xref: /linux/fs/xfs/xfs_inode.c (revision de2fe5e07d58424bc286fff3fd3c1b0bf933cd58)
1 /*
2  * Copyright (c) 2000-2003,2005 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 "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_imap.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_dir.h"
30 #include "xfs_dir2.h"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
46 #include "xfs_bmap.h"
47 #include "xfs_rw.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
52 #include "xfs_mac.h"
53 #include "xfs_acl.h"
54 
55 
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
58 kmem_zone_t *xfs_chashlist_zone;
59 
60 /*
61  * Used in xfs_itruncate().  This is the maximum number of extents
62  * freed from a file in a single transaction.
63  */
64 #define	XFS_ITRUNC_MAX_EXTENTS	2
65 
66 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
70 
71 
72 #ifdef DEBUG
73 /*
74  * Make sure that the extents in the given memory buffer
75  * are valid.
76  */
77 STATIC void
78 xfs_validate_extents(
79 	xfs_ifork_t		*ifp,
80 	int			nrecs,
81 	int			disk,
82 	xfs_exntfmt_t		fmt)
83 {
84 	xfs_bmbt_rec_t		*ep;
85 	xfs_bmbt_irec_t		irec;
86 	xfs_bmbt_rec_t		rec;
87 	int			i;
88 
89 	for (i = 0; i < nrecs; i++) {
90 		ep = xfs_iext_get_ext(ifp, i);
91 		rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
92 		rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
93 		if (disk)
94 			xfs_bmbt_disk_get_all(&rec, &irec);
95 		else
96 			xfs_bmbt_get_all(&rec, &irec);
97 		if (fmt == XFS_EXTFMT_NOSTATE)
98 			ASSERT(irec.br_state == XFS_EXT_NORM);
99 	}
100 }
101 #else /* DEBUG */
102 #define xfs_validate_extents(ifp, nrecs, disk, fmt)
103 #endif /* DEBUG */
104 
105 /*
106  * Check that none of the inode's in the buffer have a next
107  * unlinked field of 0.
108  */
109 #if defined(DEBUG)
110 void
111 xfs_inobp_check(
112 	xfs_mount_t	*mp,
113 	xfs_buf_t	*bp)
114 {
115 	int		i;
116 	int		j;
117 	xfs_dinode_t	*dip;
118 
119 	j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
120 
121 	for (i = 0; i < j; i++) {
122 		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
123 					i * mp->m_sb.sb_inodesize);
124 		if (!dip->di_next_unlinked)  {
125 			xfs_fs_cmn_err(CE_ALERT, mp,
126 				"Detected a bogus zero next_unlinked field in incore inode buffer 0x%p.  About to pop an ASSERT.",
127 				bp);
128 			ASSERT(dip->di_next_unlinked);
129 		}
130 	}
131 }
132 #endif
133 
134 /*
135  * This routine is called to map an inode number within a file
136  * system to the buffer containing the on-disk version of the
137  * inode.  It returns a pointer to the buffer containing the
138  * on-disk inode in the bpp parameter, and in the dip parameter
139  * it returns a pointer to the on-disk inode within that buffer.
140  *
141  * If a non-zero error is returned, then the contents of bpp and
142  * dipp are undefined.
143  *
144  * Use xfs_imap() to determine the size and location of the
145  * buffer to read from disk.
146  */
147 STATIC int
148 xfs_inotobp(
149 	xfs_mount_t	*mp,
150 	xfs_trans_t	*tp,
151 	xfs_ino_t	ino,
152 	xfs_dinode_t	**dipp,
153 	xfs_buf_t	**bpp,
154 	int		*offset)
155 {
156 	int		di_ok;
157 	xfs_imap_t	imap;
158 	xfs_buf_t	*bp;
159 	int		error;
160 	xfs_dinode_t	*dip;
161 
162 	/*
163 	 * Call the space management code to find the location of the
164 	 * inode on disk.
165 	 */
166 	imap.im_blkno = 0;
167 	error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
168 	if (error != 0) {
169 		cmn_err(CE_WARN,
170 	"xfs_inotobp: xfs_imap()  returned an "
171 	"error %d on %s.  Returning error.", error, mp->m_fsname);
172 		return error;
173 	}
174 
175 	/*
176 	 * If the inode number maps to a block outside the bounds of the
177 	 * file system then return NULL rather than calling read_buf
178 	 * and panicing when we get an error from the driver.
179 	 */
180 	if ((imap.im_blkno + imap.im_len) >
181 	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
182 		cmn_err(CE_WARN,
183 	"xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
184 	"of the file system %s.  Returning EINVAL.",
185 			(unsigned long long)imap.im_blkno,
186 			imap.im_len, mp->m_fsname);
187 		return XFS_ERROR(EINVAL);
188 	}
189 
190 	/*
191 	 * Read in the buffer.  If tp is NULL, xfs_trans_read_buf() will
192 	 * default to just a read_buf() call.
193 	 */
194 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
195 				   (int)imap.im_len, XFS_BUF_LOCK, &bp);
196 
197 	if (error) {
198 		cmn_err(CE_WARN,
199 	"xfs_inotobp: xfs_trans_read_buf()  returned an "
200 	"error %d on %s.  Returning error.", error, mp->m_fsname);
201 		return error;
202 	}
203 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
204 	di_ok =
205 		INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
206 		XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
207 	if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
208 			XFS_RANDOM_ITOBP_INOTOBP))) {
209 		XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
210 		xfs_trans_brelse(tp, bp);
211 		cmn_err(CE_WARN,
212 	"xfs_inotobp: XFS_TEST_ERROR()  returned an "
213 	"error on %s.  Returning EFSCORRUPTED.",  mp->m_fsname);
214 		return XFS_ERROR(EFSCORRUPTED);
215 	}
216 
217 	xfs_inobp_check(mp, bp);
218 
219 	/*
220 	 * Set *dipp to point to the on-disk inode in the buffer.
221 	 */
222 	*dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
223 	*bpp = bp;
224 	*offset = imap.im_boffset;
225 	return 0;
226 }
227 
228 
229 /*
230  * This routine is called to map an inode to the buffer containing
231  * the on-disk version of the inode.  It returns a pointer to the
232  * buffer containing the on-disk inode in the bpp parameter, and in
233  * the dip parameter it returns a pointer to the on-disk inode within
234  * that buffer.
235  *
236  * If a non-zero error is returned, then the contents of bpp and
237  * dipp are undefined.
238  *
239  * If the inode is new and has not yet been initialized, use xfs_imap()
240  * to determine the size and location of the buffer to read from disk.
241  * If the inode has already been mapped to its buffer and read in once,
242  * then use the mapping information stored in the inode rather than
243  * calling xfs_imap().  This allows us to avoid the overhead of looking
244  * at the inode btree for small block file systems (see xfs_dilocate()).
245  * We can tell whether the inode has been mapped in before by comparing
246  * its disk block address to 0.  Only uninitialized inodes will have
247  * 0 for the disk block address.
248  */
249 int
250 xfs_itobp(
251 	xfs_mount_t	*mp,
252 	xfs_trans_t	*tp,
253 	xfs_inode_t	*ip,
254 	xfs_dinode_t	**dipp,
255 	xfs_buf_t	**bpp,
256 	xfs_daddr_t	bno,
257 	uint		imap_flags)
258 {
259 	xfs_buf_t	*bp;
260 	int		error;
261 	xfs_imap_t	imap;
262 #ifdef __KERNEL__
263 	int		i;
264 	int		ni;
265 #endif
266 
267 	if (ip->i_blkno == (xfs_daddr_t)0) {
268 		/*
269 		 * Call the space management code to find the location of the
270 		 * inode on disk.
271 		 */
272 		imap.im_blkno = bno;
273 		if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
274 					XFS_IMAP_LOOKUP | imap_flags)))
275 			return error;
276 
277 		/*
278 		 * If the inode number maps to a block outside the bounds
279 		 * of the file system then return NULL rather than calling
280 		 * read_buf and panicing when we get an error from the
281 		 * driver.
282 		 */
283 		if ((imap.im_blkno + imap.im_len) >
284 		    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
285 #ifdef DEBUG
286 			xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
287 					"(imap.im_blkno (0x%llx) "
288 					"+ imap.im_len (0x%llx)) > "
289 					" XFS_FSB_TO_BB(mp, "
290 					"mp->m_sb.sb_dblocks) (0x%llx)",
291 					(unsigned long long) imap.im_blkno,
292 					(unsigned long long) imap.im_len,
293 					XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
294 #endif /* DEBUG */
295 			return XFS_ERROR(EINVAL);
296 		}
297 
298 		/*
299 		 * Fill in the fields in the inode that will be used to
300 		 * map the inode to its buffer from now on.
301 		 */
302 		ip->i_blkno = imap.im_blkno;
303 		ip->i_len = imap.im_len;
304 		ip->i_boffset = imap.im_boffset;
305 	} else {
306 		/*
307 		 * We've already mapped the inode once, so just use the
308 		 * mapping that we saved the first time.
309 		 */
310 		imap.im_blkno = ip->i_blkno;
311 		imap.im_len = ip->i_len;
312 		imap.im_boffset = ip->i_boffset;
313 	}
314 	ASSERT(bno == 0 || bno == imap.im_blkno);
315 
316 	/*
317 	 * Read in the buffer.  If tp is NULL, xfs_trans_read_buf() will
318 	 * default to just a read_buf() call.
319 	 */
320 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
321 				   (int)imap.im_len, XFS_BUF_LOCK, &bp);
322 
323 	if (error) {
324 #ifdef DEBUG
325 		xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
326 				"xfs_trans_read_buf() returned error %d, "
327 				"imap.im_blkno 0x%llx, imap.im_len 0x%llx",
328 				error, (unsigned long long) imap.im_blkno,
329 				(unsigned long long) imap.im_len);
330 #endif /* DEBUG */
331 		return error;
332 	}
333 #ifdef __KERNEL__
334 	/*
335 	 * Validate the magic number and version of every inode in the buffer
336 	 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
337 	 */
338 #ifdef DEBUG
339 	ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 :
340 		(BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog);
341 #else
342 	ni = (imap_flags & XFS_IMAP_BULKSTAT) ? 0 : 1;
343 #endif
344 	for (i = 0; i < ni; i++) {
345 		int		di_ok;
346 		xfs_dinode_t	*dip;
347 
348 		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
349 					(i << mp->m_sb.sb_inodelog));
350 		di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
351 			    XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
352 		if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
353 				 XFS_RANDOM_ITOBP_INOTOBP))) {
354 #ifdef DEBUG
355 			prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
356 				mp->m_ddev_targp,
357 				(unsigned long long)imap.im_blkno, i,
358 				INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
359 #endif
360 			XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
361 					     mp, dip);
362 			xfs_trans_brelse(tp, bp);
363 			return XFS_ERROR(EFSCORRUPTED);
364 		}
365 	}
366 #endif	/* __KERNEL__ */
367 
368 	xfs_inobp_check(mp, bp);
369 
370 	/*
371 	 * Mark the buffer as an inode buffer now that it looks good
372 	 */
373 	XFS_BUF_SET_VTYPE(bp, B_FS_INO);
374 
375 	/*
376 	 * Set *dipp to point to the on-disk inode in the buffer.
377 	 */
378 	*dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
379 	*bpp = bp;
380 	return 0;
381 }
382 
383 /*
384  * Move inode type and inode format specific information from the
385  * on-disk inode to the in-core inode.  For fifos, devs, and sockets
386  * this means set if_rdev to the proper value.  For files, directories,
387  * and symlinks this means to bring in the in-line data or extent
388  * pointers.  For a file in B-tree format, only the root is immediately
389  * brought in-core.  The rest will be in-lined in if_extents when it
390  * is first referenced (see xfs_iread_extents()).
391  */
392 STATIC int
393 xfs_iformat(
394 	xfs_inode_t		*ip,
395 	xfs_dinode_t		*dip)
396 {
397 	xfs_attr_shortform_t	*atp;
398 	int			size;
399 	int			error;
400 	xfs_fsize_t             di_size;
401 	ip->i_df.if_ext_max =
402 		XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
403 	error = 0;
404 
405 	if (unlikely(
406 	    INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
407 		INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
408 	    INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
409 		xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
410 			"corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
411 			(unsigned long long)ip->i_ino,
412 			(int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
413 			    + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
414 			(unsigned long long)
415 			INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
416 		XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
417 				     ip->i_mount, dip);
418 		return XFS_ERROR(EFSCORRUPTED);
419 	}
420 
421 	if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
422 		xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
423 			"corrupt dinode %Lu, forkoff = 0x%x.",
424 			(unsigned long long)ip->i_ino,
425 			(int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
426 		XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
427 				     ip->i_mount, dip);
428 		return XFS_ERROR(EFSCORRUPTED);
429 	}
430 
431 	switch (ip->i_d.di_mode & S_IFMT) {
432 	case S_IFIFO:
433 	case S_IFCHR:
434 	case S_IFBLK:
435 	case S_IFSOCK:
436 		if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
437 			XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
438 					      ip->i_mount, dip);
439 			return XFS_ERROR(EFSCORRUPTED);
440 		}
441 		ip->i_d.di_size = 0;
442 		ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
443 		break;
444 
445 	case S_IFREG:
446 	case S_IFLNK:
447 	case S_IFDIR:
448 		switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
449 		case XFS_DINODE_FMT_LOCAL:
450 			/*
451 			 * no local regular files yet
452 			 */
453 			if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
454 				xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
455 					"corrupt inode %Lu "
456 					"(local format for regular file).",
457 					(unsigned long long) ip->i_ino);
458 				XFS_CORRUPTION_ERROR("xfs_iformat(4)",
459 						     XFS_ERRLEVEL_LOW,
460 						     ip->i_mount, dip);
461 				return XFS_ERROR(EFSCORRUPTED);
462 			}
463 
464 			di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
465 			if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
466 				xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
467 					"corrupt inode %Lu "
468 					"(bad size %Ld for local inode).",
469 					(unsigned long long) ip->i_ino,
470 					(long long) di_size);
471 				XFS_CORRUPTION_ERROR("xfs_iformat(5)",
472 						     XFS_ERRLEVEL_LOW,
473 						     ip->i_mount, dip);
474 				return XFS_ERROR(EFSCORRUPTED);
475 			}
476 
477 			size = (int)di_size;
478 			error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
479 			break;
480 		case XFS_DINODE_FMT_EXTENTS:
481 			error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
482 			break;
483 		case XFS_DINODE_FMT_BTREE:
484 			error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
485 			break;
486 		default:
487 			XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
488 					 ip->i_mount);
489 			return XFS_ERROR(EFSCORRUPTED);
490 		}
491 		break;
492 
493 	default:
494 		XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
495 		return XFS_ERROR(EFSCORRUPTED);
496 	}
497 	if (error) {
498 		return error;
499 	}
500 	if (!XFS_DFORK_Q(dip))
501 		return 0;
502 	ASSERT(ip->i_afp == NULL);
503 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
504 	ip->i_afp->if_ext_max =
505 		XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
506 	switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
507 	case XFS_DINODE_FMT_LOCAL:
508 		atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
509 		size = be16_to_cpu(atp->hdr.totsize);
510 		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
511 		break;
512 	case XFS_DINODE_FMT_EXTENTS:
513 		error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
514 		break;
515 	case XFS_DINODE_FMT_BTREE:
516 		error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
517 		break;
518 	default:
519 		error = XFS_ERROR(EFSCORRUPTED);
520 		break;
521 	}
522 	if (error) {
523 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
524 		ip->i_afp = NULL;
525 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
526 	}
527 	return error;
528 }
529 
530 /*
531  * The file is in-lined in the on-disk inode.
532  * If it fits into if_inline_data, then copy
533  * it there, otherwise allocate a buffer for it
534  * and copy the data there.  Either way, set
535  * if_data to point at the data.
536  * If we allocate a buffer for the data, make
537  * sure that its size is a multiple of 4 and
538  * record the real size in i_real_bytes.
539  */
540 STATIC int
541 xfs_iformat_local(
542 	xfs_inode_t	*ip,
543 	xfs_dinode_t	*dip,
544 	int		whichfork,
545 	int		size)
546 {
547 	xfs_ifork_t	*ifp;
548 	int		real_size;
549 
550 	/*
551 	 * If the size is unreasonable, then something
552 	 * is wrong and we just bail out rather than crash in
553 	 * kmem_alloc() or memcpy() below.
554 	 */
555 	if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
556 		xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
557 			"corrupt inode %Lu "
558 			"(bad size %d for local fork, size = %d).",
559 			(unsigned long long) ip->i_ino, size,
560 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
561 		XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
562 				     ip->i_mount, dip);
563 		return XFS_ERROR(EFSCORRUPTED);
564 	}
565 	ifp = XFS_IFORK_PTR(ip, whichfork);
566 	real_size = 0;
567 	if (size == 0)
568 		ifp->if_u1.if_data = NULL;
569 	else if (size <= sizeof(ifp->if_u2.if_inline_data))
570 		ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
571 	else {
572 		real_size = roundup(size, 4);
573 		ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
574 	}
575 	ifp->if_bytes = size;
576 	ifp->if_real_bytes = real_size;
577 	if (size)
578 		memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
579 	ifp->if_flags &= ~XFS_IFEXTENTS;
580 	ifp->if_flags |= XFS_IFINLINE;
581 	return 0;
582 }
583 
584 /*
585  * The file consists of a set of extents all
586  * of which fit into the on-disk inode.
587  * If there are few enough extents to fit into
588  * the if_inline_ext, then copy them there.
589  * Otherwise allocate a buffer for them and copy
590  * them into it.  Either way, set if_extents
591  * to point at the extents.
592  */
593 STATIC int
594 xfs_iformat_extents(
595 	xfs_inode_t	*ip,
596 	xfs_dinode_t	*dip,
597 	int		whichfork)
598 {
599 	xfs_bmbt_rec_t	*ep, *dp;
600 	xfs_ifork_t	*ifp;
601 	int		nex;
602 	int		size;
603 	int		i;
604 
605 	ifp = XFS_IFORK_PTR(ip, whichfork);
606 	nex = XFS_DFORK_NEXTENTS(dip, whichfork);
607 	size = nex * (uint)sizeof(xfs_bmbt_rec_t);
608 
609 	/*
610 	 * If the number of extents is unreasonable, then something
611 	 * is wrong and we just bail out rather than crash in
612 	 * kmem_alloc() or memcpy() below.
613 	 */
614 	if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
615 		xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
616 			"corrupt inode %Lu ((a)extents = %d).",
617 			(unsigned long long) ip->i_ino, nex);
618 		XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
619 				     ip->i_mount, dip);
620 		return XFS_ERROR(EFSCORRUPTED);
621 	}
622 
623 	ifp->if_real_bytes = 0;
624 	if (nex == 0)
625 		ifp->if_u1.if_extents = NULL;
626 	else if (nex <= XFS_INLINE_EXTS)
627 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
628 	else
629 		xfs_iext_add(ifp, 0, nex);
630 
631 	ifp->if_bytes = size;
632 	if (size) {
633 		dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
634 		xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
635 		for (i = 0; i < nex; i++, dp++) {
636 			ep = xfs_iext_get_ext(ifp, i);
637 			ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
638 								ARCH_CONVERT);
639 			ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
640 								ARCH_CONVERT);
641 		}
642 		xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
643 			whichfork);
644 		if (whichfork != XFS_DATA_FORK ||
645 			XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
646 				if (unlikely(xfs_check_nostate_extents(
647 				    ifp, 0, nex))) {
648 					XFS_ERROR_REPORT("xfs_iformat_extents(2)",
649 							 XFS_ERRLEVEL_LOW,
650 							 ip->i_mount);
651 					return XFS_ERROR(EFSCORRUPTED);
652 				}
653 	}
654 	ifp->if_flags |= XFS_IFEXTENTS;
655 	return 0;
656 }
657 
658 /*
659  * The file has too many extents to fit into
660  * the inode, so they are in B-tree format.
661  * Allocate a buffer for the root of the B-tree
662  * and copy the root into it.  The i_extents
663  * field will remain NULL until all of the
664  * extents are read in (when they are needed).
665  */
666 STATIC int
667 xfs_iformat_btree(
668 	xfs_inode_t		*ip,
669 	xfs_dinode_t		*dip,
670 	int			whichfork)
671 {
672 	xfs_bmdr_block_t	*dfp;
673 	xfs_ifork_t		*ifp;
674 	/* REFERENCED */
675 	int			nrecs;
676 	int			size;
677 
678 	ifp = XFS_IFORK_PTR(ip, whichfork);
679 	dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
680 	size = XFS_BMAP_BROOT_SPACE(dfp);
681 	nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
682 
683 	/*
684 	 * blow out if -- fork has less extents than can fit in
685 	 * fork (fork shouldn't be a btree format), root btree
686 	 * block has more records than can fit into the fork,
687 	 * or the number of extents is greater than the number of
688 	 * blocks.
689 	 */
690 	if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
691 	    || XFS_BMDR_SPACE_CALC(nrecs) >
692 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
693 	    || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
694 		xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
695 			"corrupt inode %Lu (btree).",
696 			(unsigned long long) ip->i_ino);
697 		XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
698 				 ip->i_mount);
699 		return XFS_ERROR(EFSCORRUPTED);
700 	}
701 
702 	ifp->if_broot_bytes = size;
703 	ifp->if_broot = kmem_alloc(size, KM_SLEEP);
704 	ASSERT(ifp->if_broot != NULL);
705 	/*
706 	 * Copy and convert from the on-disk structure
707 	 * to the in-memory structure.
708 	 */
709 	xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
710 		ifp->if_broot, size);
711 	ifp->if_flags &= ~XFS_IFEXTENTS;
712 	ifp->if_flags |= XFS_IFBROOT;
713 
714 	return 0;
715 }
716 
717 /*
718  * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
719  * and native format
720  *
721  * buf  = on-disk representation
722  * dip  = native representation
723  * dir  = direction - +ve -> disk to native
724  *                    -ve -> native to disk
725  */
726 void
727 xfs_xlate_dinode_core(
728 	xfs_caddr_t		buf,
729 	xfs_dinode_core_t	*dip,
730 	int			dir)
731 {
732 	xfs_dinode_core_t	*buf_core = (xfs_dinode_core_t *)buf;
733 	xfs_dinode_core_t	*mem_core = (xfs_dinode_core_t *)dip;
734 	xfs_arch_t		arch = ARCH_CONVERT;
735 
736 	ASSERT(dir);
737 
738 	INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
739 	INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
740 	INT_XLATE(buf_core->di_version,	mem_core->di_version, dir, arch);
741 	INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
742 	INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
743 	INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
744 	INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
745 	INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
746 	INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
747 
748 	if (dir > 0) {
749 		memcpy(mem_core->di_pad, buf_core->di_pad,
750 			sizeof(buf_core->di_pad));
751 	} else {
752 		memcpy(buf_core->di_pad, mem_core->di_pad,
753 			sizeof(buf_core->di_pad));
754 	}
755 
756 	INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
757 
758 	INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
759 			dir, arch);
760 	INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
761 			dir, arch);
762 	INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
763 			dir, arch);
764 	INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
765 			dir, arch);
766 	INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
767 			dir, arch);
768 	INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
769 			dir, arch);
770 	INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
771 	INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
772 	INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
773 	INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
774 	INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
775 	INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
776 	INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
777 	INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
778 	INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
779 	INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
780 	INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
781 }
782 
783 STATIC uint
784 _xfs_dic2xflags(
785 	xfs_dinode_core_t	*dic,
786 	__uint16_t		di_flags)
787 {
788 	uint			flags = 0;
789 
790 	if (di_flags & XFS_DIFLAG_ANY) {
791 		if (di_flags & XFS_DIFLAG_REALTIME)
792 			flags |= XFS_XFLAG_REALTIME;
793 		if (di_flags & XFS_DIFLAG_PREALLOC)
794 			flags |= XFS_XFLAG_PREALLOC;
795 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
796 			flags |= XFS_XFLAG_IMMUTABLE;
797 		if (di_flags & XFS_DIFLAG_APPEND)
798 			flags |= XFS_XFLAG_APPEND;
799 		if (di_flags & XFS_DIFLAG_SYNC)
800 			flags |= XFS_XFLAG_SYNC;
801 		if (di_flags & XFS_DIFLAG_NOATIME)
802 			flags |= XFS_XFLAG_NOATIME;
803 		if (di_flags & XFS_DIFLAG_NODUMP)
804 			flags |= XFS_XFLAG_NODUMP;
805 		if (di_flags & XFS_DIFLAG_RTINHERIT)
806 			flags |= XFS_XFLAG_RTINHERIT;
807 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
808 			flags |= XFS_XFLAG_PROJINHERIT;
809 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
810 			flags |= XFS_XFLAG_NOSYMLINKS;
811 		if (di_flags & XFS_DIFLAG_EXTSIZE)
812 			flags |= XFS_XFLAG_EXTSIZE;
813 		if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
814 			flags |= XFS_XFLAG_EXTSZINHERIT;
815 	}
816 
817 	return flags;
818 }
819 
820 uint
821 xfs_ip2xflags(
822 	xfs_inode_t		*ip)
823 {
824 	xfs_dinode_core_t	*dic = &ip->i_d;
825 
826 	return _xfs_dic2xflags(dic, dic->di_flags) |
827 		(XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
828 }
829 
830 uint
831 xfs_dic2xflags(
832 	xfs_dinode_core_t	*dic)
833 {
834 	return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
835 		(XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
836 }
837 
838 /*
839  * Given a mount structure and an inode number, return a pointer
840  * to a newly allocated in-core inode corresponding to the given
841  * inode number.
842  *
843  * Initialize the inode's attributes and extent pointers if it
844  * already has them (it will not if the inode has no links).
845  */
846 int
847 xfs_iread(
848 	xfs_mount_t	*mp,
849 	xfs_trans_t	*tp,
850 	xfs_ino_t	ino,
851 	xfs_inode_t	**ipp,
852 	xfs_daddr_t	bno)
853 {
854 	xfs_buf_t	*bp;
855 	xfs_dinode_t	*dip;
856 	xfs_inode_t	*ip;
857 	int		error;
858 
859 	ASSERT(xfs_inode_zone != NULL);
860 
861 	ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
862 	ip->i_ino = ino;
863 	ip->i_mount = mp;
864 
865 	/*
866 	 * Get pointer's to the on-disk inode and the buffer containing it.
867 	 * If the inode number refers to a block outside the file system
868 	 * then xfs_itobp() will return NULL.  In this case we should
869 	 * return NULL as well.  Set i_blkno to 0 so that xfs_itobp() will
870 	 * know that this is a new incore inode.
871 	 */
872 	error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0);
873 	if (error) {
874 		kmem_zone_free(xfs_inode_zone, ip);
875 		return error;
876 	}
877 
878 	/*
879 	 * Initialize inode's trace buffers.
880 	 * Do this before xfs_iformat in case it adds entries.
881 	 */
882 #ifdef XFS_BMAP_TRACE
883 	ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
884 #endif
885 #ifdef XFS_BMBT_TRACE
886 	ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
887 #endif
888 #ifdef XFS_RW_TRACE
889 	ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
890 #endif
891 #ifdef XFS_ILOCK_TRACE
892 	ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
893 #endif
894 #ifdef XFS_DIR2_TRACE
895 	ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
896 #endif
897 
898 	/*
899 	 * If we got something that isn't an inode it means someone
900 	 * (nfs or dmi) has a stale handle.
901 	 */
902 	if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
903 		kmem_zone_free(xfs_inode_zone, ip);
904 		xfs_trans_brelse(tp, bp);
905 #ifdef DEBUG
906 		xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
907 				"dip->di_core.di_magic (0x%x) != "
908 				"XFS_DINODE_MAGIC (0x%x)",
909 				INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
910 				XFS_DINODE_MAGIC);
911 #endif /* DEBUG */
912 		return XFS_ERROR(EINVAL);
913 	}
914 
915 	/*
916 	 * If the on-disk inode is already linked to a directory
917 	 * entry, copy all of the inode into the in-core inode.
918 	 * xfs_iformat() handles copying in the inode format
919 	 * specific information.
920 	 * Otherwise, just get the truly permanent information.
921 	 */
922 	if (dip->di_core.di_mode) {
923 		xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
924 		     &(ip->i_d), 1);
925 		error = xfs_iformat(ip, dip);
926 		if (error)  {
927 			kmem_zone_free(xfs_inode_zone, ip);
928 			xfs_trans_brelse(tp, bp);
929 #ifdef DEBUG
930 			xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
931 					"xfs_iformat() returned error %d",
932 					error);
933 #endif /* DEBUG */
934 			return error;
935 		}
936 	} else {
937 		ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
938 		ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
939 		ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
940 		ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
941 		/*
942 		 * Make sure to pull in the mode here as well in
943 		 * case the inode is released without being used.
944 		 * This ensures that xfs_inactive() will see that
945 		 * the inode is already free and not try to mess
946 		 * with the uninitialized part of it.
947 		 */
948 		ip->i_d.di_mode = 0;
949 		/*
950 		 * Initialize the per-fork minima and maxima for a new
951 		 * inode here.  xfs_iformat will do it for old inodes.
952 		 */
953 		ip->i_df.if_ext_max =
954 			XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
955 	}
956 
957 	INIT_LIST_HEAD(&ip->i_reclaim);
958 
959 	/*
960 	 * The inode format changed when we moved the link count and
961 	 * made it 32 bits long.  If this is an old format inode,
962 	 * convert it in memory to look like a new one.  If it gets
963 	 * flushed to disk we will convert back before flushing or
964 	 * logging it.  We zero out the new projid field and the old link
965 	 * count field.  We'll handle clearing the pad field (the remains
966 	 * of the old uuid field) when we actually convert the inode to
967 	 * the new format. We don't change the version number so that we
968 	 * can distinguish this from a real new format inode.
969 	 */
970 	if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
971 		ip->i_d.di_nlink = ip->i_d.di_onlink;
972 		ip->i_d.di_onlink = 0;
973 		ip->i_d.di_projid = 0;
974 	}
975 
976 	ip->i_delayed_blks = 0;
977 
978 	/*
979 	 * Mark the buffer containing the inode as something to keep
980 	 * around for a while.  This helps to keep recently accessed
981 	 * meta-data in-core longer.
982 	 */
983 	 XFS_BUF_SET_REF(bp, XFS_INO_REF);
984 
985 	/*
986 	 * Use xfs_trans_brelse() to release the buffer containing the
987 	 * on-disk inode, because it was acquired with xfs_trans_read_buf()
988 	 * in xfs_itobp() above.  If tp is NULL, this is just a normal
989 	 * brelse().  If we're within a transaction, then xfs_trans_brelse()
990 	 * will only release the buffer if it is not dirty within the
991 	 * transaction.  It will be OK to release the buffer in this case,
992 	 * because inodes on disk are never destroyed and we will be
993 	 * locking the new in-core inode before putting it in the hash
994 	 * table where other processes can find it.  Thus we don't have
995 	 * to worry about the inode being changed just because we released
996 	 * the buffer.
997 	 */
998 	xfs_trans_brelse(tp, bp);
999 	*ipp = ip;
1000 	return 0;
1001 }
1002 
1003 /*
1004  * Read in extents from a btree-format inode.
1005  * Allocate and fill in if_extents.  Real work is done in xfs_bmap.c.
1006  */
1007 int
1008 xfs_iread_extents(
1009 	xfs_trans_t	*tp,
1010 	xfs_inode_t	*ip,
1011 	int		whichfork)
1012 {
1013 	int		error;
1014 	xfs_ifork_t	*ifp;
1015 	xfs_extnum_t	nextents;
1016 	size_t		size;
1017 
1018 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1019 		XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1020 				 ip->i_mount);
1021 		return XFS_ERROR(EFSCORRUPTED);
1022 	}
1023 	nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1024 	size = nextents * sizeof(xfs_bmbt_rec_t);
1025 	ifp = XFS_IFORK_PTR(ip, whichfork);
1026 
1027 	/*
1028 	 * We know that the size is valid (it's checked in iformat_btree)
1029 	 */
1030 	ifp->if_lastex = NULLEXTNUM;
1031 	ifp->if_bytes = ifp->if_real_bytes = 0;
1032 	ifp->if_flags |= XFS_IFEXTENTS;
1033 	xfs_iext_add(ifp, 0, nextents);
1034 	error = xfs_bmap_read_extents(tp, ip, whichfork);
1035 	if (error) {
1036 		xfs_iext_destroy(ifp);
1037 		ifp->if_flags &= ~XFS_IFEXTENTS;
1038 		return error;
1039 	}
1040 	xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
1041 	return 0;
1042 }
1043 
1044 /*
1045  * Allocate an inode on disk and return a copy of its in-core version.
1046  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
1047  * appropriately within the inode.  The uid and gid for the inode are
1048  * set according to the contents of the given cred structure.
1049  *
1050  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1051  * has a free inode available, call xfs_iget()
1052  * to obtain the in-core version of the allocated inode.  Finally,
1053  * fill in the inode and log its initial contents.  In this case,
1054  * ialloc_context would be set to NULL and call_again set to false.
1055  *
1056  * If xfs_dialloc() does not have an available inode,
1057  * it will replenish its supply by doing an allocation. Since we can
1058  * only do one allocation within a transaction without deadlocks, we
1059  * must commit the current transaction before returning the inode itself.
1060  * In this case, therefore, we will set call_again to true and return.
1061  * The caller should then commit the current transaction, start a new
1062  * transaction, and call xfs_ialloc() again to actually get the inode.
1063  *
1064  * To ensure that some other process does not grab the inode that
1065  * was allocated during the first call to xfs_ialloc(), this routine
1066  * also returns the [locked] bp pointing to the head of the freelist
1067  * as ialloc_context.  The caller should hold this buffer across
1068  * the commit and pass it back into this routine on the second call.
1069  */
1070 int
1071 xfs_ialloc(
1072 	xfs_trans_t	*tp,
1073 	xfs_inode_t	*pip,
1074 	mode_t		mode,
1075 	xfs_nlink_t	nlink,
1076 	xfs_dev_t	rdev,
1077 	cred_t		*cr,
1078 	xfs_prid_t	prid,
1079 	int		okalloc,
1080 	xfs_buf_t	**ialloc_context,
1081 	boolean_t	*call_again,
1082 	xfs_inode_t	**ipp)
1083 {
1084 	xfs_ino_t	ino;
1085 	xfs_inode_t	*ip;
1086 	vnode_t		*vp;
1087 	uint		flags;
1088 	int		error;
1089 
1090 	/*
1091 	 * Call the space management code to pick
1092 	 * the on-disk inode to be allocated.
1093 	 */
1094 	error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1095 			    ialloc_context, call_again, &ino);
1096 	if (error != 0) {
1097 		return error;
1098 	}
1099 	if (*call_again || ino == NULLFSINO) {
1100 		*ipp = NULL;
1101 		return 0;
1102 	}
1103 	ASSERT(*ialloc_context == NULL);
1104 
1105 	/*
1106 	 * Get the in-core inode with the lock held exclusively.
1107 	 * This is because we're setting fields here we need
1108 	 * to prevent others from looking at until we're done.
1109 	 */
1110 	error = xfs_trans_iget(tp->t_mountp, tp, ino,
1111 			IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1112 	if (error != 0) {
1113 		return error;
1114 	}
1115 	ASSERT(ip != NULL);
1116 
1117 	vp = XFS_ITOV(ip);
1118 	ip->i_d.di_mode = (__uint16_t)mode;
1119 	ip->i_d.di_onlink = 0;
1120 	ip->i_d.di_nlink = nlink;
1121 	ASSERT(ip->i_d.di_nlink == nlink);
1122 	ip->i_d.di_uid = current_fsuid(cr);
1123 	ip->i_d.di_gid = current_fsgid(cr);
1124 	ip->i_d.di_projid = prid;
1125 	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1126 
1127 	/*
1128 	 * If the superblock version is up to where we support new format
1129 	 * inodes and this is currently an old format inode, then change
1130 	 * the inode version number now.  This way we only do the conversion
1131 	 * here rather than here and in the flush/logging code.
1132 	 */
1133 	if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1134 	    ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1135 		ip->i_d.di_version = XFS_DINODE_VERSION_2;
1136 		/*
1137 		 * We've already zeroed the old link count, the projid field,
1138 		 * and the pad field.
1139 		 */
1140 	}
1141 
1142 	/*
1143 	 * Project ids won't be stored on disk if we are using a version 1 inode.
1144 	 */
1145 	if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1146 		xfs_bump_ino_vers2(tp, ip);
1147 
1148 	if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1149 		ip->i_d.di_gid = pip->i_d.di_gid;
1150 		if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1151 			ip->i_d.di_mode |= S_ISGID;
1152 		}
1153 	}
1154 
1155 	/*
1156 	 * If the group ID of the new file does not match the effective group
1157 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1158 	 * (and only if the irix_sgid_inherit compatibility variable is set).
1159 	 */
1160 	if ((irix_sgid_inherit) &&
1161 	    (ip->i_d.di_mode & S_ISGID) &&
1162 	    (!in_group_p((gid_t)ip->i_d.di_gid))) {
1163 		ip->i_d.di_mode &= ~S_ISGID;
1164 	}
1165 
1166 	ip->i_d.di_size = 0;
1167 	ip->i_d.di_nextents = 0;
1168 	ASSERT(ip->i_d.di_nblocks == 0);
1169 	xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1170 	/*
1171 	 * di_gen will have been taken care of in xfs_iread.
1172 	 */
1173 	ip->i_d.di_extsize = 0;
1174 	ip->i_d.di_dmevmask = 0;
1175 	ip->i_d.di_dmstate = 0;
1176 	ip->i_d.di_flags = 0;
1177 	flags = XFS_ILOG_CORE;
1178 	switch (mode & S_IFMT) {
1179 	case S_IFIFO:
1180 	case S_IFCHR:
1181 	case S_IFBLK:
1182 	case S_IFSOCK:
1183 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1184 		ip->i_df.if_u2.if_rdev = rdev;
1185 		ip->i_df.if_flags = 0;
1186 		flags |= XFS_ILOG_DEV;
1187 		break;
1188 	case S_IFREG:
1189 	case S_IFDIR:
1190 		if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1191 			uint	di_flags = 0;
1192 
1193 			if ((mode & S_IFMT) == S_IFDIR) {
1194 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1195 					di_flags |= XFS_DIFLAG_RTINHERIT;
1196 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1197 					di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1198 					ip->i_d.di_extsize = pip->i_d.di_extsize;
1199 				}
1200 			} else if ((mode & S_IFMT) == S_IFREG) {
1201 				if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1202 					di_flags |= XFS_DIFLAG_REALTIME;
1203 					ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1204 				}
1205 				if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1206 					di_flags |= XFS_DIFLAG_EXTSIZE;
1207 					ip->i_d.di_extsize = pip->i_d.di_extsize;
1208 				}
1209 			}
1210 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1211 			    xfs_inherit_noatime)
1212 				di_flags |= XFS_DIFLAG_NOATIME;
1213 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1214 			    xfs_inherit_nodump)
1215 				di_flags |= XFS_DIFLAG_NODUMP;
1216 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1217 			    xfs_inherit_sync)
1218 				di_flags |= XFS_DIFLAG_SYNC;
1219 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1220 			    xfs_inherit_nosymlinks)
1221 				di_flags |= XFS_DIFLAG_NOSYMLINKS;
1222 			if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1223 				di_flags |= XFS_DIFLAG_PROJINHERIT;
1224 			ip->i_d.di_flags |= di_flags;
1225 		}
1226 		/* FALLTHROUGH */
1227 	case S_IFLNK:
1228 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1229 		ip->i_df.if_flags = XFS_IFEXTENTS;
1230 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1231 		ip->i_df.if_u1.if_extents = NULL;
1232 		break;
1233 	default:
1234 		ASSERT(0);
1235 	}
1236 	/*
1237 	 * Attribute fork settings for new inode.
1238 	 */
1239 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1240 	ip->i_d.di_anextents = 0;
1241 
1242 	/*
1243 	 * Log the new values stuffed into the inode.
1244 	 */
1245 	xfs_trans_log_inode(tp, ip, flags);
1246 
1247 	/* now that we have an i_mode  we can set Linux inode ops (& unlock) */
1248 	VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1249 
1250 	*ipp = ip;
1251 	return 0;
1252 }
1253 
1254 /*
1255  * Check to make sure that there are no blocks allocated to the
1256  * file beyond the size of the file.  We don't check this for
1257  * files with fixed size extents or real time extents, but we
1258  * at least do it for regular files.
1259  */
1260 #ifdef DEBUG
1261 void
1262 xfs_isize_check(
1263 	xfs_mount_t	*mp,
1264 	xfs_inode_t	*ip,
1265 	xfs_fsize_t	isize)
1266 {
1267 	xfs_fileoff_t	map_first;
1268 	int		nimaps;
1269 	xfs_bmbt_irec_t	imaps[2];
1270 
1271 	if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1272 		return;
1273 
1274 	if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
1275 		return;
1276 
1277 	nimaps = 2;
1278 	map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1279 	/*
1280 	 * The filesystem could be shutting down, so bmapi may return
1281 	 * an error.
1282 	 */
1283 	if (xfs_bmapi(NULL, ip, map_first,
1284 			 (XFS_B_TO_FSB(mp,
1285 				       (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1286 			  map_first),
1287 			 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1288 			 NULL))
1289 	    return;
1290 	ASSERT(nimaps == 1);
1291 	ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1292 }
1293 #endif	/* DEBUG */
1294 
1295 /*
1296  * Calculate the last possible buffered byte in a file.  This must
1297  * include data that was buffered beyond the EOF by the write code.
1298  * This also needs to deal with overflowing the xfs_fsize_t type
1299  * which can happen for sizes near the limit.
1300  *
1301  * We also need to take into account any blocks beyond the EOF.  It
1302  * may be the case that they were buffered by a write which failed.
1303  * In that case the pages will still be in memory, but the inode size
1304  * will never have been updated.
1305  */
1306 xfs_fsize_t
1307 xfs_file_last_byte(
1308 	xfs_inode_t	*ip)
1309 {
1310 	xfs_mount_t	*mp;
1311 	xfs_fsize_t	last_byte;
1312 	xfs_fileoff_t	last_block;
1313 	xfs_fileoff_t	size_last_block;
1314 	int		error;
1315 
1316 	ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1317 
1318 	mp = ip->i_mount;
1319 	/*
1320 	 * Only check for blocks beyond the EOF if the extents have
1321 	 * been read in.  This eliminates the need for the inode lock,
1322 	 * and it also saves us from looking when it really isn't
1323 	 * necessary.
1324 	 */
1325 	if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1326 		error = xfs_bmap_last_offset(NULL, ip, &last_block,
1327 			XFS_DATA_FORK);
1328 		if (error) {
1329 			last_block = 0;
1330 		}
1331 	} else {
1332 		last_block = 0;
1333 	}
1334 	size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1335 	last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1336 
1337 	last_byte = XFS_FSB_TO_B(mp, last_block);
1338 	if (last_byte < 0) {
1339 		return XFS_MAXIOFFSET(mp);
1340 	}
1341 	last_byte += (1 << mp->m_writeio_log);
1342 	if (last_byte < 0) {
1343 		return XFS_MAXIOFFSET(mp);
1344 	}
1345 	return last_byte;
1346 }
1347 
1348 #if defined(XFS_RW_TRACE)
1349 STATIC void
1350 xfs_itrunc_trace(
1351 	int		tag,
1352 	xfs_inode_t	*ip,
1353 	int		flag,
1354 	xfs_fsize_t	new_size,
1355 	xfs_off_t	toss_start,
1356 	xfs_off_t	toss_finish)
1357 {
1358 	if (ip->i_rwtrace == NULL) {
1359 		return;
1360 	}
1361 
1362 	ktrace_enter(ip->i_rwtrace,
1363 		     (void*)((long)tag),
1364 		     (void*)ip,
1365 		     (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1366 		     (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1367 		     (void*)((long)flag),
1368 		     (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1369 		     (void*)(unsigned long)(new_size & 0xffffffff),
1370 		     (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1371 		     (void*)(unsigned long)(toss_start & 0xffffffff),
1372 		     (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1373 		     (void*)(unsigned long)(toss_finish & 0xffffffff),
1374 		     (void*)(unsigned long)current_cpu(),
1375 		     (void*)(unsigned long)current_pid(),
1376 		     (void*)NULL,
1377 		     (void*)NULL,
1378 		     (void*)NULL);
1379 }
1380 #else
1381 #define	xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1382 #endif
1383 
1384 /*
1385  * Start the truncation of the file to new_size.  The new size
1386  * must be smaller than the current size.  This routine will
1387  * clear the buffer and page caches of file data in the removed
1388  * range, and xfs_itruncate_finish() will remove the underlying
1389  * disk blocks.
1390  *
1391  * The inode must have its I/O lock locked EXCLUSIVELY, and it
1392  * must NOT have the inode lock held at all.  This is because we're
1393  * calling into the buffer/page cache code and we can't hold the
1394  * inode lock when we do so.
1395  *
1396  * We need to wait for any direct I/Os in flight to complete before we
1397  * proceed with the truncate. This is needed to prevent the extents
1398  * being read or written by the direct I/Os from being removed while the
1399  * I/O is in flight as there is no other method of synchronising
1400  * direct I/O with the truncate operation.  Also, because we hold
1401  * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1402  * started until the truncate completes and drops the lock. Essentially,
1403  * the vn_iowait() call forms an I/O barrier that provides strict ordering
1404  * between direct I/Os and the truncate operation.
1405  *
1406  * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1407  * or XFS_ITRUNC_MAYBE.  The XFS_ITRUNC_MAYBE value should be used
1408  * in the case that the caller is locking things out of order and
1409  * may not be able to call xfs_itruncate_finish() with the inode lock
1410  * held without dropping the I/O lock.  If the caller must drop the
1411  * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1412  * must be called again with all the same restrictions as the initial
1413  * call.
1414  */
1415 void
1416 xfs_itruncate_start(
1417 	xfs_inode_t	*ip,
1418 	uint		flags,
1419 	xfs_fsize_t	new_size)
1420 {
1421 	xfs_fsize_t	last_byte;
1422 	xfs_off_t	toss_start;
1423 	xfs_mount_t	*mp;
1424 	vnode_t		*vp;
1425 
1426 	ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1427 	ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1428 	ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1429 	       (flags == XFS_ITRUNC_MAYBE));
1430 
1431 	mp = ip->i_mount;
1432 	vp = XFS_ITOV(ip);
1433 
1434 	vn_iowait(vp);  /* wait for the completion of any pending DIOs */
1435 
1436 	/*
1437 	 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1438 	 * overlapping the region being removed.  We have to use
1439 	 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1440 	 * caller may not be able to finish the truncate without
1441 	 * dropping the inode's I/O lock.  Make sure
1442 	 * to catch any pages brought in by buffers overlapping
1443 	 * the EOF by searching out beyond the isize by our
1444 	 * block size. We round new_size up to a block boundary
1445 	 * so that we don't toss things on the same block as
1446 	 * new_size but before it.
1447 	 *
1448 	 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1449 	 * call remapf() over the same region if the file is mapped.
1450 	 * This frees up mapped file references to the pages in the
1451 	 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1452 	 * that we get the latest mapped changes flushed out.
1453 	 */
1454 	toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1455 	toss_start = XFS_FSB_TO_B(mp, toss_start);
1456 	if (toss_start < 0) {
1457 		/*
1458 		 * The place to start tossing is beyond our maximum
1459 		 * file size, so there is no way that the data extended
1460 		 * out there.
1461 		 */
1462 		return;
1463 	}
1464 	last_byte = xfs_file_last_byte(ip);
1465 	xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1466 			 last_byte);
1467 	if (last_byte > toss_start) {
1468 		if (flags & XFS_ITRUNC_DEFINITE) {
1469 			VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1470 		} else {
1471 			VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1472 		}
1473 	}
1474 
1475 #ifdef DEBUG
1476 	if (new_size == 0) {
1477 		ASSERT(VN_CACHED(vp) == 0);
1478 	}
1479 #endif
1480 }
1481 
1482 /*
1483  * Shrink the file to the given new_size.  The new
1484  * size must be smaller than the current size.
1485  * This will free up the underlying blocks
1486  * in the removed range after a call to xfs_itruncate_start()
1487  * or xfs_atruncate_start().
1488  *
1489  * The transaction passed to this routine must have made
1490  * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1491  * This routine may commit the given transaction and
1492  * start new ones, so make sure everything involved in
1493  * the transaction is tidy before calling here.
1494  * Some transaction will be returned to the caller to be
1495  * committed.  The incoming transaction must already include
1496  * the inode, and both inode locks must be held exclusively.
1497  * The inode must also be "held" within the transaction.  On
1498  * return the inode will be "held" within the returned transaction.
1499  * This routine does NOT require any disk space to be reserved
1500  * for it within the transaction.
1501  *
1502  * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1503  * and it indicates the fork which is to be truncated.  For the
1504  * attribute fork we only support truncation to size 0.
1505  *
1506  * We use the sync parameter to indicate whether or not the first
1507  * transaction we perform might have to be synchronous.  For the attr fork,
1508  * it needs to be so if the unlink of the inode is not yet known to be
1509  * permanent in the log.  This keeps us from freeing and reusing the
1510  * blocks of the attribute fork before the unlink of the inode becomes
1511  * permanent.
1512  *
1513  * For the data fork, we normally have to run synchronously if we're
1514  * being called out of the inactive path or we're being called
1515  * out of the create path where we're truncating an existing file.
1516  * Either way, the truncate needs to be sync so blocks don't reappear
1517  * in the file with altered data in case of a crash.  wsync filesystems
1518  * can run the first case async because anything that shrinks the inode
1519  * has to run sync so by the time we're called here from inactive, the
1520  * inode size is permanently set to 0.
1521  *
1522  * Calls from the truncate path always need to be sync unless we're
1523  * in a wsync filesystem and the file has already been unlinked.
1524  *
1525  * The caller is responsible for correctly setting the sync parameter.
1526  * It gets too hard for us to guess here which path we're being called
1527  * out of just based on inode state.
1528  */
1529 int
1530 xfs_itruncate_finish(
1531 	xfs_trans_t	**tp,
1532 	xfs_inode_t	*ip,
1533 	xfs_fsize_t	new_size,
1534 	int		fork,
1535 	int		sync)
1536 {
1537 	xfs_fsblock_t	first_block;
1538 	xfs_fileoff_t	first_unmap_block;
1539 	xfs_fileoff_t	last_block;
1540 	xfs_filblks_t	unmap_len=0;
1541 	xfs_mount_t	*mp;
1542 	xfs_trans_t	*ntp;
1543 	int		done;
1544 	int		committed;
1545 	xfs_bmap_free_t	free_list;
1546 	int		error;
1547 
1548 	ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1549 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1550 	ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1551 	ASSERT(*tp != NULL);
1552 	ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1553 	ASSERT(ip->i_transp == *tp);
1554 	ASSERT(ip->i_itemp != NULL);
1555 	ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1556 
1557 
1558 	ntp = *tp;
1559 	mp = (ntp)->t_mountp;
1560 	ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1561 
1562 	/*
1563 	 * We only support truncating the entire attribute fork.
1564 	 */
1565 	if (fork == XFS_ATTR_FORK) {
1566 		new_size = 0LL;
1567 	}
1568 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1569 	xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1570 	/*
1571 	 * The first thing we do is set the size to new_size permanently
1572 	 * on disk.  This way we don't have to worry about anyone ever
1573 	 * being able to look at the data being freed even in the face
1574 	 * of a crash.  What we're getting around here is the case where
1575 	 * we free a block, it is allocated to another file, it is written
1576 	 * to, and then we crash.  If the new data gets written to the
1577 	 * file but the log buffers containing the free and reallocation
1578 	 * don't, then we'd end up with garbage in the blocks being freed.
1579 	 * As long as we make the new_size permanent before actually
1580 	 * freeing any blocks it doesn't matter if they get writtten to.
1581 	 *
1582 	 * The callers must signal into us whether or not the size
1583 	 * setting here must be synchronous.  There are a few cases
1584 	 * where it doesn't have to be synchronous.  Those cases
1585 	 * occur if the file is unlinked and we know the unlink is
1586 	 * permanent or if the blocks being truncated are guaranteed
1587 	 * to be beyond the inode eof (regardless of the link count)
1588 	 * and the eof value is permanent.  Both of these cases occur
1589 	 * only on wsync-mounted filesystems.  In those cases, we're
1590 	 * guaranteed that no user will ever see the data in the blocks
1591 	 * that are being truncated so the truncate can run async.
1592 	 * In the free beyond eof case, the file may wind up with
1593 	 * more blocks allocated to it than it needs if we crash
1594 	 * and that won't get fixed until the next time the file
1595 	 * is re-opened and closed but that's ok as that shouldn't
1596 	 * be too many blocks.
1597 	 *
1598 	 * However, we can't just make all wsync xactions run async
1599 	 * because there's one call out of the create path that needs
1600 	 * to run sync where it's truncating an existing file to size
1601 	 * 0 whose size is > 0.
1602 	 *
1603 	 * It's probably possible to come up with a test in this
1604 	 * routine that would correctly distinguish all the above
1605 	 * cases from the values of the function parameters and the
1606 	 * inode state but for sanity's sake, I've decided to let the
1607 	 * layers above just tell us.  It's simpler to correctly figure
1608 	 * out in the layer above exactly under what conditions we
1609 	 * can run async and I think it's easier for others read and
1610 	 * follow the logic in case something has to be changed.
1611 	 * cscope is your friend -- rcc.
1612 	 *
1613 	 * The attribute fork is much simpler.
1614 	 *
1615 	 * For the attribute fork we allow the caller to tell us whether
1616 	 * the unlink of the inode that led to this call is yet permanent
1617 	 * in the on disk log.  If it is not and we will be freeing extents
1618 	 * in this inode then we make the first transaction synchronous
1619 	 * to make sure that the unlink is permanent by the time we free
1620 	 * the blocks.
1621 	 */
1622 	if (fork == XFS_DATA_FORK) {
1623 		if (ip->i_d.di_nextents > 0) {
1624 			ip->i_d.di_size = new_size;
1625 			xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1626 		}
1627 	} else if (sync) {
1628 		ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1629 		if (ip->i_d.di_anextents > 0)
1630 			xfs_trans_set_sync(ntp);
1631 	}
1632 	ASSERT(fork == XFS_DATA_FORK ||
1633 		(fork == XFS_ATTR_FORK &&
1634 			((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1635 			 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1636 
1637 	/*
1638 	 * Since it is possible for space to become allocated beyond
1639 	 * the end of the file (in a crash where the space is allocated
1640 	 * but the inode size is not yet updated), simply remove any
1641 	 * blocks which show up between the new EOF and the maximum
1642 	 * possible file size.  If the first block to be removed is
1643 	 * beyond the maximum file size (ie it is the same as last_block),
1644 	 * then there is nothing to do.
1645 	 */
1646 	last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1647 	ASSERT(first_unmap_block <= last_block);
1648 	done = 0;
1649 	if (last_block == first_unmap_block) {
1650 		done = 1;
1651 	} else {
1652 		unmap_len = last_block - first_unmap_block + 1;
1653 	}
1654 	while (!done) {
1655 		/*
1656 		 * Free up up to XFS_ITRUNC_MAX_EXTENTS.  xfs_bunmapi()
1657 		 * will tell us whether it freed the entire range or
1658 		 * not.  If this is a synchronous mount (wsync),
1659 		 * then we can tell bunmapi to keep all the
1660 		 * transactions asynchronous since the unlink
1661 		 * transaction that made this inode inactive has
1662 		 * already hit the disk.  There's no danger of
1663 		 * the freed blocks being reused, there being a
1664 		 * crash, and the reused blocks suddenly reappearing
1665 		 * in this file with garbage in them once recovery
1666 		 * runs.
1667 		 */
1668 		XFS_BMAP_INIT(&free_list, &first_block);
1669 		error = xfs_bunmapi(ntp, ip, first_unmap_block,
1670 				    unmap_len,
1671 				    XFS_BMAPI_AFLAG(fork) |
1672 				      (sync ? 0 : XFS_BMAPI_ASYNC),
1673 				    XFS_ITRUNC_MAX_EXTENTS,
1674 				    &first_block, &free_list, &done);
1675 		if (error) {
1676 			/*
1677 			 * If the bunmapi call encounters an error,
1678 			 * return to the caller where the transaction
1679 			 * can be properly aborted.  We just need to
1680 			 * make sure we're not holding any resources
1681 			 * that we were not when we came in.
1682 			 */
1683 			xfs_bmap_cancel(&free_list);
1684 			return error;
1685 		}
1686 
1687 		/*
1688 		 * Duplicate the transaction that has the permanent
1689 		 * reservation and commit the old transaction.
1690 		 */
1691 		error = xfs_bmap_finish(tp, &free_list, first_block,
1692 					&committed);
1693 		ntp = *tp;
1694 		if (error) {
1695 			/*
1696 			 * If the bmap finish call encounters an error,
1697 			 * return to the caller where the transaction
1698 			 * can be properly aborted.  We just need to
1699 			 * make sure we're not holding any resources
1700 			 * that we were not when we came in.
1701 			 *
1702 			 * Aborting from this point might lose some
1703 			 * blocks in the file system, but oh well.
1704 			 */
1705 			xfs_bmap_cancel(&free_list);
1706 			if (committed) {
1707 				/*
1708 				 * If the passed in transaction committed
1709 				 * in xfs_bmap_finish(), then we want to
1710 				 * add the inode to this one before returning.
1711 				 * This keeps things simple for the higher
1712 				 * level code, because it always knows that
1713 				 * the inode is locked and held in the
1714 				 * transaction that returns to it whether
1715 				 * errors occur or not.  We don't mark the
1716 				 * inode dirty so that this transaction can
1717 				 * be easily aborted if possible.
1718 				 */
1719 				xfs_trans_ijoin(ntp, ip,
1720 					XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1721 				xfs_trans_ihold(ntp, ip);
1722 			}
1723 			return error;
1724 		}
1725 
1726 		if (committed) {
1727 			/*
1728 			 * The first xact was committed,
1729 			 * so add the inode to the new one.
1730 			 * Mark it dirty so it will be logged
1731 			 * and moved forward in the log as
1732 			 * part of every commit.
1733 			 */
1734 			xfs_trans_ijoin(ntp, ip,
1735 					XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1736 			xfs_trans_ihold(ntp, ip);
1737 			xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1738 		}
1739 		ntp = xfs_trans_dup(ntp);
1740 		(void) xfs_trans_commit(*tp, 0, NULL);
1741 		*tp = ntp;
1742 		error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1743 					  XFS_TRANS_PERM_LOG_RES,
1744 					  XFS_ITRUNCATE_LOG_COUNT);
1745 		/*
1746 		 * Add the inode being truncated to the next chained
1747 		 * transaction.
1748 		 */
1749 		xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1750 		xfs_trans_ihold(ntp, ip);
1751 		if (error)
1752 			return (error);
1753 	}
1754 	/*
1755 	 * Only update the size in the case of the data fork, but
1756 	 * always re-log the inode so that our permanent transaction
1757 	 * can keep on rolling it forward in the log.
1758 	 */
1759 	if (fork == XFS_DATA_FORK) {
1760 		xfs_isize_check(mp, ip, new_size);
1761 		ip->i_d.di_size = new_size;
1762 	}
1763 	xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1764 	ASSERT((new_size != 0) ||
1765 	       (fork == XFS_ATTR_FORK) ||
1766 	       (ip->i_delayed_blks == 0));
1767 	ASSERT((new_size != 0) ||
1768 	       (fork == XFS_ATTR_FORK) ||
1769 	       (ip->i_d.di_nextents == 0));
1770 	xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1771 	return 0;
1772 }
1773 
1774 
1775 /*
1776  * xfs_igrow_start
1777  *
1778  * Do the first part of growing a file: zero any data in the last
1779  * block that is beyond the old EOF.  We need to do this before
1780  * the inode is joined to the transaction to modify the i_size.
1781  * That way we can drop the inode lock and call into the buffer
1782  * cache to get the buffer mapping the EOF.
1783  */
1784 int
1785 xfs_igrow_start(
1786 	xfs_inode_t	*ip,
1787 	xfs_fsize_t	new_size,
1788 	cred_t		*credp)
1789 {
1790 	int		error;
1791 
1792 	ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1793 	ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1794 	ASSERT(new_size > ip->i_d.di_size);
1795 
1796 	/*
1797 	 * Zero any pages that may have been created by
1798 	 * xfs_write_file() beyond the end of the file
1799 	 * and any blocks between the old and new file sizes.
1800 	 */
1801 	error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1802 			     ip->i_d.di_size, new_size);
1803 	return error;
1804 }
1805 
1806 /*
1807  * xfs_igrow_finish
1808  *
1809  * This routine is called to extend the size of a file.
1810  * The inode must have both the iolock and the ilock locked
1811  * for update and it must be a part of the current transaction.
1812  * The xfs_igrow_start() function must have been called previously.
1813  * If the change_flag is not zero, the inode change timestamp will
1814  * be updated.
1815  */
1816 void
1817 xfs_igrow_finish(
1818 	xfs_trans_t	*tp,
1819 	xfs_inode_t	*ip,
1820 	xfs_fsize_t	new_size,
1821 	int		change_flag)
1822 {
1823 	ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1824 	ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1825 	ASSERT(ip->i_transp == tp);
1826 	ASSERT(new_size > ip->i_d.di_size);
1827 
1828 	/*
1829 	 * Update the file size.  Update the inode change timestamp
1830 	 * if change_flag set.
1831 	 */
1832 	ip->i_d.di_size = new_size;
1833 	if (change_flag)
1834 		xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1835 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1836 
1837 }
1838 
1839 
1840 /*
1841  * This is called when the inode's link count goes to 0.
1842  * We place the on-disk inode on a list in the AGI.  It
1843  * will be pulled from this list when the inode is freed.
1844  */
1845 int
1846 xfs_iunlink(
1847 	xfs_trans_t	*tp,
1848 	xfs_inode_t	*ip)
1849 {
1850 	xfs_mount_t	*mp;
1851 	xfs_agi_t	*agi;
1852 	xfs_dinode_t	*dip;
1853 	xfs_buf_t	*agibp;
1854 	xfs_buf_t	*ibp;
1855 	xfs_agnumber_t	agno;
1856 	xfs_daddr_t	agdaddr;
1857 	xfs_agino_t	agino;
1858 	short		bucket_index;
1859 	int		offset;
1860 	int		error;
1861 	int		agi_ok;
1862 
1863 	ASSERT(ip->i_d.di_nlink == 0);
1864 	ASSERT(ip->i_d.di_mode != 0);
1865 	ASSERT(ip->i_transp == tp);
1866 
1867 	mp = tp->t_mountp;
1868 
1869 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1870 	agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1871 
1872 	/*
1873 	 * Get the agi buffer first.  It ensures lock ordering
1874 	 * on the list.
1875 	 */
1876 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1877 				   XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1878 	if (error) {
1879 		return error;
1880 	}
1881 	/*
1882 	 * Validate the magic number of the agi block.
1883 	 */
1884 	agi = XFS_BUF_TO_AGI(agibp);
1885 	agi_ok =
1886 		be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1887 		XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1888 	if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1889 			XFS_RANDOM_IUNLINK))) {
1890 		XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1891 		xfs_trans_brelse(tp, agibp);
1892 		return XFS_ERROR(EFSCORRUPTED);
1893 	}
1894 	/*
1895 	 * Get the index into the agi hash table for the
1896 	 * list this inode will go on.
1897 	 */
1898 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1899 	ASSERT(agino != 0);
1900 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1901 	ASSERT(agi->agi_unlinked[bucket_index]);
1902 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1903 
1904 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1905 		/*
1906 		 * There is already another inode in the bucket we need
1907 		 * to add ourselves to.  Add us at the front of the list.
1908 		 * Here we put the head pointer into our next pointer,
1909 		 * and then we fall through to point the head at us.
1910 		 */
1911 		error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
1912 		if (error) {
1913 			return error;
1914 		}
1915 		ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1916 		ASSERT(dip->di_next_unlinked);
1917 		/* both on-disk, don't endian flip twice */
1918 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1919 		offset = ip->i_boffset +
1920 			offsetof(xfs_dinode_t, di_next_unlinked);
1921 		xfs_trans_inode_buf(tp, ibp);
1922 		xfs_trans_log_buf(tp, ibp, offset,
1923 				  (offset + sizeof(xfs_agino_t) - 1));
1924 		xfs_inobp_check(mp, ibp);
1925 	}
1926 
1927 	/*
1928 	 * Point the bucket head pointer at the inode being inserted.
1929 	 */
1930 	ASSERT(agino != 0);
1931 	agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1932 	offset = offsetof(xfs_agi_t, agi_unlinked) +
1933 		(sizeof(xfs_agino_t) * bucket_index);
1934 	xfs_trans_log_buf(tp, agibp, offset,
1935 			  (offset + sizeof(xfs_agino_t) - 1));
1936 	return 0;
1937 }
1938 
1939 /*
1940  * Pull the on-disk inode from the AGI unlinked list.
1941  */
1942 STATIC int
1943 xfs_iunlink_remove(
1944 	xfs_trans_t	*tp,
1945 	xfs_inode_t	*ip)
1946 {
1947 	xfs_ino_t	next_ino;
1948 	xfs_mount_t	*mp;
1949 	xfs_agi_t	*agi;
1950 	xfs_dinode_t	*dip;
1951 	xfs_buf_t	*agibp;
1952 	xfs_buf_t	*ibp;
1953 	xfs_agnumber_t	agno;
1954 	xfs_daddr_t	agdaddr;
1955 	xfs_agino_t	agino;
1956 	xfs_agino_t	next_agino;
1957 	xfs_buf_t	*last_ibp;
1958 	xfs_dinode_t	*last_dip;
1959 	short		bucket_index;
1960 	int		offset, last_offset;
1961 	int		error;
1962 	int		agi_ok;
1963 
1964 	/*
1965 	 * First pull the on-disk inode from the AGI unlinked list.
1966 	 */
1967 	mp = tp->t_mountp;
1968 
1969 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1970 	agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1971 
1972 	/*
1973 	 * Get the agi buffer first.  It ensures lock ordering
1974 	 * on the list.
1975 	 */
1976 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1977 				   XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1978 	if (error) {
1979 		cmn_err(CE_WARN,
1980 			"xfs_iunlink_remove: xfs_trans_read_buf()  returned an error %d on %s.  Returning error.",
1981 			error, mp->m_fsname);
1982 		return error;
1983 	}
1984 	/*
1985 	 * Validate the magic number of the agi block.
1986 	 */
1987 	agi = XFS_BUF_TO_AGI(agibp);
1988 	agi_ok =
1989 		be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1990 		XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1991 	if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1992 			XFS_RANDOM_IUNLINK_REMOVE))) {
1993 		XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1994 				     mp, agi);
1995 		xfs_trans_brelse(tp, agibp);
1996 		cmn_err(CE_WARN,
1997 			"xfs_iunlink_remove: XFS_TEST_ERROR()  returned an error on %s.  Returning EFSCORRUPTED.",
1998 			 mp->m_fsname);
1999 		return XFS_ERROR(EFSCORRUPTED);
2000 	}
2001 	/*
2002 	 * Get the index into the agi hash table for the
2003 	 * list this inode will go on.
2004 	 */
2005 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2006 	ASSERT(agino != 0);
2007 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2008 	ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
2009 	ASSERT(agi->agi_unlinked[bucket_index]);
2010 
2011 	if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2012 		/*
2013 		 * We're at the head of the list.  Get the inode's
2014 		 * on-disk buffer to see if there is anyone after us
2015 		 * on the list.  Only modify our next pointer if it
2016 		 * is not already NULLAGINO.  This saves us the overhead
2017 		 * of dealing with the buffer when there is no need to
2018 		 * change it.
2019 		 */
2020 		error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2021 		if (error) {
2022 			cmn_err(CE_WARN,
2023 				"xfs_iunlink_remove: xfs_itobp()  returned an error %d on %s.  Returning error.",
2024 				error, mp->m_fsname);
2025 			return error;
2026 		}
2027 		next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2028 		ASSERT(next_agino != 0);
2029 		if (next_agino != NULLAGINO) {
2030 			INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2031 			offset = ip->i_boffset +
2032 				offsetof(xfs_dinode_t, di_next_unlinked);
2033 			xfs_trans_inode_buf(tp, ibp);
2034 			xfs_trans_log_buf(tp, ibp, offset,
2035 					  (offset + sizeof(xfs_agino_t) - 1));
2036 			xfs_inobp_check(mp, ibp);
2037 		} else {
2038 			xfs_trans_brelse(tp, ibp);
2039 		}
2040 		/*
2041 		 * Point the bucket head pointer at the next inode.
2042 		 */
2043 		ASSERT(next_agino != 0);
2044 		ASSERT(next_agino != agino);
2045 		agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2046 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2047 			(sizeof(xfs_agino_t) * bucket_index);
2048 		xfs_trans_log_buf(tp, agibp, offset,
2049 				  (offset + sizeof(xfs_agino_t) - 1));
2050 	} else {
2051 		/*
2052 		 * We need to search the list for the inode being freed.
2053 		 */
2054 		next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2055 		last_ibp = NULL;
2056 		while (next_agino != agino) {
2057 			/*
2058 			 * If the last inode wasn't the one pointing to
2059 			 * us, then release its buffer since we're not
2060 			 * going to do anything with it.
2061 			 */
2062 			if (last_ibp != NULL) {
2063 				xfs_trans_brelse(tp, last_ibp);
2064 			}
2065 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2066 			error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2067 					    &last_ibp, &last_offset);
2068 			if (error) {
2069 				cmn_err(CE_WARN,
2070 			"xfs_iunlink_remove: xfs_inotobp()  returned an error %d on %s.  Returning error.",
2071 					error, mp->m_fsname);
2072 				return error;
2073 			}
2074 			next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2075 			ASSERT(next_agino != NULLAGINO);
2076 			ASSERT(next_agino != 0);
2077 		}
2078 		/*
2079 		 * Now last_ibp points to the buffer previous to us on
2080 		 * the unlinked list.  Pull us from the list.
2081 		 */
2082 		error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
2083 		if (error) {
2084 			cmn_err(CE_WARN,
2085 				"xfs_iunlink_remove: xfs_itobp()  returned an error %d on %s.  Returning error.",
2086 				error, mp->m_fsname);
2087 			return error;
2088 		}
2089 		next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2090 		ASSERT(next_agino != 0);
2091 		ASSERT(next_agino != agino);
2092 		if (next_agino != NULLAGINO) {
2093 			INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2094 			offset = ip->i_boffset +
2095 				offsetof(xfs_dinode_t, di_next_unlinked);
2096 			xfs_trans_inode_buf(tp, ibp);
2097 			xfs_trans_log_buf(tp, ibp, offset,
2098 					  (offset + sizeof(xfs_agino_t) - 1));
2099 			xfs_inobp_check(mp, ibp);
2100 		} else {
2101 			xfs_trans_brelse(tp, ibp);
2102 		}
2103 		/*
2104 		 * Point the previous inode on the list to the next inode.
2105 		 */
2106 		INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2107 		ASSERT(next_agino != 0);
2108 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2109 		xfs_trans_inode_buf(tp, last_ibp);
2110 		xfs_trans_log_buf(tp, last_ibp, offset,
2111 				  (offset + sizeof(xfs_agino_t) - 1));
2112 		xfs_inobp_check(mp, last_ibp);
2113 	}
2114 	return 0;
2115 }
2116 
2117 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2118 {
2119 	return (((ip->i_itemp == NULL) ||
2120 		!(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2121 		(ip->i_update_core == 0));
2122 }
2123 
2124 STATIC void
2125 xfs_ifree_cluster(
2126 	xfs_inode_t	*free_ip,
2127 	xfs_trans_t	*tp,
2128 	xfs_ino_t	inum)
2129 {
2130 	xfs_mount_t		*mp = free_ip->i_mount;
2131 	int			blks_per_cluster;
2132 	int			nbufs;
2133 	int			ninodes;
2134 	int			i, j, found, pre_flushed;
2135 	xfs_daddr_t		blkno;
2136 	xfs_buf_t		*bp;
2137 	xfs_ihash_t		*ih;
2138 	xfs_inode_t		*ip, **ip_found;
2139 	xfs_inode_log_item_t	*iip;
2140 	xfs_log_item_t		*lip;
2141 	SPLDECL(s);
2142 
2143 	if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2144 		blks_per_cluster = 1;
2145 		ninodes = mp->m_sb.sb_inopblock;
2146 		nbufs = XFS_IALLOC_BLOCKS(mp);
2147 	} else {
2148 		blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2149 					mp->m_sb.sb_blocksize;
2150 		ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2151 		nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2152 	}
2153 
2154 	ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2155 
2156 	for (j = 0; j < nbufs; j++, inum += ninodes) {
2157 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2158 					 XFS_INO_TO_AGBNO(mp, inum));
2159 
2160 
2161 		/*
2162 		 * Look for each inode in memory and attempt to lock it,
2163 		 * we can be racing with flush and tail pushing here.
2164 		 * any inode we get the locks on, add to an array of
2165 		 * inode items to process later.
2166 		 *
2167 		 * The get the buffer lock, we could beat a flush
2168 		 * or tail pushing thread to the lock here, in which
2169 		 * case they will go looking for the inode buffer
2170 		 * and fail, we need some other form of interlock
2171 		 * here.
2172 		 */
2173 		found = 0;
2174 		for (i = 0; i < ninodes; i++) {
2175 			ih = XFS_IHASH(mp, inum + i);
2176 			read_lock(&ih->ih_lock);
2177 			for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2178 				if (ip->i_ino == inum + i)
2179 					break;
2180 			}
2181 
2182 			/* Inode not in memory or we found it already,
2183 			 * nothing to do
2184 			 */
2185 			if (!ip || (ip->i_flags & XFS_ISTALE)) {
2186 				read_unlock(&ih->ih_lock);
2187 				continue;
2188 			}
2189 
2190 			if (xfs_inode_clean(ip)) {
2191 				read_unlock(&ih->ih_lock);
2192 				continue;
2193 			}
2194 
2195 			/* If we can get the locks then add it to the
2196 			 * list, otherwise by the time we get the bp lock
2197 			 * below it will already be attached to the
2198 			 * inode buffer.
2199 			 */
2200 
2201 			/* This inode will already be locked - by us, lets
2202 			 * keep it that way.
2203 			 */
2204 
2205 			if (ip == free_ip) {
2206 				if (xfs_iflock_nowait(ip)) {
2207 					ip->i_flags |= XFS_ISTALE;
2208 
2209 					if (xfs_inode_clean(ip)) {
2210 						xfs_ifunlock(ip);
2211 					} else {
2212 						ip_found[found++] = ip;
2213 					}
2214 				}
2215 				read_unlock(&ih->ih_lock);
2216 				continue;
2217 			}
2218 
2219 			if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2220 				if (xfs_iflock_nowait(ip)) {
2221 					ip->i_flags |= XFS_ISTALE;
2222 
2223 					if (xfs_inode_clean(ip)) {
2224 						xfs_ifunlock(ip);
2225 						xfs_iunlock(ip, XFS_ILOCK_EXCL);
2226 					} else {
2227 						ip_found[found++] = ip;
2228 					}
2229 				} else {
2230 					xfs_iunlock(ip, XFS_ILOCK_EXCL);
2231 				}
2232 			}
2233 
2234 			read_unlock(&ih->ih_lock);
2235 		}
2236 
2237 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2238 					mp->m_bsize * blks_per_cluster,
2239 					XFS_BUF_LOCK);
2240 
2241 		pre_flushed = 0;
2242 		lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2243 		while (lip) {
2244 			if (lip->li_type == XFS_LI_INODE) {
2245 				iip = (xfs_inode_log_item_t *)lip;
2246 				ASSERT(iip->ili_logged == 1);
2247 				lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2248 				AIL_LOCK(mp,s);
2249 				iip->ili_flush_lsn = iip->ili_item.li_lsn;
2250 				AIL_UNLOCK(mp, s);
2251 				iip->ili_inode->i_flags |= XFS_ISTALE;
2252 				pre_flushed++;
2253 			}
2254 			lip = lip->li_bio_list;
2255 		}
2256 
2257 		for (i = 0; i < found; i++) {
2258 			ip = ip_found[i];
2259 			iip = ip->i_itemp;
2260 
2261 			if (!iip) {
2262 				ip->i_update_core = 0;
2263 				xfs_ifunlock(ip);
2264 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2265 				continue;
2266 			}
2267 
2268 			iip->ili_last_fields = iip->ili_format.ilf_fields;
2269 			iip->ili_format.ilf_fields = 0;
2270 			iip->ili_logged = 1;
2271 			AIL_LOCK(mp,s);
2272 			iip->ili_flush_lsn = iip->ili_item.li_lsn;
2273 			AIL_UNLOCK(mp, s);
2274 
2275 			xfs_buf_attach_iodone(bp,
2276 				(void(*)(xfs_buf_t*,xfs_log_item_t*))
2277 				xfs_istale_done, (xfs_log_item_t *)iip);
2278 			if (ip != free_ip) {
2279 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2280 			}
2281 		}
2282 
2283 		if (found || pre_flushed)
2284 			xfs_trans_stale_inode_buf(tp, bp);
2285 		xfs_trans_binval(tp, bp);
2286 	}
2287 
2288 	kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2289 }
2290 
2291 /*
2292  * This is called to return an inode to the inode free list.
2293  * The inode should already be truncated to 0 length and have
2294  * no pages associated with it.  This routine also assumes that
2295  * the inode is already a part of the transaction.
2296  *
2297  * The on-disk copy of the inode will have been added to the list
2298  * of unlinked inodes in the AGI. We need to remove the inode from
2299  * that list atomically with respect to freeing it here.
2300  */
2301 int
2302 xfs_ifree(
2303 	xfs_trans_t	*tp,
2304 	xfs_inode_t	*ip,
2305 	xfs_bmap_free_t	*flist)
2306 {
2307 	int			error;
2308 	int			delete;
2309 	xfs_ino_t		first_ino;
2310 
2311 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2312 	ASSERT(ip->i_transp == tp);
2313 	ASSERT(ip->i_d.di_nlink == 0);
2314 	ASSERT(ip->i_d.di_nextents == 0);
2315 	ASSERT(ip->i_d.di_anextents == 0);
2316 	ASSERT((ip->i_d.di_size == 0) ||
2317 	       ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2318 	ASSERT(ip->i_d.di_nblocks == 0);
2319 
2320 	/*
2321 	 * Pull the on-disk inode from the AGI unlinked list.
2322 	 */
2323 	error = xfs_iunlink_remove(tp, ip);
2324 	if (error != 0) {
2325 		return error;
2326 	}
2327 
2328 	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2329 	if (error != 0) {
2330 		return error;
2331 	}
2332 	ip->i_d.di_mode = 0;		/* mark incore inode as free */
2333 	ip->i_d.di_flags = 0;
2334 	ip->i_d.di_dmevmask = 0;
2335 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2336 	ip->i_df.if_ext_max =
2337 		XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2338 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2339 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2340 	/*
2341 	 * Bump the generation count so no one will be confused
2342 	 * by reincarnations of this inode.
2343 	 */
2344 	ip->i_d.di_gen++;
2345 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2346 
2347 	if (delete) {
2348 		xfs_ifree_cluster(ip, tp, first_ino);
2349 	}
2350 
2351 	return 0;
2352 }
2353 
2354 /*
2355  * Reallocate the space for if_broot based on the number of records
2356  * being added or deleted as indicated in rec_diff.  Move the records
2357  * and pointers in if_broot to fit the new size.  When shrinking this
2358  * will eliminate holes between the records and pointers created by
2359  * the caller.  When growing this will create holes to be filled in
2360  * by the caller.
2361  *
2362  * The caller must not request to add more records than would fit in
2363  * the on-disk inode root.  If the if_broot is currently NULL, then
2364  * if we adding records one will be allocated.  The caller must also
2365  * not request that the number of records go below zero, although
2366  * it can go to zero.
2367  *
2368  * ip -- the inode whose if_broot area is changing
2369  * ext_diff -- the change in the number of records, positive or negative,
2370  *	 requested for the if_broot array.
2371  */
2372 void
2373 xfs_iroot_realloc(
2374 	xfs_inode_t		*ip,
2375 	int			rec_diff,
2376 	int			whichfork)
2377 {
2378 	int			cur_max;
2379 	xfs_ifork_t		*ifp;
2380 	xfs_bmbt_block_t	*new_broot;
2381 	int			new_max;
2382 	size_t			new_size;
2383 	char			*np;
2384 	char			*op;
2385 
2386 	/*
2387 	 * Handle the degenerate case quietly.
2388 	 */
2389 	if (rec_diff == 0) {
2390 		return;
2391 	}
2392 
2393 	ifp = XFS_IFORK_PTR(ip, whichfork);
2394 	if (rec_diff > 0) {
2395 		/*
2396 		 * If there wasn't any memory allocated before, just
2397 		 * allocate it now and get out.
2398 		 */
2399 		if (ifp->if_broot_bytes == 0) {
2400 			new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2401 			ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2402 								     KM_SLEEP);
2403 			ifp->if_broot_bytes = (int)new_size;
2404 			return;
2405 		}
2406 
2407 		/*
2408 		 * If there is already an existing if_broot, then we need
2409 		 * to realloc() it and shift the pointers to their new
2410 		 * location.  The records don't change location because
2411 		 * they are kept butted up against the btree block header.
2412 		 */
2413 		cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2414 		new_max = cur_max + rec_diff;
2415 		new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2416 		ifp->if_broot = (xfs_bmbt_block_t *)
2417 		  kmem_realloc(ifp->if_broot,
2418 				new_size,
2419 				(size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2420 				KM_SLEEP);
2421 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2422 						      ifp->if_broot_bytes);
2423 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2424 						      (int)new_size);
2425 		ifp->if_broot_bytes = (int)new_size;
2426 		ASSERT(ifp->if_broot_bytes <=
2427 			XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2428 		memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2429 		return;
2430 	}
2431 
2432 	/*
2433 	 * rec_diff is less than 0.  In this case, we are shrinking the
2434 	 * if_broot buffer.  It must already exist.  If we go to zero
2435 	 * records, just get rid of the root and clear the status bit.
2436 	 */
2437 	ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2438 	cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2439 	new_max = cur_max + rec_diff;
2440 	ASSERT(new_max >= 0);
2441 	if (new_max > 0)
2442 		new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2443 	else
2444 		new_size = 0;
2445 	if (new_size > 0) {
2446 		new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2447 		/*
2448 		 * First copy over the btree block header.
2449 		 */
2450 		memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2451 	} else {
2452 		new_broot = NULL;
2453 		ifp->if_flags &= ~XFS_IFBROOT;
2454 	}
2455 
2456 	/*
2457 	 * Only copy the records and pointers if there are any.
2458 	 */
2459 	if (new_max > 0) {
2460 		/*
2461 		 * First copy the records.
2462 		 */
2463 		op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2464 						     ifp->if_broot_bytes);
2465 		np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2466 						     (int)new_size);
2467 		memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2468 
2469 		/*
2470 		 * Then copy the pointers.
2471 		 */
2472 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2473 						     ifp->if_broot_bytes);
2474 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2475 						     (int)new_size);
2476 		memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2477 	}
2478 	kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2479 	ifp->if_broot = new_broot;
2480 	ifp->if_broot_bytes = (int)new_size;
2481 	ASSERT(ifp->if_broot_bytes <=
2482 		XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2483 	return;
2484 }
2485 
2486 
2487 /*
2488  * This is called when the amount of space needed for if_data
2489  * is increased or decreased.  The change in size is indicated by
2490  * the number of bytes that need to be added or deleted in the
2491  * byte_diff parameter.
2492  *
2493  * If the amount of space needed has decreased below the size of the
2494  * inline buffer, then switch to using the inline buffer.  Otherwise,
2495  * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2496  * to what is needed.
2497  *
2498  * ip -- the inode whose if_data area is changing
2499  * byte_diff -- the change in the number of bytes, positive or negative,
2500  *	 requested for the if_data array.
2501  */
2502 void
2503 xfs_idata_realloc(
2504 	xfs_inode_t	*ip,
2505 	int		byte_diff,
2506 	int		whichfork)
2507 {
2508 	xfs_ifork_t	*ifp;
2509 	int		new_size;
2510 	int		real_size;
2511 
2512 	if (byte_diff == 0) {
2513 		return;
2514 	}
2515 
2516 	ifp = XFS_IFORK_PTR(ip, whichfork);
2517 	new_size = (int)ifp->if_bytes + byte_diff;
2518 	ASSERT(new_size >= 0);
2519 
2520 	if (new_size == 0) {
2521 		if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2522 			kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2523 		}
2524 		ifp->if_u1.if_data = NULL;
2525 		real_size = 0;
2526 	} else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2527 		/*
2528 		 * If the valid extents/data can fit in if_inline_ext/data,
2529 		 * copy them from the malloc'd vector and free it.
2530 		 */
2531 		if (ifp->if_u1.if_data == NULL) {
2532 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2533 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2534 			ASSERT(ifp->if_real_bytes != 0);
2535 			memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2536 			      new_size);
2537 			kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2538 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2539 		}
2540 		real_size = 0;
2541 	} else {
2542 		/*
2543 		 * Stuck with malloc/realloc.
2544 		 * For inline data, the underlying buffer must be
2545 		 * a multiple of 4 bytes in size so that it can be
2546 		 * logged and stay on word boundaries.  We enforce
2547 		 * that here.
2548 		 */
2549 		real_size = roundup(new_size, 4);
2550 		if (ifp->if_u1.if_data == NULL) {
2551 			ASSERT(ifp->if_real_bytes == 0);
2552 			ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2553 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2554 			/*
2555 			 * Only do the realloc if the underlying size
2556 			 * is really changing.
2557 			 */
2558 			if (ifp->if_real_bytes != real_size) {
2559 				ifp->if_u1.if_data =
2560 					kmem_realloc(ifp->if_u1.if_data,
2561 							real_size,
2562 							ifp->if_real_bytes,
2563 							KM_SLEEP);
2564 			}
2565 		} else {
2566 			ASSERT(ifp->if_real_bytes == 0);
2567 			ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2568 			memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2569 				ifp->if_bytes);
2570 		}
2571 	}
2572 	ifp->if_real_bytes = real_size;
2573 	ifp->if_bytes = new_size;
2574 	ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2575 }
2576 
2577 
2578 
2579 
2580 /*
2581  * Map inode to disk block and offset.
2582  *
2583  * mp -- the mount point structure for the current file system
2584  * tp -- the current transaction
2585  * ino -- the inode number of the inode to be located
2586  * imap -- this structure is filled in with the information necessary
2587  *	 to retrieve the given inode from disk
2588  * flags -- flags to pass to xfs_dilocate indicating whether or not
2589  *	 lookups in the inode btree were OK or not
2590  */
2591 int
2592 xfs_imap(
2593 	xfs_mount_t	*mp,
2594 	xfs_trans_t	*tp,
2595 	xfs_ino_t	ino,
2596 	xfs_imap_t	*imap,
2597 	uint		flags)
2598 {
2599 	xfs_fsblock_t	fsbno;
2600 	int		len;
2601 	int		off;
2602 	int		error;
2603 
2604 	fsbno = imap->im_blkno ?
2605 		XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2606 	error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2607 	if (error != 0) {
2608 		return error;
2609 	}
2610 	imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2611 	imap->im_len = XFS_FSB_TO_BB(mp, len);
2612 	imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2613 	imap->im_ioffset = (ushort)off;
2614 	imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2615 	return 0;
2616 }
2617 
2618 void
2619 xfs_idestroy_fork(
2620 	xfs_inode_t	*ip,
2621 	int		whichfork)
2622 {
2623 	xfs_ifork_t	*ifp;
2624 
2625 	ifp = XFS_IFORK_PTR(ip, whichfork);
2626 	if (ifp->if_broot != NULL) {
2627 		kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2628 		ifp->if_broot = NULL;
2629 	}
2630 
2631 	/*
2632 	 * If the format is local, then we can't have an extents
2633 	 * array so just look for an inline data array.  If we're
2634 	 * not local then we may or may not have an extents list,
2635 	 * so check and free it up if we do.
2636 	 */
2637 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2638 		if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2639 		    (ifp->if_u1.if_data != NULL)) {
2640 			ASSERT(ifp->if_real_bytes != 0);
2641 			kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2642 			ifp->if_u1.if_data = NULL;
2643 			ifp->if_real_bytes = 0;
2644 		}
2645 	} else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2646 		   ((ifp->if_flags & XFS_IFEXTIREC) ||
2647 		    ((ifp->if_u1.if_extents != NULL) &&
2648 		     (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2649 		ASSERT(ifp->if_real_bytes != 0);
2650 		xfs_iext_destroy(ifp);
2651 	}
2652 	ASSERT(ifp->if_u1.if_extents == NULL ||
2653 	       ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2654 	ASSERT(ifp->if_real_bytes == 0);
2655 	if (whichfork == XFS_ATTR_FORK) {
2656 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2657 		ip->i_afp = NULL;
2658 	}
2659 }
2660 
2661 /*
2662  * This is called free all the memory associated with an inode.
2663  * It must free the inode itself and any buffers allocated for
2664  * if_extents/if_data and if_broot.  It must also free the lock
2665  * associated with the inode.
2666  */
2667 void
2668 xfs_idestroy(
2669 	xfs_inode_t	*ip)
2670 {
2671 
2672 	switch (ip->i_d.di_mode & S_IFMT) {
2673 	case S_IFREG:
2674 	case S_IFDIR:
2675 	case S_IFLNK:
2676 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
2677 		break;
2678 	}
2679 	if (ip->i_afp)
2680 		xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2681 	mrfree(&ip->i_lock);
2682 	mrfree(&ip->i_iolock);
2683 	freesema(&ip->i_flock);
2684 #ifdef XFS_BMAP_TRACE
2685 	ktrace_free(ip->i_xtrace);
2686 #endif
2687 #ifdef XFS_BMBT_TRACE
2688 	ktrace_free(ip->i_btrace);
2689 #endif
2690 #ifdef XFS_RW_TRACE
2691 	ktrace_free(ip->i_rwtrace);
2692 #endif
2693 #ifdef XFS_ILOCK_TRACE
2694 	ktrace_free(ip->i_lock_trace);
2695 #endif
2696 #ifdef XFS_DIR2_TRACE
2697 	ktrace_free(ip->i_dir_trace);
2698 #endif
2699 	if (ip->i_itemp) {
2700 		/* XXXdpd should be able to assert this but shutdown
2701 		 * is leaving the AIL behind. */
2702 		ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2703 		       XFS_FORCED_SHUTDOWN(ip->i_mount));
2704 		xfs_inode_item_destroy(ip);
2705 	}
2706 	kmem_zone_free(xfs_inode_zone, ip);
2707 }
2708 
2709 
2710 /*
2711  * Increment the pin count of the given buffer.
2712  * This value is protected by ipinlock spinlock in the mount structure.
2713  */
2714 void
2715 xfs_ipin(
2716 	xfs_inode_t	*ip)
2717 {
2718 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2719 
2720 	atomic_inc(&ip->i_pincount);
2721 }
2722 
2723 /*
2724  * Decrement the pin count of the given inode, and wake up
2725  * anyone in xfs_iwait_unpin() if the count goes to 0.  The
2726  * inode must have been previously pinned with a call to xfs_ipin().
2727  */
2728 void
2729 xfs_iunpin(
2730 	xfs_inode_t	*ip)
2731 {
2732 	ASSERT(atomic_read(&ip->i_pincount) > 0);
2733 
2734 	if (atomic_dec_and_test(&ip->i_pincount)) {
2735 		vnode_t	*vp = XFS_ITOV_NULL(ip);
2736 
2737 		/* make sync come back and flush this inode */
2738 		if (vp) {
2739 			struct inode	*inode = vn_to_inode(vp);
2740 
2741 			if (!(inode->i_state & I_NEW))
2742 				mark_inode_dirty_sync(inode);
2743 		}
2744 
2745 		wake_up(&ip->i_ipin_wait);
2746 	}
2747 }
2748 
2749 /*
2750  * This is called to wait for the given inode to be unpinned.
2751  * It will sleep until this happens.  The caller must have the
2752  * inode locked in at least shared mode so that the buffer cannot
2753  * be subsequently pinned once someone is waiting for it to be
2754  * unpinned.
2755  */
2756 STATIC void
2757 xfs_iunpin_wait(
2758 	xfs_inode_t	*ip)
2759 {
2760 	xfs_inode_log_item_t	*iip;
2761 	xfs_lsn_t	lsn;
2762 
2763 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2764 
2765 	if (atomic_read(&ip->i_pincount) == 0) {
2766 		return;
2767 	}
2768 
2769 	iip = ip->i_itemp;
2770 	if (iip && iip->ili_last_lsn) {
2771 		lsn = iip->ili_last_lsn;
2772 	} else {
2773 		lsn = (xfs_lsn_t)0;
2774 	}
2775 
2776 	/*
2777 	 * Give the log a push so we don't wait here too long.
2778 	 */
2779 	xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2780 
2781 	wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2782 }
2783 
2784 
2785 /*
2786  * xfs_iextents_copy()
2787  *
2788  * This is called to copy the REAL extents (as opposed to the delayed
2789  * allocation extents) from the inode into the given buffer.  It
2790  * returns the number of bytes copied into the buffer.
2791  *
2792  * If there are no delayed allocation extents, then we can just
2793  * memcpy() the extents into the buffer.  Otherwise, we need to
2794  * examine each extent in turn and skip those which are delayed.
2795  */
2796 int
2797 xfs_iextents_copy(
2798 	xfs_inode_t		*ip,
2799 	xfs_bmbt_rec_t		*buffer,
2800 	int			whichfork)
2801 {
2802 	int			copied;
2803 	xfs_bmbt_rec_t		*dest_ep;
2804 	xfs_bmbt_rec_t		*ep;
2805 #ifdef XFS_BMAP_TRACE
2806 	static char		fname[] = "xfs_iextents_copy";
2807 #endif
2808 	int			i;
2809 	xfs_ifork_t		*ifp;
2810 	int			nrecs;
2811 	xfs_fsblock_t		start_block;
2812 
2813 	ifp = XFS_IFORK_PTR(ip, whichfork);
2814 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2815 	ASSERT(ifp->if_bytes > 0);
2816 
2817 	nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2818 	xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2819 	ASSERT(nrecs > 0);
2820 
2821 	/*
2822 	 * There are some delayed allocation extents in the
2823 	 * inode, so copy the extents one at a time and skip
2824 	 * the delayed ones.  There must be at least one
2825 	 * non-delayed extent.
2826 	 */
2827 	dest_ep = buffer;
2828 	copied = 0;
2829 	for (i = 0; i < nrecs; i++) {
2830 		ep = xfs_iext_get_ext(ifp, i);
2831 		start_block = xfs_bmbt_get_startblock(ep);
2832 		if (ISNULLSTARTBLOCK(start_block)) {
2833 			/*
2834 			 * It's a delayed allocation extent, so skip it.
2835 			 */
2836 			continue;
2837 		}
2838 
2839 		/* Translate to on disk format */
2840 		put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2841 			      (__uint64_t*)&dest_ep->l0);
2842 		put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2843 			      (__uint64_t*)&dest_ep->l1);
2844 		dest_ep++;
2845 		copied++;
2846 	}
2847 	ASSERT(copied != 0);
2848 	xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
2849 
2850 	return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2851 }
2852 
2853 /*
2854  * Each of the following cases stores data into the same region
2855  * of the on-disk inode, so only one of them can be valid at
2856  * any given time. While it is possible to have conflicting formats
2857  * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2858  * in EXTENTS format, this can only happen when the fork has
2859  * changed formats after being modified but before being flushed.
2860  * In these cases, the format always takes precedence, because the
2861  * format indicates the current state of the fork.
2862  */
2863 /*ARGSUSED*/
2864 STATIC int
2865 xfs_iflush_fork(
2866 	xfs_inode_t		*ip,
2867 	xfs_dinode_t		*dip,
2868 	xfs_inode_log_item_t	*iip,
2869 	int			whichfork,
2870 	xfs_buf_t		*bp)
2871 {
2872 	char			*cp;
2873 	xfs_ifork_t		*ifp;
2874 	xfs_mount_t		*mp;
2875 #ifdef XFS_TRANS_DEBUG
2876 	int			first;
2877 #endif
2878 	static const short	brootflag[2] =
2879 		{ XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2880 	static const short	dataflag[2] =
2881 		{ XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2882 	static const short	extflag[2] =
2883 		{ XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2884 
2885 	if (iip == NULL)
2886 		return 0;
2887 	ifp = XFS_IFORK_PTR(ip, whichfork);
2888 	/*
2889 	 * This can happen if we gave up in iformat in an error path,
2890 	 * for the attribute fork.
2891 	 */
2892 	if (ifp == NULL) {
2893 		ASSERT(whichfork == XFS_ATTR_FORK);
2894 		return 0;
2895 	}
2896 	cp = XFS_DFORK_PTR(dip, whichfork);
2897 	mp = ip->i_mount;
2898 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2899 	case XFS_DINODE_FMT_LOCAL:
2900 		if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2901 		    (ifp->if_bytes > 0)) {
2902 			ASSERT(ifp->if_u1.if_data != NULL);
2903 			ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2904 			memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2905 		}
2906 		if (whichfork == XFS_DATA_FORK) {
2907 			if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2908 				XFS_ERROR_REPORT("xfs_iflush_fork",
2909 						 XFS_ERRLEVEL_LOW, mp);
2910 				return XFS_ERROR(EFSCORRUPTED);
2911 			}
2912 		}
2913 		break;
2914 
2915 	case XFS_DINODE_FMT_EXTENTS:
2916 		ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2917 		       !(iip->ili_format.ilf_fields & extflag[whichfork]));
2918 		ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2919 			(ifp->if_bytes == 0));
2920 		ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2921 			(ifp->if_bytes > 0));
2922 		if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2923 		    (ifp->if_bytes > 0)) {
2924 			ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2925 			(void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2926 				whichfork);
2927 		}
2928 		break;
2929 
2930 	case XFS_DINODE_FMT_BTREE:
2931 		if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2932 		    (ifp->if_broot_bytes > 0)) {
2933 			ASSERT(ifp->if_broot != NULL);
2934 			ASSERT(ifp->if_broot_bytes <=
2935 			       (XFS_IFORK_SIZE(ip, whichfork) +
2936 				XFS_BROOT_SIZE_ADJ));
2937 			xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2938 				(xfs_bmdr_block_t *)cp,
2939 				XFS_DFORK_SIZE(dip, mp, whichfork));
2940 		}
2941 		break;
2942 
2943 	case XFS_DINODE_FMT_DEV:
2944 		if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2945 			ASSERT(whichfork == XFS_DATA_FORK);
2946 			INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2947 		}
2948 		break;
2949 
2950 	case XFS_DINODE_FMT_UUID:
2951 		if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2952 			ASSERT(whichfork == XFS_DATA_FORK);
2953 			memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2954 				sizeof(uuid_t));
2955 		}
2956 		break;
2957 
2958 	default:
2959 		ASSERT(0);
2960 		break;
2961 	}
2962 
2963 	return 0;
2964 }
2965 
2966 /*
2967  * xfs_iflush() will write a modified inode's changes out to the
2968  * inode's on disk home.  The caller must have the inode lock held
2969  * in at least shared mode and the inode flush semaphore must be
2970  * held as well.  The inode lock will still be held upon return from
2971  * the call and the caller is free to unlock it.
2972  * The inode flush lock will be unlocked when the inode reaches the disk.
2973  * The flags indicate how the inode's buffer should be written out.
2974  */
2975 int
2976 xfs_iflush(
2977 	xfs_inode_t		*ip,
2978 	uint			flags)
2979 {
2980 	xfs_inode_log_item_t	*iip;
2981 	xfs_buf_t		*bp;
2982 	xfs_dinode_t		*dip;
2983 	xfs_mount_t		*mp;
2984 	int			error;
2985 	/* REFERENCED */
2986 	xfs_chash_t		*ch;
2987 	xfs_inode_t		*iq;
2988 	int			clcount;	/* count of inodes clustered */
2989 	int			bufwasdelwri;
2990 	enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
2991 	SPLDECL(s);
2992 
2993 	XFS_STATS_INC(xs_iflush_count);
2994 
2995 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2996 	ASSERT(valusema(&ip->i_flock) <= 0);
2997 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2998 	       ip->i_d.di_nextents > ip->i_df.if_ext_max);
2999 
3000 	iip = ip->i_itemp;
3001 	mp = ip->i_mount;
3002 
3003 	/*
3004 	 * If the inode isn't dirty, then just release the inode
3005 	 * flush lock and do nothing.
3006 	 */
3007 	if ((ip->i_update_core == 0) &&
3008 	    ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3009 		ASSERT((iip != NULL) ?
3010 			 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3011 		xfs_ifunlock(ip);
3012 		return 0;
3013 	}
3014 
3015 	/*
3016 	 * We can't flush the inode until it is unpinned, so
3017 	 * wait for it.  We know noone new can pin it, because
3018 	 * we are holding the inode lock shared and you need
3019 	 * to hold it exclusively to pin the inode.
3020 	 */
3021 	xfs_iunpin_wait(ip);
3022 
3023 	/*
3024 	 * This may have been unpinned because the filesystem is shutting
3025 	 * down forcibly. If that's the case we must not write this inode
3026 	 * to disk, because the log record didn't make it to disk!
3027 	 */
3028 	if (XFS_FORCED_SHUTDOWN(mp)) {
3029 		ip->i_update_core = 0;
3030 		if (iip)
3031 			iip->ili_format.ilf_fields = 0;
3032 		xfs_ifunlock(ip);
3033 		return XFS_ERROR(EIO);
3034 	}
3035 
3036 	/*
3037 	 * Get the buffer containing the on-disk inode.
3038 	 */
3039 	error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3040 	if (error) {
3041 		xfs_ifunlock(ip);
3042 		return error;
3043 	}
3044 
3045 	/*
3046 	 * Decide how buffer will be flushed out.  This is done before
3047 	 * the call to xfs_iflush_int because this field is zeroed by it.
3048 	 */
3049 	if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3050 		/*
3051 		 * Flush out the inode buffer according to the directions
3052 		 * of the caller.  In the cases where the caller has given
3053 		 * us a choice choose the non-delwri case.  This is because
3054 		 * the inode is in the AIL and we need to get it out soon.
3055 		 */
3056 		switch (flags) {
3057 		case XFS_IFLUSH_SYNC:
3058 		case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3059 			flags = 0;
3060 			break;
3061 		case XFS_IFLUSH_ASYNC:
3062 		case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3063 			flags = INT_ASYNC;
3064 			break;
3065 		case XFS_IFLUSH_DELWRI:
3066 			flags = INT_DELWRI;
3067 			break;
3068 		default:
3069 			ASSERT(0);
3070 			flags = 0;
3071 			break;
3072 		}
3073 	} else {
3074 		switch (flags) {
3075 		case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3076 		case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3077 		case XFS_IFLUSH_DELWRI:
3078 			flags = INT_DELWRI;
3079 			break;
3080 		case XFS_IFLUSH_ASYNC:
3081 			flags = INT_ASYNC;
3082 			break;
3083 		case XFS_IFLUSH_SYNC:
3084 			flags = 0;
3085 			break;
3086 		default:
3087 			ASSERT(0);
3088 			flags = 0;
3089 			break;
3090 		}
3091 	}
3092 
3093 	/*
3094 	 * First flush out the inode that xfs_iflush was called with.
3095 	 */
3096 	error = xfs_iflush_int(ip, bp);
3097 	if (error) {
3098 		goto corrupt_out;
3099 	}
3100 
3101 	/*
3102 	 * inode clustering:
3103 	 * see if other inodes can be gathered into this write
3104 	 */
3105 
3106 	ip->i_chash->chl_buf = bp;
3107 
3108 	ch = XFS_CHASH(mp, ip->i_blkno);
3109 	s = mutex_spinlock(&ch->ch_lock);
3110 
3111 	clcount = 0;
3112 	for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3113 		/*
3114 		 * Do an un-protected check to see if the inode is dirty and
3115 		 * is a candidate for flushing.  These checks will be repeated
3116 		 * later after the appropriate locks are acquired.
3117 		 */
3118 		iip = iq->i_itemp;
3119 		if ((iq->i_update_core == 0) &&
3120 		    ((iip == NULL) ||
3121 		     !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3122 		      xfs_ipincount(iq) == 0) {
3123 			continue;
3124 		}
3125 
3126 		/*
3127 		 * Try to get locks.  If any are unavailable,
3128 		 * then this inode cannot be flushed and is skipped.
3129 		 */
3130 
3131 		/* get inode locks (just i_lock) */
3132 		if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3133 			/* get inode flush lock */
3134 			if (xfs_iflock_nowait(iq)) {
3135 				/* check if pinned */
3136 				if (xfs_ipincount(iq) == 0) {
3137 					/* arriving here means that
3138 					 * this inode can be flushed.
3139 					 * first re-check that it's
3140 					 * dirty
3141 					 */
3142 					iip = iq->i_itemp;
3143 					if ((iq->i_update_core != 0)||
3144 					    ((iip != NULL) &&
3145 					     (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3146 						clcount++;
3147 						error = xfs_iflush_int(iq, bp);
3148 						if (error) {
3149 							xfs_iunlock(iq,
3150 								    XFS_ILOCK_SHARED);
3151 							goto cluster_corrupt_out;
3152 						}
3153 					} else {
3154 						xfs_ifunlock(iq);
3155 					}
3156 				} else {
3157 					xfs_ifunlock(iq);
3158 				}
3159 			}
3160 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
3161 		}
3162 	}
3163 	mutex_spinunlock(&ch->ch_lock, s);
3164 
3165 	if (clcount) {
3166 		XFS_STATS_INC(xs_icluster_flushcnt);
3167 		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3168 	}
3169 
3170 	/*
3171 	 * If the buffer is pinned then push on the log so we won't
3172 	 * get stuck waiting in the write for too long.
3173 	 */
3174 	if (XFS_BUF_ISPINNED(bp)){
3175 		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3176 	}
3177 
3178 	if (flags & INT_DELWRI) {
3179 		xfs_bdwrite(mp, bp);
3180 	} else if (flags & INT_ASYNC) {
3181 		xfs_bawrite(mp, bp);
3182 	} else {
3183 		error = xfs_bwrite(mp, bp);
3184 	}
3185 	return error;
3186 
3187 corrupt_out:
3188 	xfs_buf_relse(bp);
3189 	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3190 	xfs_iflush_abort(ip);
3191 	/*
3192 	 * Unlocks the flush lock
3193 	 */
3194 	return XFS_ERROR(EFSCORRUPTED);
3195 
3196 cluster_corrupt_out:
3197 	/* Corruption detected in the clustering loop.  Invalidate the
3198 	 * inode buffer and shut down the filesystem.
3199 	 */
3200 	mutex_spinunlock(&ch->ch_lock, s);
3201 
3202 	/*
3203 	 * Clean up the buffer.  If it was B_DELWRI, just release it --
3204 	 * brelse can handle it with no problems.  If not, shut down the
3205 	 * filesystem before releasing the buffer.
3206 	 */
3207 	if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3208 		xfs_buf_relse(bp);
3209 	}
3210 
3211 	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3212 
3213 	if(!bufwasdelwri)  {
3214 		/*
3215 		 * Just like incore_relse: if we have b_iodone functions,
3216 		 * mark the buffer as an error and call them.  Otherwise
3217 		 * mark it as stale and brelse.
3218 		 */
3219 		if (XFS_BUF_IODONE_FUNC(bp)) {
3220 			XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3221 			XFS_BUF_UNDONE(bp);
3222 			XFS_BUF_STALE(bp);
3223 			XFS_BUF_SHUT(bp);
3224 			XFS_BUF_ERROR(bp,EIO);
3225 			xfs_biodone(bp);
3226 		} else {
3227 			XFS_BUF_STALE(bp);
3228 			xfs_buf_relse(bp);
3229 		}
3230 	}
3231 
3232 	xfs_iflush_abort(iq);
3233 	/*
3234 	 * Unlocks the flush lock
3235 	 */
3236 	return XFS_ERROR(EFSCORRUPTED);
3237 }
3238 
3239 
3240 STATIC int
3241 xfs_iflush_int(
3242 	xfs_inode_t		*ip,
3243 	xfs_buf_t		*bp)
3244 {
3245 	xfs_inode_log_item_t	*iip;
3246 	xfs_dinode_t		*dip;
3247 	xfs_mount_t		*mp;
3248 #ifdef XFS_TRANS_DEBUG
3249 	int			first;
3250 #endif
3251 	SPLDECL(s);
3252 
3253 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3254 	ASSERT(valusema(&ip->i_flock) <= 0);
3255 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3256 	       ip->i_d.di_nextents > ip->i_df.if_ext_max);
3257 
3258 	iip = ip->i_itemp;
3259 	mp = ip->i_mount;
3260 
3261 
3262 	/*
3263 	 * If the inode isn't dirty, then just release the inode
3264 	 * flush lock and do nothing.
3265 	 */
3266 	if ((ip->i_update_core == 0) &&
3267 	    ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3268 		xfs_ifunlock(ip);
3269 		return 0;
3270 	}
3271 
3272 	/* set *dip = inode's place in the buffer */
3273 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3274 
3275 	/*
3276 	 * Clear i_update_core before copying out the data.
3277 	 * This is for coordination with our timestamp updates
3278 	 * that don't hold the inode lock. They will always
3279 	 * update the timestamps BEFORE setting i_update_core,
3280 	 * so if we clear i_update_core after they set it we
3281 	 * are guaranteed to see their updates to the timestamps.
3282 	 * I believe that this depends on strongly ordered memory
3283 	 * semantics, but we have that.  We use the SYNCHRONIZE
3284 	 * macro to make sure that the compiler does not reorder
3285 	 * the i_update_core access below the data copy below.
3286 	 */
3287 	ip->i_update_core = 0;
3288 	SYNCHRONIZE();
3289 
3290 	/*
3291 	 * Make sure to get the latest atime from the Linux inode.
3292 	 */
3293 	xfs_synchronize_atime(ip);
3294 
3295 	if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3296 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3297 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3298 		    "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3299 			ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3300 		goto corrupt_out;
3301 	}
3302 	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3303 				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3304 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3305 			"xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3306 			ip->i_ino, ip, ip->i_d.di_magic);
3307 		goto corrupt_out;
3308 	}
3309 	if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3310 		if (XFS_TEST_ERROR(
3311 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3312 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3313 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3314 			xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3315 				"xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3316 				ip->i_ino, ip);
3317 			goto corrupt_out;
3318 		}
3319 	} else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3320 		if (XFS_TEST_ERROR(
3321 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3322 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3323 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3324 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3325 			xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3326 				"xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3327 				ip->i_ino, ip);
3328 			goto corrupt_out;
3329 		}
3330 	}
3331 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3332 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3333 				XFS_RANDOM_IFLUSH_5)) {
3334 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3335 			"xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3336 			ip->i_ino,
3337 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3338 			ip->i_d.di_nblocks,
3339 			ip);
3340 		goto corrupt_out;
3341 	}
3342 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3343 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3344 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3345 			"xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3346 			ip->i_ino, ip->i_d.di_forkoff, ip);
3347 		goto corrupt_out;
3348 	}
3349 	/*
3350 	 * bump the flush iteration count, used to detect flushes which
3351 	 * postdate a log record during recovery.
3352 	 */
3353 
3354 	ip->i_d.di_flushiter++;
3355 
3356 	/*
3357 	 * Copy the dirty parts of the inode into the on-disk
3358 	 * inode.  We always copy out the core of the inode,
3359 	 * because if the inode is dirty at all the core must
3360 	 * be.
3361 	 */
3362 	xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3363 
3364 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3365 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3366 		ip->i_d.di_flushiter = 0;
3367 
3368 	/*
3369 	 * If this is really an old format inode and the superblock version
3370 	 * has not been updated to support only new format inodes, then
3371 	 * convert back to the old inode format.  If the superblock version
3372 	 * has been updated, then make the conversion permanent.
3373 	 */
3374 	ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3375 	       XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3376 	if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3377 		if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3378 			/*
3379 			 * Convert it back.
3380 			 */
3381 			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3382 			INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3383 		} else {
3384 			/*
3385 			 * The superblock version has already been bumped,
3386 			 * so just make the conversion to the new inode
3387 			 * format permanent.
3388 			 */
3389 			ip->i_d.di_version = XFS_DINODE_VERSION_2;
3390 			INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3391 			ip->i_d.di_onlink = 0;
3392 			dip->di_core.di_onlink = 0;
3393 			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3394 			memset(&(dip->di_core.di_pad[0]), 0,
3395 			      sizeof(dip->di_core.di_pad));
3396 			ASSERT(ip->i_d.di_projid == 0);
3397 		}
3398 	}
3399 
3400 	if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3401 		goto corrupt_out;
3402 	}
3403 
3404 	if (XFS_IFORK_Q(ip)) {
3405 		/*
3406 		 * The only error from xfs_iflush_fork is on the data fork.
3407 		 */
3408 		(void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3409 	}
3410 	xfs_inobp_check(mp, bp);
3411 
3412 	/*
3413 	 * We've recorded everything logged in the inode, so we'd
3414 	 * like to clear the ilf_fields bits so we don't log and
3415 	 * flush things unnecessarily.  However, we can't stop
3416 	 * logging all this information until the data we've copied
3417 	 * into the disk buffer is written to disk.  If we did we might
3418 	 * overwrite the copy of the inode in the log with all the
3419 	 * data after re-logging only part of it, and in the face of
3420 	 * a crash we wouldn't have all the data we need to recover.
3421 	 *
3422 	 * What we do is move the bits to the ili_last_fields field.
3423 	 * When logging the inode, these bits are moved back to the
3424 	 * ilf_fields field.  In the xfs_iflush_done() routine we
3425 	 * clear ili_last_fields, since we know that the information
3426 	 * those bits represent is permanently on disk.  As long as
3427 	 * the flush completes before the inode is logged again, then
3428 	 * both ilf_fields and ili_last_fields will be cleared.
3429 	 *
3430 	 * We can play with the ilf_fields bits here, because the inode
3431 	 * lock must be held exclusively in order to set bits there
3432 	 * and the flush lock protects the ili_last_fields bits.
3433 	 * Set ili_logged so the flush done
3434 	 * routine can tell whether or not to look in the AIL.
3435 	 * Also, store the current LSN of the inode so that we can tell
3436 	 * whether the item has moved in the AIL from xfs_iflush_done().
3437 	 * In order to read the lsn we need the AIL lock, because
3438 	 * it is a 64 bit value that cannot be read atomically.
3439 	 */
3440 	if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3441 		iip->ili_last_fields = iip->ili_format.ilf_fields;
3442 		iip->ili_format.ilf_fields = 0;
3443 		iip->ili_logged = 1;
3444 
3445 		ASSERT(sizeof(xfs_lsn_t) == 8);	/* don't lock if it shrinks */
3446 		AIL_LOCK(mp,s);
3447 		iip->ili_flush_lsn = iip->ili_item.li_lsn;
3448 		AIL_UNLOCK(mp, s);
3449 
3450 		/*
3451 		 * Attach the function xfs_iflush_done to the inode's
3452 		 * buffer.  This will remove the inode from the AIL
3453 		 * and unlock the inode's flush lock when the inode is
3454 		 * completely written to disk.
3455 		 */
3456 		xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3457 				      xfs_iflush_done, (xfs_log_item_t *)iip);
3458 
3459 		ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3460 		ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3461 	} else {
3462 		/*
3463 		 * We're flushing an inode which is not in the AIL and has
3464 		 * not been logged but has i_update_core set.  For this
3465 		 * case we can use a B_DELWRI flush and immediately drop
3466 		 * the inode flush lock because we can avoid the whole
3467 		 * AIL state thing.  It's OK to drop the flush lock now,
3468 		 * because we've already locked the buffer and to do anything
3469 		 * you really need both.
3470 		 */
3471 		if (iip != NULL) {
3472 			ASSERT(iip->ili_logged == 0);
3473 			ASSERT(iip->ili_last_fields == 0);
3474 			ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3475 		}
3476 		xfs_ifunlock(ip);
3477 	}
3478 
3479 	return 0;
3480 
3481 corrupt_out:
3482 	return XFS_ERROR(EFSCORRUPTED);
3483 }
3484 
3485 
3486 /*
3487  * Flush all inactive inodes in mp.
3488  */
3489 void
3490 xfs_iflush_all(
3491 	xfs_mount_t	*mp)
3492 {
3493 	xfs_inode_t	*ip;
3494 	vnode_t		*vp;
3495 
3496  again:
3497 	XFS_MOUNT_ILOCK(mp);
3498 	ip = mp->m_inodes;
3499 	if (ip == NULL)
3500 		goto out;
3501 
3502 	do {
3503 		/* Make sure we skip markers inserted by sync */
3504 		if (ip->i_mount == NULL) {
3505 			ip = ip->i_mnext;
3506 			continue;
3507 		}
3508 
3509 		vp = XFS_ITOV_NULL(ip);
3510 		if (!vp) {
3511 			XFS_MOUNT_IUNLOCK(mp);
3512 			xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3513 			goto again;
3514 		}
3515 
3516 		ASSERT(vn_count(vp) == 0);
3517 
3518 		ip = ip->i_mnext;
3519 	} while (ip != mp->m_inodes);
3520  out:
3521 	XFS_MOUNT_IUNLOCK(mp);
3522 }
3523 
3524 /*
3525  * xfs_iaccess: check accessibility of inode for mode.
3526  */
3527 int
3528 xfs_iaccess(
3529 	xfs_inode_t	*ip,
3530 	mode_t		mode,
3531 	cred_t		*cr)
3532 {
3533 	int		error;
3534 	mode_t		orgmode = mode;
3535 	struct inode	*inode = vn_to_inode(XFS_ITOV(ip));
3536 
3537 	if (mode & S_IWUSR) {
3538 		umode_t		imode = inode->i_mode;
3539 
3540 		if (IS_RDONLY(inode) &&
3541 		    (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3542 			return XFS_ERROR(EROFS);
3543 
3544 		if (IS_IMMUTABLE(inode))
3545 			return XFS_ERROR(EACCES);
3546 	}
3547 
3548 	/*
3549 	 * If there's an Access Control List it's used instead of
3550 	 * the mode bits.
3551 	 */
3552 	if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3553 		return error ? XFS_ERROR(error) : 0;
3554 
3555 	if (current_fsuid(cr) != ip->i_d.di_uid) {
3556 		mode >>= 3;
3557 		if (!in_group_p((gid_t)ip->i_d.di_gid))
3558 			mode >>= 3;
3559 	}
3560 
3561 	/*
3562 	 * If the DACs are ok we don't need any capability check.
3563 	 */
3564 	if ((ip->i_d.di_mode & mode) == mode)
3565 		return 0;
3566 	/*
3567 	 * Read/write DACs are always overridable.
3568 	 * Executable DACs are overridable if at least one exec bit is set.
3569 	 */
3570 	if (!(orgmode & S_IXUSR) ||
3571 	    (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3572 		if (capable_cred(cr, CAP_DAC_OVERRIDE))
3573 			return 0;
3574 
3575 	if ((orgmode == S_IRUSR) ||
3576 	    (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3577 		if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3578 			return 0;
3579 #ifdef	NOISE
3580 		cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3581 #endif	/* NOISE */
3582 		return XFS_ERROR(EACCES);
3583 	}
3584 	return XFS_ERROR(EACCES);
3585 }
3586 
3587 /*
3588  * xfs_iroundup: round up argument to next power of two
3589  */
3590 uint
3591 xfs_iroundup(
3592 	uint	v)
3593 {
3594 	int i;
3595 	uint m;
3596 
3597 	if ((v & (v - 1)) == 0)
3598 		return v;
3599 	ASSERT((v & 0x80000000) == 0);
3600 	if ((v & (v + 1)) == 0)
3601 		return v + 1;
3602 	for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3603 		if (v & m)
3604 			continue;
3605 		v |= m;
3606 		if ((v & (v + 1)) == 0)
3607 			return v + 1;
3608 	}
3609 	ASSERT(0);
3610 	return( 0 );
3611 }
3612 
3613 #ifdef XFS_ILOCK_TRACE
3614 ktrace_t	*xfs_ilock_trace_buf;
3615 
3616 void
3617 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3618 {
3619 	ktrace_enter(ip->i_lock_trace,
3620 		     (void *)ip,
3621 		     (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3622 		     (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3623 		     (void *)ra,		/* caller of ilock */
3624 		     (void *)(unsigned long)current_cpu(),
3625 		     (void *)(unsigned long)current_pid(),
3626 		     NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3627 }
3628 #endif
3629 
3630 /*
3631  * Return a pointer to the extent record at file index idx.
3632  */
3633 xfs_bmbt_rec_t *
3634 xfs_iext_get_ext(
3635 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3636 	xfs_extnum_t	idx)		/* index of target extent */
3637 {
3638 	ASSERT(idx >= 0);
3639 	if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3640 		return ifp->if_u1.if_ext_irec->er_extbuf;
3641 	} else if (ifp->if_flags & XFS_IFEXTIREC) {
3642 		xfs_ext_irec_t	*erp;		/* irec pointer */
3643 		int		erp_idx = 0;	/* irec index */
3644 		xfs_extnum_t	page_idx = idx;	/* ext index in target list */
3645 
3646 		erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3647 		return &erp->er_extbuf[page_idx];
3648 	} else if (ifp->if_bytes) {
3649 		return &ifp->if_u1.if_extents[idx];
3650 	} else {
3651 		return NULL;
3652 	}
3653 }
3654 
3655 /*
3656  * Insert new item(s) into the extent records for incore inode
3657  * fork 'ifp'.  'count' new items are inserted at index 'idx'.
3658  */
3659 void
3660 xfs_iext_insert(
3661 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3662 	xfs_extnum_t	idx,		/* starting index of new items */
3663 	xfs_extnum_t	count,		/* number of inserted items */
3664 	xfs_bmbt_irec_t	*new)		/* items to insert */
3665 {
3666 	xfs_bmbt_rec_t	*ep;		/* extent record pointer */
3667 	xfs_extnum_t	i;		/* extent record index */
3668 
3669 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3670 	xfs_iext_add(ifp, idx, count);
3671 	for (i = idx; i < idx + count; i++, new++) {
3672 		ep = xfs_iext_get_ext(ifp, i);
3673 		xfs_bmbt_set_all(ep, new);
3674 	}
3675 }
3676 
3677 /*
3678  * This is called when the amount of space required for incore file
3679  * extents needs to be increased. The ext_diff parameter stores the
3680  * number of new extents being added and the idx parameter contains
3681  * the extent index where the new extents will be added. If the new
3682  * extents are being appended, then we just need to (re)allocate and
3683  * initialize the space. Otherwise, if the new extents are being
3684  * inserted into the middle of the existing entries, a bit more work
3685  * is required to make room for the new extents to be inserted. The
3686  * caller is responsible for filling in the new extent entries upon
3687  * return.
3688  */
3689 void
3690 xfs_iext_add(
3691 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3692 	xfs_extnum_t	idx,		/* index to begin adding exts */
3693 	int		ext_diff)	/* number of extents to add */
3694 {
3695 	int		byte_diff;	/* new bytes being added */
3696 	int		new_size;	/* size of extents after adding */
3697 	xfs_extnum_t	nextents;	/* number of extents in file */
3698 
3699 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3700 	ASSERT((idx >= 0) && (idx <= nextents));
3701 	byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3702 	new_size = ifp->if_bytes + byte_diff;
3703 	/*
3704 	 * If the new number of extents (nextents + ext_diff)
3705 	 * fits inside the inode, then continue to use the inline
3706 	 * extent buffer.
3707 	 */
3708 	if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3709 		if (idx < nextents) {
3710 			memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3711 				&ifp->if_u2.if_inline_ext[idx],
3712 				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
3713 			memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3714 		}
3715 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3716 		ifp->if_real_bytes = 0;
3717 		ifp->if_lastex = nextents + ext_diff;
3718 	}
3719 	/*
3720 	 * Otherwise use a linear (direct) extent list.
3721 	 * If the extents are currently inside the inode,
3722 	 * xfs_iext_realloc_direct will switch us from
3723 	 * inline to direct extent allocation mode.
3724 	 */
3725 	else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3726 		xfs_iext_realloc_direct(ifp, new_size);
3727 		if (idx < nextents) {
3728 			memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3729 				&ifp->if_u1.if_extents[idx],
3730 				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
3731 			memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3732 		}
3733 	}
3734 	/* Indirection array */
3735 	else {
3736 		xfs_ext_irec_t	*erp;
3737 		int		erp_idx = 0;
3738 		int		page_idx = idx;
3739 
3740 		ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3741 		if (ifp->if_flags & XFS_IFEXTIREC) {
3742 			erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3743 		} else {
3744 			xfs_iext_irec_init(ifp);
3745 			ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3746 			erp = ifp->if_u1.if_ext_irec;
3747 		}
3748 		/* Extents fit in target extent page */
3749 		if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3750 			if (page_idx < erp->er_extcount) {
3751 				memmove(&erp->er_extbuf[page_idx + ext_diff],
3752 					&erp->er_extbuf[page_idx],
3753 					(erp->er_extcount - page_idx) *
3754 					sizeof(xfs_bmbt_rec_t));
3755 				memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3756 			}
3757 			erp->er_extcount += ext_diff;
3758 			xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3759 		}
3760 		/* Insert a new extent page */
3761 		else if (erp) {
3762 			xfs_iext_add_indirect_multi(ifp,
3763 				erp_idx, page_idx, ext_diff);
3764 		}
3765 		/*
3766 		 * If extent(s) are being appended to the last page in
3767 		 * the indirection array and the new extent(s) don't fit
3768 		 * in the page, then erp is NULL and erp_idx is set to
3769 		 * the next index needed in the indirection array.
3770 		 */
3771 		else {
3772 			int	count = ext_diff;
3773 
3774 			while (count) {
3775 				erp = xfs_iext_irec_new(ifp, erp_idx);
3776 				erp->er_extcount = count;
3777 				count -= MIN(count, (int)XFS_LINEAR_EXTS);
3778 				if (count) {
3779 					erp_idx++;
3780 				}
3781 			}
3782 		}
3783 	}
3784 	ifp->if_bytes = new_size;
3785 }
3786 
3787 /*
3788  * This is called when incore extents are being added to the indirection
3789  * array and the new extents do not fit in the target extent list. The
3790  * erp_idx parameter contains the irec index for the target extent list
3791  * in the indirection array, and the idx parameter contains the extent
3792  * index within the list. The number of extents being added is stored
3793  * in the count parameter.
3794  *
3795  *    |-------|   |-------|
3796  *    |       |   |       |    idx - number of extents before idx
3797  *    |  idx  |   | count |
3798  *    |       |   |       |    count - number of extents being inserted at idx
3799  *    |-------|   |-------|
3800  *    | count |   | nex2  |    nex2 - number of extents after idx + count
3801  *    |-------|   |-------|
3802  */
3803 void
3804 xfs_iext_add_indirect_multi(
3805 	xfs_ifork_t	*ifp,			/* inode fork pointer */
3806 	int		erp_idx,		/* target extent irec index */
3807 	xfs_extnum_t	idx,			/* index within target list */
3808 	int		count)			/* new extents being added */
3809 {
3810 	int		byte_diff;		/* new bytes being added */
3811 	xfs_ext_irec_t	*erp;			/* pointer to irec entry */
3812 	xfs_extnum_t	ext_diff;		/* number of extents to add */
3813 	xfs_extnum_t	ext_cnt;		/* new extents still needed */
3814 	xfs_extnum_t	nex2;			/* extents after idx + count */
3815 	xfs_bmbt_rec_t	*nex2_ep = NULL;	/* temp list for nex2 extents */
3816 	int		nlists;			/* number of irec's (lists) */
3817 
3818 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3819 	erp = &ifp->if_u1.if_ext_irec[erp_idx];
3820 	nex2 = erp->er_extcount - idx;
3821 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3822 
3823 	/*
3824 	 * Save second part of target extent list
3825 	 * (all extents past */
3826 	if (nex2) {
3827 		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3828 		nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3829 		memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3830 		erp->er_extcount -= nex2;
3831 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3832 		memset(&erp->er_extbuf[idx], 0, byte_diff);
3833 	}
3834 
3835 	/*
3836 	 * Add the new extents to the end of the target
3837 	 * list, then allocate new irec record(s) and
3838 	 * extent buffer(s) as needed to store the rest
3839 	 * of the new extents.
3840 	 */
3841 	ext_cnt = count;
3842 	ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3843 	if (ext_diff) {
3844 		erp->er_extcount += ext_diff;
3845 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3846 		ext_cnt -= ext_diff;
3847 	}
3848 	while (ext_cnt) {
3849 		erp_idx++;
3850 		erp = xfs_iext_irec_new(ifp, erp_idx);
3851 		ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3852 		erp->er_extcount = ext_diff;
3853 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3854 		ext_cnt -= ext_diff;
3855 	}
3856 
3857 	/* Add nex2 extents back to indirection array */
3858 	if (nex2) {
3859 		xfs_extnum_t	ext_avail;
3860 		int		i;
3861 
3862 		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3863 		ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3864 		i = 0;
3865 		/*
3866 		 * If nex2 extents fit in the current page, append
3867 		 * nex2_ep after the new extents.
3868 		 */
3869 		if (nex2 <= ext_avail) {
3870 			i = erp->er_extcount;
3871 		}
3872 		/*
3873 		 * Otherwise, check if space is available in the
3874 		 * next page.
3875 		 */
3876 		else if ((erp_idx < nlists - 1) &&
3877 			 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3878 			  ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3879 			erp_idx++;
3880 			erp++;
3881 			/* Create a hole for nex2 extents */
3882 			memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3883 				erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3884 		}
3885 		/*
3886 		 * Final choice, create a new extent page for
3887 		 * nex2 extents.
3888 		 */
3889 		else {
3890 			erp_idx++;
3891 			erp = xfs_iext_irec_new(ifp, erp_idx);
3892 		}
3893 		memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3894 		kmem_free(nex2_ep, byte_diff);
3895 		erp->er_extcount += nex2;
3896 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3897 	}
3898 }
3899 
3900 /*
3901  * This is called when the amount of space required for incore file
3902  * extents needs to be decreased. The ext_diff parameter stores the
3903  * number of extents to be removed and the idx parameter contains
3904  * the extent index where the extents will be removed from.
3905  *
3906  * If the amount of space needed has decreased below the linear
3907  * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3908  * extent array.  Otherwise, use kmem_realloc() to adjust the
3909  * size to what is needed.
3910  */
3911 void
3912 xfs_iext_remove(
3913 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3914 	xfs_extnum_t	idx,		/* index to begin removing exts */
3915 	int		ext_diff)	/* number of extents to remove */
3916 {
3917 	xfs_extnum_t	nextents;	/* number of extents in file */
3918 	int		new_size;	/* size of extents after removal */
3919 
3920 	ASSERT(ext_diff > 0);
3921 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3922 	new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3923 
3924 	if (new_size == 0) {
3925 		xfs_iext_destroy(ifp);
3926 	} else if (ifp->if_flags & XFS_IFEXTIREC) {
3927 		xfs_iext_remove_indirect(ifp, idx, ext_diff);
3928 	} else if (ifp->if_real_bytes) {
3929 		xfs_iext_remove_direct(ifp, idx, ext_diff);
3930 	} else {
3931 		xfs_iext_remove_inline(ifp, idx, ext_diff);
3932 	}
3933 	ifp->if_bytes = new_size;
3934 }
3935 
3936 /*
3937  * This removes ext_diff extents from the inline buffer, beginning
3938  * at extent index idx.
3939  */
3940 void
3941 xfs_iext_remove_inline(
3942 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3943 	xfs_extnum_t	idx,		/* index to begin removing exts */
3944 	int		ext_diff)	/* number of extents to remove */
3945 {
3946 	int		nextents;	/* number of extents in file */
3947 
3948 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3949 	ASSERT(idx < XFS_INLINE_EXTS);
3950 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3951 	ASSERT(((nextents - ext_diff) > 0) &&
3952 		(nextents - ext_diff) < XFS_INLINE_EXTS);
3953 
3954 	if (idx + ext_diff < nextents) {
3955 		memmove(&ifp->if_u2.if_inline_ext[idx],
3956 			&ifp->if_u2.if_inline_ext[idx + ext_diff],
3957 			(nextents - (idx + ext_diff)) *
3958 			 sizeof(xfs_bmbt_rec_t));
3959 		memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3960 			0, ext_diff * sizeof(xfs_bmbt_rec_t));
3961 	} else {
3962 		memset(&ifp->if_u2.if_inline_ext[idx], 0,
3963 			ext_diff * sizeof(xfs_bmbt_rec_t));
3964 	}
3965 }
3966 
3967 /*
3968  * This removes ext_diff extents from a linear (direct) extent list,
3969  * beginning at extent index idx. If the extents are being removed
3970  * from the end of the list (ie. truncate) then we just need to re-
3971  * allocate the list to remove the extra space. Otherwise, if the
3972  * extents are being removed from the middle of the existing extent
3973  * entries, then we first need to move the extent records beginning
3974  * at idx + ext_diff up in the list to overwrite the records being
3975  * removed, then remove the extra space via kmem_realloc.
3976  */
3977 void
3978 xfs_iext_remove_direct(
3979 	xfs_ifork_t	*ifp,		/* inode fork pointer */
3980 	xfs_extnum_t	idx,		/* index to begin removing exts */
3981 	int		ext_diff)	/* number of extents to remove */
3982 {
3983 	xfs_extnum_t	nextents;	/* number of extents in file */
3984 	int		new_size;	/* size of extents after removal */
3985 
3986 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3987 	new_size = ifp->if_bytes -
3988 		(ext_diff * sizeof(xfs_bmbt_rec_t));
3989 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3990 
3991 	if (new_size == 0) {
3992 		xfs_iext_destroy(ifp);
3993 		return;
3994 	}
3995 	/* Move extents up in the list (if needed) */
3996 	if (idx + ext_diff < nextents) {
3997 		memmove(&ifp->if_u1.if_extents[idx],
3998 			&ifp->if_u1.if_extents[idx + ext_diff],
3999 			(nextents - (idx + ext_diff)) *
4000 			 sizeof(xfs_bmbt_rec_t));
4001 	}
4002 	memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4003 		0, ext_diff * sizeof(xfs_bmbt_rec_t));
4004 	/*
4005 	 * Reallocate the direct extent list. If the extents
4006 	 * will fit inside the inode then xfs_iext_realloc_direct
4007 	 * will switch from direct to inline extent allocation
4008 	 * mode for us.
4009 	 */
4010 	xfs_iext_realloc_direct(ifp, new_size);
4011 	ifp->if_bytes = new_size;
4012 }
4013 
4014 /*
4015  * This is called when incore extents are being removed from the
4016  * indirection array and the extents being removed span multiple extent
4017  * buffers. The idx parameter contains the file extent index where we
4018  * want to begin removing extents, and the count parameter contains
4019  * how many extents need to be removed.
4020  *
4021  *    |-------|   |-------|
4022  *    | nex1  |   |       |    nex1 - number of extents before idx
4023  *    |-------|   | count |
4024  *    |       |   |       |    count - number of extents being removed at idx
4025  *    | count |   |-------|
4026  *    |       |   | nex2  |    nex2 - number of extents after idx + count
4027  *    |-------|   |-------|
4028  */
4029 void
4030 xfs_iext_remove_indirect(
4031 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4032 	xfs_extnum_t	idx,		/* index to begin removing extents */
4033 	int		count)		/* number of extents to remove */
4034 {
4035 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
4036 	int		erp_idx = 0;	/* indirection array index */
4037 	xfs_extnum_t	ext_cnt;	/* extents left to remove */
4038 	xfs_extnum_t	ext_diff;	/* extents to remove in current list */
4039 	xfs_extnum_t	nex1;		/* number of extents before idx */
4040 	xfs_extnum_t	nex2;		/* extents after idx + count */
4041 	int		nlists;		/* entries in indirection array */
4042 	int		page_idx = idx;	/* index in target extent list */
4043 
4044 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4045 	erp = xfs_iext_idx_to_irec(ifp,  &page_idx, &erp_idx, 0);
4046 	ASSERT(erp != NULL);
4047 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4048 	nex1 = page_idx;
4049 	ext_cnt = count;
4050 	while (ext_cnt) {
4051 		nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4052 		ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4053 		/*
4054 		 * Check for deletion of entire list;
4055 		 * xfs_iext_irec_remove() updates extent offsets.
4056 		 */
4057 		if (ext_diff == erp->er_extcount) {
4058 			xfs_iext_irec_remove(ifp, erp_idx);
4059 			ext_cnt -= ext_diff;
4060 			nex1 = 0;
4061 			if (ext_cnt) {
4062 				ASSERT(erp_idx < ifp->if_real_bytes /
4063 					XFS_IEXT_BUFSZ);
4064 				erp = &ifp->if_u1.if_ext_irec[erp_idx];
4065 				nex1 = 0;
4066 				continue;
4067 			} else {
4068 				break;
4069 			}
4070 		}
4071 		/* Move extents up (if needed) */
4072 		if (nex2) {
4073 			memmove(&erp->er_extbuf[nex1],
4074 				&erp->er_extbuf[nex1 + ext_diff],
4075 				nex2 * sizeof(xfs_bmbt_rec_t));
4076 		}
4077 		/* Zero out rest of page */
4078 		memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4079 			((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4080 		/* Update remaining counters */
4081 		erp->er_extcount -= ext_diff;
4082 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4083 		ext_cnt -= ext_diff;
4084 		nex1 = 0;
4085 		erp_idx++;
4086 		erp++;
4087 	}
4088 	ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4089 	xfs_iext_irec_compact(ifp);
4090 }
4091 
4092 /*
4093  * Create, destroy, or resize a linear (direct) block of extents.
4094  */
4095 void
4096 xfs_iext_realloc_direct(
4097 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4098 	int		new_size)	/* new size of extents */
4099 {
4100 	int		rnew_size;	/* real new size of extents */
4101 
4102 	rnew_size = new_size;
4103 
4104 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4105 		((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4106 		 (new_size != ifp->if_real_bytes)));
4107 
4108 	/* Free extent records */
4109 	if (new_size == 0) {
4110 		xfs_iext_destroy(ifp);
4111 	}
4112 	/* Resize direct extent list and zero any new bytes */
4113 	else if (ifp->if_real_bytes) {
4114 		/* Check if extents will fit inside the inode */
4115 		if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4116 			xfs_iext_direct_to_inline(ifp, new_size /
4117 				(uint)sizeof(xfs_bmbt_rec_t));
4118 			ifp->if_bytes = new_size;
4119 			return;
4120 		}
4121 		if ((new_size & (new_size - 1)) != 0) {
4122 			rnew_size = xfs_iroundup(new_size);
4123 		}
4124 		if (rnew_size != ifp->if_real_bytes) {
4125 			ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4126 				kmem_realloc(ifp->if_u1.if_extents,
4127 						rnew_size,
4128 						ifp->if_real_bytes,
4129 						KM_SLEEP);
4130 		}
4131 		if (rnew_size > ifp->if_real_bytes) {
4132 			memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4133 				(uint)sizeof(xfs_bmbt_rec_t)], 0,
4134 				rnew_size - ifp->if_real_bytes);
4135 		}
4136 	}
4137 	/*
4138 	 * Switch from the inline extent buffer to a direct
4139 	 * extent list. Be sure to include the inline extent
4140 	 * bytes in new_size.
4141 	 */
4142 	else {
4143 		new_size += ifp->if_bytes;
4144 		if ((new_size & (new_size - 1)) != 0) {
4145 			rnew_size = xfs_iroundup(new_size);
4146 		}
4147 		xfs_iext_inline_to_direct(ifp, rnew_size);
4148 	}
4149 	ifp->if_real_bytes = rnew_size;
4150 	ifp->if_bytes = new_size;
4151 }
4152 
4153 /*
4154  * Switch from linear (direct) extent records to inline buffer.
4155  */
4156 void
4157 xfs_iext_direct_to_inline(
4158 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4159 	xfs_extnum_t	nextents)	/* number of extents in file */
4160 {
4161 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4162 	ASSERT(nextents <= XFS_INLINE_EXTS);
4163 	/*
4164 	 * The inline buffer was zeroed when we switched
4165 	 * from inline to direct extent allocation mode,
4166 	 * so we don't need to clear it here.
4167 	 */
4168 	memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4169 		nextents * sizeof(xfs_bmbt_rec_t));
4170 	kmem_free(ifp->if_u1.if_extents, KM_SLEEP);
4171 	ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4172 	ifp->if_real_bytes = 0;
4173 }
4174 
4175 /*
4176  * Switch from inline buffer to linear (direct) extent records.
4177  * new_size should already be rounded up to the next power of 2
4178  * by the caller (when appropriate), so use new_size as it is.
4179  * However, since new_size may be rounded up, we can't update
4180  * if_bytes here. It is the caller's responsibility to update
4181  * if_bytes upon return.
4182  */
4183 void
4184 xfs_iext_inline_to_direct(
4185 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4186 	int		new_size)	/* number of extents in file */
4187 {
4188 	ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4189 		kmem_alloc(new_size, KM_SLEEP);
4190 	memset(ifp->if_u1.if_extents, 0, new_size);
4191 	if (ifp->if_bytes) {
4192 		memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4193 			ifp->if_bytes);
4194 		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4195 			sizeof(xfs_bmbt_rec_t));
4196 	}
4197 	ifp->if_real_bytes = new_size;
4198 }
4199 
4200 /*
4201  * Resize an extent indirection array to new_size bytes.
4202  */
4203 void
4204 xfs_iext_realloc_indirect(
4205 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4206 	int		new_size)	/* new indirection array size */
4207 {
4208 	int		nlists;		/* number of irec's (ex lists) */
4209 	int		size;		/* current indirection array size */
4210 
4211 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4212 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4213 	size = nlists * sizeof(xfs_ext_irec_t);
4214 	ASSERT(ifp->if_real_bytes);
4215 	ASSERT((new_size >= 0) && (new_size != size));
4216 	if (new_size == 0) {
4217 		xfs_iext_destroy(ifp);
4218 	} else {
4219 		ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4220 			kmem_realloc(ifp->if_u1.if_ext_irec,
4221 				new_size, size, KM_SLEEP);
4222 	}
4223 }
4224 
4225 /*
4226  * Switch from indirection array to linear (direct) extent allocations.
4227  */
4228 void
4229 xfs_iext_indirect_to_direct(
4230 	 xfs_ifork_t	*ifp)		/* inode fork pointer */
4231 {
4232 	xfs_bmbt_rec_t	*ep;		/* extent record pointer */
4233 	xfs_extnum_t	nextents;	/* number of extents in file */
4234 	int		size;		/* size of file extents */
4235 
4236 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4237 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4238 	ASSERT(nextents <= XFS_LINEAR_EXTS);
4239 	size = nextents * sizeof(xfs_bmbt_rec_t);
4240 
4241 	xfs_iext_irec_compact_full(ifp);
4242 	ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4243 
4244 	ep = ifp->if_u1.if_ext_irec->er_extbuf;
4245 	kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4246 	ifp->if_flags &= ~XFS_IFEXTIREC;
4247 	ifp->if_u1.if_extents = ep;
4248 	ifp->if_bytes = size;
4249 	if (nextents < XFS_LINEAR_EXTS) {
4250 		xfs_iext_realloc_direct(ifp, size);
4251 	}
4252 }
4253 
4254 /*
4255  * Free incore file extents.
4256  */
4257 void
4258 xfs_iext_destroy(
4259 	xfs_ifork_t	*ifp)		/* inode fork pointer */
4260 {
4261 	if (ifp->if_flags & XFS_IFEXTIREC) {
4262 		int	erp_idx;
4263 		int	nlists;
4264 
4265 		nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4266 		for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4267 			xfs_iext_irec_remove(ifp, erp_idx);
4268 		}
4269 		ifp->if_flags &= ~XFS_IFEXTIREC;
4270 	} else if (ifp->if_real_bytes) {
4271 		kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4272 	} else if (ifp->if_bytes) {
4273 		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4274 			sizeof(xfs_bmbt_rec_t));
4275 	}
4276 	ifp->if_u1.if_extents = NULL;
4277 	ifp->if_real_bytes = 0;
4278 	ifp->if_bytes = 0;
4279 }
4280 
4281 /*
4282  * Return a pointer to the extent record for file system block bno.
4283  */
4284 xfs_bmbt_rec_t *			/* pointer to found extent record */
4285 xfs_iext_bno_to_ext(
4286 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4287 	xfs_fileoff_t	bno,		/* block number to search for */
4288 	xfs_extnum_t	*idxp)		/* index of target extent */
4289 {
4290 	xfs_bmbt_rec_t	*base;		/* pointer to first extent */
4291 	xfs_filblks_t	blockcount = 0;	/* number of blocks in extent */
4292 	xfs_bmbt_rec_t	*ep = NULL;	/* pointer to target extent */
4293 	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
4294 	int		high;		/* upper boundary in search */
4295 	xfs_extnum_t	idx = 0;	/* index of target extent */
4296 	int		low;		/* lower boundary in search */
4297 	xfs_extnum_t	nextents;	/* number of file extents */
4298 	xfs_fileoff_t	startoff = 0;	/* start offset of extent */
4299 
4300 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4301 	if (nextents == 0) {
4302 		*idxp = 0;
4303 		return NULL;
4304 	}
4305 	low = 0;
4306 	if (ifp->if_flags & XFS_IFEXTIREC) {
4307 		/* Find target extent list */
4308 		int	erp_idx = 0;
4309 		erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4310 		base = erp->er_extbuf;
4311 		high = erp->er_extcount - 1;
4312 	} else {
4313 		base = ifp->if_u1.if_extents;
4314 		high = nextents - 1;
4315 	}
4316 	/* Binary search extent records */
4317 	while (low <= high) {
4318 		idx = (low + high) >> 1;
4319 		ep = base + idx;
4320 		startoff = xfs_bmbt_get_startoff(ep);
4321 		blockcount = xfs_bmbt_get_blockcount(ep);
4322 		if (bno < startoff) {
4323 			high = idx - 1;
4324 		} else if (bno >= startoff + blockcount) {
4325 			low = idx + 1;
4326 		} else {
4327 			/* Convert back to file-based extent index */
4328 			if (ifp->if_flags & XFS_IFEXTIREC) {
4329 				idx += erp->er_extoff;
4330 			}
4331 			*idxp = idx;
4332 			return ep;
4333 		}
4334 	}
4335 	/* Convert back to file-based extent index */
4336 	if (ifp->if_flags & XFS_IFEXTIREC) {
4337 		idx += erp->er_extoff;
4338 	}
4339 	if (bno >= startoff + blockcount) {
4340 		if (++idx == nextents) {
4341 			ep = NULL;
4342 		} else {
4343 			ep = xfs_iext_get_ext(ifp, idx);
4344 		}
4345 	}
4346 	*idxp = idx;
4347 	return ep;
4348 }
4349 
4350 /*
4351  * Return a pointer to the indirection array entry containing the
4352  * extent record for filesystem block bno. Store the index of the
4353  * target irec in *erp_idxp.
4354  */
4355 xfs_ext_irec_t *			/* pointer to found extent record */
4356 xfs_iext_bno_to_irec(
4357 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4358 	xfs_fileoff_t	bno,		/* block number to search for */
4359 	int		*erp_idxp)	/* irec index of target ext list */
4360 {
4361 	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
4362 	xfs_ext_irec_t	*erp_next;	/* next indirection array entry */
4363 	int		erp_idx;	/* indirection array index */
4364 	int		nlists;		/* number of extent irec's (lists) */
4365 	int		high;		/* binary search upper limit */
4366 	int		low;		/* binary search lower limit */
4367 
4368 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4369 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4370 	erp_idx = 0;
4371 	low = 0;
4372 	high = nlists - 1;
4373 	while (low <= high) {
4374 		erp_idx = (low + high) >> 1;
4375 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
4376 		erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4377 		if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4378 			high = erp_idx - 1;
4379 		} else if (erp_next && bno >=
4380 			   xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4381 			low = erp_idx + 1;
4382 		} else {
4383 			break;
4384 		}
4385 	}
4386 	*erp_idxp = erp_idx;
4387 	return erp;
4388 }
4389 
4390 /*
4391  * Return a pointer to the indirection array entry containing the
4392  * extent record at file extent index *idxp. Store the index of the
4393  * target irec in *erp_idxp and store the page index of the target
4394  * extent record in *idxp.
4395  */
4396 xfs_ext_irec_t *
4397 xfs_iext_idx_to_irec(
4398 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4399 	xfs_extnum_t	*idxp,		/* extent index (file -> page) */
4400 	int		*erp_idxp,	/* pointer to target irec */
4401 	int		realloc)	/* new bytes were just added */
4402 {
4403 	xfs_ext_irec_t	*prev;		/* pointer to previous irec */
4404 	xfs_ext_irec_t	*erp = NULL;	/* pointer to current irec */
4405 	int		erp_idx;	/* indirection array index */
4406 	int		nlists;		/* number of irec's (ex lists) */
4407 	int		high;		/* binary search upper limit */
4408 	int		low;		/* binary search lower limit */
4409 	xfs_extnum_t	page_idx = *idxp; /* extent index in target list */
4410 
4411 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4412 	ASSERT(page_idx >= 0 && page_idx <=
4413 		ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4414 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4415 	erp_idx = 0;
4416 	low = 0;
4417 	high = nlists - 1;
4418 
4419 	/* Binary search extent irec's */
4420 	while (low <= high) {
4421 		erp_idx = (low + high) >> 1;
4422 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
4423 		prev = erp_idx > 0 ? erp - 1 : NULL;
4424 		if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4425 		     realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4426 			high = erp_idx - 1;
4427 		} else if (page_idx > erp->er_extoff + erp->er_extcount ||
4428 			   (page_idx == erp->er_extoff + erp->er_extcount &&
4429 			    !realloc)) {
4430 			low = erp_idx + 1;
4431 		} else if (page_idx == erp->er_extoff + erp->er_extcount &&
4432 			   erp->er_extcount == XFS_LINEAR_EXTS) {
4433 			ASSERT(realloc);
4434 			page_idx = 0;
4435 			erp_idx++;
4436 			erp = erp_idx < nlists ? erp + 1 : NULL;
4437 			break;
4438 		} else {
4439 			page_idx -= erp->er_extoff;
4440 			break;
4441 		}
4442 	}
4443 	*idxp = page_idx;
4444 	*erp_idxp = erp_idx;
4445 	return(erp);
4446 }
4447 
4448 /*
4449  * Allocate and initialize an indirection array once the space needed
4450  * for incore extents increases above XFS_IEXT_BUFSZ.
4451  */
4452 void
4453 xfs_iext_irec_init(
4454 	xfs_ifork_t	*ifp)		/* inode fork pointer */
4455 {
4456 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
4457 	xfs_extnum_t	nextents;	/* number of extents in file */
4458 
4459 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4460 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4461 	ASSERT(nextents <= XFS_LINEAR_EXTS);
4462 
4463 	erp = (xfs_ext_irec_t *)
4464 		kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4465 
4466 	if (nextents == 0) {
4467 		ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4468 			kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4469 	} else if (!ifp->if_real_bytes) {
4470 		xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4471 	} else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4472 		xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4473 	}
4474 	erp->er_extbuf = ifp->if_u1.if_extents;
4475 	erp->er_extcount = nextents;
4476 	erp->er_extoff = 0;
4477 
4478 	ifp->if_flags |= XFS_IFEXTIREC;
4479 	ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4480 	ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4481 	ifp->if_u1.if_ext_irec = erp;
4482 
4483 	return;
4484 }
4485 
4486 /*
4487  * Allocate and initialize a new entry in the indirection array.
4488  */
4489 xfs_ext_irec_t *
4490 xfs_iext_irec_new(
4491 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4492 	int		erp_idx)	/* index for new irec */
4493 {
4494 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
4495 	int		i;		/* loop counter */
4496 	int		nlists;		/* number of irec's (ex lists) */
4497 
4498 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4499 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4500 
4501 	/* Resize indirection array */
4502 	xfs_iext_realloc_indirect(ifp, ++nlists *
4503 				  sizeof(xfs_ext_irec_t));
4504 	/*
4505 	 * Move records down in the array so the
4506 	 * new page can use erp_idx.
4507 	 */
4508 	erp = ifp->if_u1.if_ext_irec;
4509 	for (i = nlists - 1; i > erp_idx; i--) {
4510 		memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4511 	}
4512 	ASSERT(i == erp_idx);
4513 
4514 	/* Initialize new extent record */
4515 	erp = ifp->if_u1.if_ext_irec;
4516 	erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4517 		kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4518 	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4519 	memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4520 	erp[erp_idx].er_extcount = 0;
4521 	erp[erp_idx].er_extoff = erp_idx > 0 ?
4522 		erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4523 	return (&erp[erp_idx]);
4524 }
4525 
4526 /*
4527  * Remove a record from the indirection array.
4528  */
4529 void
4530 xfs_iext_irec_remove(
4531 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4532 	int		erp_idx)	/* irec index to remove */
4533 {
4534 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
4535 	int		i;		/* loop counter */
4536 	int		nlists;		/* number of irec's (ex lists) */
4537 
4538 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4539 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4540 	erp = &ifp->if_u1.if_ext_irec[erp_idx];
4541 	if (erp->er_extbuf) {
4542 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4543 			-erp->er_extcount);
4544 		kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4545 	}
4546 	/* Compact extent records */
4547 	erp = ifp->if_u1.if_ext_irec;
4548 	for (i = erp_idx; i < nlists - 1; i++) {
4549 		memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4550 	}
4551 	/*
4552 	 * Manually free the last extent record from the indirection
4553 	 * array.  A call to xfs_iext_realloc_indirect() with a size
4554 	 * of zero would result in a call to xfs_iext_destroy() which
4555 	 * would in turn call this function again, creating a nasty
4556 	 * infinite loop.
4557 	 */
4558 	if (--nlists) {
4559 		xfs_iext_realloc_indirect(ifp,
4560 			nlists * sizeof(xfs_ext_irec_t));
4561 	} else {
4562 		kmem_free(ifp->if_u1.if_ext_irec,
4563 			sizeof(xfs_ext_irec_t));
4564 	}
4565 	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4566 }
4567 
4568 /*
4569  * This is called to clean up large amounts of unused memory allocated
4570  * by the indirection array.  Before compacting anything though, verify
4571  * that the indirection array is still needed and switch back to the
4572  * linear extent list (or even the inline buffer) if possible.  The
4573  * compaction policy is as follows:
4574  *
4575  *    Full Compaction: Extents fit into a single page (or inline buffer)
4576  *    Full Compaction: Extents occupy less than 10% of allocated space
4577  * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4578  *      No Compaction: Extents occupy at least 50% of allocated space
4579  */
4580 void
4581 xfs_iext_irec_compact(
4582 	xfs_ifork_t	*ifp)		/* inode fork pointer */
4583 {
4584 	xfs_extnum_t	nextents;	/* number of extents in file */
4585 	int		nlists;		/* number of irec's (ex lists) */
4586 
4587 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4588 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4589 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4590 
4591 	if (nextents == 0) {
4592 		xfs_iext_destroy(ifp);
4593 	} else if (nextents <= XFS_INLINE_EXTS) {
4594 		xfs_iext_indirect_to_direct(ifp);
4595 		xfs_iext_direct_to_inline(ifp, nextents);
4596 	} else if (nextents <= XFS_LINEAR_EXTS) {
4597 		xfs_iext_indirect_to_direct(ifp);
4598 	} else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4599 		xfs_iext_irec_compact_full(ifp);
4600 	} else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4601 		xfs_iext_irec_compact_pages(ifp);
4602 	}
4603 }
4604 
4605 /*
4606  * Combine extents from neighboring extent pages.
4607  */
4608 void
4609 xfs_iext_irec_compact_pages(
4610 	xfs_ifork_t	*ifp)		/* inode fork pointer */
4611 {
4612 	xfs_ext_irec_t	*erp, *erp_next;/* pointers to irec entries */
4613 	int		erp_idx = 0;	/* indirection array index */
4614 	int		nlists;		/* number of irec's (ex lists) */
4615 
4616 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4617 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4618 	while (erp_idx < nlists - 1) {
4619 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
4620 		erp_next = erp + 1;
4621 		if (erp_next->er_extcount <=
4622 		    (XFS_LINEAR_EXTS - erp->er_extcount)) {
4623 			memmove(&erp->er_extbuf[erp->er_extcount],
4624 				erp_next->er_extbuf, erp_next->er_extcount *
4625 				sizeof(xfs_bmbt_rec_t));
4626 			erp->er_extcount += erp_next->er_extcount;
4627 			/*
4628 			 * Free page before removing extent record
4629 			 * so er_extoffs don't get modified in
4630 			 * xfs_iext_irec_remove.
4631 			 */
4632 			kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4633 			erp_next->er_extbuf = NULL;
4634 			xfs_iext_irec_remove(ifp, erp_idx + 1);
4635 			nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4636 		} else {
4637 			erp_idx++;
4638 		}
4639 	}
4640 }
4641 
4642 /*
4643  * Fully compact the extent records managed by the indirection array.
4644  */
4645 void
4646 xfs_iext_irec_compact_full(
4647 	xfs_ifork_t	*ifp)			/* inode fork pointer */
4648 {
4649 	xfs_bmbt_rec_t	*ep, *ep_next;		/* extent record pointers */
4650 	xfs_ext_irec_t	*erp, *erp_next;	/* extent irec pointers */
4651 	int		erp_idx = 0;		/* extent irec index */
4652 	int		ext_avail;		/* empty entries in ex list */
4653 	int		ext_diff;		/* number of exts to add */
4654 	int		nlists;			/* number of irec's (ex lists) */
4655 
4656 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4657 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4658 	erp = ifp->if_u1.if_ext_irec;
4659 	ep = &erp->er_extbuf[erp->er_extcount];
4660 	erp_next = erp + 1;
4661 	ep_next = erp_next->er_extbuf;
4662 	while (erp_idx < nlists - 1) {
4663 		ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4664 		ext_diff = MIN(ext_avail, erp_next->er_extcount);
4665 		memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4666 		erp->er_extcount += ext_diff;
4667 		erp_next->er_extcount -= ext_diff;
4668 		/* Remove next page */
4669 		if (erp_next->er_extcount == 0) {
4670 			/*
4671 			 * Free page before removing extent record
4672 			 * so er_extoffs don't get modified in
4673 			 * xfs_iext_irec_remove.
4674 			 */
4675 			kmem_free(erp_next->er_extbuf,
4676 				erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4677 			erp_next->er_extbuf = NULL;
4678 			xfs_iext_irec_remove(ifp, erp_idx + 1);
4679 			erp = &ifp->if_u1.if_ext_irec[erp_idx];
4680 			nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4681 		/* Update next page */
4682 		} else {
4683 			/* Move rest of page up to become next new page */
4684 			memmove(erp_next->er_extbuf, ep_next,
4685 				erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4686 			ep_next = erp_next->er_extbuf;
4687 			memset(&ep_next[erp_next->er_extcount], 0,
4688 				(XFS_LINEAR_EXTS - erp_next->er_extcount) *
4689 				sizeof(xfs_bmbt_rec_t));
4690 		}
4691 		if (erp->er_extcount == XFS_LINEAR_EXTS) {
4692 			erp_idx++;
4693 			if (erp_idx < nlists)
4694 				erp = &ifp->if_u1.if_ext_irec[erp_idx];
4695 			else
4696 				break;
4697 		}
4698 		ep = &erp->er_extbuf[erp->er_extcount];
4699 		erp_next = erp + 1;
4700 		ep_next = erp_next->er_extbuf;
4701 	}
4702 }
4703 
4704 /*
4705  * This is called to update the er_extoff field in the indirection
4706  * array when extents have been added or removed from one of the
4707  * extent lists. erp_idx contains the irec index to begin updating
4708  * at and ext_diff contains the number of extents that were added
4709  * or removed.
4710  */
4711 void
4712 xfs_iext_irec_update_extoffs(
4713 	xfs_ifork_t	*ifp,		/* inode fork pointer */
4714 	int		erp_idx,	/* irec index to update */
4715 	int		ext_diff)	/* number of new extents */
4716 {
4717 	int		i;		/* loop counter */
4718 	int		nlists;		/* number of irec's (ex lists */
4719 
4720 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4721 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4722 	for (i = erp_idx; i < nlists; i++) {
4723 		ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4724 	}
4725 }
4726