 	 * array so just look for an inline data array.  If we're
2315 	 * not local then we may or may not have an extents list,
2316 	 * so check and free it up if we do.
2317 	 */
2318 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2319 		if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2320 		    (ifp->if_u1.if_data != NULL)) {
2321 			ASSERT(ifp->if_real_bytes != 0);
2322 			kmem_free(ifp->if_u1.if_data);
2323 			ifp->if_u1.if_data = NULL;
2324 			ifp->if_real_bytes = 0;
2325 		}
2326 	} else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2327 		   ((ifp->if_flags & XFS_IFEXTIREC) ||
2328 		    ((ifp->if_u1.if_extents != NULL) &&
2329 		     (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2330 		ASSERT(ifp->if_real_bytes != 0);
2331 		xfs_iext_destroy(ifp);
2332 	}
2333 	ASSERT(ifp->if_u1.if_extents == NULL ||
2334 	       ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2335 	ASSERT(ifp->if_real_bytes == 0);
2336 	if (whichfork == XFS_ATTR_FORK) {
2337 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2338 		ip->i_afp = NULL;
2339 	}
2340 }
2341 
2342 /*
2343  * This is called to unpin an inode.  The caller must have the inode locked
2344  * in at least shared mode so that the buffer cannot be subsequently pinned
2345  * once someone is waiting for it to be unpinned.
2346  */
2347 static void
2348 xfs_iunpin(
2349 	struct xfs_inode	*ip)
2350 {
2351 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2352 
2353 	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2354 
2355 	/* Give the log a push to start the unpinning I/O */
2356 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2357 
2358 }
2359 
2360 static void
2361 __xfs_iunpin_wait(
2362 	struct xfs_inode	*ip)
2363 {
2364 	wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2365 	DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2366 
2367 	xfs_iunpin(ip);
2368 
2369 	do {
2370 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2371 		if (xfs_ipincount(ip))
2372 			io_schedule();
2373 	} while (xfs_ipincount(ip));
2374 	finish_wait(wq, &wait.wait);
2375 }
2376 
2377 void
2378 xfs_iunpin_wait(
2379 	struct xfs_inode	*ip)
2380 {
2381 	if (xfs_ipincount(ip))
2382 		__xfs_iunpin_wait(ip);
2383 }
2384 
2385 /*
2386  * xfs_iextents_copy()
2387  *
2388  * This is called to copy the REAL extents (as opposed to the delayed
2389  * allocation extents) from the inode into the given buffer.  It
2390  * returns the number of bytes copied into the buffer.
2391  *
2392  * If there are no delayed allocation extents, then we can just
2393  * memcpy() the extents into the buffer.  Otherwise, we need to
2394  * examine each extent in turn and skip those which are delayed.
2395  */
2396 int
2397 xfs_iextents_copy(
2398 	xfs_inode_t		*ip,
2399 	xfs_bmbt_rec_t		*dp,
2400 	int			whichfork)
2401 {
2402 	int			copied;
2403 	int			i;
2404 	xfs_ifork_t		*ifp;
2405 	int			nrecs;
2406 	xfs_fsblock_t		start_block;
2407 
2408 	ifp = XFS_IFORK_PTR(ip, whichfork);
2409 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2410 	ASSERT(ifp->if_bytes > 0);
2411 
2412 	nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2413 	XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2414 	ASSERT(nrecs > 0);
2415 
2416 	/*
2417 	 * There are some delayed allocation extents in the
2418 	 * inode, so copy the extents one at a time and skip
2419 	 * the delayed ones.  There must be at least one
2420 	 * non-delayed extent.
2421 	 */
2422 	copied = 0;
2423 	for (i = 0; i < nrecs; i++) {
2424 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2425 		start_block = xfs_bmbt_get_startblock(ep);
2426 		if (isnullstartblock(start_block)) {
2427 			/*
2428 			 * It's a delayed allocation extent, so skip it.
2429 			 */
2430 			continue;
2431 		}
2432 
2433 		/* Translate to on disk format */
2434 		put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2435 		put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2436 		dp++;
2437 		copied++;
2438 	}
2439 	ASSERT(copied != 0);
2440 	xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2441 
2442 	return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2443 }
2444 
2445 /*
2446  * Each of the following cases stores data into the same region
2447  * of the on-disk inode, so only one of them can be valid at
2448  * any given time. While it is possible to have conflicting formats
2449  * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2450  * in EXTENTS format, this can only happen when the fork has
2451  * changed formats after being modified but before being flushed.
2452  * In these cases, the format always takes precedence, because the
2453  * format indicates the current state of the fork.
2454  */
2455 /*ARGSUSED*/
2456 STATIC void
2457 xfs_iflush_fork(
2458 	xfs_inode_t		*ip,
2459 	xfs_dinode_t		*dip,
2460 	xfs_inode_log_item_t	*iip,
2461 	int			whichfork,
2462 	xfs_buf_t		*bp)
2463 {
2464 	char			*cp;
2465 	xfs_ifork_t		*ifp;
2466 	xfs_mount_t		*mp;
2467 	static const short	brootflag[2] =
2468 		{ XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2469 	static const short	dataflag[2] =
2470 		{ XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2471 	static const short	extflag[2] =
2472 		{ XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2473 
2474 	if (!iip)
2475 		return;
2476 	ifp = XFS_IFORK_PTR(ip, whichfork);
2477 	/*
2478 	 * This can happen if we gave up in iformat in an error path,
2479 	 * for the attribute fork.
2480 	 */
2481 	if (!ifp) {
2482 		ASSERT(whichfork == XFS_ATTR_FORK);
2483 		return;
2484 	}
2485 	cp = XFS_DFORK_PTR(dip, whichfork);
2486 	mp = ip->i_mount;
2487 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2488 	case XFS_DINODE_FMT_LOCAL:
2489 		if ((iip->ili_fields & dataflag[whichfork]) &&
2490 		    (ifp->if_bytes > 0)) {
2491 			ASSERT(ifp->if_u1.if_data != NULL);
2492 			ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2493 			memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2494 		}
2495 		break;
2496 
2497 	case XFS_DINODE_FMT_EXTENTS:
2498 		ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2499 		       !(iip->ili_fields & extflag[whichfork]));
2500 		if ((iip->ili_fields & extflag[whichfork]) &&
2501 		    (ifp->if_bytes > 0)) {
2502 			ASSERT(xfs_iext_get_ext(ifp, 0));
2503 			ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2504 			(void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2505 				whichfork);
2506 		}
2507 		break;
2508 
2509 	case XFS_DINODE_FMT_BTREE:
2510 		if ((iip->ili_fields & brootflag[whichfork]) &&
2511 		    (ifp->if_broot_bytes > 0)) {
2512 			ASSERT(ifp->if_broot != NULL);
2513 			ASSERT(ifp->if_broot_bytes <=
2514 			       (XFS_IFORK_SIZE(ip, whichfork) +
2515 				XFS_BROOT_SIZE_ADJ(ip)));
2516 			xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2517 				(xfs_bmdr_block_t *)cp,
2518 				XFS_DFORK_SIZE(dip, mp, whichfork));
2519 		}
2520 		break;
2521 
2522 	case XFS_DINODE_FMT_DEV:
2523 		if (iip->ili_fields & XFS_ILOG_DEV) {
2524 			ASSERT(whichfork == XFS_DATA_FORK);
2525 			xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2526 		}
2527 		break;
2528 
2529 	case XFS_DINODE_FMT_UUID:
2530 		if (iip->ili_fields & XFS_ILOG_UUID) {
2531 			ASSERT(whichfork == XFS_DATA_FORK);
2532 			memcpy(XFS_DFORK_DPTR(dip),
2533 			       &ip->i_df.if_u2.if_uuid,
2534 			       sizeof(uuid_t));
2535 		}
2536 		break;
2537 
2538 	default:
2539 		ASSERT(0);
2540 		break;
2541 	}
2542 }
2543 
2544 STATIC int
2545 xfs_iflush_cluster(
2546 	xfs_inode_t	*ip,
2547 	xfs_buf_t	*bp)
2548 {
2549 	xfs_mount_t		*mp = ip->i_mount;
2550 	struct xfs_perag	*pag;
2551 	unsigned long		first_index, mask;
2552 	unsigned long		inodes_per_cluster;
2553 	int			ilist_size;
2554 	xfs_inode_t		**ilist;
2555 	xfs_inode_t		*iq;
2556 	int			nr_found;
2557 	int			clcount = 0;
2558 	int			bufwasdelwri;
2559 	int			i;
2560 
2561 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2562 
2563 	inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2564 	ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2565 	ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2566 	if (!ilist)
2567 		goto out_put;
2568 
2569 	mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2570 	first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2571 	rcu_read_lock();
2572 	/* really need a gang lookup range call here */
2573 	nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2574 					first_index, inodes_per_cluster);
2575 	if (nr_found == 0)
2576 		goto out_free;
2577 
2578 	for (i = 0; i < nr_found; i++) {
2579 		iq = ilist[i];
2580 		if (iq == ip)
2581 			continue;
2582 
2583 		/*
2584 		 * because this is an RCU protected lookup, we could find a
2585 		 * recently freed or even reallocated inode during the lookup.
2586 		 * We need to check under the i_flags_lock for a valid inode
2587 		 * here. Skip it if it is not valid or the wrong inode.
2588 		 */
2589 		spin_lock(&ip->i_flags_lock);
2590 		if (!ip->i_ino ||
2591 		    (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2592 			spin_unlock(&ip->i_flags_lock);
2593 			continue;
2594 		}
2595 		spin_unlock(&ip->i_flags_lock);
2596 
2597 		/*
2598 		 * Do an un-protected check to see if the inode is dirty and
2599 		 * is a candidate for flushing.  These checks will be repeated
2600 		 * later after the appropriate locks are acquired.
2601 		 */
2602 		if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2603 			continue;
2604 
2605 		/*
2606 		 * Try to get locks.  If any are unavailable or it is pinned,
2607 		 * then this inode cannot be flushed and is skipped.
2608 		 */
2609 
2610 		if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2611 			continue;
2612 		if (!xfs_iflock_nowait(iq)) {
2613 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
2614 			continue;
2615 		}
2616 		if (xfs_ipincount(iq)) {
2617 			xfs_ifunlock(iq);
2618 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
2619 			continue;
2620 		}
2621 
2622 		/*
2623 		 * arriving here means that this inode can be flushed.  First
2624 		 * re-check that it's dirty before flushing.
2625 		 */
2626 		if (!xfs_inode_clean(iq)) {
2627 			int	error;
2628 			error = xfs_iflush_int(iq, bp);
2629 			if (error) {
2630 				xfs_iunlock(iq, XFS_ILOCK_SHARED);
2631 				goto cluster_corrupt_out;
2632 			}
2633 			clcount++;
2634 		} else {
2635 			xfs_ifunlock(iq);
2636 		}
2637 		xfs_iunlock(iq, XFS_ILOCK_SHARED);
2638 	}
2639 
2640 	if (clcount) {
2641 		XFS_STATS_INC(xs_icluster_flushcnt);
2642 		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2643 	}
2644 
2645 out_free:
2646 	rcu_read_unlock();
2647 	kmem_free(ilist);
2648 out_put:
2649 	xfs_perag_put(pag);
2650 	return 0;
2651 
2652 
2653 cluster_corrupt_out:
2654 	/*
2655 	 * Corruption detected in the clustering loop.  Invalidate the
2656 	 * inode buffer and shut down the filesystem.
2657 	 */
2658 	rcu_read_unlock();
2659 	/*
2660 	 * Clean up the buffer.  If it was delwri, just release it --
2661 	 * brelse can handle it with no problems.  If not, shut down the
2662 	 * filesystem before releasing the buffer.
2663 	 */
2664 	bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
2665 	if (bufwasdelwri)
2666 		xfs_buf_relse(bp);
2667 
2668 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2669 
2670 	if (!bufwasdelwri) {
2671 		/*
2672 		 * Just like incore_relse: if we have b_iodone functions,
2673 		 * mark the buffer as an error and call them.  Otherwise
2674 		 * mark it as stale and brelse.
2675 		 */
2676 		if (bp->b_iodone) {
2677 			XFS_BUF_UNDONE(bp);
2678 			xfs_buf_stale(bp);
2679 			xfs_buf_ioerror(bp, EIO);
2680 			xfs_buf_ioend(bp, 0);
2681 		} else {
2682 			xfs_buf_stale(bp);
2683 			xfs_buf_relse(bp);
2684 		}
2685 	}
2686 
2687 	/*
2688 	 * Unlocks the flush lock
2689 	 */
2690 	xfs_iflush_abort(iq, false);
2691 	kmem_free(ilist);
2692 	xfs_perag_put(pag);
2693 	return XFS_ERROR(EFSCORRUPTED);
2694 }
2695 
2696 /*
2697  * Flush dirty inode metadata into the backing buffer.
2698  *
2699  * The caller must have the inode lock and the inode flush lock held.  The
2700  * inode lock will still be held upon return to the caller, and the inode
2701  * flush lock will be released after the inode has reached the disk.
2702  *
2703  * The caller must write out the buffer returned in *bpp and release it.
2704  */
2705 int
2706 xfs_iflush(
2707 	struct xfs_inode	*ip,
2708 	struct xfs_buf		**bpp)
2709 {
2710 	struct xfs_mount	*mp = ip->i_mount;
2711 	struct xfs_buf		*bp;
2712 	struct xfs_dinode	*dip;
2713 	int			error;
2714 
2715 	XFS_STATS_INC(xs_iflush_count);
2716 
2717 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2718 	ASSERT(xfs_isiflocked(ip));
2719 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2720 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2721 
2722 	*bpp = NULL;
2723 
2724 	xfs_iunpin_wait(ip);
2725 
2726 	/*
2727 	 * For stale inodes we cannot rely on the backing buffer remaining
2728 	 * stale in cache for the remaining life of the stale inode and so
2729 	 * xfs_imap_to_bp() below may give us a buffer that no longer contains
2730 	 * inodes below. We have to check this after ensuring the inode is
2731 	 * unpinned so that it is safe to reclaim the stale inode after the
2732 	 * flush call.
2733 	 */
2734 	if (xfs_iflags_test(ip, XFS_ISTALE)) {
2735 		xfs_ifunlock(ip);
2736 		return 0;
2737 	}
2738 
2739 	/*
2740 	 * This may have been unpinned because the filesystem is shutting
2741 	 * down forcibly. If that's the case we must not write this inode
2742 	 * to disk, because the log record didn't make it to disk.
2743 	 *
2744 	 * We also have to remove the log item from the AIL in this case,
2745 	 * as we wait for an empty AIL as part of the unmount process.
2746 	 */
2747 	if (XFS_FORCED_SHUTDOWN(mp)) {
2748 		error = XFS_ERROR(EIO);
2749 		goto abort_out;
2750 	}
2751 
2752 	/*
2753 	 * Get the buffer containing the on-disk inode.
2754 	 */
2755 	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
2756 			       0);
2757 	if (error || !bp) {
2758 		xfs_ifunlock(ip);
2759 		return error;
2760 	}
2761 
2762 	/*
2763 	 * First flush out the inode that xfs_iflush was called with.
2764 	 */
2765 	error = xfs_iflush_int(ip, bp);
2766 	if (error)
2767 		goto corrupt_out;
2768 
2769 	/*
2770 	 * If the buffer is pinned then push on the log now so we won't
2771 	 * get stuck waiting in the write for too long.
2772 	 */
2773 	if (xfs_buf_ispinned(bp))
2774 		xfs_log_force(mp, 0);
2775 
2776 	/*
2777 	 * inode clustering:
2778 	 * see if other inodes can be gathered into this write
2779 	 */
2780 	error = xfs_iflush_cluster(ip, bp);
2781 	if (error)
2782 		goto cluster_corrupt_out;
2783 
2784 	*bpp = bp;
2785 	return 0;
2786 
2787 corrupt_out:
2788 	xfs_buf_relse(bp);
2789 	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2790 cluster_corrupt_out:
2791 	error = XFS_ERROR(EFSCORRUPTED);
2792 abort_out:
2793 	/*
2794 	 * Unlocks the flush lock
2795 	 */
2796 	xfs_iflush_abort(ip, false);
2797 	return error;
2798 }
2799 
2800 
2801 STATIC int
2802 xfs_iflush_int(
2803 	struct xfs_inode	*ip,
2804 	struct xfs_buf		*bp)
2805 {
2806 	struct xfs_inode_log_item *iip = ip->i_itemp;
2807 	struct xfs_dinode	*dip;
2808 	struct xfs_mount	*mp = ip->i_mount;
2809 
2810 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2811 	ASSERT(xfs_isiflocked(ip));
2812 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2813 	       ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2814 	ASSERT(iip != NULL && iip->ili_fields != 0);
2815 
2816 	/* set *dip = inode's place in the buffer */
2817 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2818 
2819 	if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2820 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2821 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2822 			"%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2823 			__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2824 		goto corrupt_out;
2825 	}
2826 	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2827 				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2828 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2829 			"%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2830 			__func__, ip->i_ino, ip, ip->i_d.di_magic);
2831 		goto corrupt_out;
2832 	}
2833 	if (S_ISREG(ip->i_d.di_mode)) {
2834 		if (XFS_TEST_ERROR(
2835 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2836 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2837 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2838 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2839 				"%s: Bad regular inode %Lu, ptr 0x%p",
2840 				__func__, ip->i_ino, ip);
2841 			goto corrupt_out;
2842 		}
2843 	} else if (S_ISDIR(ip->i_d.di_mode)) {
2844 		if (XFS_TEST_ERROR(
2845 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2846 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2847 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2848 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2849 			xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2850 				"%s: Bad directory inode %Lu, ptr 0x%p",
2851 				__func__, ip->i_ino, ip);
2852 			goto corrupt_out;
2853 		}
2854 	}
2855 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2856 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2857 				XFS_RANDOM_IFLUSH_5)) {
2858 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2859 			"%s: detected corrupt incore inode %Lu, "
2860 			"total extents = %d, nblocks = %Ld, ptr 0x%p",
2861 			__func__, ip->i_ino,
2862 			ip->i_d.di_nextents + ip->i_d.di_anextents,
2863 			ip->i_d.di_nblocks, ip);
2864 		goto corrupt_out;
2865 	}
2866 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2867 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2868 		xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2869 			"%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2870 			__func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2871 		goto corrupt_out;
2872 	}
2873 	/*
2874 	 * bump the flush iteration count, used to detect flushes which
2875 	 * postdate a log record during recovery. This is redundant as we now
2876 	 * log every change and hence this can't happen. Still, it doesn't hurt.
2877 	 */
2878 	ip->i_d.di_flushiter++;
2879 
2880 	/*
2881 	 * Copy the dirty parts of the inode into the on-disk
2882 	 * inode.  We always copy out the core of the inode,
2883 	 * because if the inode is dirty at all the core must
2884 	 * be.
2885 	 */
2886 	xfs_dinode_to_disk(dip, &ip->i_d);
2887 
2888 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
2889 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2890 		ip->i_d.di_flushiter = 0;
2891 
2892 	/*
2893 	 * If this is really an old format inode and the superblock version
2894 	 * has not been updated to support only new format inodes, then
2895 	 * convert back to the old inode format.  If the superblock version
2896 	 * has been updated, then make the conversion permanent.
2897 	 */
2898 	ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2899 	if (ip->i_d.di_version == 1) {
2900 		if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2901 			/*
2902 			 * Convert it back.
2903 			 */
2904 			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2905 			dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2906 		} else {
2907 			/*
2908 			 * The superblock version has already been bumped,
2909 			 * so just make the conversion to the new inode
2910 			 * format permanent.
2911 			 */
2912 			ip->i_d.di_version = 2;
2913 			dip->di_version = 2;
2914 			ip->i_d.di_onlink = 0;
2915 			dip->di_onlink = 0;
2916 			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2917 			memset(&(dip->di_pad[0]), 0,
2918 			      sizeof(dip->di_pad));
2919 			ASSERT(xfs_get_projid(ip) == 0);
2920 		}
2921 	}
2922 
2923 	xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2924 	if (XFS_IFORK_Q(ip))
2925 		xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2926 	xfs_inobp_check(mp, bp);
2927 
2928 	/*
2929 	 * We've recorded everything logged in the inode, so we'd like to clear
2930 	 * the ili_fields bits so we don't log and flush things unnecessarily.
2931 	 * However, we can't stop logging all this information until the data
2932 	 * we've copied into the disk buffer is written to disk.  If we did we
2933 	 * might overwrite the copy of the inode in the log with all the data
2934 	 * after re-logging only part of it, and in the face of a crash we
2935 	 * wouldn't have all the data we need to recover.
2936 	 *
2937 	 * What we do is move the bits to the ili_last_fields field.  When
2938 	 * logging the inode, these bits are moved back to the ili_fields field.
2939 	 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2940 	 * know that the information those bits represent is permanently on
2941 	 * disk.  As long as the flush completes before the inode is logged
2942 	 * again, then both ili_fields and ili_last_fields will be cleared.
2943 	 *
2944 	 * We can play with the ili_fields bits here, because the inode lock
2945 	 * must be held exclusively in order to set bits there and the flush
2946 	 * lock protects the ili_last_fields bits.  Set ili_logged so the flush
2947 	 * done routine can tell whether or not to look in the AIL.  Also, store
2948 	 * the current LSN of the inode so that we can tell whether the item has
2949 	 * moved in the AIL from xfs_iflush_done().  In order to read the lsn we
2950 	 * need the AIL lock, because it is a 64 bit value that cannot be read
2951 	 * atomically.
2952 	 */
2953 	iip->ili_last_fields = iip->ili_fields;
2954 	iip->ili_fields = 0;
2955 	iip->ili_logged = 1;
2956 
2957 	xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2958 				&iip->ili_item.li_lsn);
2959 
2960 	/*
2961 	 * Attach the function xfs_iflush_done to the inode's
2962 	 * buffer.  This will remove the inode from the AIL
2963 	 * and unlock the inode's flush lock when the inode is
2964 	 * completely written to disk.
2965 	 */
2966 	xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
2967 
2968 	/* update the lsn in the on disk inode if required */
2969 	if (ip->i_d.di_version == 3)
2970 		dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
2971 
2972 	/* generate the checksum. */
2973 	xfs_dinode_calc_crc(mp, dip);
2974 
2975 	ASSERT(bp->b_fspriv != NULL);
2976 	ASSERT(bp->b_iodone != NULL);
2977 	return 0;
2978 
2979 corrupt_out:
2980 	return XFS_ERROR(EFSCORRUPTED);
2981 }
2982 
2983 /*
2984  * Return a pointer to the extent record at file index idx.
2985  */
2986 xfs_bmbt_rec_host_t *
2987 xfs_iext_get_ext(
2988 	xfs_ifork_t	*ifp,		/* inode fork pointer */
2989 	xfs_extnum_t	idx)		/* index of target extent */
2990 {
2991 	ASSERT(idx >= 0);
2992 	ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2993 
2994 	if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2995 		return ifp->if_u1.if_ext_irec->er_extbuf;
2996 	} else if (ifp->if_flags & XFS_IFEXTIREC) {
2997 		xfs_ext_irec_t	*erp;		/* irec pointer */
2998 		int		erp_idx = 0;	/* irec index */
2999 		xfs_extnum_t	page_idx = idx;	/* ext index in target list */
3000 
3001 		erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3002 		return &erp->er_extbuf[page_idx];
3003 	} else if (ifp->if_bytes) {
3004 		return &ifp->if_u1.if_extents[idx];
3005 	} else {
3006 		return NULL;
3007 	}
3008 }
3009 
3010 /*
3011  * Insert new item(s) into the extent records for incore inode
3012  * fork 'ifp'.  'count' new items are inserted at index 'idx'.
3013  */
3014 void
3015 xfs_iext_insert(
3016 	xfs_inode_t	*ip,		/* incore inode pointer */
3017 	xfs_extnum_t	idx,		/* starting index of new items */
3018 	xfs_extnum_t	count,		/* number of inserted items */
3019 	xfs_bmbt_irec_t	*new,		/* items to insert */
3020 	int		state)		/* type of extent conversion */
3021 {
3022 	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3023 	xfs_extnum_t	i;		/* extent record index */
3024 
3025 	trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3026 
3027 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3028 	xfs_iext_add(ifp, idx, count);
3029 	for (i = idx; i < idx + count; i++, new++)
3030 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3031 }
3032 
3033 /*
3034  * This is called when the amount of space required for incore file
3035  * extents needs to be increased. The ext_diff parameter stores the
3036  * number of new extents being added and the idx parameter contains
3037  * the extent index where the new extents will be added. If the new
3038  * extents are being appended, then we just need to (re)allocate and
3039  * initialize the space. Otherwise, if the new extents are being
3040  * inserted into the middle of the existing entries, a bit more work
3041  * is required to make room for the new extents to be inserted. The
3042  * caller is responsible for filling in the new extent entries upon
3043  * return.
3044  */
3045 void
3046 xfs_iext_add(
3047 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3048 	xfs_extnum_t	idx,		/* index to begin adding exts */
3049 	int		ext_diff)	/* number of extents to add */
3050 {
3051 	int		byte_diff;	/* new bytes being added */
3052 	int		new_size;	/* size of extents after adding */
3053 	xfs_extnum_t	nextents;	/* number of extents in file */
3054 
3055 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3056 	ASSERT((idx >= 0) && (idx <= nextents));
3057 	byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3058 	new_size = ifp->if_bytes + byte_diff;
3059 	/*
3060 	 * If the new number of extents (nextents + ext_diff)
3061 	 * fits inside the inode, then continue to use the inline
3062 	 * extent buffer.
3063 	 */
3064 	if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3065 		if (idx < nextents) {
3066 			memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3067 				&ifp->if_u2.if_inline_ext[idx],
3068 				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
3069 			memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3070 		}
3071 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3072 		ifp->if_real_bytes = 0;
3073 	}
3074 	/*
3075 	 * Otherwise use a linear (direct) extent list.
3076 	 * If the extents are currently inside the inode,
3077 	 * xfs_iext_realloc_direct will switch us from
3078 	 * inline to direct extent allocation mode.
3079 	 */
3080 	else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3081 		xfs_iext_realloc_direct(ifp, new_size);
3082 		if (idx < nextents) {
3083 			memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3084 				&ifp->if_u1.if_extents[idx],
3085 				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
3086 			memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3087 		}
3088 	}
3089 	/* Indirection array */
3090 	else {
3091 		xfs_ext_irec_t	*erp;
3092 		int		erp_idx = 0;
3093 		int		page_idx = idx;
3094 
3095 		ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3096 		if (ifp->if_flags & XFS_IFEXTIREC) {
3097 			erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3098 		} else {
3099 			xfs_iext_irec_init(ifp);
3100 			ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3101 			erp = ifp->if_u1.if_ext_irec;
3102 		}
3103 		/* Extents fit in target extent page */
3104 		if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3105 			if (page_idx < erp->er_extcount) {
3106 				memmove(&erp->er_extbuf[page_idx + ext_diff],
3107 					&erp->er_extbuf[page_idx],
3108 					(erp->er_extcount - page_idx) *
3109 					sizeof(xfs_bmbt_rec_t));
3110 				memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3111 			}
3112 			erp->er_extcount += ext_diff;
3113 			xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3114 		}
3115 		/* Insert a new extent page */
3116 		else if (erp) {
3117 			xfs_iext_add_indirect_multi(ifp,
3118 				erp_idx, page_idx, ext_diff);
3119 		}
3120 		/*
3121 		 * If extent(s) are being appended to the last page in
3122 		 * the indirection array and the new extent(s) don't fit
3123 		 * in the page, then erp is NULL and erp_idx is set to
3124 		 * the next index needed in the indirection array.
3125 		 */
3126 		else {
3127 			int	count = ext_diff;
3128 
3129 			while (count) {
3130 				erp = xfs_iext_irec_new(ifp, erp_idx);
3131 				erp->er_extcount = count;
3132 				count -= MIN(count, (int)XFS_LINEAR_EXTS);
3133 				if (count) {
3134 					erp_idx++;
3135 				}
3136 			}
3137 		}
3138 	}
3139 	ifp->if_bytes = new_size;
3140 }
3141 
3142 /*
3143  * This is called when incore extents are being added to the indirection
3144  * array and the new extents do not fit in the target extent list. The
3145  * erp_idx parameter contains the irec index for the target extent list
3146  * in the indirection array, and the idx parameter contains the extent
3147  * index within the list. The number of extents being added is stored
3148  * in the count parameter.
3149  *
3150  *    |-------|   |-------|
3151  *    |       |   |       |    idx - number of extents before idx
3152  *    |  idx  |   | count |
3153  *    |       |   |       |    count - number of extents being inserted at idx
3154  *    |-------|   |-------|
3155  *    | count |   | nex2  |    nex2 - number of extents after idx + count
3156  *    |-------|   |-------|
3157  */
3158 void
3159 xfs_iext_add_indirect_multi(
3160 	xfs_ifork_t	*ifp,			/* inode fork pointer */
3161 	int		erp_idx,		/* target extent irec index */
3162 	xfs_extnum_t	idx,			/* index within target list */
3163 	int		count)			/* new extents being added */
3164 {
3165 	int		byte_diff;		/* new bytes being added */
3166 	xfs_ext_irec_t	*erp;			/* pointer to irec entry */
3167 	xfs_extnum_t	ext_diff;		/* number of extents to add */
3168 	xfs_extnum_t	ext_cnt;		/* new extents still needed */
3169 	xfs_extnum_t	nex2;			/* extents after idx + count */
3170 	xfs_bmbt_rec_t	*nex2_ep = NULL;	/* temp list for nex2 extents */
3171 	int		nlists;			/* number of irec's (lists) */
3172 
3173 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3174 	erp = &ifp->if_u1.if_ext_irec[erp_idx];
3175 	nex2 = erp->er_extcount - idx;
3176 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3177 
3178 	/*
3179 	 * Save second part of target extent list
3180 	 * (all extents past */
3181 	if (nex2) {
3182 		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3183 		nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3184 		memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3185 		erp->er_extcount -= nex2;
3186 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3187 		memset(&erp->er_extbuf[idx], 0, byte_diff);
3188 	}
3189 
3190 	/*
3191 	 * Add the new extents to the end of the target
3192 	 * list, then allocate new irec record(s) and
3193 	 * extent buffer(s) as needed to store the rest
3194 	 * of the new extents.
3195 	 */
3196 	ext_cnt = count;
3197 	ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3198 	if (ext_diff) {
3199 		erp->er_extcount += ext_diff;
3200 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3201 		ext_cnt -= ext_diff;
3202 	}
3203 	while (ext_cnt) {
3204 		erp_idx++;
3205 		erp = xfs_iext_irec_new(ifp, erp_idx);
3206 		ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3207 		erp->er_extcount = ext_diff;
3208 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3209 		ext_cnt -= ext_diff;
3210 	}
3211 
3212 	/* Add nex2 extents back to indirection array */
3213 	if (nex2) {
3214 		xfs_extnum_t	ext_avail;
3215 		int		i;
3216 
3217 		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3218 		ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3219 		i = 0;
3220 		/*
3221 		 * If nex2 extents fit in the current page, append
3222 		 * nex2_ep after the new extents.
3223 		 */
3224 		if (nex2 <= ext_avail) {
3225 			i = erp->er_extcount;
3226 		}
3227 		/*
3228 		 * Otherwise, check if space is available in the
3229 		 * next page.
3230 		 */
3231 		else if ((erp_idx < nlists - 1) &&
3232 			 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3233 			  ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3234 			erp_idx++;
3235 			erp++;
3236 			/* Create a hole for nex2 extents */
3237 			memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3238 				erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3239 		}
3240 		/*
3241 		 * Final choice, create a new extent page for
3242 		 * nex2 extents.
3243 		 */
3244 		else {
3245 			erp_idx++;
3246 			erp = xfs_iext_irec_new(ifp, erp_idx);
3247 		}
3248 		memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3249 		kmem_free(nex2_ep);
3250 		erp->er_extcount += nex2;
3251 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3252 	}
3253 }
3254 
3255 /*
3256  * This is called when the amount of space required for incore file
3257  * extents needs to be decreased. The ext_diff parameter stores the
3258  * number of extents to be removed and the idx parameter contains
3259  * the extent index where the extents will be removed from.
3260  *
3261  * If the amount of space needed has decreased below the linear
3262  * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3263  * extent array.  Otherwise, use kmem_realloc() to adjust the
3264  * size to what is needed.
3265  */
3266 void
3267 xfs_iext_remove(
3268 	xfs_inode_t	*ip,		/* incore inode pointer */
3269 	xfs_extnum_t	idx,		/* index to begin removing exts */
3270 	int		ext_diff,	/* number of extents to remove */
3271 	int		state)		/* type of extent conversion */
3272 {
3273 	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3274 	xfs_extnum_t	nextents;	/* number of extents in file */
3275 	int		new_size;	/* size of extents after removal */
3276 
3277 	trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3278 
3279 	ASSERT(ext_diff > 0);
3280 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3281 	new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3282 
3283 	if (new_size == 0) {
3284 		xfs_iext_destroy(ifp);
3285 	} else if (ifp->if_flags & XFS_IFEXTIREC) {
3286 		xfs_iext_remove_indirect(ifp, idx, ext_diff);
3287 	} else if (ifp->if_real_bytes) {
3288 		xfs_iext_remove_direct(ifp, idx, ext_diff);
3289 	} else {
3290 		xfs_iext_remove_inline(ifp, idx, ext_diff);
3291 	}
3292 	ifp->if_bytes = new_size;
3293 }
3294 
3295 /*
3296  * This removes ext_diff extents from the inline buffer, beginning
3297  * at extent index idx.
3298  */
3299 void
3300 xfs_iext_remove_inline(
3301 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3302 	xfs_extnum_t	idx,		/* index to begin removing exts */
3303 	int		ext_diff)	/* number of extents to remove */
3304 {
3305 	int		nextents;	/* number of extents in file */
3306 
3307 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3308 	ASSERT(idx < XFS_INLINE_EXTS);
3309 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3310 	ASSERT(((nextents - ext_diff) > 0) &&
3311 		(nextents - ext_diff) < XFS_INLINE_EXTS);
3312 
3313 	if (idx + ext_diff < nextents) {
3314 		memmove(&ifp->if_u2.if_inline_ext[idx],
3315 			&ifp->if_u2.if_inline_ext[idx + ext_diff],
3316 			(nextents - (idx + ext_diff)) *
3317 			 sizeof(xfs_bmbt_rec_t));
3318 		memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3319 			0, ext_diff * sizeof(xfs_bmbt_rec_t));
3320 	} else {
3321 		memset(&ifp->if_u2.if_inline_ext[idx], 0,
3322 			ext_diff * sizeof(xfs_bmbt_rec_t));
3323 	}
3324 }
3325 
3326 /*
3327  * This removes ext_diff extents from a linear (direct) extent list,
3328  * beginning at extent index idx. If the extents are being removed
3329  * from the end of the list (ie. truncate) then we just need to re-
3330  * allocate the list to remove the extra space. Otherwise, if the
3331  * extents are being removed from the middle of the existing extent
3332  * entries, then we first need to move the extent records beginning
3333  * at idx + ext_diff up in the list to overwrite the records being
3334  * removed, then remove the extra space via kmem_realloc.
3335  */
3336 void
3337 xfs_iext_remove_direct(
3338 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3339 	xfs_extnum_t	idx,		/* index to begin removing exts */
3340 	int		ext_diff)	/* number of extents to remove */
3341 {
3342 	xfs_extnum_t	nextents;	/* number of extents in file */
3343 	int		new_size;	/* size of extents after removal */
3344 
3345 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3346 	new_size = ifp->if_bytes -
3347 		(ext_diff * sizeof(xfs_bmbt_rec_t));
3348 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3349 
3350 	if (new_size == 0) {
3351 		xfs_iext_destroy(ifp);
3352 		return;
3353 	}
3354 	/* Move extents up in the list (if needed) */
3355 	if (idx + ext_diff < nextents) {
3356 		memmove(&ifp->if_u1.if_extents[idx],
3357 			&ifp->if_u1.if_extents[idx + ext_diff],
3358 			(nextents - (idx + ext_diff)) *
3359 			 sizeof(xfs_bmbt_rec_t));
3360 	}
3361 	memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3362 		0, ext_diff * sizeof(xfs_bmbt_rec_t));
3363 	/*
3364 	 * Reallocate the direct extent list. If the extents
3365 	 * will fit inside the inode then xfs_iext_realloc_direct
3366 	 * will switch from direct to inline extent allocation
3367 	 * mode for us.
3368 	 */
3369 	xfs_iext_realloc_direct(ifp, new_size);
3370 	ifp->if_bytes = new_size;
3371 }
3372 
3373 /*
3374  * This is called when incore extents are being removed from the
3375  * indirection array and the extents being removed span multiple extent
3376  * buffers. The idx parameter contains the file extent index where we
3377  * want to begin removing extents, and the count parameter contains
3378  * how many extents need to be removed.
3379  *
3380  *    |-------|   |-------|
3381  *    | nex1  |   |       |    nex1 - number of extents before idx
3382  *    |-------|   | count |
3383  *    |       |   |       |    count - number of extents being removed at idx
3384  *    | count |   |-------|
3385  *    |       |   | nex2  |    nex2 - number of extents after idx + count
3386  *    |-------|   |-------|
3387  */
3388 void
3389 xfs_iext_remove_indirect(
3390 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3391 	xfs_extnum_t	idx,		/* index to begin removing extents */
3392 	int		count)		/* number of extents to remove */
3393 {
3394 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3395 	int		erp_idx = 0;	/* indirection array index */
3396 	xfs_extnum_t	ext_cnt;	/* extents left to remove */
3397 	xfs_extnum_t	ext_diff;	/* extents to remove in current list */
3398 	xfs_extnum_t	nex1;		/* number of extents before idx */
3399 	xfs_extnum_t	nex2;		/* extents after idx + count */
3400 	int		page_idx = idx;	/* index in target extent list */
3401 
3402 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3403 	erp = xfs_iext_idx_to_irec(ifp,  &page_idx, &erp_idx, 0);
3404 	ASSERT(erp != NULL);
3405 	nex1 = page_idx;
3406 	ext_cnt = count;
3407 	while (ext_cnt) {
3408 		nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3409 		ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3410 		/*
3411 		 * Check for deletion of entire list;
3412 		 * xfs_iext_irec_remove() updates extent offsets.
3413 		 */
3414 		if (ext_diff == erp->er_extcount) {
3415 			xfs_iext_irec_remove(ifp, erp_idx);
3416 			ext_cnt -= ext_diff;
3417 			nex1 = 0;
3418 			if (ext_cnt) {
3419 				ASSERT(erp_idx < ifp->if_real_bytes /
3420 					XFS_IEXT_BUFSZ);
3421 				erp = &ifp->if_u1.if_ext_irec[erp_idx];
3422 				nex1 = 0;
3423 				continue;
3424 			} else {
3425 				break;
3426 			}
3427 		}
3428 		/* Move extents up (if needed) */
3429 		if (nex2) {
3430 			memmove(&erp->er_extbuf[nex1],
3431 				&erp->er_extbuf[nex1 + ext_diff],
3432 				nex2 * sizeof(xfs_bmbt_rec_t));
3433 		}
3434 		/* Zero out rest of page */
3435 		memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3436 			((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3437 		/* Update remaining counters */
3438 		erp->er_extcount -= ext_diff;
3439 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3440 		ext_cnt -= ext_diff;
3441 		nex1 = 0;
3442 		erp_idx++;
3443 		erp++;
3444 	}
3445 	ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3446 	xfs_iext_irec_compact(ifp);
3447 }
3448 
3449 /*
3450  * Create, destroy, or resize a linear (direct) block of extents.
3451  */
3452 void
3453 xfs_iext_realloc_direct(
3454 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3455 	int		new_size)	/* new size of extents */
3456 {
3457 	int		rnew_size;	/* real new size of extents */
3458 
3459 	rnew_size = new_size;
3460 
3461 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3462 		((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3463 		 (new_size != ifp->if_real_bytes)));
3464 
3465 	/* Free extent records */
3466 	if (new_size == 0) {
3467 		xfs_iext_destroy(ifp);
3468 	}
3469 	/* Resize direct extent list and zero any new bytes */
3470 	else if (ifp->if_real_bytes) {
3471 		/* Check if extents will fit inside the inode */
3472 		if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3473 			xfs_iext_direct_to_inline(ifp, new_size /
3474 				(uint)sizeof(xfs_bmbt_rec_t));
3475 			ifp->if_bytes = new_size;
3476 			return;
3477 		}
3478 		if (!is_power_of_2(new_size)){
3479 			rnew_size = roundup_pow_of_two(new_size);
3480 		}
3481 		if (rnew_size != ifp->if_real_bytes) {
3482 			ifp->if_u1.if_extents =
3483 				kmem_realloc(ifp->if_u1.if_extents,
3484 						rnew_size,
3485 						ifp->if_real_bytes, KM_NOFS);
3486 		}
3487 		if (rnew_size > ifp->if_real_bytes) {
3488 			memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3489 				(uint)sizeof(xfs_bmbt_rec_t)], 0,
3490 				rnew_size - ifp->if_real_bytes);
3491 		}
3492 	}
3493 	/*
3494 	 * Switch from the inline extent buffer to a direct
3495 	 * extent list. Be sure to include the inline extent
3496 	 * bytes in new_size.
3497 	 */
3498 	else {
3499 		new_size += ifp->if_bytes;
3500 		if (!is_power_of_2(new_size)) {
3501 			rnew_size = roundup_pow_of_two(new_size);
3502 		}
3503 		xfs_iext_inline_to_direct(ifp, rnew_size);
3504 	}
3505 	ifp->if_real_bytes = rnew_size;
3506 	ifp->if_bytes = new_size;
3507 }
3508 
3509 /*
3510  * Switch from linear (direct) extent records to inline buffer.
3511  */
3512 void
3513 xfs_iext_direct_to_inline(
3514 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3515 	xfs_extnum_t	nextents)	/* number of extents in file */
3516 {
3517 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3518 	ASSERT(nextents <= XFS_INLINE_EXTS);
3519 	/*
3520 	 * The inline buffer was zeroed when we switched
3521 	 * from inline to direct extent allocation mode,
3522 	 * so we don't need to clear it here.
3523 	 */
3524 	memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3525 		nextents * sizeof(xfs_bmbt_rec_t));
3526 	kmem_free(ifp->if_u1.if_extents);
3527 	ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3528 	ifp->if_real_bytes = 0;
3529 }
3530 
3531 /*
3532  * Switch from inline buffer to linear (direct) extent records.
3533  * new_size should already be rounded up to the next power of 2
3534  * by the caller (when appropriate), so use new_size as it is.
3535  * However, since new_size may be rounded up, we can't update
3536  * if_bytes here. It is the caller's responsibility to update
3537  * if_bytes upon return.
3538  */
3539 void
3540 xfs_iext_inline_to_direct(
3541 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3542 	int		new_size)	/* number of extents in file */
3543 {
3544 	ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3545 	memset(ifp->if_u1.if_extents, 0, new_size);
3546 	if (ifp->if_bytes) {
3547 		memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3548 			ifp->if_bytes);
3549 		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3550 			sizeof(xfs_bmbt_rec_t));
3551 	}
3552 	ifp->if_real_bytes = new_size;
3553 }
3554 
3555 /*
3556  * Resize an extent indirection array to new_size bytes.
3557  */
3558 STATIC void
3559 xfs_iext_realloc_indirect(
3560 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3561 	int		new_size)	/* new indirection array size */
3562 {
3563 	int		nlists;		/* number of irec's (ex lists) */
3564 	int		size;		/* current indirection array size */
3565 
3566 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3567 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3568 	size = nlists * sizeof(xfs_ext_irec_t);
3569 	ASSERT(ifp->if_real_bytes);
3570 	ASSERT((new_size >= 0) && (new_size != size));
3571 	if (new_size == 0) {
3572 		xfs_iext_destroy(ifp);
3573 	} else {
3574 		ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3575 			kmem_realloc(ifp->if_u1.if_ext_irec,
3576 				new_size, size, KM_NOFS);
3577 	}
3578 }
3579 
3580 /*
3581  * Switch from indirection array to linear (direct) extent allocations.
3582  */
3583 STATIC void
3584 xfs_iext_indirect_to_direct(
3585 	 xfs_ifork_t	*ifp)		/* inode fork pointer */
3586 {
3587 	xfs_bmbt_rec_host_t *ep;	/* extent record pointer */
3588 	xfs_extnum_t	nextents;	/* number of extents in file */
3589 	int		size;		/* size of file extents */
3590 
3591 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3592 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3593 	ASSERT(nextents <= XFS_LINEAR_EXTS);
3594 	size = nextents * sizeof(xfs_bmbt_rec_t);
3595 
3596 	xfs_iext_irec_compact_pages(ifp);
3597 	ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3598 
3599 	ep = ifp->if_u1.if_ext_irec->er_extbuf;
3600 	kmem_free(ifp->if_u1.if_ext_irec);
3601 	ifp->if_flags &= ~XFS_IFEXTIREC;
3602 	ifp->if_u1.if_extents = ep;
3603 	ifp->if_bytes = size;
3604 	if (nextents < XFS_LINEAR_EXTS) {
3605 		xfs_iext_realloc_direct(ifp, size);
3606 	}
3607 }
3608 
3609 /*
3610  * Free incore file extents.
3611  */
3612 void
3613 xfs_iext_destroy(
3614 	xfs_ifork_t	*ifp)		/* inode fork pointer */
3615 {
3616 	if (ifp->if_flags & XFS_IFEXTIREC) {
3617 		int	erp_idx;
3618 		int	nlists;
3619 
3620 		nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3621 		for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3622 			xfs_iext_irec_remove(ifp, erp_idx);
3623 		}
3624 		ifp->if_flags &= ~XFS_IFEXTIREC;
3625 	} else if (ifp->if_real_bytes) {
3626 		kmem_free(ifp->if_u1.if_extents);
3627 	} else if (ifp->if_bytes) {
3628 		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3629 			sizeof(xfs_bmbt_rec_t));
3630 	}
3631 	ifp->if_u1.if_extents = NULL;
3632 	ifp->if_real_bytes = 0;
3633 	ifp->if_bytes = 0;
3634 }
3635 
3636 /*
3637  * Return a pointer to the extent record for file system block bno.
3638  */
3639 xfs_bmbt_rec_host_t *			/* pointer to found extent record */
3640 xfs_iext_bno_to_ext(
3641 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3642 	xfs_fileoff_t	bno,		/* block number to search for */
3643 	xfs_extnum_t	*idxp)		/* index of target extent */
3644 {
3645 	xfs_bmbt_rec_host_t *base;	/* pointer to first extent */
3646 	xfs_filblks_t	blockcount = 0;	/* number of blocks in extent */
3647 	xfs_bmbt_rec_host_t *ep = NULL;	/* pointer to target extent */
3648 	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
3649 	int		high;		/* upper boundary in search */
3650 	xfs_extnum_t	idx = 0;	/* index of target extent */
3651 	int		low;		/* lower boundary in search */
3652 	xfs_extnum_t	nextents;	/* number of file extents */
3653 	xfs_fileoff_t	startoff = 0;	/* start offset of extent */
3654 
3655 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3656 	if (nextents == 0) {
3657 		*idxp = 0;
3658 		return NULL;
3659 	}
3660 	low = 0;
3661 	if (ifp->if_flags & XFS_IFEXTIREC) {
3662 		/* Find target extent list */
3663 		int	erp_idx = 0;
3664 		erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3665 		base = erp->er_extbuf;
3666 		high = erp->er_extcount - 1;
3667 	} else {
3668 		base = ifp->if_u1.if_extents;
3669 		high = nextents - 1;
3670 	}
3671 	/* Binary search extent records */
3672 	while (low <= high) {
3673 		idx = (low + high) >> 1;
3674 		ep = base + idx;
3675 		startoff = xfs_bmbt_get_startoff(ep);
3676 		blockcount = xfs_bmbt_get_blockcount(ep);
3677 		if (bno < startoff) {
3678 			high = idx - 1;
3679 		} else if (bno >= startoff + blockcount) {
3680 			low = idx + 1;
3681 		} else {
3682 			/* Convert back to file-based extent index */
3683 			if (ifp->if_flags & XFS_IFEXTIREC) {
3684 				idx += erp->er_extoff;
3685 			}
3686 			*idxp = idx;
3687 			return ep;
3688 		}
3689 	}
3690 	/* Convert back to file-based extent index */
3691 	if (ifp->if_flags & XFS_IFEXTIREC) {
3692 		idx += erp->er_extoff;
3693 	}
3694 	if (bno >= startoff + blockcount) {
3695 		if (++idx == nextents) {
3696 			ep = NULL;
3697 		} else {
3698 			ep = xfs_iext_get_ext(ifp, idx);
3699 		}
3700 	}
3701 	*idxp = idx;
3702 	return ep;
3703 }
3704 
3705 /*
3706  * Return a pointer to the indirection array entry containing the
3707  * extent record for filesystem block bno. Store the index of the
3708  * target irec in *erp_idxp.
3709  */
3710 xfs_ext_irec_t *			/* pointer to found extent record */
3711 xfs_iext_bno_to_irec(
3712 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3713 	xfs_fileoff_t	bno,		/* block number to search for */
3714 	int		*erp_idxp)	/* irec index of target ext list */
3715 {
3716 	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
3717 	xfs_ext_irec_t	*erp_next;	/* next indirection array entry */
3718 	int		erp_idx;	/* indirection array index */
3719 	int		nlists;		/* number of extent irec's (lists) */
3720 	int		high;		/* binary search upper limit */
3721 	int		low;		/* binary search lower limit */
3722 
3723 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3724 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3725 	erp_idx = 0;
3726 	low = 0;
3727 	high = nlists - 1;
3728 	while (low <= high) {
3729 		erp_idx = (low + high) >> 1;
3730 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
3731 		erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3732 		if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3733 			high = erp_idx - 1;
3734 		} else if (erp_next && bno >=
3735 			   xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3736 			low = erp_idx + 1;
3737 		} else {
3738 			break;
3739 		}
3740 	}
3741 	*erp_idxp = erp_idx;
3742 	return erp;
3743 }
3744 
3745 /*
3746  * Return a pointer to the indirection array entry containing the
3747  * extent record at file extent index *idxp. Store the index of the
3748  * target irec in *erp_idxp and store the page index of the target
3749  * extent record in *idxp.
3750  */
3751 xfs_ext_irec_t *
3752 xfs_iext_idx_to_irec(
3753 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3754 	xfs_extnum_t	*idxp,		/* extent index (file -> page) */
3755 	int		*erp_idxp,	/* pointer to target irec */
3756 	int		realloc)	/* new bytes were just added */
3757 {
3758 	xfs_ext_irec_t	*prev;		/* pointer to previous irec */
3759 	xfs_ext_irec_t	*erp = NULL;	/* pointer to current irec */
3760 	int		erp_idx;	/* indirection array index */
3761 	int		nlists;		/* number of irec's (ex lists) */
3762 	int		high;		/* binary search upper limit */
3763 	int		low;		/* binary search lower limit */
3764 	xfs_extnum_t	page_idx = *idxp; /* extent index in target list */
3765 
3766 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3767 	ASSERT(page_idx >= 0);
3768 	ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3769 	ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3770 
3771 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3772 	erp_idx = 0;
3773 	low = 0;
3774 	high = nlists - 1;
3775 
3776 	/* Binary search extent irec's */
3777 	while (low <= high) {
3778 		erp_idx = (low + high) >> 1;
3779 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
3780 		prev = erp_idx > 0 ? erp - 1 : NULL;
3781 		if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3782 		     realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3783 			high = erp_idx - 1;
3784 		} else if (page_idx > erp->er_extoff + erp->er_extcount ||
3785 			   (page_idx == erp->er_extoff + erp->er_extcount &&
3786 			    !realloc)) {
3787 			low = erp_idx + 1;
3788 		} else if (page_idx == erp->er_extoff + erp->er_extcount &&
3789 			   erp->er_extcount == XFS_LINEAR_EXTS) {
3790 			ASSERT(realloc);
3791 			page_idx = 0;
3792 			erp_idx++;
3793 			erp = erp_idx < nlists ? erp + 1 : NULL;
3794 			break;
3795 		} else {
3796 			page_idx -= erp->er_extoff;
3797 			break;
3798 		}
3799 	}
3800 	*idxp = page_idx;
3801 	*erp_idxp = erp_idx;
3802 	return(erp);
3803 }
3804 
3805 /*
3806  * Allocate and initialize an indirection array once the space needed
3807  * for incore extents increases above XFS_IEXT_BUFSZ.
3808  */
3809 void
3810 xfs_iext_irec_init(
3811 	xfs_ifork_t	*ifp)		/* inode fork pointer */
3812 {
3813 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3814 	xfs_extnum_t	nextents;	/* number of extents in file */
3815 
3816 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3817 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3818 	ASSERT(nextents <= XFS_LINEAR_EXTS);
3819 
3820 	erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3821 
3822 	if (nextents == 0) {
3823 		ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3824 	} else if (!ifp->if_real_bytes) {
3825 		xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3826 	} else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3827 		xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3828 	}
3829 	erp->er_extbuf = ifp->if_u1.if_extents;
3830 	erp->er_extcount = nextents;
3831 	erp->er_extoff = 0;
3832 
3833 	ifp->if_flags |= XFS_IFEXTIREC;
3834 	ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3835 	ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3836 	ifp->if_u1.if_ext_irec = erp;
3837 
3838 	return;
3839 }
3840 
3841 /*
3842  * Allocate and initialize a new entry in the indirection array.
3843  */
3844 xfs_ext_irec_t *
3845 xfs_iext_irec_new(
3846 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3847 	int		erp_idx)	/* index for new irec */
3848 {
3849 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3850 	int		i;		/* loop counter */
3851 	int		nlists;		/* number of irec's (ex lists) */
3852 
3853 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3854 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3855 
3856 	/* Resize indirection array */
3857 	xfs_iext_realloc_indirect(ifp, ++nlists *
3858 				  sizeof(xfs_ext_irec_t));
3859 	/*
3860 	 * Move records down in the array so the
3861 	 * new page can use erp_idx.
3862 	 */
3863 	erp = ifp->if_u1.if_ext_irec;
3864 	for (i = nlists - 1; i > erp_idx; i--) {
3865 		memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3866 	}
3867 	ASSERT(i == erp_idx);
3868 
3869 	/* Initialize new extent record */
3870 	erp = ifp->if_u1.if_ext_irec;
3871 	erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3872 	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3873 	memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3874 	erp[erp_idx].er_extcount = 0;
3875 	erp[erp_idx].er_extoff = erp_idx > 0 ?
3876 		erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3877 	return (&erp[erp_idx]);
3878 }
3879 
3880 /*
3881  * Remove a record from the indirection array.
3882  */
3883 void
3884 xfs_iext_irec_remove(
3885 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3886 	int		erp_idx)	/* irec index to remove */
3887 {
3888 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
3889 	int		i;		/* loop counter */
3890 	int		nlists;		/* number of irec's (ex lists) */
3891 
3892 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3893 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3894 	erp = &ifp->if_u1.if_ext_irec[erp_idx];
3895 	if (erp->er_extbuf) {
3896 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3897 			-erp->er_extcount);
3898 		kmem_free(erp->er_extbuf);
3899 	}
3900 	/* Compact extent records */
3901 	erp = ifp->if_u1.if_ext_irec;
3902 	for (i = erp_idx; i < nlists - 1; i++) {
3903 		memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3904 	}
3905 	/*
3906 	 * Manually free the last extent record from the indirection
3907 	 * array.  A call to xfs_iext_realloc_indirect() with a size
3908 	 * of zero would result in a call to xfs_iext_destroy() which
3909 	 * would in turn call this function again, creating a nasty
3910 	 * infinite loop.
3911 	 */
3912 	if (--nlists) {
3913 		xfs_iext_realloc_indirect(ifp,
3914 			nlists * sizeof(xfs_ext_irec_t));
3915 	} else {
3916 		kmem_free(ifp->if_u1.if_ext_irec);
3917 	}
3918 	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3919 }
3920 
3921 /*
3922  * This is called to clean up large amounts of unused memory allocated
3923  * by the indirection array.  Before compacting anything though, verify
3924  * that the indirection array is still needed and switch back to the
3925  * linear extent list (or even the inline buffer) if possible.  The
3926  * compaction policy is as follows:
3927  *
3928  *    Full Compaction: Extents fit into a single page (or inline buffer)
3929  * Partial Compaction: Extents occupy less than 50% of allocated space
3930  *      No Compaction: Extents occupy at least 50% of allocated space
3931  */
3932 void
3933 xfs_iext_irec_compact(
3934 	xfs_ifork_t	*ifp)		/* inode fork pointer */
3935 {
3936 	xfs_extnum_t	nextents;	/* number of extents in file */
3937 	int		nlists;		/* number of irec's (ex lists) */
3938 
3939 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3940 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3941 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3942 
3943 	if (nextents == 0) {
3944 		xfs_iext_destroy(ifp);
3945 	} else if (nextents <= XFS_INLINE_EXTS) {
3946 		xfs_iext_indirect_to_direct(ifp);
3947 		xfs_iext_direct_to_inline(ifp, nextents);
3948 	} else if (nextents <= XFS_LINEAR_EXTS) {
3949 		xfs_iext_indirect_to_direct(ifp);
3950 	} else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3951 		xfs_iext_irec_compact_pages(ifp);
3952 	}
3953 }
3954 
3955 /*
3956  * Combine extents from neighboring extent pages.
3957  */
3958 void
3959 xfs_iext_irec_compact_pages(
3960 	xfs_ifork_t	*ifp)		/* inode fork pointer */
3961 {
3962 	xfs_ext_irec_t	*erp, *erp_next;/* pointers to irec entries */
3963 	int		erp_idx = 0;	/* indirection array index */
3964 	int		nlists;		/* number of irec's (ex lists) */
3965 
3966 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3967 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3968 	while (erp_idx < nlists - 1) {
3969 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
3970 		erp_next = erp + 1;
3971 		if (erp_next->er_extcount <=
3972 		    (XFS_LINEAR_EXTS - erp->er_extcount)) {
3973 			memcpy(&erp->er_extbuf[erp->er_extcount],
3974 				erp_next->er_extbuf, erp_next->er_extcount *
3975 				sizeof(xfs_bmbt_rec_t));
3976 			erp->er_extcount += erp_next->er_extcount;
3977 			/*
3978 			 * Free page before removing extent record
3979 			 * so er_extoffs don't get modified in
3980 			 * xfs_iext_irec_remove.
3981 			 */
3982 			kmem_free(erp_next->er_extbuf);
3983 			erp_next->er_extbuf = NULL;
3984 			xfs_iext_irec_remove(ifp, erp_idx + 1);
3985 			nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3986 		} else {
3987 			erp_idx++;
3988 		}
3989 	}
3990 }
3991 
3992 /*
3993  * This is called to update the er_extoff field in the indirection
3994  * array when extents have been added or removed from one of the
3995  * extent lists. erp_idx contains the irec index to begin updating
3996  * at and ext_diff contains the number of extents that were added
3997  * or removed.
3998  */
3999 void
4000 xfs_iext_irec_update_extoffs(
4001 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4002 	int		erp_idx,	/* irec index to update */
4003 	int		ext_diff)	/* number of new extents */
4004 {
4005 	int		i;		/* loop counter */
4006 	int		nlists;		/* number of irec's (ex lists */
4007 
4008 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4009 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4010 	for (i = erp_idx; i < nlists; i++) {
4011 		ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4012 	}
4013 }
4014 
4015 /*
4016  * Test whether it is appropriate to check an inode for and free post EOF
4017  * blocks. The 'force' parameter determines whether we should also consider
4018  * regular files that are marked preallocated or append-only.
4019  */
4020 bool
4021 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
4022 {
4023 	/* prealloc/delalloc exists only on regular files */
4024 	if (!S_ISREG(ip->i_d.di_mode))
4025 		return false;
4026 
4027 	/*
4028 	 * Zero sized files with no cached pages and delalloc blocks will not
4029 	 * have speculative prealloc/delalloc blocks to remove.
4030 	 */
4031 	if (VFS_I(ip)->i_size == 0 &&
4032 	    VN_CACHED(VFS_I(ip)) == 0 &&
4033 	    ip->i_delayed_blks == 0)
4034 		return false;
4035 
4036 	/* If we haven't read in the extent list, then don't do it now. */
4037 	if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
4038 		return false;
4039 
4040 	/*
4041 	 * Do not free real preallocated or append-only files unless the file
4042 	 * has delalloc blocks and we are forced to remove them.
4043 	 */
4044 	if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
4045 		if (!force || ip->i_delayed_blks == 0)
4046 			return false;
4047 
4048 	return true;
4049 }
4050 
4051