xref: /openbmc/linux/fs/xfs/xfs_inode.c (revision 1da177e4)
1 /*
2  * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 
33 #include "xfs.h"
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
36 #include "xfs_inum.h"
37 #include "xfs_log.h"
38 #include "xfs_trans.h"
39 #include "xfs_trans_priv.h"
40 #include "xfs_sb.h"
41 #include "xfs_ag.h"
42 #include "xfs_dir.h"
43 #include "xfs_dir2.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_alloc_btree.h"
47 #include "xfs_bmap_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_imap.h"
51 #include "xfs_alloc.h"
52 #include "xfs_ialloc.h"
53 #include "xfs_attr_sf.h"
54 #include "xfs_dir_sf.h"
55 #include "xfs_dir2_sf.h"
56 #include "xfs_dinode.h"
57 #include "xfs_inode_item.h"
58 #include "xfs_inode.h"
59 #include "xfs_bmap.h"
60 #include "xfs_buf_item.h"
61 #include "xfs_rw.h"
62 #include "xfs_error.h"
63 #include "xfs_bit.h"
64 #include "xfs_utils.h"
65 #include "xfs_dir2_trace.h"
66 #include "xfs_quota.h"
67 #include "xfs_mac.h"
68 #include "xfs_acl.h"
69 
70 
71 kmem_zone_t *xfs_ifork_zone;
72 kmem_zone_t *xfs_inode_zone;
73 kmem_zone_t *xfs_chashlist_zone;
74 
75 /*
76  * Used in xfs_itruncate().  This is the maximum number of extents
77  * freed from a file in a single transaction.
78  */
79 #define	XFS_ITRUNC_MAX_EXTENTS	2
80 
81 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
82 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
83 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
84 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
85 
86 
87 #ifdef DEBUG
88 /*
89  * Make sure that the extents in the given memory buffer
90  * are valid.
91  */
92 STATIC void
93 xfs_validate_extents(
94 	xfs_bmbt_rec_t		*ep,
95 	int			nrecs,
96 	int			disk,
97 	xfs_exntfmt_t		fmt)
98 {
99 	xfs_bmbt_irec_t		irec;
100 	xfs_bmbt_rec_t		rec;
101 	int			i;
102 
103 	for (i = 0; i < nrecs; i++) {
104 		rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
105 		rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
106 		if (disk)
107 			xfs_bmbt_disk_get_all(&rec, &irec);
108 		else
109 			xfs_bmbt_get_all(&rec, &irec);
110 		if (fmt == XFS_EXTFMT_NOSTATE)
111 			ASSERT(irec.br_state == XFS_EXT_NORM);
112 		ep++;
113 	}
114 }
115 #else /* DEBUG */
116 #define xfs_validate_extents(ep, nrecs, disk, fmt)
117 #endif /* DEBUG */
118 
119 /*
120  * Check that none of the inode's in the buffer have a next
121  * unlinked field of 0.
122  */
123 #if defined(DEBUG)
124 void
125 xfs_inobp_check(
126 	xfs_mount_t	*mp,
127 	xfs_buf_t	*bp)
128 {
129 	int		i;
130 	int		j;
131 	xfs_dinode_t	*dip;
132 
133 	j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
134 
135 	for (i = 0; i < j; i++) {
136 		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
137 					i * mp->m_sb.sb_inodesize);
138 		if (!dip->di_next_unlinked)  {
139 			xfs_fs_cmn_err(CE_ALERT, mp,
140 				"Detected a bogus zero next_unlinked field in incore inode buffer 0x%p.  About to pop an ASSERT.",
141 				bp);
142 			ASSERT(dip->di_next_unlinked);
143 		}
144 	}
145 }
146 #endif
147 
148 /*
149  * called from bwrite on xfs inode buffers
150  */
151 void
152 xfs_inobp_bwcheck(xfs_buf_t *bp)
153 {
154 	xfs_mount_t	*mp;
155 	int		i;
156 	int		j;
157 	xfs_dinode_t	*dip;
158 
159 	ASSERT(XFS_BUF_FSPRIVATE3(bp, void *) != NULL);
160 
161 	mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
162 
163 
164 	j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
165 
166 	for (i = 0; i < j; i++)  {
167 		dip = (xfs_dinode_t *) xfs_buf_offset(bp,
168 						i * mp->m_sb.sb_inodesize);
169 		if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
170 			cmn_err(CE_WARN,
171 "Bad magic # 0x%x in XFS inode buffer 0x%Lx, starting blockno %Ld, offset 0x%x",
172 				INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
173 				(__uint64_t)(__psunsigned_t) bp,
174 				(__int64_t) XFS_BUF_ADDR(bp),
175 				xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
176 			xfs_fs_cmn_err(CE_WARN, mp,
177 				"corrupt, unmount and run xfs_repair");
178 		}
179 		if (!dip->di_next_unlinked)  {
180 			cmn_err(CE_WARN,
181 "Bad next_unlinked field (0) in XFS inode buffer 0x%p, starting blockno %Ld, offset 0x%x",
182 				(__uint64_t)(__psunsigned_t) bp,
183 				(__int64_t) XFS_BUF_ADDR(bp),
184 				xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
185 			xfs_fs_cmn_err(CE_WARN, mp,
186 				"corrupt, unmount and run xfs_repair");
187 		}
188 	}
189 
190 	return;
191 }
192 
193 /*
194  * This routine is called to map an inode number within a file
195  * system to the buffer containing the on-disk version of the
196  * inode.  It returns a pointer to the buffer containing the
197  * on-disk inode in the bpp parameter, and in the dip parameter
198  * it returns a pointer to the on-disk inode within that buffer.
199  *
200  * If a non-zero error is returned, then the contents of bpp and
201  * dipp are undefined.
202  *
203  * Use xfs_imap() to determine the size and location of the
204  * buffer to read from disk.
205  */
206 int
207 xfs_inotobp(
208 	xfs_mount_t	*mp,
209 	xfs_trans_t	*tp,
210 	xfs_ino_t	ino,
211 	xfs_dinode_t	**dipp,
212 	xfs_buf_t	**bpp,
213 	int		*offset)
214 {
215 	int		di_ok;
216 	xfs_imap_t	imap;
217 	xfs_buf_t	*bp;
218 	int		error;
219 	xfs_dinode_t	*dip;
220 
221 	/*
222 	 * Call the space managment code to find the location of the
223 	 * inode on disk.
224 	 */
225 	imap.im_blkno = 0;
226 	error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
227 	if (error != 0) {
228 		cmn_err(CE_WARN,
229 	"xfs_inotobp: xfs_imap()  returned an "
230 	"error %d on %s.  Returning error.", error, mp->m_fsname);
231 		return error;
232 	}
233 
234 	/*
235 	 * If the inode number maps to a block outside the bounds of the
236 	 * file system then return NULL rather than calling read_buf
237 	 * and panicing when we get an error from the driver.
238 	 */
239 	if ((imap.im_blkno + imap.im_len) >
240 	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
241 		cmn_err(CE_WARN,
242 	"xfs_inotobp: inode number (%d + %d) maps to a block outside the bounds "
243 	"of the file system %s.  Returning EINVAL.",
244 			imap.im_blkno, imap.im_len,mp->m_fsname);
245 		return XFS_ERROR(EINVAL);
246 	}
247 
248 	/*
249 	 * Read in the buffer.  If tp is NULL, xfs_trans_read_buf() will
250 	 * default to just a read_buf() call.
251 	 */
252 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
253 				   (int)imap.im_len, XFS_BUF_LOCK, &bp);
254 
255 	if (error) {
256 		cmn_err(CE_WARN,
257 	"xfs_inotobp: xfs_trans_read_buf()  returned an "
258 	"error %d on %s.  Returning error.", error, mp->m_fsname);
259 		return error;
260 	}
261 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
262 	di_ok =
263 		INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
264 		XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
265 	if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
266 			XFS_RANDOM_ITOBP_INOTOBP))) {
267 		XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
268 		xfs_trans_brelse(tp, bp);
269 		cmn_err(CE_WARN,
270 	"xfs_inotobp: XFS_TEST_ERROR()  returned an "
271 	"error on %s.  Returning EFSCORRUPTED.",  mp->m_fsname);
272 		return XFS_ERROR(EFSCORRUPTED);
273 	}
274 
275 	xfs_inobp_check(mp, bp);
276 
277 	/*
278 	 * Set *dipp to point to the on-disk inode in the buffer.
279 	 */
280 	*dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
281 	*bpp = bp;
282 	*offset = imap.im_boffset;
283 	return 0;
284 }
285 
286 
287 /*
288  * This routine is called to map an inode to the buffer containing
289  * the on-disk version of the inode.  It returns a pointer to the
290  * buffer containing the on-disk inode in the bpp parameter, and in
291  * the dip parameter it returns a pointer to the on-disk inode within
292  * that buffer.
293  *
294  * If a non-zero error is returned, then the contents of bpp and
295  * dipp are undefined.
296  *
297  * If the inode is new and has not yet been initialized, use xfs_imap()
298  * to determine the size and location of the buffer to read from disk.
299  * If the inode has already been mapped to its buffer and read in once,
300  * then use the mapping information stored in the inode rather than
301  * calling xfs_imap().  This allows us to avoid the overhead of looking
302  * at the inode btree for small block file systems (see xfs_dilocate()).
303  * We can tell whether the inode has been mapped in before by comparing
304  * its disk block address to 0.  Only uninitialized inodes will have
305  * 0 for the disk block address.
306  */
307 int
308 xfs_itobp(
309 	xfs_mount_t	*mp,
310 	xfs_trans_t	*tp,
311 	xfs_inode_t	*ip,
312 	xfs_dinode_t	**dipp,
313 	xfs_buf_t	**bpp,
314 	xfs_daddr_t	bno)
315 {
316 	xfs_buf_t	*bp;
317 	int		error;
318 	xfs_imap_t	imap;
319 #ifdef __KERNEL__
320 	int		i;
321 	int		ni;
322 #endif
323 
324 	if (ip->i_blkno == (xfs_daddr_t)0) {
325 		/*
326 		 * Call the space management code to find the location of the
327 		 * inode on disk.
328 		 */
329 		imap.im_blkno = bno;
330 		error = xfs_imap(mp, tp, ip->i_ino, &imap, XFS_IMAP_LOOKUP);
331 		if (error != 0) {
332 			return error;
333 		}
334 
335 		/*
336 		 * If the inode number maps to a block outside the bounds
337 		 * of the file system then return NULL rather than calling
338 		 * read_buf and panicing when we get an error from the
339 		 * driver.
340 		 */
341 		if ((imap.im_blkno + imap.im_len) >
342 		    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
343 #ifdef DEBUG
344 			xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
345 					"(imap.im_blkno (0x%llx) "
346 					"+ imap.im_len (0x%llx)) > "
347 					" XFS_FSB_TO_BB(mp, "
348 					"mp->m_sb.sb_dblocks) (0x%llx)",
349 					(unsigned long long) imap.im_blkno,
350 					(unsigned long long) imap.im_len,
351 					XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
352 #endif /* DEBUG */
353 			return XFS_ERROR(EINVAL);
354 		}
355 
356 		/*
357 		 * Fill in the fields in the inode that will be used to
358 		 * map the inode to its buffer from now on.
359 		 */
360 		ip->i_blkno = imap.im_blkno;
361 		ip->i_len = imap.im_len;
362 		ip->i_boffset = imap.im_boffset;
363 	} else {
364 		/*
365 		 * We've already mapped the inode once, so just use the
366 		 * mapping that we saved the first time.
367 		 */
368 		imap.im_blkno = ip->i_blkno;
369 		imap.im_len = ip->i_len;
370 		imap.im_boffset = ip->i_boffset;
371 	}
372 	ASSERT(bno == 0 || bno == imap.im_blkno);
373 
374 	/*
375 	 * Read in the buffer.  If tp is NULL, xfs_trans_read_buf() will
376 	 * default to just a read_buf() call.
377 	 */
378 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
379 				   (int)imap.im_len, XFS_BUF_LOCK, &bp);
380 
381 	if (error) {
382 #ifdef DEBUG
383 		xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
384 				"xfs_trans_read_buf() returned error %d, "
385 				"imap.im_blkno 0x%llx, imap.im_len 0x%llx",
386 				error, (unsigned long long) imap.im_blkno,
387 				(unsigned long long) imap.im_len);
388 #endif /* DEBUG */
389 		return error;
390 	}
391 #ifdef __KERNEL__
392 	/*
393 	 * Validate the magic number and version of every inode in the buffer
394 	 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
395 	 */
396 #ifdef DEBUG
397 	ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
398 #else
399 	ni = 1;
400 #endif
401 	for (i = 0; i < ni; i++) {
402 		int		di_ok;
403 		xfs_dinode_t	*dip;
404 
405 		dip = (xfs_dinode_t *)xfs_buf_offset(bp,
406 					(i << mp->m_sb.sb_inodelog));
407 		di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
408 			    XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
409 		if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
410 				 XFS_RANDOM_ITOBP_INOTOBP))) {
411 #ifdef DEBUG
412 			prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
413 				mp->m_ddev_targp,
414 				(unsigned long long)imap.im_blkno, i,
415 				INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
416 #endif
417 			XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
418 					     mp, dip);
419 			xfs_trans_brelse(tp, bp);
420 			return XFS_ERROR(EFSCORRUPTED);
421 		}
422 	}
423 #endif	/* __KERNEL__ */
424 
425 	xfs_inobp_check(mp, bp);
426 
427 	/*
428 	 * Mark the buffer as an inode buffer now that it looks good
429 	 */
430 	XFS_BUF_SET_VTYPE(bp, B_FS_INO);
431 
432 	/*
433 	 * Set *dipp to point to the on-disk inode in the buffer.
434 	 */
435 	*dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
436 	*bpp = bp;
437 	return 0;
438 }
439 
440 /*
441  * Move inode type and inode format specific information from the
442  * on-disk inode to the in-core inode.  For fifos, devs, and sockets
443  * this means set if_rdev to the proper value.  For files, directories,
444  * and symlinks this means to bring in the in-line data or extent
445  * pointers.  For a file in B-tree format, only the root is immediately
446  * brought in-core.  The rest will be in-lined in if_extents when it
447  * is first referenced (see xfs_iread_extents()).
448  */
449 STATIC int
450 xfs_iformat(
451 	xfs_inode_t		*ip,
452 	xfs_dinode_t		*dip)
453 {
454 	xfs_attr_shortform_t	*atp;
455 	int			size;
456 	int			error;
457 	xfs_fsize_t             di_size;
458 	ip->i_df.if_ext_max =
459 		XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
460 	error = 0;
461 
462 	if (unlikely(
463 	    INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
464 		INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
465 	    INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
466 		xfs_fs_cmn_err(CE_WARN, ip->i_mount,
467 			"corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
468 			"  Unmount and run xfs_repair.",
469 			(unsigned long long)ip->i_ino,
470 			(int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
471 			    + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
472 			(unsigned long long)
473 			INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
474 		XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
475 				     ip->i_mount, dip);
476 		return XFS_ERROR(EFSCORRUPTED);
477 	}
478 
479 	if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
480 		xfs_fs_cmn_err(CE_WARN, ip->i_mount,
481 			"corrupt dinode %Lu, forkoff = 0x%x."
482 			"  Unmount and run xfs_repair.",
483 			(unsigned long long)ip->i_ino,
484 			(int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
485 		XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
486 				     ip->i_mount, dip);
487 		return XFS_ERROR(EFSCORRUPTED);
488 	}
489 
490 	switch (ip->i_d.di_mode & S_IFMT) {
491 	case S_IFIFO:
492 	case S_IFCHR:
493 	case S_IFBLK:
494 	case S_IFSOCK:
495 		if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
496 			XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
497 					      ip->i_mount, dip);
498 			return XFS_ERROR(EFSCORRUPTED);
499 		}
500 		ip->i_d.di_size = 0;
501 		ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
502 		break;
503 
504 	case S_IFREG:
505 	case S_IFLNK:
506 	case S_IFDIR:
507 		switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
508 		case XFS_DINODE_FMT_LOCAL:
509 			/*
510 			 * no local regular files yet
511 			 */
512 			if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
513 				xfs_fs_cmn_err(CE_WARN, ip->i_mount,
514 					"corrupt inode (local format for regular file) %Lu.  Unmount and run xfs_repair.",
515 					(unsigned long long) ip->i_ino);
516 				XFS_CORRUPTION_ERROR("xfs_iformat(4)",
517 						     XFS_ERRLEVEL_LOW,
518 						     ip->i_mount, dip);
519 				return XFS_ERROR(EFSCORRUPTED);
520 			}
521 
522 			di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
523 			if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
524 				xfs_fs_cmn_err(CE_WARN, ip->i_mount,
525 					"corrupt inode %Lu (bad size %Ld for local inode).  Unmount and run xfs_repair.",
526 					(unsigned long long) ip->i_ino,
527 					(long long) di_size);
528 				XFS_CORRUPTION_ERROR("xfs_iformat(5)",
529 						     XFS_ERRLEVEL_LOW,
530 						     ip->i_mount, dip);
531 				return XFS_ERROR(EFSCORRUPTED);
532 			}
533 
534 			size = (int)di_size;
535 			error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
536 			break;
537 		case XFS_DINODE_FMT_EXTENTS:
538 			error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
539 			break;
540 		case XFS_DINODE_FMT_BTREE:
541 			error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
542 			break;
543 		default:
544 			XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
545 					 ip->i_mount);
546 			return XFS_ERROR(EFSCORRUPTED);
547 		}
548 		break;
549 
550 	default:
551 		XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
552 		return XFS_ERROR(EFSCORRUPTED);
553 	}
554 	if (error) {
555 		return error;
556 	}
557 	if (!XFS_DFORK_Q(dip))
558 		return 0;
559 	ASSERT(ip->i_afp == NULL);
560 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
561 	ip->i_afp->if_ext_max =
562 		XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
563 	switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
564 	case XFS_DINODE_FMT_LOCAL:
565 		atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
566 		size = (int)INT_GET(atp->hdr.totsize, ARCH_CONVERT);
567 		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
568 		break;
569 	case XFS_DINODE_FMT_EXTENTS:
570 		error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
571 		break;
572 	case XFS_DINODE_FMT_BTREE:
573 		error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
574 		break;
575 	default:
576 		error = XFS_ERROR(EFSCORRUPTED);
577 		break;
578 	}
579 	if (error) {
580 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
581 		ip->i_afp = NULL;
582 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
583 	}
584 	return error;
585 }
586 
587 /*
588  * The file is in-lined in the on-disk inode.
589  * If it fits into if_inline_data, then copy
590  * it there, otherwise allocate a buffer for it
591  * and copy the data there.  Either way, set
592  * if_data to point at the data.
593  * If we allocate a buffer for the data, make
594  * sure that its size is a multiple of 4 and
595  * record the real size in i_real_bytes.
596  */
597 STATIC int
598 xfs_iformat_local(
599 	xfs_inode_t	*ip,
600 	xfs_dinode_t	*dip,
601 	int		whichfork,
602 	int		size)
603 {
604 	xfs_ifork_t	*ifp;
605 	int		real_size;
606 
607 	/*
608 	 * If the size is unreasonable, then something
609 	 * is wrong and we just bail out rather than crash in
610 	 * kmem_alloc() or memcpy() below.
611 	 */
612 	if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
613 		xfs_fs_cmn_err(CE_WARN, ip->i_mount,
614 			"corrupt inode %Lu (bad size %d for local fork, size = %d).  Unmount and run xfs_repair.",
615 			(unsigned long long) ip->i_ino, size,
616 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
617 		XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
618 				     ip->i_mount, dip);
619 		return XFS_ERROR(EFSCORRUPTED);
620 	}
621 	ifp = XFS_IFORK_PTR(ip, whichfork);
622 	real_size = 0;
623 	if (size == 0)
624 		ifp->if_u1.if_data = NULL;
625 	else if (size <= sizeof(ifp->if_u2.if_inline_data))
626 		ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
627 	else {
628 		real_size = roundup(size, 4);
629 		ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
630 	}
631 	ifp->if_bytes = size;
632 	ifp->if_real_bytes = real_size;
633 	if (size)
634 		memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
635 	ifp->if_flags &= ~XFS_IFEXTENTS;
636 	ifp->if_flags |= XFS_IFINLINE;
637 	return 0;
638 }
639 
640 /*
641  * The file consists of a set of extents all
642  * of which fit into the on-disk inode.
643  * If there are few enough extents to fit into
644  * the if_inline_ext, then copy them there.
645  * Otherwise allocate a buffer for them and copy
646  * them into it.  Either way, set if_extents
647  * to point at the extents.
648  */
649 STATIC int
650 xfs_iformat_extents(
651 	xfs_inode_t	*ip,
652 	xfs_dinode_t	*dip,
653 	int		whichfork)
654 {
655 	xfs_bmbt_rec_t	*ep, *dp;
656 	xfs_ifork_t	*ifp;
657 	int		nex;
658 	int		real_size;
659 	int		size;
660 	int		i;
661 
662 	ifp = XFS_IFORK_PTR(ip, whichfork);
663 	nex = XFS_DFORK_NEXTENTS(dip, whichfork);
664 	size = nex * (uint)sizeof(xfs_bmbt_rec_t);
665 
666 	/*
667 	 * If the number of extents is unreasonable, then something
668 	 * is wrong and we just bail out rather than crash in
669 	 * kmem_alloc() or memcpy() below.
670 	 */
671 	if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
672 		xfs_fs_cmn_err(CE_WARN, ip->i_mount,
673 			"corrupt inode %Lu ((a)extents = %d).  Unmount and run xfs_repair.",
674 			(unsigned long long) ip->i_ino, nex);
675 		XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
676 				     ip->i_mount, dip);
677 		return XFS_ERROR(EFSCORRUPTED);
678 	}
679 
680 	real_size = 0;
681 	if (nex == 0)
682 		ifp->if_u1.if_extents = NULL;
683 	else if (nex <= XFS_INLINE_EXTS)
684 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
685 	else {
686 		ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
687 		ASSERT(ifp->if_u1.if_extents != NULL);
688 		real_size = size;
689 	}
690 	ifp->if_bytes = size;
691 	ifp->if_real_bytes = real_size;
692 	if (size) {
693 		dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
694 		xfs_validate_extents(dp, nex, 1, XFS_EXTFMT_INODE(ip));
695 		ep = ifp->if_u1.if_extents;
696 		for (i = 0; i < nex; i++, ep++, dp++) {
697 			ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
698 								ARCH_CONVERT);
699 			ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
700 								ARCH_CONVERT);
701 		}
702 		xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
703 			whichfork);
704 		if (whichfork != XFS_DATA_FORK ||
705 			XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
706 				if (unlikely(xfs_check_nostate_extents(
707 				    ifp->if_u1.if_extents, nex))) {
708 					XFS_ERROR_REPORT("xfs_iformat_extents(2)",
709 							 XFS_ERRLEVEL_LOW,
710 							 ip->i_mount);
711 					return XFS_ERROR(EFSCORRUPTED);
712 				}
713 	}
714 	ifp->if_flags |= XFS_IFEXTENTS;
715 	return 0;
716 }
717 
718 /*
719  * The file has too many extents to fit into
720  * the inode, so they are in B-tree format.
721  * Allocate a buffer for the root of the B-tree
722  * and copy the root into it.  The i_extents
723  * field will remain NULL until all of the
724  * extents are read in (when they are needed).
725  */
726 STATIC int
727 xfs_iformat_btree(
728 	xfs_inode_t		*ip,
729 	xfs_dinode_t		*dip,
730 	int			whichfork)
731 {
732 	xfs_bmdr_block_t	*dfp;
733 	xfs_ifork_t		*ifp;
734 	/* REFERENCED */
735 	int			nrecs;
736 	int			size;
737 
738 	ifp = XFS_IFORK_PTR(ip, whichfork);
739 	dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
740 	size = XFS_BMAP_BROOT_SPACE(dfp);
741 	nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
742 
743 	/*
744 	 * blow out if -- fork has less extents than can fit in
745 	 * fork (fork shouldn't be a btree format), root btree
746 	 * block has more records than can fit into the fork,
747 	 * or the number of extents is greater than the number of
748 	 * blocks.
749 	 */
750 	if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
751 	    || XFS_BMDR_SPACE_CALC(nrecs) >
752 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
753 	    || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
754 		xfs_fs_cmn_err(CE_WARN, ip->i_mount,
755 			"corrupt inode %Lu (btree).  Unmount and run xfs_repair.",
756 			(unsigned long long) ip->i_ino);
757 		XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
758 				 ip->i_mount);
759 		return XFS_ERROR(EFSCORRUPTED);
760 	}
761 
762 	ifp->if_broot_bytes = size;
763 	ifp->if_broot = kmem_alloc(size, KM_SLEEP);
764 	ASSERT(ifp->if_broot != NULL);
765 	/*
766 	 * Copy and convert from the on-disk structure
767 	 * to the in-memory structure.
768 	 */
769 	xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
770 		ifp->if_broot, size);
771 	ifp->if_flags &= ~XFS_IFEXTENTS;
772 	ifp->if_flags |= XFS_IFBROOT;
773 
774 	return 0;
775 }
776 
777 /*
778  * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
779  * and native format
780  *
781  * buf  = on-disk representation
782  * dip  = native representation
783  * dir  = direction - +ve -> disk to native
784  *                    -ve -> native to disk
785  */
786 void
787 xfs_xlate_dinode_core(
788 	xfs_caddr_t		buf,
789 	xfs_dinode_core_t	*dip,
790 	int			dir)
791 {
792 	xfs_dinode_core_t	*buf_core = (xfs_dinode_core_t *)buf;
793 	xfs_dinode_core_t	*mem_core = (xfs_dinode_core_t *)dip;
794 	xfs_arch_t		arch = ARCH_CONVERT;
795 
796 	ASSERT(dir);
797 
798 	INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
799 	INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
800 	INT_XLATE(buf_core->di_version,	mem_core->di_version, dir, arch);
801 	INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
802 	INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
803 	INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
804 	INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
805 	INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
806 	INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
807 
808 	if (dir > 0) {
809 		memcpy(mem_core->di_pad, buf_core->di_pad,
810 			sizeof(buf_core->di_pad));
811 	} else {
812 		memcpy(buf_core->di_pad, mem_core->di_pad,
813 			sizeof(buf_core->di_pad));
814 	}
815 
816 	INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
817 
818 	INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
819 			dir, arch);
820 	INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
821 			dir, arch);
822 	INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
823 			dir, arch);
824 	INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
825 			dir, arch);
826 	INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
827 			dir, arch);
828 	INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
829 			dir, arch);
830 	INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
831 	INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
832 	INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
833 	INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
834 	INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
835 	INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
836 	INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
837 	INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
838 	INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
839 	INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
840 	INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
841 }
842 
843 STATIC uint
844 _xfs_dic2xflags(
845 	xfs_dinode_core_t	*dic,
846 	__uint16_t		di_flags)
847 {
848 	uint			flags = 0;
849 
850 	if (di_flags & XFS_DIFLAG_ANY) {
851 		if (di_flags & XFS_DIFLAG_REALTIME)
852 			flags |= XFS_XFLAG_REALTIME;
853 		if (di_flags & XFS_DIFLAG_PREALLOC)
854 			flags |= XFS_XFLAG_PREALLOC;
855 		if (di_flags & XFS_DIFLAG_IMMUTABLE)
856 			flags |= XFS_XFLAG_IMMUTABLE;
857 		if (di_flags & XFS_DIFLAG_APPEND)
858 			flags |= XFS_XFLAG_APPEND;
859 		if (di_flags & XFS_DIFLAG_SYNC)
860 			flags |= XFS_XFLAG_SYNC;
861 		if (di_flags & XFS_DIFLAG_NOATIME)
862 			flags |= XFS_XFLAG_NOATIME;
863 		if (di_flags & XFS_DIFLAG_NODUMP)
864 			flags |= XFS_XFLAG_NODUMP;
865 		if (di_flags & XFS_DIFLAG_RTINHERIT)
866 			flags |= XFS_XFLAG_RTINHERIT;
867 		if (di_flags & XFS_DIFLAG_PROJINHERIT)
868 			flags |= XFS_XFLAG_PROJINHERIT;
869 		if (di_flags & XFS_DIFLAG_NOSYMLINKS)
870 			flags |= XFS_XFLAG_NOSYMLINKS;
871 	}
872 
873 	return flags;
874 }
875 
876 uint
877 xfs_ip2xflags(
878 	xfs_inode_t		*ip)
879 {
880 	xfs_dinode_core_t	*dic = &ip->i_d;
881 
882 	return _xfs_dic2xflags(dic, dic->di_flags) |
883 		(XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
884 }
885 
886 uint
887 xfs_dic2xflags(
888 	xfs_dinode_core_t	*dic)
889 {
890 	return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
891 		(XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
892 }
893 
894 /*
895  * Given a mount structure and an inode number, return a pointer
896  * to a newly allocated in-core inode coresponding to the given
897  * inode number.
898  *
899  * Initialize the inode's attributes and extent pointers if it
900  * already has them (it will not if the inode has no links).
901  */
902 int
903 xfs_iread(
904 	xfs_mount_t	*mp,
905 	xfs_trans_t	*tp,
906 	xfs_ino_t	ino,
907 	xfs_inode_t	**ipp,
908 	xfs_daddr_t	bno)
909 {
910 	xfs_buf_t	*bp;
911 	xfs_dinode_t	*dip;
912 	xfs_inode_t	*ip;
913 	int		error;
914 
915 	ASSERT(xfs_inode_zone != NULL);
916 
917 	ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
918 	ip->i_ino = ino;
919 	ip->i_mount = mp;
920 
921 	/*
922 	 * Get pointer's to the on-disk inode and the buffer containing it.
923 	 * If the inode number refers to a block outside the file system
924 	 * then xfs_itobp() will return NULL.  In this case we should
925 	 * return NULL as well.  Set i_blkno to 0 so that xfs_itobp() will
926 	 * know that this is a new incore inode.
927 	 */
928 	error = xfs_itobp(mp, tp, ip, &dip, &bp, bno);
929 
930 	if (error != 0) {
931 		kmem_zone_free(xfs_inode_zone, ip);
932 		return error;
933 	}
934 
935 	/*
936 	 * Initialize inode's trace buffers.
937 	 * Do this before xfs_iformat in case it adds entries.
938 	 */
939 #ifdef XFS_BMAP_TRACE
940 	ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
941 #endif
942 #ifdef XFS_BMBT_TRACE
943 	ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
944 #endif
945 #ifdef XFS_RW_TRACE
946 	ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
947 #endif
948 #ifdef XFS_ILOCK_TRACE
949 	ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
950 #endif
951 #ifdef XFS_DIR2_TRACE
952 	ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
953 #endif
954 
955 	/*
956 	 * If we got something that isn't an inode it means someone
957 	 * (nfs or dmi) has a stale handle.
958 	 */
959 	if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
960 		kmem_zone_free(xfs_inode_zone, ip);
961 		xfs_trans_brelse(tp, bp);
962 #ifdef DEBUG
963 		xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
964 				"dip->di_core.di_magic (0x%x) != "
965 				"XFS_DINODE_MAGIC (0x%x)",
966 				INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
967 				XFS_DINODE_MAGIC);
968 #endif /* DEBUG */
969 		return XFS_ERROR(EINVAL);
970 	}
971 
972 	/*
973 	 * If the on-disk inode is already linked to a directory
974 	 * entry, copy all of the inode into the in-core inode.
975 	 * xfs_iformat() handles copying in the inode format
976 	 * specific information.
977 	 * Otherwise, just get the truly permanent information.
978 	 */
979 	if (dip->di_core.di_mode) {
980 		xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
981 		     &(ip->i_d), 1);
982 		error = xfs_iformat(ip, dip);
983 		if (error)  {
984 			kmem_zone_free(xfs_inode_zone, ip);
985 			xfs_trans_brelse(tp, bp);
986 #ifdef DEBUG
987 			xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
988 					"xfs_iformat() returned error %d",
989 					error);
990 #endif /* DEBUG */
991 			return error;
992 		}
993 	} else {
994 		ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
995 		ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
996 		ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
997 		ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
998 		/*
999 		 * Make sure to pull in the mode here as well in
1000 		 * case the inode is released without being used.
1001 		 * This ensures that xfs_inactive() will see that
1002 		 * the inode is already free and not try to mess
1003 		 * with the uninitialized part of it.
1004 		 */
1005 		ip->i_d.di_mode = 0;
1006 		/*
1007 		 * Initialize the per-fork minima and maxima for a new
1008 		 * inode here.  xfs_iformat will do it for old inodes.
1009 		 */
1010 		ip->i_df.if_ext_max =
1011 			XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
1012 	}
1013 
1014 	INIT_LIST_HEAD(&ip->i_reclaim);
1015 
1016 	/*
1017 	 * The inode format changed when we moved the link count and
1018 	 * made it 32 bits long.  If this is an old format inode,
1019 	 * convert it in memory to look like a new one.  If it gets
1020 	 * flushed to disk we will convert back before flushing or
1021 	 * logging it.  We zero out the new projid field and the old link
1022 	 * count field.  We'll handle clearing the pad field (the remains
1023 	 * of the old uuid field) when we actually convert the inode to
1024 	 * the new format. We don't change the version number so that we
1025 	 * can distinguish this from a real new format inode.
1026 	 */
1027 	if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1028 		ip->i_d.di_nlink = ip->i_d.di_onlink;
1029 		ip->i_d.di_onlink = 0;
1030 		ip->i_d.di_projid = 0;
1031 	}
1032 
1033 	ip->i_delayed_blks = 0;
1034 
1035 	/*
1036 	 * Mark the buffer containing the inode as something to keep
1037 	 * around for a while.  This helps to keep recently accessed
1038 	 * meta-data in-core longer.
1039 	 */
1040 	 XFS_BUF_SET_REF(bp, XFS_INO_REF);
1041 
1042 	/*
1043 	 * Use xfs_trans_brelse() to release the buffer containing the
1044 	 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1045 	 * in xfs_itobp() above.  If tp is NULL, this is just a normal
1046 	 * brelse().  If we're within a transaction, then xfs_trans_brelse()
1047 	 * will only release the buffer if it is not dirty within the
1048 	 * transaction.  It will be OK to release the buffer in this case,
1049 	 * because inodes on disk are never destroyed and we will be
1050 	 * locking the new in-core inode before putting it in the hash
1051 	 * table where other processes can find it.  Thus we don't have
1052 	 * to worry about the inode being changed just because we released
1053 	 * the buffer.
1054 	 */
1055 	xfs_trans_brelse(tp, bp);
1056 	*ipp = ip;
1057 	return 0;
1058 }
1059 
1060 /*
1061  * Read in extents from a btree-format inode.
1062  * Allocate and fill in if_extents.  Real work is done in xfs_bmap.c.
1063  */
1064 int
1065 xfs_iread_extents(
1066 	xfs_trans_t	*tp,
1067 	xfs_inode_t	*ip,
1068 	int		whichfork)
1069 {
1070 	int		error;
1071 	xfs_ifork_t	*ifp;
1072 	size_t		size;
1073 
1074 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1075 		XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1076 				 ip->i_mount);
1077 		return XFS_ERROR(EFSCORRUPTED);
1078 	}
1079 	size = XFS_IFORK_NEXTENTS(ip, whichfork) * (uint)sizeof(xfs_bmbt_rec_t);
1080 	ifp = XFS_IFORK_PTR(ip, whichfork);
1081 	/*
1082 	 * We know that the size is valid (it's checked in iformat_btree)
1083 	 */
1084 	ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
1085 	ASSERT(ifp->if_u1.if_extents != NULL);
1086 	ifp->if_lastex = NULLEXTNUM;
1087 	ifp->if_bytes = ifp->if_real_bytes = (int)size;
1088 	ifp->if_flags |= XFS_IFEXTENTS;
1089 	error = xfs_bmap_read_extents(tp, ip, whichfork);
1090 	if (error) {
1091 		kmem_free(ifp->if_u1.if_extents, size);
1092 		ifp->if_u1.if_extents = NULL;
1093 		ifp->if_bytes = ifp->if_real_bytes = 0;
1094 		ifp->if_flags &= ~XFS_IFEXTENTS;
1095 		return error;
1096 	}
1097 	xfs_validate_extents((xfs_bmbt_rec_t *)ifp->if_u1.if_extents,
1098 		XFS_IFORK_NEXTENTS(ip, whichfork), 0, XFS_EXTFMT_INODE(ip));
1099 	return 0;
1100 }
1101 
1102 /*
1103  * Allocate an inode on disk and return a copy of its in-core version.
1104  * The in-core inode is locked exclusively.  Set mode, nlink, and rdev
1105  * appropriately within the inode.  The uid and gid for the inode are
1106  * set according to the contents of the given cred structure.
1107  *
1108  * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1109  * has a free inode available, call xfs_iget()
1110  * to obtain the in-core version of the allocated inode.  Finally,
1111  * fill in the inode and log its initial contents.  In this case,
1112  * ialloc_context would be set to NULL and call_again set to false.
1113  *
1114  * If xfs_dialloc() does not have an available inode,
1115  * it will replenish its supply by doing an allocation. Since we can
1116  * only do one allocation within a transaction without deadlocks, we
1117  * must commit the current transaction before returning the inode itself.
1118  * In this case, therefore, we will set call_again to true and return.
1119  * The caller should then commit the current transaction, start a new
1120  * transaction, and call xfs_ialloc() again to actually get the inode.
1121  *
1122  * To ensure that some other process does not grab the inode that
1123  * was allocated during the first call to xfs_ialloc(), this routine
1124  * also returns the [locked] bp pointing to the head of the freelist
1125  * as ialloc_context.  The caller should hold this buffer across
1126  * the commit and pass it back into this routine on the second call.
1127  */
1128 int
1129 xfs_ialloc(
1130 	xfs_trans_t	*tp,
1131 	xfs_inode_t	*pip,
1132 	mode_t		mode,
1133 	nlink_t		nlink,
1134 	xfs_dev_t	rdev,
1135 	cred_t		*cr,
1136 	xfs_prid_t	prid,
1137 	int		okalloc,
1138 	xfs_buf_t	**ialloc_context,
1139 	boolean_t	*call_again,
1140 	xfs_inode_t	**ipp)
1141 {
1142 	xfs_ino_t	ino;
1143 	xfs_inode_t	*ip;
1144 	vnode_t		*vp;
1145 	uint		flags;
1146 	int		error;
1147 
1148 	/*
1149 	 * Call the space management code to pick
1150 	 * the on-disk inode to be allocated.
1151 	 */
1152 	error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1153 			    ialloc_context, call_again, &ino);
1154 	if (error != 0) {
1155 		return error;
1156 	}
1157 	if (*call_again || ino == NULLFSINO) {
1158 		*ipp = NULL;
1159 		return 0;
1160 	}
1161 	ASSERT(*ialloc_context == NULL);
1162 
1163 	/*
1164 	 * Get the in-core inode with the lock held exclusively.
1165 	 * This is because we're setting fields here we need
1166 	 * to prevent others from looking at until we're done.
1167 	 */
1168 	error = xfs_trans_iget(tp->t_mountp, tp, ino,
1169 			IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1170 	if (error != 0) {
1171 		return error;
1172 	}
1173 	ASSERT(ip != NULL);
1174 
1175 	vp = XFS_ITOV(ip);
1176 	vp->v_type = IFTOVT(mode);
1177 	ip->i_d.di_mode = (__uint16_t)mode;
1178 	ip->i_d.di_onlink = 0;
1179 	ip->i_d.di_nlink = nlink;
1180 	ASSERT(ip->i_d.di_nlink == nlink);
1181 	ip->i_d.di_uid = current_fsuid(cr);
1182 	ip->i_d.di_gid = current_fsgid(cr);
1183 	ip->i_d.di_projid = prid;
1184 	memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1185 
1186 	/*
1187 	 * If the superblock version is up to where we support new format
1188 	 * inodes and this is currently an old format inode, then change
1189 	 * the inode version number now.  This way we only do the conversion
1190 	 * here rather than here and in the flush/logging code.
1191 	 */
1192 	if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1193 	    ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1194 		ip->i_d.di_version = XFS_DINODE_VERSION_2;
1195 		/*
1196 		 * We've already zeroed the old link count, the projid field,
1197 		 * and the pad field.
1198 		 */
1199 	}
1200 
1201 	/*
1202 	 * Project ids won't be stored on disk if we are using a version 1 inode.
1203 	 */
1204 	if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1205 		xfs_bump_ino_vers2(tp, ip);
1206 
1207 	if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1208 		ip->i_d.di_gid = pip->i_d.di_gid;
1209 		if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1210 			ip->i_d.di_mode |= S_ISGID;
1211 		}
1212 	}
1213 
1214 	/*
1215 	 * If the group ID of the new file does not match the effective group
1216 	 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1217 	 * (and only if the irix_sgid_inherit compatibility variable is set).
1218 	 */
1219 	if ((irix_sgid_inherit) &&
1220 	    (ip->i_d.di_mode & S_ISGID) &&
1221 	    (!in_group_p((gid_t)ip->i_d.di_gid))) {
1222 		ip->i_d.di_mode &= ~S_ISGID;
1223 	}
1224 
1225 	ip->i_d.di_size = 0;
1226 	ip->i_d.di_nextents = 0;
1227 	ASSERT(ip->i_d.di_nblocks == 0);
1228 	xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1229 	/*
1230 	 * di_gen will have been taken care of in xfs_iread.
1231 	 */
1232 	ip->i_d.di_extsize = 0;
1233 	ip->i_d.di_dmevmask = 0;
1234 	ip->i_d.di_dmstate = 0;
1235 	ip->i_d.di_flags = 0;
1236 	flags = XFS_ILOG_CORE;
1237 	switch (mode & S_IFMT) {
1238 	case S_IFIFO:
1239 	case S_IFCHR:
1240 	case S_IFBLK:
1241 	case S_IFSOCK:
1242 		ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1243 		ip->i_df.if_u2.if_rdev = rdev;
1244 		ip->i_df.if_flags = 0;
1245 		flags |= XFS_ILOG_DEV;
1246 		break;
1247 	case S_IFREG:
1248 	case S_IFDIR:
1249 		if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1250 			if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1251 				if ((mode & S_IFMT) == S_IFDIR) {
1252 					ip->i_d.di_flags |= XFS_DIFLAG_RTINHERIT;
1253 				} else {
1254 					ip->i_d.di_flags |= XFS_DIFLAG_REALTIME;
1255 					ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1256 				}
1257 			}
1258 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1259 			    xfs_inherit_noatime)
1260 				ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
1261 			if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1262 			    xfs_inherit_nodump)
1263 				ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
1264 			if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1265 			    xfs_inherit_sync)
1266 				ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
1267 			if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1268 			    xfs_inherit_nosymlinks)
1269 				ip->i_d.di_flags |= XFS_DIFLAG_NOSYMLINKS;
1270 		}
1271 		/* FALLTHROUGH */
1272 	case S_IFLNK:
1273 		ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1274 		ip->i_df.if_flags = XFS_IFEXTENTS;
1275 		ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1276 		ip->i_df.if_u1.if_extents = NULL;
1277 		break;
1278 	default:
1279 		ASSERT(0);
1280 	}
1281 	/*
1282 	 * Attribute fork settings for new inode.
1283 	 */
1284 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1285 	ip->i_d.di_anextents = 0;
1286 
1287 	/*
1288 	 * Log the new values stuffed into the inode.
1289 	 */
1290 	xfs_trans_log_inode(tp, ip, flags);
1291 
1292 	/* now that we have a v_type we can set Linux inode ops (& unlock) */
1293 	VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1294 
1295 	*ipp = ip;
1296 	return 0;
1297 }
1298 
1299 /*
1300  * Check to make sure that there are no blocks allocated to the
1301  * file beyond the size of the file.  We don't check this for
1302  * files with fixed size extents or real time extents, but we
1303  * at least do it for regular files.
1304  */
1305 #ifdef DEBUG
1306 void
1307 xfs_isize_check(
1308 	xfs_mount_t	*mp,
1309 	xfs_inode_t	*ip,
1310 	xfs_fsize_t	isize)
1311 {
1312 	xfs_fileoff_t	map_first;
1313 	int		nimaps;
1314 	xfs_bmbt_irec_t	imaps[2];
1315 
1316 	if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1317 		return;
1318 
1319 	if ( ip->i_d.di_flags & XFS_DIFLAG_REALTIME )
1320 		return;
1321 
1322 	nimaps = 2;
1323 	map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1324 	/*
1325 	 * The filesystem could be shutting down, so bmapi may return
1326 	 * an error.
1327 	 */
1328 	if (xfs_bmapi(NULL, ip, map_first,
1329 			 (XFS_B_TO_FSB(mp,
1330 				       (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1331 			  map_first),
1332 			 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1333 			 NULL))
1334 	    return;
1335 	ASSERT(nimaps == 1);
1336 	ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1337 }
1338 #endif	/* DEBUG */
1339 
1340 /*
1341  * Calculate the last possible buffered byte in a file.  This must
1342  * include data that was buffered beyond the EOF by the write code.
1343  * This also needs to deal with overflowing the xfs_fsize_t type
1344  * which can happen for sizes near the limit.
1345  *
1346  * We also need to take into account any blocks beyond the EOF.  It
1347  * may be the case that they were buffered by a write which failed.
1348  * In that case the pages will still be in memory, but the inode size
1349  * will never have been updated.
1350  */
1351 xfs_fsize_t
1352 xfs_file_last_byte(
1353 	xfs_inode_t	*ip)
1354 {
1355 	xfs_mount_t	*mp;
1356 	xfs_fsize_t	last_byte;
1357 	xfs_fileoff_t	last_block;
1358 	xfs_fileoff_t	size_last_block;
1359 	int		error;
1360 
1361 	ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1362 
1363 	mp = ip->i_mount;
1364 	/*
1365 	 * Only check for blocks beyond the EOF if the extents have
1366 	 * been read in.  This eliminates the need for the inode lock,
1367 	 * and it also saves us from looking when it really isn't
1368 	 * necessary.
1369 	 */
1370 	if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1371 		error = xfs_bmap_last_offset(NULL, ip, &last_block,
1372 			XFS_DATA_FORK);
1373 		if (error) {
1374 			last_block = 0;
1375 		}
1376 	} else {
1377 		last_block = 0;
1378 	}
1379 	size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1380 	last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1381 
1382 	last_byte = XFS_FSB_TO_B(mp, last_block);
1383 	if (last_byte < 0) {
1384 		return XFS_MAXIOFFSET(mp);
1385 	}
1386 	last_byte += (1 << mp->m_writeio_log);
1387 	if (last_byte < 0) {
1388 		return XFS_MAXIOFFSET(mp);
1389 	}
1390 	return last_byte;
1391 }
1392 
1393 #if defined(XFS_RW_TRACE)
1394 STATIC void
1395 xfs_itrunc_trace(
1396 	int		tag,
1397 	xfs_inode_t	*ip,
1398 	int		flag,
1399 	xfs_fsize_t	new_size,
1400 	xfs_off_t	toss_start,
1401 	xfs_off_t	toss_finish)
1402 {
1403 	if (ip->i_rwtrace == NULL) {
1404 		return;
1405 	}
1406 
1407 	ktrace_enter(ip->i_rwtrace,
1408 		     (void*)((long)tag),
1409 		     (void*)ip,
1410 		     (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1411 		     (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1412 		     (void*)((long)flag),
1413 		     (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1414 		     (void*)(unsigned long)(new_size & 0xffffffff),
1415 		     (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1416 		     (void*)(unsigned long)(toss_start & 0xffffffff),
1417 		     (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1418 		     (void*)(unsigned long)(toss_finish & 0xffffffff),
1419 		     (void*)(unsigned long)current_cpu(),
1420 		     (void*)0,
1421 		     (void*)0,
1422 		     (void*)0,
1423 		     (void*)0);
1424 }
1425 #else
1426 #define	xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1427 #endif
1428 
1429 /*
1430  * Start the truncation of the file to new_size.  The new size
1431  * must be smaller than the current size.  This routine will
1432  * clear the buffer and page caches of file data in the removed
1433  * range, and xfs_itruncate_finish() will remove the underlying
1434  * disk blocks.
1435  *
1436  * The inode must have its I/O lock locked EXCLUSIVELY, and it
1437  * must NOT have the inode lock held at all.  This is because we're
1438  * calling into the buffer/page cache code and we can't hold the
1439  * inode lock when we do so.
1440  *
1441  * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1442  * or XFS_ITRUNC_MAYBE.  The XFS_ITRUNC_MAYBE value should be used
1443  * in the case that the caller is locking things out of order and
1444  * may not be able to call xfs_itruncate_finish() with the inode lock
1445  * held without dropping the I/O lock.  If the caller must drop the
1446  * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1447  * must be called again with all the same restrictions as the initial
1448  * call.
1449  */
1450 void
1451 xfs_itruncate_start(
1452 	xfs_inode_t	*ip,
1453 	uint		flags,
1454 	xfs_fsize_t	new_size)
1455 {
1456 	xfs_fsize_t	last_byte;
1457 	xfs_off_t	toss_start;
1458 	xfs_mount_t	*mp;
1459 	vnode_t		*vp;
1460 
1461 	ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1462 	ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1463 	ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1464 	       (flags == XFS_ITRUNC_MAYBE));
1465 
1466 	mp = ip->i_mount;
1467 	vp = XFS_ITOV(ip);
1468 	/*
1469 	 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1470 	 * overlapping the region being removed.  We have to use
1471 	 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1472 	 * caller may not be able to finish the truncate without
1473 	 * dropping the inode's I/O lock.  Make sure
1474 	 * to catch any pages brought in by buffers overlapping
1475 	 * the EOF by searching out beyond the isize by our
1476 	 * block size. We round new_size up to a block boundary
1477 	 * so that we don't toss things on the same block as
1478 	 * new_size but before it.
1479 	 *
1480 	 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1481 	 * call remapf() over the same region if the file is mapped.
1482 	 * This frees up mapped file references to the pages in the
1483 	 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1484 	 * that we get the latest mapped changes flushed out.
1485 	 */
1486 	toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1487 	toss_start = XFS_FSB_TO_B(mp, toss_start);
1488 	if (toss_start < 0) {
1489 		/*
1490 		 * The place to start tossing is beyond our maximum
1491 		 * file size, so there is no way that the data extended
1492 		 * out there.
1493 		 */
1494 		return;
1495 	}
1496 	last_byte = xfs_file_last_byte(ip);
1497 	xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1498 			 last_byte);
1499 	if (last_byte > toss_start) {
1500 		if (flags & XFS_ITRUNC_DEFINITE) {
1501 			VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1502 		} else {
1503 			VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1504 		}
1505 	}
1506 
1507 #ifdef DEBUG
1508 	if (new_size == 0) {
1509 		ASSERT(VN_CACHED(vp) == 0);
1510 	}
1511 #endif
1512 }
1513 
1514 /*
1515  * Shrink the file to the given new_size.  The new
1516  * size must be smaller than the current size.
1517  * This will free up the underlying blocks
1518  * in the removed range after a call to xfs_itruncate_start()
1519  * or xfs_atruncate_start().
1520  *
1521  * The transaction passed to this routine must have made
1522  * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1523  * This routine may commit the given transaction and
1524  * start new ones, so make sure everything involved in
1525  * the transaction is tidy before calling here.
1526  * Some transaction will be returned to the caller to be
1527  * committed.  The incoming transaction must already include
1528  * the inode, and both inode locks must be held exclusively.
1529  * The inode must also be "held" within the transaction.  On
1530  * return the inode will be "held" within the returned transaction.
1531  * This routine does NOT require any disk space to be reserved
1532  * for it within the transaction.
1533  *
1534  * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1535  * and it indicates the fork which is to be truncated.  For the
1536  * attribute fork we only support truncation to size 0.
1537  *
1538  * We use the sync parameter to indicate whether or not the first
1539  * transaction we perform might have to be synchronous.  For the attr fork,
1540  * it needs to be so if the unlink of the inode is not yet known to be
1541  * permanent in the log.  This keeps us from freeing and reusing the
1542  * blocks of the attribute fork before the unlink of the inode becomes
1543  * permanent.
1544  *
1545  * For the data fork, we normally have to run synchronously if we're
1546  * being called out of the inactive path or we're being called
1547  * out of the create path where we're truncating an existing file.
1548  * Either way, the truncate needs to be sync so blocks don't reappear
1549  * in the file with altered data in case of a crash.  wsync filesystems
1550  * can run the first case async because anything that shrinks the inode
1551  * has to run sync so by the time we're called here from inactive, the
1552  * inode size is permanently set to 0.
1553  *
1554  * Calls from the truncate path always need to be sync unless we're
1555  * in a wsync filesystem and the file has already been unlinked.
1556  *
1557  * The caller is responsible for correctly setting the sync parameter.
1558  * It gets too hard for us to guess here which path we're being called
1559  * out of just based on inode state.
1560  */
1561 int
1562 xfs_itruncate_finish(
1563 	xfs_trans_t	**tp,
1564 	xfs_inode_t	*ip,
1565 	xfs_fsize_t	new_size,
1566 	int		fork,
1567 	int		sync)
1568 {
1569 	xfs_fsblock_t	first_block;
1570 	xfs_fileoff_t	first_unmap_block;
1571 	xfs_fileoff_t	last_block;
1572 	xfs_filblks_t	unmap_len=0;
1573 	xfs_mount_t	*mp;
1574 	xfs_trans_t	*ntp;
1575 	int		done;
1576 	int		committed;
1577 	xfs_bmap_free_t	free_list;
1578 	int		error;
1579 
1580 	ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1581 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1582 	ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1583 	ASSERT(*tp != NULL);
1584 	ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1585 	ASSERT(ip->i_transp == *tp);
1586 	ASSERT(ip->i_itemp != NULL);
1587 	ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1588 
1589 
1590 	ntp = *tp;
1591 	mp = (ntp)->t_mountp;
1592 	ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1593 
1594 	/*
1595 	 * We only support truncating the entire attribute fork.
1596 	 */
1597 	if (fork == XFS_ATTR_FORK) {
1598 		new_size = 0LL;
1599 	}
1600 	first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1601 	xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1602 	/*
1603 	 * The first thing we do is set the size to new_size permanently
1604 	 * on disk.  This way we don't have to worry about anyone ever
1605 	 * being able to look at the data being freed even in the face
1606 	 * of a crash.  What we're getting around here is the case where
1607 	 * we free a block, it is allocated to another file, it is written
1608 	 * to, and then we crash.  If the new data gets written to the
1609 	 * file but the log buffers containing the free and reallocation
1610 	 * don't, then we'd end up with garbage in the blocks being freed.
1611 	 * As long as we make the new_size permanent before actually
1612 	 * freeing any blocks it doesn't matter if they get writtten to.
1613 	 *
1614 	 * The callers must signal into us whether or not the size
1615 	 * setting here must be synchronous.  There are a few cases
1616 	 * where it doesn't have to be synchronous.  Those cases
1617 	 * occur if the file is unlinked and we know the unlink is
1618 	 * permanent or if the blocks being truncated are guaranteed
1619 	 * to be beyond the inode eof (regardless of the link count)
1620 	 * and the eof value is permanent.  Both of these cases occur
1621 	 * only on wsync-mounted filesystems.  In those cases, we're
1622 	 * guaranteed that no user will ever see the data in the blocks
1623 	 * that are being truncated so the truncate can run async.
1624 	 * In the free beyond eof case, the file may wind up with
1625 	 * more blocks allocated to it than it needs if we crash
1626 	 * and that won't get fixed until the next time the file
1627 	 * is re-opened and closed but that's ok as that shouldn't
1628 	 * be too many blocks.
1629 	 *
1630 	 * However, we can't just make all wsync xactions run async
1631 	 * because there's one call out of the create path that needs
1632 	 * to run sync where it's truncating an existing file to size
1633 	 * 0 whose size is > 0.
1634 	 *
1635 	 * It's probably possible to come up with a test in this
1636 	 * routine that would correctly distinguish all the above
1637 	 * cases from the values of the function parameters and the
1638 	 * inode state but for sanity's sake, I've decided to let the
1639 	 * layers above just tell us.  It's simpler to correctly figure
1640 	 * out in the layer above exactly under what conditions we
1641 	 * can run async and I think it's easier for others read and
1642 	 * follow the logic in case something has to be changed.
1643 	 * cscope is your friend -- rcc.
1644 	 *
1645 	 * The attribute fork is much simpler.
1646 	 *
1647 	 * For the attribute fork we allow the caller to tell us whether
1648 	 * the unlink of the inode that led to this call is yet permanent
1649 	 * in the on disk log.  If it is not and we will be freeing extents
1650 	 * in this inode then we make the first transaction synchronous
1651 	 * to make sure that the unlink is permanent by the time we free
1652 	 * the blocks.
1653 	 */
1654 	if (fork == XFS_DATA_FORK) {
1655 		if (ip->i_d.di_nextents > 0) {
1656 			ip->i_d.di_size = new_size;
1657 			xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1658 		}
1659 	} else if (sync) {
1660 		ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1661 		if (ip->i_d.di_anextents > 0)
1662 			xfs_trans_set_sync(ntp);
1663 	}
1664 	ASSERT(fork == XFS_DATA_FORK ||
1665 		(fork == XFS_ATTR_FORK &&
1666 			((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1667 			 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1668 
1669 	/*
1670 	 * Since it is possible for space to become allocated beyond
1671 	 * the end of the file (in a crash where the space is allocated
1672 	 * but the inode size is not yet updated), simply remove any
1673 	 * blocks which show up between the new EOF and the maximum
1674 	 * possible file size.  If the first block to be removed is
1675 	 * beyond the maximum file size (ie it is the same as last_block),
1676 	 * then there is nothing to do.
1677 	 */
1678 	last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1679 	ASSERT(first_unmap_block <= last_block);
1680 	done = 0;
1681 	if (last_block == first_unmap_block) {
1682 		done = 1;
1683 	} else {
1684 		unmap_len = last_block - first_unmap_block + 1;
1685 	}
1686 	while (!done) {
1687 		/*
1688 		 * Free up up to XFS_ITRUNC_MAX_EXTENTS.  xfs_bunmapi()
1689 		 * will tell us whether it freed the entire range or
1690 		 * not.  If this is a synchronous mount (wsync),
1691 		 * then we can tell bunmapi to keep all the
1692 		 * transactions asynchronous since the unlink
1693 		 * transaction that made this inode inactive has
1694 		 * already hit the disk.  There's no danger of
1695 		 * the freed blocks being reused, there being a
1696 		 * crash, and the reused blocks suddenly reappearing
1697 		 * in this file with garbage in them once recovery
1698 		 * runs.
1699 		 */
1700 		XFS_BMAP_INIT(&free_list, &first_block);
1701 		error = xfs_bunmapi(ntp, ip, first_unmap_block,
1702 				    unmap_len,
1703 				    XFS_BMAPI_AFLAG(fork) |
1704 				      (sync ? 0 : XFS_BMAPI_ASYNC),
1705 				    XFS_ITRUNC_MAX_EXTENTS,
1706 				    &first_block, &free_list, &done);
1707 		if (error) {
1708 			/*
1709 			 * If the bunmapi call encounters an error,
1710 			 * return to the caller where the transaction
1711 			 * can be properly aborted.  We just need to
1712 			 * make sure we're not holding any resources
1713 			 * that we were not when we came in.
1714 			 */
1715 			xfs_bmap_cancel(&free_list);
1716 			return error;
1717 		}
1718 
1719 		/*
1720 		 * Duplicate the transaction that has the permanent
1721 		 * reservation and commit the old transaction.
1722 		 */
1723 		error = xfs_bmap_finish(tp, &free_list, first_block,
1724 					&committed);
1725 		ntp = *tp;
1726 		if (error) {
1727 			/*
1728 			 * If the bmap finish call encounters an error,
1729 			 * return to the caller where the transaction
1730 			 * can be properly aborted.  We just need to
1731 			 * make sure we're not holding any resources
1732 			 * that we were not when we came in.
1733 			 *
1734 			 * Aborting from this point might lose some
1735 			 * blocks in the file system, but oh well.
1736 			 */
1737 			xfs_bmap_cancel(&free_list);
1738 			if (committed) {
1739 				/*
1740 				 * If the passed in transaction committed
1741 				 * in xfs_bmap_finish(), then we want to
1742 				 * add the inode to this one before returning.
1743 				 * This keeps things simple for the higher
1744 				 * level code, because it always knows that
1745 				 * the inode is locked and held in the
1746 				 * transaction that returns to it whether
1747 				 * errors occur or not.  We don't mark the
1748 				 * inode dirty so that this transaction can
1749 				 * be easily aborted if possible.
1750 				 */
1751 				xfs_trans_ijoin(ntp, ip,
1752 					XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1753 				xfs_trans_ihold(ntp, ip);
1754 			}
1755 			return error;
1756 		}
1757 
1758 		if (committed) {
1759 			/*
1760 			 * The first xact was committed,
1761 			 * so add the inode to the new one.
1762 			 * Mark it dirty so it will be logged
1763 			 * and moved forward in the log as
1764 			 * part of every commit.
1765 			 */
1766 			xfs_trans_ijoin(ntp, ip,
1767 					XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1768 			xfs_trans_ihold(ntp, ip);
1769 			xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1770 		}
1771 		ntp = xfs_trans_dup(ntp);
1772 		(void) xfs_trans_commit(*tp, 0, NULL);
1773 		*tp = ntp;
1774 		error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1775 					  XFS_TRANS_PERM_LOG_RES,
1776 					  XFS_ITRUNCATE_LOG_COUNT);
1777 		/*
1778 		 * Add the inode being truncated to the next chained
1779 		 * transaction.
1780 		 */
1781 		xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1782 		xfs_trans_ihold(ntp, ip);
1783 		if (error)
1784 			return (error);
1785 	}
1786 	/*
1787 	 * Only update the size in the case of the data fork, but
1788 	 * always re-log the inode so that our permanent transaction
1789 	 * can keep on rolling it forward in the log.
1790 	 */
1791 	if (fork == XFS_DATA_FORK) {
1792 		xfs_isize_check(mp, ip, new_size);
1793 		ip->i_d.di_size = new_size;
1794 	}
1795 	xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1796 	ASSERT((new_size != 0) ||
1797 	       (fork == XFS_ATTR_FORK) ||
1798 	       (ip->i_delayed_blks == 0));
1799 	ASSERT((new_size != 0) ||
1800 	       (fork == XFS_ATTR_FORK) ||
1801 	       (ip->i_d.di_nextents == 0));
1802 	xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1803 	return 0;
1804 }
1805 
1806 
1807 /*
1808  * xfs_igrow_start
1809  *
1810  * Do the first part of growing a file: zero any data in the last
1811  * block that is beyond the old EOF.  We need to do this before
1812  * the inode is joined to the transaction to modify the i_size.
1813  * That way we can drop the inode lock and call into the buffer
1814  * cache to get the buffer mapping the EOF.
1815  */
1816 int
1817 xfs_igrow_start(
1818 	xfs_inode_t	*ip,
1819 	xfs_fsize_t	new_size,
1820 	cred_t		*credp)
1821 {
1822 	xfs_fsize_t	isize;
1823 	int		error;
1824 
1825 	ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1826 	ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1827 	ASSERT(new_size > ip->i_d.di_size);
1828 
1829 	error = 0;
1830 	isize = ip->i_d.di_size;
1831 	/*
1832 	 * Zero any pages that may have been created by
1833 	 * xfs_write_file() beyond the end of the file
1834 	 * and any blocks between the old and new file sizes.
1835 	 */
1836 	error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size, isize,
1837 				new_size);
1838 	return error;
1839 }
1840 
1841 /*
1842  * xfs_igrow_finish
1843  *
1844  * This routine is called to extend the size of a file.
1845  * The inode must have both the iolock and the ilock locked
1846  * for update and it must be a part of the current transaction.
1847  * The xfs_igrow_start() function must have been called previously.
1848  * If the change_flag is not zero, the inode change timestamp will
1849  * be updated.
1850  */
1851 void
1852 xfs_igrow_finish(
1853 	xfs_trans_t	*tp,
1854 	xfs_inode_t	*ip,
1855 	xfs_fsize_t	new_size,
1856 	int		change_flag)
1857 {
1858 	ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1859 	ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1860 	ASSERT(ip->i_transp == tp);
1861 	ASSERT(new_size > ip->i_d.di_size);
1862 
1863 	/*
1864 	 * Update the file size.  Update the inode change timestamp
1865 	 * if change_flag set.
1866 	 */
1867 	ip->i_d.di_size = new_size;
1868 	if (change_flag)
1869 		xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1870 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1871 
1872 }
1873 
1874 
1875 /*
1876  * This is called when the inode's link count goes to 0.
1877  * We place the on-disk inode on a list in the AGI.  It
1878  * will be pulled from this list when the inode is freed.
1879  */
1880 int
1881 xfs_iunlink(
1882 	xfs_trans_t	*tp,
1883 	xfs_inode_t	*ip)
1884 {
1885 	xfs_mount_t	*mp;
1886 	xfs_agi_t	*agi;
1887 	xfs_dinode_t	*dip;
1888 	xfs_buf_t	*agibp;
1889 	xfs_buf_t	*ibp;
1890 	xfs_agnumber_t	agno;
1891 	xfs_daddr_t	agdaddr;
1892 	xfs_agino_t	agino;
1893 	short		bucket_index;
1894 	int		offset;
1895 	int		error;
1896 	int		agi_ok;
1897 
1898 	ASSERT(ip->i_d.di_nlink == 0);
1899 	ASSERT(ip->i_d.di_mode != 0);
1900 	ASSERT(ip->i_transp == tp);
1901 
1902 	mp = tp->t_mountp;
1903 
1904 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1905 	agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1906 
1907 	/*
1908 	 * Get the agi buffer first.  It ensures lock ordering
1909 	 * on the list.
1910 	 */
1911 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1912 				   XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1913 	if (error) {
1914 		return error;
1915 	}
1916 	/*
1917 	 * Validate the magic number of the agi block.
1918 	 */
1919 	agi = XFS_BUF_TO_AGI(agibp);
1920 	agi_ok =
1921 		INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC &&
1922 		XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT));
1923 	if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1924 			XFS_RANDOM_IUNLINK))) {
1925 		XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1926 		xfs_trans_brelse(tp, agibp);
1927 		return XFS_ERROR(EFSCORRUPTED);
1928 	}
1929 	/*
1930 	 * Get the index into the agi hash table for the
1931 	 * list this inode will go on.
1932 	 */
1933 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1934 	ASSERT(agino != 0);
1935 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1936 	ASSERT(agi->agi_unlinked[bucket_index]);
1937 	ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != agino);
1938 
1939 	if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO) {
1940 		/*
1941 		 * There is already another inode in the bucket we need
1942 		 * to add ourselves to.  Add us at the front of the list.
1943 		 * Here we put the head pointer into our next pointer,
1944 		 * and then we fall through to point the head at us.
1945 		 */
1946 		error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1947 		if (error) {
1948 			return error;
1949 		}
1950 		ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1951 		ASSERT(dip->di_next_unlinked);
1952 		/* both on-disk, don't endian flip twice */
1953 		dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1954 		offset = ip->i_boffset +
1955 			offsetof(xfs_dinode_t, di_next_unlinked);
1956 		xfs_trans_inode_buf(tp, ibp);
1957 		xfs_trans_log_buf(tp, ibp, offset,
1958 				  (offset + sizeof(xfs_agino_t) - 1));
1959 		xfs_inobp_check(mp, ibp);
1960 	}
1961 
1962 	/*
1963 	 * Point the bucket head pointer at the inode being inserted.
1964 	 */
1965 	ASSERT(agino != 0);
1966 	INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, agino);
1967 	offset = offsetof(xfs_agi_t, agi_unlinked) +
1968 		(sizeof(xfs_agino_t) * bucket_index);
1969 	xfs_trans_log_buf(tp, agibp, offset,
1970 			  (offset + sizeof(xfs_agino_t) - 1));
1971 	return 0;
1972 }
1973 
1974 /*
1975  * Pull the on-disk inode from the AGI unlinked list.
1976  */
1977 STATIC int
1978 xfs_iunlink_remove(
1979 	xfs_trans_t	*tp,
1980 	xfs_inode_t	*ip)
1981 {
1982 	xfs_ino_t	next_ino;
1983 	xfs_mount_t	*mp;
1984 	xfs_agi_t	*agi;
1985 	xfs_dinode_t	*dip;
1986 	xfs_buf_t	*agibp;
1987 	xfs_buf_t	*ibp;
1988 	xfs_agnumber_t	agno;
1989 	xfs_daddr_t	agdaddr;
1990 	xfs_agino_t	agino;
1991 	xfs_agino_t	next_agino;
1992 	xfs_buf_t	*last_ibp;
1993 	xfs_dinode_t	*last_dip;
1994 	short		bucket_index;
1995 	int		offset, last_offset;
1996 	int		error;
1997 	int		agi_ok;
1998 
1999 	/*
2000 	 * First pull the on-disk inode from the AGI unlinked list.
2001 	 */
2002 	mp = tp->t_mountp;
2003 
2004 	agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2005 	agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
2006 
2007 	/*
2008 	 * Get the agi buffer first.  It ensures lock ordering
2009 	 * on the list.
2010 	 */
2011 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
2012 				   XFS_FSS_TO_BB(mp, 1), 0, &agibp);
2013 	if (error) {
2014 		cmn_err(CE_WARN,
2015 			"xfs_iunlink_remove: xfs_trans_read_buf()  returned an error %d on %s.  Returning error.",
2016 			error, mp->m_fsname);
2017 		return error;
2018 	}
2019 	/*
2020 	 * Validate the magic number of the agi block.
2021 	 */
2022 	agi = XFS_BUF_TO_AGI(agibp);
2023 	agi_ok =
2024 		INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC &&
2025 		XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT));
2026 	if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2027 			XFS_RANDOM_IUNLINK_REMOVE))) {
2028 		XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2029 				     mp, agi);
2030 		xfs_trans_brelse(tp, agibp);
2031 		cmn_err(CE_WARN,
2032 			"xfs_iunlink_remove: XFS_TEST_ERROR()  returned an error on %s.  Returning EFSCORRUPTED.",
2033 			 mp->m_fsname);
2034 		return XFS_ERROR(EFSCORRUPTED);
2035 	}
2036 	/*
2037 	 * Get the index into the agi hash table for the
2038 	 * list this inode will go on.
2039 	 */
2040 	agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2041 	ASSERT(agino != 0);
2042 	bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2043 	ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO);
2044 	ASSERT(agi->agi_unlinked[bucket_index]);
2045 
2046 	if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) == agino) {
2047 		/*
2048 		 * We're at the head of the list.  Get the inode's
2049 		 * on-disk buffer to see if there is anyone after us
2050 		 * on the list.  Only modify our next pointer if it
2051 		 * is not already NULLAGINO.  This saves us the overhead
2052 		 * of dealing with the buffer when there is no need to
2053 		 * change it.
2054 		 */
2055 		error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2056 		if (error) {
2057 			cmn_err(CE_WARN,
2058 				"xfs_iunlink_remove: xfs_itobp()  returned an error %d on %s.  Returning error.",
2059 				error, mp->m_fsname);
2060 			return error;
2061 		}
2062 		next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2063 		ASSERT(next_agino != 0);
2064 		if (next_agino != NULLAGINO) {
2065 			INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2066 			offset = ip->i_boffset +
2067 				offsetof(xfs_dinode_t, di_next_unlinked);
2068 			xfs_trans_inode_buf(tp, ibp);
2069 			xfs_trans_log_buf(tp, ibp, offset,
2070 					  (offset + sizeof(xfs_agino_t) - 1));
2071 			xfs_inobp_check(mp, ibp);
2072 		} else {
2073 			xfs_trans_brelse(tp, ibp);
2074 		}
2075 		/*
2076 		 * Point the bucket head pointer at the next inode.
2077 		 */
2078 		ASSERT(next_agino != 0);
2079 		ASSERT(next_agino != agino);
2080 		INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, next_agino);
2081 		offset = offsetof(xfs_agi_t, agi_unlinked) +
2082 			(sizeof(xfs_agino_t) * bucket_index);
2083 		xfs_trans_log_buf(tp, agibp, offset,
2084 				  (offset + sizeof(xfs_agino_t) - 1));
2085 	} else {
2086 		/*
2087 		 * We need to search the list for the inode being freed.
2088 		 */
2089 		next_agino = INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT);
2090 		last_ibp = NULL;
2091 		while (next_agino != agino) {
2092 			/*
2093 			 * If the last inode wasn't the one pointing to
2094 			 * us, then release its buffer since we're not
2095 			 * going to do anything with it.
2096 			 */
2097 			if (last_ibp != NULL) {
2098 				xfs_trans_brelse(tp, last_ibp);
2099 			}
2100 			next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2101 			error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2102 					    &last_ibp, &last_offset);
2103 			if (error) {
2104 				cmn_err(CE_WARN,
2105 			"xfs_iunlink_remove: xfs_inotobp()  returned an error %d on %s.  Returning error.",
2106 					error, mp->m_fsname);
2107 				return error;
2108 			}
2109 			next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2110 			ASSERT(next_agino != NULLAGINO);
2111 			ASSERT(next_agino != 0);
2112 		}
2113 		/*
2114 		 * Now last_ibp points to the buffer previous to us on
2115 		 * the unlinked list.  Pull us from the list.
2116 		 */
2117 		error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2118 		if (error) {
2119 			cmn_err(CE_WARN,
2120 				"xfs_iunlink_remove: xfs_itobp()  returned an error %d on %s.  Returning error.",
2121 				error, mp->m_fsname);
2122 			return error;
2123 		}
2124 		next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2125 		ASSERT(next_agino != 0);
2126 		ASSERT(next_agino != agino);
2127 		if (next_agino != NULLAGINO) {
2128 			INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2129 			offset = ip->i_boffset +
2130 				offsetof(xfs_dinode_t, di_next_unlinked);
2131 			xfs_trans_inode_buf(tp, ibp);
2132 			xfs_trans_log_buf(tp, ibp, offset,
2133 					  (offset + sizeof(xfs_agino_t) - 1));
2134 			xfs_inobp_check(mp, ibp);
2135 		} else {
2136 			xfs_trans_brelse(tp, ibp);
2137 		}
2138 		/*
2139 		 * Point the previous inode on the list to the next inode.
2140 		 */
2141 		INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2142 		ASSERT(next_agino != 0);
2143 		offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2144 		xfs_trans_inode_buf(tp, last_ibp);
2145 		xfs_trans_log_buf(tp, last_ibp, offset,
2146 				  (offset + sizeof(xfs_agino_t) - 1));
2147 		xfs_inobp_check(mp, last_ibp);
2148 	}
2149 	return 0;
2150 }
2151 
2152 static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2153 {
2154 	return (((ip->i_itemp == NULL) ||
2155 		!(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2156 		(ip->i_update_core == 0));
2157 }
2158 
2159 void
2160 xfs_ifree_cluster(
2161 	xfs_inode_t	*free_ip,
2162 	xfs_trans_t	*tp,
2163 	xfs_ino_t	inum)
2164 {
2165 	xfs_mount_t		*mp = free_ip->i_mount;
2166 	int			blks_per_cluster;
2167 	int			nbufs;
2168 	int			ninodes;
2169 	int			i, j, found, pre_flushed;
2170 	xfs_daddr_t		blkno;
2171 	xfs_buf_t		*bp;
2172 	xfs_ihash_t		*ih;
2173 	xfs_inode_t		*ip, **ip_found;
2174 	xfs_inode_log_item_t	*iip;
2175 	xfs_log_item_t		*lip;
2176 	SPLDECL(s);
2177 
2178 	if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2179 		blks_per_cluster = 1;
2180 		ninodes = mp->m_sb.sb_inopblock;
2181 		nbufs = XFS_IALLOC_BLOCKS(mp);
2182 	} else {
2183 		blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2184 					mp->m_sb.sb_blocksize;
2185 		ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2186 		nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2187 	}
2188 
2189 	ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2190 
2191 	for (j = 0; j < nbufs; j++, inum += ninodes) {
2192 		blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2193 					 XFS_INO_TO_AGBNO(mp, inum));
2194 
2195 
2196 		/*
2197 		 * Look for each inode in memory and attempt to lock it,
2198 		 * we can be racing with flush and tail pushing here.
2199 		 * any inode we get the locks on, add to an array of
2200 		 * inode items to process later.
2201 		 *
2202 		 * The get the buffer lock, we could beat a flush
2203 		 * or tail pushing thread to the lock here, in which
2204 		 * case they will go looking for the inode buffer
2205 		 * and fail, we need some other form of interlock
2206 		 * here.
2207 		 */
2208 		found = 0;
2209 		for (i = 0; i < ninodes; i++) {
2210 			ih = XFS_IHASH(mp, inum + i);
2211 			read_lock(&ih->ih_lock);
2212 			for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2213 				if (ip->i_ino == inum + i)
2214 					break;
2215 			}
2216 
2217 			/* Inode not in memory or we found it already,
2218 			 * nothing to do
2219 			 */
2220 			if (!ip || (ip->i_flags & XFS_ISTALE)) {
2221 				read_unlock(&ih->ih_lock);
2222 				continue;
2223 			}
2224 
2225 			if (xfs_inode_clean(ip)) {
2226 				read_unlock(&ih->ih_lock);
2227 				continue;
2228 			}
2229 
2230 			/* If we can get the locks then add it to the
2231 			 * list, otherwise by the time we get the bp lock
2232 			 * below it will already be attached to the
2233 			 * inode buffer.
2234 			 */
2235 
2236 			/* This inode will already be locked - by us, lets
2237 			 * keep it that way.
2238 			 */
2239 
2240 			if (ip == free_ip) {
2241 				if (xfs_iflock_nowait(ip)) {
2242 					ip->i_flags |= XFS_ISTALE;
2243 
2244 					if (xfs_inode_clean(ip)) {
2245 						xfs_ifunlock(ip);
2246 					} else {
2247 						ip_found[found++] = ip;
2248 					}
2249 				}
2250 				read_unlock(&ih->ih_lock);
2251 				continue;
2252 			}
2253 
2254 			if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2255 				if (xfs_iflock_nowait(ip)) {
2256 					ip->i_flags |= XFS_ISTALE;
2257 
2258 					if (xfs_inode_clean(ip)) {
2259 						xfs_ifunlock(ip);
2260 						xfs_iunlock(ip, XFS_ILOCK_EXCL);
2261 					} else {
2262 						ip_found[found++] = ip;
2263 					}
2264 				} else {
2265 					xfs_iunlock(ip, XFS_ILOCK_EXCL);
2266 				}
2267 			}
2268 
2269 			read_unlock(&ih->ih_lock);
2270 		}
2271 
2272 		bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2273 					mp->m_bsize * blks_per_cluster,
2274 					XFS_BUF_LOCK);
2275 
2276 		pre_flushed = 0;
2277 		lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2278 		while (lip) {
2279 			if (lip->li_type == XFS_LI_INODE) {
2280 				iip = (xfs_inode_log_item_t *)lip;
2281 				ASSERT(iip->ili_logged == 1);
2282 				lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2283 				AIL_LOCK(mp,s);
2284 				iip->ili_flush_lsn = iip->ili_item.li_lsn;
2285 				AIL_UNLOCK(mp, s);
2286 				iip->ili_inode->i_flags |= XFS_ISTALE;
2287 				pre_flushed++;
2288 			}
2289 			lip = lip->li_bio_list;
2290 		}
2291 
2292 		for (i = 0; i < found; i++) {
2293 			ip = ip_found[i];
2294 			iip = ip->i_itemp;
2295 
2296 			if (!iip) {
2297 				ip->i_update_core = 0;
2298 				xfs_ifunlock(ip);
2299 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2300 				continue;
2301 			}
2302 
2303 			iip->ili_last_fields = iip->ili_format.ilf_fields;
2304 			iip->ili_format.ilf_fields = 0;
2305 			iip->ili_logged = 1;
2306 			AIL_LOCK(mp,s);
2307 			iip->ili_flush_lsn = iip->ili_item.li_lsn;
2308 			AIL_UNLOCK(mp, s);
2309 
2310 			xfs_buf_attach_iodone(bp,
2311 				(void(*)(xfs_buf_t*,xfs_log_item_t*))
2312 				xfs_istale_done, (xfs_log_item_t *)iip);
2313 			if (ip != free_ip) {
2314 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
2315 			}
2316 		}
2317 
2318 		if (found || pre_flushed)
2319 			xfs_trans_stale_inode_buf(tp, bp);
2320 		xfs_trans_binval(tp, bp);
2321 	}
2322 
2323 	kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2324 }
2325 
2326 /*
2327  * This is called to return an inode to the inode free list.
2328  * The inode should already be truncated to 0 length and have
2329  * no pages associated with it.  This routine also assumes that
2330  * the inode is already a part of the transaction.
2331  *
2332  * The on-disk copy of the inode will have been added to the list
2333  * of unlinked inodes in the AGI. We need to remove the inode from
2334  * that list atomically with respect to freeing it here.
2335  */
2336 int
2337 xfs_ifree(
2338 	xfs_trans_t	*tp,
2339 	xfs_inode_t	*ip,
2340 	xfs_bmap_free_t	*flist)
2341 {
2342 	int			error;
2343 	int			delete;
2344 	xfs_ino_t		first_ino;
2345 
2346 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2347 	ASSERT(ip->i_transp == tp);
2348 	ASSERT(ip->i_d.di_nlink == 0);
2349 	ASSERT(ip->i_d.di_nextents == 0);
2350 	ASSERT(ip->i_d.di_anextents == 0);
2351 	ASSERT((ip->i_d.di_size == 0) ||
2352 	       ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2353 	ASSERT(ip->i_d.di_nblocks == 0);
2354 
2355 	/*
2356 	 * Pull the on-disk inode from the AGI unlinked list.
2357 	 */
2358 	error = xfs_iunlink_remove(tp, ip);
2359 	if (error != 0) {
2360 		return error;
2361 	}
2362 
2363 	error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2364 	if (error != 0) {
2365 		return error;
2366 	}
2367 	ip->i_d.di_mode = 0;		/* mark incore inode as free */
2368 	ip->i_d.di_flags = 0;
2369 	ip->i_d.di_dmevmask = 0;
2370 	ip->i_d.di_forkoff = 0;		/* mark the attr fork not in use */
2371 	ip->i_df.if_ext_max =
2372 		XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2373 	ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2374 	ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2375 	/*
2376 	 * Bump the generation count so no one will be confused
2377 	 * by reincarnations of this inode.
2378 	 */
2379 	ip->i_d.di_gen++;
2380 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2381 
2382 	if (delete) {
2383 		xfs_ifree_cluster(ip, tp, first_ino);
2384 	}
2385 
2386 	return 0;
2387 }
2388 
2389 /*
2390  * Reallocate the space for if_broot based on the number of records
2391  * being added or deleted as indicated in rec_diff.  Move the records
2392  * and pointers in if_broot to fit the new size.  When shrinking this
2393  * will eliminate holes between the records and pointers created by
2394  * the caller.  When growing this will create holes to be filled in
2395  * by the caller.
2396  *
2397  * The caller must not request to add more records than would fit in
2398  * the on-disk inode root.  If the if_broot is currently NULL, then
2399  * if we adding records one will be allocated.  The caller must also
2400  * not request that the number of records go below zero, although
2401  * it can go to zero.
2402  *
2403  * ip -- the inode whose if_broot area is changing
2404  * ext_diff -- the change in the number of records, positive or negative,
2405  *	 requested for the if_broot array.
2406  */
2407 void
2408 xfs_iroot_realloc(
2409 	xfs_inode_t		*ip,
2410 	int			rec_diff,
2411 	int			whichfork)
2412 {
2413 	int			cur_max;
2414 	xfs_ifork_t		*ifp;
2415 	xfs_bmbt_block_t	*new_broot;
2416 	int			new_max;
2417 	size_t			new_size;
2418 	char			*np;
2419 	char			*op;
2420 
2421 	/*
2422 	 * Handle the degenerate case quietly.
2423 	 */
2424 	if (rec_diff == 0) {
2425 		return;
2426 	}
2427 
2428 	ifp = XFS_IFORK_PTR(ip, whichfork);
2429 	if (rec_diff > 0) {
2430 		/*
2431 		 * If there wasn't any memory allocated before, just
2432 		 * allocate it now and get out.
2433 		 */
2434 		if (ifp->if_broot_bytes == 0) {
2435 			new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2436 			ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2437 								     KM_SLEEP);
2438 			ifp->if_broot_bytes = (int)new_size;
2439 			return;
2440 		}
2441 
2442 		/*
2443 		 * If there is already an existing if_broot, then we need
2444 		 * to realloc() it and shift the pointers to their new
2445 		 * location.  The records don't change location because
2446 		 * they are kept butted up against the btree block header.
2447 		 */
2448 		cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2449 		new_max = cur_max + rec_diff;
2450 		new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2451 		ifp->if_broot = (xfs_bmbt_block_t *)
2452 		  kmem_realloc(ifp->if_broot,
2453 				new_size,
2454 				(size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2455 				KM_SLEEP);
2456 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2457 						      ifp->if_broot_bytes);
2458 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2459 						      (int)new_size);
2460 		ifp->if_broot_bytes = (int)new_size;
2461 		ASSERT(ifp->if_broot_bytes <=
2462 			XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2463 		memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2464 		return;
2465 	}
2466 
2467 	/*
2468 	 * rec_diff is less than 0.  In this case, we are shrinking the
2469 	 * if_broot buffer.  It must already exist.  If we go to zero
2470 	 * records, just get rid of the root and clear the status bit.
2471 	 */
2472 	ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2473 	cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2474 	new_max = cur_max + rec_diff;
2475 	ASSERT(new_max >= 0);
2476 	if (new_max > 0)
2477 		new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2478 	else
2479 		new_size = 0;
2480 	if (new_size > 0) {
2481 		new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2482 		/*
2483 		 * First copy over the btree block header.
2484 		 */
2485 		memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2486 	} else {
2487 		new_broot = NULL;
2488 		ifp->if_flags &= ~XFS_IFBROOT;
2489 	}
2490 
2491 	/*
2492 	 * Only copy the records and pointers if there are any.
2493 	 */
2494 	if (new_max > 0) {
2495 		/*
2496 		 * First copy the records.
2497 		 */
2498 		op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2499 						     ifp->if_broot_bytes);
2500 		np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2501 						     (int)new_size);
2502 		memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2503 
2504 		/*
2505 		 * Then copy the pointers.
2506 		 */
2507 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2508 						     ifp->if_broot_bytes);
2509 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2510 						     (int)new_size);
2511 		memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2512 	}
2513 	kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2514 	ifp->if_broot = new_broot;
2515 	ifp->if_broot_bytes = (int)new_size;
2516 	ASSERT(ifp->if_broot_bytes <=
2517 		XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2518 	return;
2519 }
2520 
2521 
2522 /*
2523  * This is called when the amount of space needed for if_extents
2524  * is increased or decreased.  The change in size is indicated by
2525  * the number of extents that need to be added or deleted in the
2526  * ext_diff parameter.
2527  *
2528  * If the amount of space needed has decreased below the size of the
2529  * inline buffer, then switch to using the inline buffer.  Otherwise,
2530  * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2531  * to what is needed.
2532  *
2533  * ip -- the inode whose if_extents area is changing
2534  * ext_diff -- the change in the number of extents, positive or negative,
2535  *	 requested for the if_extents array.
2536  */
2537 void
2538 xfs_iext_realloc(
2539 	xfs_inode_t	*ip,
2540 	int		ext_diff,
2541 	int		whichfork)
2542 {
2543 	int		byte_diff;
2544 	xfs_ifork_t	*ifp;
2545 	int		new_size;
2546 	uint		rnew_size;
2547 
2548 	if (ext_diff == 0) {
2549 		return;
2550 	}
2551 
2552 	ifp = XFS_IFORK_PTR(ip, whichfork);
2553 	byte_diff = ext_diff * (uint)sizeof(xfs_bmbt_rec_t);
2554 	new_size = (int)ifp->if_bytes + byte_diff;
2555 	ASSERT(new_size >= 0);
2556 
2557 	if (new_size == 0) {
2558 		if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) {
2559 			ASSERT(ifp->if_real_bytes != 0);
2560 			kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
2561 		}
2562 		ifp->if_u1.if_extents = NULL;
2563 		rnew_size = 0;
2564 	} else if (new_size <= sizeof(ifp->if_u2.if_inline_ext)) {
2565 		/*
2566 		 * If the valid extents can fit in if_inline_ext,
2567 		 * copy them from the malloc'd vector and free it.
2568 		 */
2569 		if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) {
2570 			/*
2571 			 * For now, empty files are format EXTENTS,
2572 			 * so the if_extents pointer is null.
2573 			 */
2574 			if (ifp->if_u1.if_extents) {
2575 				memcpy(ifp->if_u2.if_inline_ext,
2576 					ifp->if_u1.if_extents, new_size);
2577 				kmem_free(ifp->if_u1.if_extents,
2578 					  ifp->if_real_bytes);
2579 			}
2580 			ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2581 		}
2582 		rnew_size = 0;
2583 	} else {
2584 		rnew_size = new_size;
2585 		if ((rnew_size & (rnew_size - 1)) != 0)
2586 			rnew_size = xfs_iroundup(rnew_size);
2587 		/*
2588 		 * Stuck with malloc/realloc.
2589 		 */
2590 		if (ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext) {
2591 			ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
2592 				kmem_alloc(rnew_size, KM_SLEEP);
2593 			memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
2594 			      sizeof(ifp->if_u2.if_inline_ext));
2595 		} else if (rnew_size != ifp->if_real_bytes) {
2596 			ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
2597 			  kmem_realloc(ifp->if_u1.if_extents,
2598 					rnew_size,
2599 					ifp->if_real_bytes,
2600 					KM_NOFS);
2601 		}
2602 	}
2603 	ifp->if_real_bytes = rnew_size;
2604 	ifp->if_bytes = new_size;
2605 }
2606 
2607 
2608 /*
2609  * This is called when the amount of space needed for if_data
2610  * is increased or decreased.  The change in size is indicated by
2611  * the number of bytes that need to be added or deleted in the
2612  * byte_diff parameter.
2613  *
2614  * If the amount of space needed has decreased below the size of the
2615  * inline buffer, then switch to using the inline buffer.  Otherwise,
2616  * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2617  * to what is needed.
2618  *
2619  * ip -- the inode whose if_data area is changing
2620  * byte_diff -- the change in the number of bytes, positive or negative,
2621  *	 requested for the if_data array.
2622  */
2623 void
2624 xfs_idata_realloc(
2625 	xfs_inode_t	*ip,
2626 	int		byte_diff,
2627 	int		whichfork)
2628 {
2629 	xfs_ifork_t	*ifp;
2630 	int		new_size;
2631 	int		real_size;
2632 
2633 	if (byte_diff == 0) {
2634 		return;
2635 	}
2636 
2637 	ifp = XFS_IFORK_PTR(ip, whichfork);
2638 	new_size = (int)ifp->if_bytes + byte_diff;
2639 	ASSERT(new_size >= 0);
2640 
2641 	if (new_size == 0) {
2642 		if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2643 			kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2644 		}
2645 		ifp->if_u1.if_data = NULL;
2646 		real_size = 0;
2647 	} else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2648 		/*
2649 		 * If the valid extents/data can fit in if_inline_ext/data,
2650 		 * copy them from the malloc'd vector and free it.
2651 		 */
2652 		if (ifp->if_u1.if_data == NULL) {
2653 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2654 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2655 			ASSERT(ifp->if_real_bytes != 0);
2656 			memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2657 			      new_size);
2658 			kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2659 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2660 		}
2661 		real_size = 0;
2662 	} else {
2663 		/*
2664 		 * Stuck with malloc/realloc.
2665 		 * For inline data, the underlying buffer must be
2666 		 * a multiple of 4 bytes in size so that it can be
2667 		 * logged and stay on word boundaries.  We enforce
2668 		 * that here.
2669 		 */
2670 		real_size = roundup(new_size, 4);
2671 		if (ifp->if_u1.if_data == NULL) {
2672 			ASSERT(ifp->if_real_bytes == 0);
2673 			ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2674 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2675 			/*
2676 			 * Only do the realloc if the underlying size
2677 			 * is really changing.
2678 			 */
2679 			if (ifp->if_real_bytes != real_size) {
2680 				ifp->if_u1.if_data =
2681 					kmem_realloc(ifp->if_u1.if_data,
2682 							real_size,
2683 							ifp->if_real_bytes,
2684 							KM_SLEEP);
2685 			}
2686 		} else {
2687 			ASSERT(ifp->if_real_bytes == 0);
2688 			ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2689 			memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2690 				ifp->if_bytes);
2691 		}
2692 	}
2693 	ifp->if_real_bytes = real_size;
2694 	ifp->if_bytes = new_size;
2695 	ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2696 }
2697 
2698 
2699 
2700 
2701 /*
2702  * Map inode to disk block and offset.
2703  *
2704  * mp -- the mount point structure for the current file system
2705  * tp -- the current transaction
2706  * ino -- the inode number of the inode to be located
2707  * imap -- this structure is filled in with the information necessary
2708  *	 to retrieve the given inode from disk
2709  * flags -- flags to pass to xfs_dilocate indicating whether or not
2710  *	 lookups in the inode btree were OK or not
2711  */
2712 int
2713 xfs_imap(
2714 	xfs_mount_t	*mp,
2715 	xfs_trans_t	*tp,
2716 	xfs_ino_t	ino,
2717 	xfs_imap_t	*imap,
2718 	uint		flags)
2719 {
2720 	xfs_fsblock_t	fsbno;
2721 	int		len;
2722 	int		off;
2723 	int		error;
2724 
2725 	fsbno = imap->im_blkno ?
2726 		XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2727 	error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2728 	if (error != 0) {
2729 		return error;
2730 	}
2731 	imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2732 	imap->im_len = XFS_FSB_TO_BB(mp, len);
2733 	imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2734 	imap->im_ioffset = (ushort)off;
2735 	imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2736 	return 0;
2737 }
2738 
2739 void
2740 xfs_idestroy_fork(
2741 	xfs_inode_t	*ip,
2742 	int		whichfork)
2743 {
2744 	xfs_ifork_t	*ifp;
2745 
2746 	ifp = XFS_IFORK_PTR(ip, whichfork);
2747 	if (ifp->if_broot != NULL) {
2748 		kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2749 		ifp->if_broot = NULL;
2750 	}
2751 
2752 	/*
2753 	 * If the format is local, then we can't have an extents
2754 	 * array so just look for an inline data array.  If we're
2755 	 * not local then we may or may not have an extents list,
2756 	 * so check and free it up if we do.
2757 	 */
2758 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2759 		if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2760 		    (ifp->if_u1.if_data != NULL)) {
2761 			ASSERT(ifp->if_real_bytes != 0);
2762 			kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2763 			ifp->if_u1.if_data = NULL;
2764 			ifp->if_real_bytes = 0;
2765 		}
2766 	} else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2767 		   (ifp->if_u1.if_extents != NULL) &&
2768 		   (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)) {
2769 		ASSERT(ifp->if_real_bytes != 0);
2770 		kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
2771 		ifp->if_u1.if_extents = NULL;
2772 		ifp->if_real_bytes = 0;
2773 	}
2774 	ASSERT(ifp->if_u1.if_extents == NULL ||
2775 	       ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2776 	ASSERT(ifp->if_real_bytes == 0);
2777 	if (whichfork == XFS_ATTR_FORK) {
2778 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2779 		ip->i_afp = NULL;
2780 	}
2781 }
2782 
2783 /*
2784  * This is called free all the memory associated with an inode.
2785  * It must free the inode itself and any buffers allocated for
2786  * if_extents/if_data and if_broot.  It must also free the lock
2787  * associated with the inode.
2788  */
2789 void
2790 xfs_idestroy(
2791 	xfs_inode_t	*ip)
2792 {
2793 
2794 	switch (ip->i_d.di_mode & S_IFMT) {
2795 	case S_IFREG:
2796 	case S_IFDIR:
2797 	case S_IFLNK:
2798 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
2799 		break;
2800 	}
2801 	if (ip->i_afp)
2802 		xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2803 	mrfree(&ip->i_lock);
2804 	mrfree(&ip->i_iolock);
2805 	freesema(&ip->i_flock);
2806 #ifdef XFS_BMAP_TRACE
2807 	ktrace_free(ip->i_xtrace);
2808 #endif
2809 #ifdef XFS_BMBT_TRACE
2810 	ktrace_free(ip->i_btrace);
2811 #endif
2812 #ifdef XFS_RW_TRACE
2813 	ktrace_free(ip->i_rwtrace);
2814 #endif
2815 #ifdef XFS_ILOCK_TRACE
2816 	ktrace_free(ip->i_lock_trace);
2817 #endif
2818 #ifdef XFS_DIR2_TRACE
2819 	ktrace_free(ip->i_dir_trace);
2820 #endif
2821 	if (ip->i_itemp) {
2822 		/* XXXdpd should be able to assert this but shutdown
2823 		 * is leaving the AIL behind. */
2824 		ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2825 		       XFS_FORCED_SHUTDOWN(ip->i_mount));
2826 		xfs_inode_item_destroy(ip);
2827 	}
2828 	kmem_zone_free(xfs_inode_zone, ip);
2829 }
2830 
2831 
2832 /*
2833  * Increment the pin count of the given buffer.
2834  * This value is protected by ipinlock spinlock in the mount structure.
2835  */
2836 void
2837 xfs_ipin(
2838 	xfs_inode_t	*ip)
2839 {
2840 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2841 
2842 	atomic_inc(&ip->i_pincount);
2843 }
2844 
2845 /*
2846  * Decrement the pin count of the given inode, and wake up
2847  * anyone in xfs_iwait_unpin() if the count goes to 0.  The
2848  * inode must have been previoulsy pinned with a call to xfs_ipin().
2849  */
2850 void
2851 xfs_iunpin(
2852 	xfs_inode_t	*ip)
2853 {
2854 	ASSERT(atomic_read(&ip->i_pincount) > 0);
2855 
2856 	if (atomic_dec_and_test(&ip->i_pincount)) {
2857 		vnode_t	*vp = XFS_ITOV_NULL(ip);
2858 
2859 		/* make sync come back and flush this inode */
2860 		if (vp) {
2861 			struct inode	*inode = LINVFS_GET_IP(vp);
2862 
2863 			if (!(inode->i_state & I_NEW))
2864 				mark_inode_dirty_sync(inode);
2865 		}
2866 
2867 		wake_up(&ip->i_ipin_wait);
2868 	}
2869 }
2870 
2871 /*
2872  * This is called to wait for the given inode to be unpinned.
2873  * It will sleep until this happens.  The caller must have the
2874  * inode locked in at least shared mode so that the buffer cannot
2875  * be subsequently pinned once someone is waiting for it to be
2876  * unpinned.
2877  */
2878 void
2879 xfs_iunpin_wait(
2880 	xfs_inode_t	*ip)
2881 {
2882 	xfs_inode_log_item_t	*iip;
2883 	xfs_lsn_t	lsn;
2884 
2885 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2886 
2887 	if (atomic_read(&ip->i_pincount) == 0) {
2888 		return;
2889 	}
2890 
2891 	iip = ip->i_itemp;
2892 	if (iip && iip->ili_last_lsn) {
2893 		lsn = iip->ili_last_lsn;
2894 	} else {
2895 		lsn = (xfs_lsn_t)0;
2896 	}
2897 
2898 	/*
2899 	 * Give the log a push so we don't wait here too long.
2900 	 */
2901 	xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2902 
2903 	wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2904 }
2905 
2906 
2907 /*
2908  * xfs_iextents_copy()
2909  *
2910  * This is called to copy the REAL extents (as opposed to the delayed
2911  * allocation extents) from the inode into the given buffer.  It
2912  * returns the number of bytes copied into the buffer.
2913  *
2914  * If there are no delayed allocation extents, then we can just
2915  * memcpy() the extents into the buffer.  Otherwise, we need to
2916  * examine each extent in turn and skip those which are delayed.
2917  */
2918 int
2919 xfs_iextents_copy(
2920 	xfs_inode_t		*ip,
2921 	xfs_bmbt_rec_t		*buffer,
2922 	int			whichfork)
2923 {
2924 	int			copied;
2925 	xfs_bmbt_rec_t		*dest_ep;
2926 	xfs_bmbt_rec_t		*ep;
2927 #ifdef XFS_BMAP_TRACE
2928 	static char		fname[] = "xfs_iextents_copy";
2929 #endif
2930 	int			i;
2931 	xfs_ifork_t		*ifp;
2932 	int			nrecs;
2933 	xfs_fsblock_t		start_block;
2934 
2935 	ifp = XFS_IFORK_PTR(ip, whichfork);
2936 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2937 	ASSERT(ifp->if_bytes > 0);
2938 
2939 	nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2940 	xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2941 	ASSERT(nrecs > 0);
2942 
2943 	/*
2944 	 * There are some delayed allocation extents in the
2945 	 * inode, so copy the extents one at a time and skip
2946 	 * the delayed ones.  There must be at least one
2947 	 * non-delayed extent.
2948 	 */
2949 	ep = ifp->if_u1.if_extents;
2950 	dest_ep = buffer;
2951 	copied = 0;
2952 	for (i = 0; i < nrecs; i++) {
2953 		start_block = xfs_bmbt_get_startblock(ep);
2954 		if (ISNULLSTARTBLOCK(start_block)) {
2955 			/*
2956 			 * It's a delayed allocation extent, so skip it.
2957 			 */
2958 			ep++;
2959 			continue;
2960 		}
2961 
2962 		/* Translate to on disk format */
2963 		put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2964 			      (__uint64_t*)&dest_ep->l0);
2965 		put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2966 			      (__uint64_t*)&dest_ep->l1);
2967 		dest_ep++;
2968 		ep++;
2969 		copied++;
2970 	}
2971 	ASSERT(copied != 0);
2972 	xfs_validate_extents(buffer, copied, 1, XFS_EXTFMT_INODE(ip));
2973 
2974 	return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2975 }
2976 
2977 /*
2978  * Each of the following cases stores data into the same region
2979  * of the on-disk inode, so only one of them can be valid at
2980  * any given time. While it is possible to have conflicting formats
2981  * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2982  * in EXTENTS format, this can only happen when the fork has
2983  * changed formats after being modified but before being flushed.
2984  * In these cases, the format always takes precedence, because the
2985  * format indicates the current state of the fork.
2986  */
2987 /*ARGSUSED*/
2988 STATIC int
2989 xfs_iflush_fork(
2990 	xfs_inode_t		*ip,
2991 	xfs_dinode_t		*dip,
2992 	xfs_inode_log_item_t	*iip,
2993 	int			whichfork,
2994 	xfs_buf_t		*bp)
2995 {
2996 	char			*cp;
2997 	xfs_ifork_t		*ifp;
2998 	xfs_mount_t		*mp;
2999 #ifdef XFS_TRANS_DEBUG
3000 	int			first;
3001 #endif
3002 	static const short	brootflag[2] =
3003 		{ XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
3004 	static const short	dataflag[2] =
3005 		{ XFS_ILOG_DDATA, XFS_ILOG_ADATA };
3006 	static const short	extflag[2] =
3007 		{ XFS_ILOG_DEXT, XFS_ILOG_AEXT };
3008 
3009 	if (iip == NULL)
3010 		return 0;
3011 	ifp = XFS_IFORK_PTR(ip, whichfork);
3012 	/*
3013 	 * This can happen if we gave up in iformat in an error path,
3014 	 * for the attribute fork.
3015 	 */
3016 	if (ifp == NULL) {
3017 		ASSERT(whichfork == XFS_ATTR_FORK);
3018 		return 0;
3019 	}
3020 	cp = XFS_DFORK_PTR(dip, whichfork);
3021 	mp = ip->i_mount;
3022 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
3023 	case XFS_DINODE_FMT_LOCAL:
3024 		if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
3025 		    (ifp->if_bytes > 0)) {
3026 			ASSERT(ifp->if_u1.if_data != NULL);
3027 			ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
3028 			memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
3029 		}
3030 		if (whichfork == XFS_DATA_FORK) {
3031 			if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
3032 				XFS_ERROR_REPORT("xfs_iflush_fork",
3033 						 XFS_ERRLEVEL_LOW, mp);
3034 				return XFS_ERROR(EFSCORRUPTED);
3035 			}
3036 		}
3037 		break;
3038 
3039 	case XFS_DINODE_FMT_EXTENTS:
3040 		ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
3041 		       !(iip->ili_format.ilf_fields & extflag[whichfork]));
3042 		ASSERT((ifp->if_u1.if_extents != NULL) || (ifp->if_bytes == 0));
3043 		ASSERT((ifp->if_u1.if_extents == NULL) || (ifp->if_bytes > 0));
3044 		if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
3045 		    (ifp->if_bytes > 0)) {
3046 			ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
3047 			(void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
3048 				whichfork);
3049 		}
3050 		break;
3051 
3052 	case XFS_DINODE_FMT_BTREE:
3053 		if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
3054 		    (ifp->if_broot_bytes > 0)) {
3055 			ASSERT(ifp->if_broot != NULL);
3056 			ASSERT(ifp->if_broot_bytes <=
3057 			       (XFS_IFORK_SIZE(ip, whichfork) +
3058 				XFS_BROOT_SIZE_ADJ));
3059 			xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3060 				(xfs_bmdr_block_t *)cp,
3061 				XFS_DFORK_SIZE(dip, mp, whichfork));
3062 		}
3063 		break;
3064 
3065 	case XFS_DINODE_FMT_DEV:
3066 		if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3067 			ASSERT(whichfork == XFS_DATA_FORK);
3068 			INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
3069 		}
3070 		break;
3071 
3072 	case XFS_DINODE_FMT_UUID:
3073 		if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3074 			ASSERT(whichfork == XFS_DATA_FORK);
3075 			memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3076 				sizeof(uuid_t));
3077 		}
3078 		break;
3079 
3080 	default:
3081 		ASSERT(0);
3082 		break;
3083 	}
3084 
3085 	return 0;
3086 }
3087 
3088 /*
3089  * xfs_iflush() will write a modified inode's changes out to the
3090  * inode's on disk home.  The caller must have the inode lock held
3091  * in at least shared mode and the inode flush semaphore must be
3092  * held as well.  The inode lock will still be held upon return from
3093  * the call and the caller is free to unlock it.
3094  * The inode flush lock will be unlocked when the inode reaches the disk.
3095  * The flags indicate how the inode's buffer should be written out.
3096  */
3097 int
3098 xfs_iflush(
3099 	xfs_inode_t		*ip,
3100 	uint			flags)
3101 {
3102 	xfs_inode_log_item_t	*iip;
3103 	xfs_buf_t		*bp;
3104 	xfs_dinode_t		*dip;
3105 	xfs_mount_t		*mp;
3106 	int			error;
3107 	/* REFERENCED */
3108 	xfs_chash_t		*ch;
3109 	xfs_inode_t		*iq;
3110 	int			clcount;	/* count of inodes clustered */
3111 	int			bufwasdelwri;
3112 	enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3113 	SPLDECL(s);
3114 
3115 	XFS_STATS_INC(xs_iflush_count);
3116 
3117 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3118 	ASSERT(valusema(&ip->i_flock) <= 0);
3119 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3120 	       ip->i_d.di_nextents > ip->i_df.if_ext_max);
3121 
3122 	iip = ip->i_itemp;
3123 	mp = ip->i_mount;
3124 
3125 	/*
3126 	 * If the inode isn't dirty, then just release the inode
3127 	 * flush lock and do nothing.
3128 	 */
3129 	if ((ip->i_update_core == 0) &&
3130 	    ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3131 		ASSERT((iip != NULL) ?
3132 			 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3133 		xfs_ifunlock(ip);
3134 		return 0;
3135 	}
3136 
3137 	/*
3138 	 * We can't flush the inode until it is unpinned, so
3139 	 * wait for it.  We know noone new can pin it, because
3140 	 * we are holding the inode lock shared and you need
3141 	 * to hold it exclusively to pin the inode.
3142 	 */
3143 	xfs_iunpin_wait(ip);
3144 
3145 	/*
3146 	 * This may have been unpinned because the filesystem is shutting
3147 	 * down forcibly. If that's the case we must not write this inode
3148 	 * to disk, because the log record didn't make it to disk!
3149 	 */
3150 	if (XFS_FORCED_SHUTDOWN(mp)) {
3151 		ip->i_update_core = 0;
3152 		if (iip)
3153 			iip->ili_format.ilf_fields = 0;
3154 		xfs_ifunlock(ip);
3155 		return XFS_ERROR(EIO);
3156 	}
3157 
3158 	/*
3159 	 * Get the buffer containing the on-disk inode.
3160 	 */
3161 	error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0);
3162 	if (error != 0) {
3163 		xfs_ifunlock(ip);
3164 		return error;
3165 	}
3166 
3167 	/*
3168 	 * Decide how buffer will be flushed out.  This is done before
3169 	 * the call to xfs_iflush_int because this field is zeroed by it.
3170 	 */
3171 	if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3172 		/*
3173 		 * Flush out the inode buffer according to the directions
3174 		 * of the caller.  In the cases where the caller has given
3175 		 * us a choice choose the non-delwri case.  This is because
3176 		 * the inode is in the AIL and we need to get it out soon.
3177 		 */
3178 		switch (flags) {
3179 		case XFS_IFLUSH_SYNC:
3180 		case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3181 			flags = 0;
3182 			break;
3183 		case XFS_IFLUSH_ASYNC:
3184 		case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3185 			flags = INT_ASYNC;
3186 			break;
3187 		case XFS_IFLUSH_DELWRI:
3188 			flags = INT_DELWRI;
3189 			break;
3190 		default:
3191 			ASSERT(0);
3192 			flags = 0;
3193 			break;
3194 		}
3195 	} else {
3196 		switch (flags) {
3197 		case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3198 		case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3199 		case XFS_IFLUSH_DELWRI:
3200 			flags = INT_DELWRI;
3201 			break;
3202 		case XFS_IFLUSH_ASYNC:
3203 			flags = INT_ASYNC;
3204 			break;
3205 		case XFS_IFLUSH_SYNC:
3206 			flags = 0;
3207 			break;
3208 		default:
3209 			ASSERT(0);
3210 			flags = 0;
3211 			break;
3212 		}
3213 	}
3214 
3215 	/*
3216 	 * First flush out the inode that xfs_iflush was called with.
3217 	 */
3218 	error = xfs_iflush_int(ip, bp);
3219 	if (error) {
3220 		goto corrupt_out;
3221 	}
3222 
3223 	/*
3224 	 * inode clustering:
3225 	 * see if other inodes can be gathered into this write
3226 	 */
3227 
3228 	ip->i_chash->chl_buf = bp;
3229 
3230 	ch = XFS_CHASH(mp, ip->i_blkno);
3231 	s = mutex_spinlock(&ch->ch_lock);
3232 
3233 	clcount = 0;
3234 	for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3235 		/*
3236 		 * Do an un-protected check to see if the inode is dirty and
3237 		 * is a candidate for flushing.  These checks will be repeated
3238 		 * later after the appropriate locks are acquired.
3239 		 */
3240 		iip = iq->i_itemp;
3241 		if ((iq->i_update_core == 0) &&
3242 		    ((iip == NULL) ||
3243 		     !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3244 		      xfs_ipincount(iq) == 0) {
3245 			continue;
3246 		}
3247 
3248 		/*
3249 		 * Try to get locks.  If any are unavailable,
3250 		 * then this inode cannot be flushed and is skipped.
3251 		 */
3252 
3253 		/* get inode locks (just i_lock) */
3254 		if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3255 			/* get inode flush lock */
3256 			if (xfs_iflock_nowait(iq)) {
3257 				/* check if pinned */
3258 				if (xfs_ipincount(iq) == 0) {
3259 					/* arriving here means that
3260 					 * this inode can be flushed.
3261 					 * first re-check that it's
3262 					 * dirty
3263 					 */
3264 					iip = iq->i_itemp;
3265 					if ((iq->i_update_core != 0)||
3266 					    ((iip != NULL) &&
3267 					     (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3268 						clcount++;
3269 						error = xfs_iflush_int(iq, bp);
3270 						if (error) {
3271 							xfs_iunlock(iq,
3272 								    XFS_ILOCK_SHARED);
3273 							goto cluster_corrupt_out;
3274 						}
3275 					} else {
3276 						xfs_ifunlock(iq);
3277 					}
3278 				} else {
3279 					xfs_ifunlock(iq);
3280 				}
3281 			}
3282 			xfs_iunlock(iq, XFS_ILOCK_SHARED);
3283 		}
3284 	}
3285 	mutex_spinunlock(&ch->ch_lock, s);
3286 
3287 	if (clcount) {
3288 		XFS_STATS_INC(xs_icluster_flushcnt);
3289 		XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3290 	}
3291 
3292 	/*
3293 	 * If the buffer is pinned then push on the log so we won't
3294 	 * get stuck waiting in the write for too long.
3295 	 */
3296 	if (XFS_BUF_ISPINNED(bp)){
3297 		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3298 	}
3299 
3300 	if (flags & INT_DELWRI) {
3301 		xfs_bdwrite(mp, bp);
3302 	} else if (flags & INT_ASYNC) {
3303 		xfs_bawrite(mp, bp);
3304 	} else {
3305 		error = xfs_bwrite(mp, bp);
3306 	}
3307 	return error;
3308 
3309 corrupt_out:
3310 	xfs_buf_relse(bp);
3311 	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3312 	xfs_iflush_abort(ip);
3313 	/*
3314 	 * Unlocks the flush lock
3315 	 */
3316 	return XFS_ERROR(EFSCORRUPTED);
3317 
3318 cluster_corrupt_out:
3319 	/* Corruption detected in the clustering loop.  Invalidate the
3320 	 * inode buffer and shut down the filesystem.
3321 	 */
3322 	mutex_spinunlock(&ch->ch_lock, s);
3323 
3324 	/*
3325 	 * Clean up the buffer.  If it was B_DELWRI, just release it --
3326 	 * brelse can handle it with no problems.  If not, shut down the
3327 	 * filesystem before releasing the buffer.
3328 	 */
3329 	if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3330 		xfs_buf_relse(bp);
3331 	}
3332 
3333 	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3334 
3335 	if(!bufwasdelwri)  {
3336 		/*
3337 		 * Just like incore_relse: if we have b_iodone functions,
3338 		 * mark the buffer as an error and call them.  Otherwise
3339 		 * mark it as stale and brelse.
3340 		 */
3341 		if (XFS_BUF_IODONE_FUNC(bp)) {
3342 			XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3343 			XFS_BUF_UNDONE(bp);
3344 			XFS_BUF_STALE(bp);
3345 			XFS_BUF_SHUT(bp);
3346 			XFS_BUF_ERROR(bp,EIO);
3347 			xfs_biodone(bp);
3348 		} else {
3349 			XFS_BUF_STALE(bp);
3350 			xfs_buf_relse(bp);
3351 		}
3352 	}
3353 
3354 	xfs_iflush_abort(iq);
3355 	/*
3356 	 * Unlocks the flush lock
3357 	 */
3358 	return XFS_ERROR(EFSCORRUPTED);
3359 }
3360 
3361 
3362 STATIC int
3363 xfs_iflush_int(
3364 	xfs_inode_t		*ip,
3365 	xfs_buf_t		*bp)
3366 {
3367 	xfs_inode_log_item_t	*iip;
3368 	xfs_dinode_t		*dip;
3369 	xfs_mount_t		*mp;
3370 #ifdef XFS_TRANS_DEBUG
3371 	int			first;
3372 #endif
3373 	SPLDECL(s);
3374 
3375 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3376 	ASSERT(valusema(&ip->i_flock) <= 0);
3377 	ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3378 	       ip->i_d.di_nextents > ip->i_df.if_ext_max);
3379 
3380 	iip = ip->i_itemp;
3381 	mp = ip->i_mount;
3382 
3383 
3384 	/*
3385 	 * If the inode isn't dirty, then just release the inode
3386 	 * flush lock and do nothing.
3387 	 */
3388 	if ((ip->i_update_core == 0) &&
3389 	    ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3390 		xfs_ifunlock(ip);
3391 		return 0;
3392 	}
3393 
3394 	/* set *dip = inode's place in the buffer */
3395 	dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3396 
3397 	/*
3398 	 * Clear i_update_core before copying out the data.
3399 	 * This is for coordination with our timestamp updates
3400 	 * that don't hold the inode lock. They will always
3401 	 * update the timestamps BEFORE setting i_update_core,
3402 	 * so if we clear i_update_core after they set it we
3403 	 * are guaranteed to see their updates to the timestamps.
3404 	 * I believe that this depends on strongly ordered memory
3405 	 * semantics, but we have that.  We use the SYNCHRONIZE
3406 	 * macro to make sure that the compiler does not reorder
3407 	 * the i_update_core access below the data copy below.
3408 	 */
3409 	ip->i_update_core = 0;
3410 	SYNCHRONIZE();
3411 
3412 	if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3413 			       mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3414 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3415 		    "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3416 			ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3417 		goto corrupt_out;
3418 	}
3419 	if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3420 				mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3421 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3422 			"xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3423 			ip->i_ino, ip, ip->i_d.di_magic);
3424 		goto corrupt_out;
3425 	}
3426 	if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3427 		if (XFS_TEST_ERROR(
3428 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3429 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3430 		    mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3431 			xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3432 				"xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3433 				ip->i_ino, ip);
3434 			goto corrupt_out;
3435 		}
3436 	} else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3437 		if (XFS_TEST_ERROR(
3438 		    (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3439 		    (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3440 		    (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3441 		    mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3442 			xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3443 				"xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3444 				ip->i_ino, ip);
3445 			goto corrupt_out;
3446 		}
3447 	}
3448 	if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3449 				ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3450 				XFS_RANDOM_IFLUSH_5)) {
3451 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3452 			"xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3453 			ip->i_ino,
3454 			ip->i_d.di_nextents + ip->i_d.di_anextents,
3455 			ip->i_d.di_nblocks,
3456 			ip);
3457 		goto corrupt_out;
3458 	}
3459 	if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3460 				mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3461 		xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3462 			"xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3463 			ip->i_ino, ip->i_d.di_forkoff, ip);
3464 		goto corrupt_out;
3465 	}
3466 	/*
3467 	 * bump the flush iteration count, used to detect flushes which
3468 	 * postdate a log record during recovery.
3469 	 */
3470 
3471 	ip->i_d.di_flushiter++;
3472 
3473 	/*
3474 	 * Copy the dirty parts of the inode into the on-disk
3475 	 * inode.  We always copy out the core of the inode,
3476 	 * because if the inode is dirty at all the core must
3477 	 * be.
3478 	 */
3479 	xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3480 
3481 	/* Wrap, we never let the log put out DI_MAX_FLUSH */
3482 	if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3483 		ip->i_d.di_flushiter = 0;
3484 
3485 	/*
3486 	 * If this is really an old format inode and the superblock version
3487 	 * has not been updated to support only new format inodes, then
3488 	 * convert back to the old inode format.  If the superblock version
3489 	 * has been updated, then make the conversion permanent.
3490 	 */
3491 	ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3492 	       XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3493 	if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3494 		if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3495 			/*
3496 			 * Convert it back.
3497 			 */
3498 			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3499 			INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3500 		} else {
3501 			/*
3502 			 * The superblock version has already been bumped,
3503 			 * so just make the conversion to the new inode
3504 			 * format permanent.
3505 			 */
3506 			ip->i_d.di_version = XFS_DINODE_VERSION_2;
3507 			INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3508 			ip->i_d.di_onlink = 0;
3509 			dip->di_core.di_onlink = 0;
3510 			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3511 			memset(&(dip->di_core.di_pad[0]), 0,
3512 			      sizeof(dip->di_core.di_pad));
3513 			ASSERT(ip->i_d.di_projid == 0);
3514 		}
3515 	}
3516 
3517 	if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3518 		goto corrupt_out;
3519 	}
3520 
3521 	if (XFS_IFORK_Q(ip)) {
3522 		/*
3523 		 * The only error from xfs_iflush_fork is on the data fork.
3524 		 */
3525 		(void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3526 	}
3527 	xfs_inobp_check(mp, bp);
3528 
3529 	/*
3530 	 * We've recorded everything logged in the inode, so we'd
3531 	 * like to clear the ilf_fields bits so we don't log and
3532 	 * flush things unnecessarily.  However, we can't stop
3533 	 * logging all this information until the data we've copied
3534 	 * into the disk buffer is written to disk.  If we did we might
3535 	 * overwrite the copy of the inode in the log with all the
3536 	 * data after re-logging only part of it, and in the face of
3537 	 * a crash we wouldn't have all the data we need to recover.
3538 	 *
3539 	 * What we do is move the bits to the ili_last_fields field.
3540 	 * When logging the inode, these bits are moved back to the
3541 	 * ilf_fields field.  In the xfs_iflush_done() routine we
3542 	 * clear ili_last_fields, since we know that the information
3543 	 * those bits represent is permanently on disk.  As long as
3544 	 * the flush completes before the inode is logged again, then
3545 	 * both ilf_fields and ili_last_fields will be cleared.
3546 	 *
3547 	 * We can play with the ilf_fields bits here, because the inode
3548 	 * lock must be held exclusively in order to set bits there
3549 	 * and the flush lock protects the ili_last_fields bits.
3550 	 * Set ili_logged so the flush done
3551 	 * routine can tell whether or not to look in the AIL.
3552 	 * Also, store the current LSN of the inode so that we can tell
3553 	 * whether the item has moved in the AIL from xfs_iflush_done().
3554 	 * In order to read the lsn we need the AIL lock, because
3555 	 * it is a 64 bit value that cannot be read atomically.
3556 	 */
3557 	if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3558 		iip->ili_last_fields = iip->ili_format.ilf_fields;
3559 		iip->ili_format.ilf_fields = 0;
3560 		iip->ili_logged = 1;
3561 
3562 		ASSERT(sizeof(xfs_lsn_t) == 8);	/* don't lock if it shrinks */
3563 		AIL_LOCK(mp,s);
3564 		iip->ili_flush_lsn = iip->ili_item.li_lsn;
3565 		AIL_UNLOCK(mp, s);
3566 
3567 		/*
3568 		 * Attach the function xfs_iflush_done to the inode's
3569 		 * buffer.  This will remove the inode from the AIL
3570 		 * and unlock the inode's flush lock when the inode is
3571 		 * completely written to disk.
3572 		 */
3573 		xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3574 				      xfs_iflush_done, (xfs_log_item_t *)iip);
3575 
3576 		ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3577 		ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3578 	} else {
3579 		/*
3580 		 * We're flushing an inode which is not in the AIL and has
3581 		 * not been logged but has i_update_core set.  For this
3582 		 * case we can use a B_DELWRI flush and immediately drop
3583 		 * the inode flush lock because we can avoid the whole
3584 		 * AIL state thing.  It's OK to drop the flush lock now,
3585 		 * because we've already locked the buffer and to do anything
3586 		 * you really need both.
3587 		 */
3588 		if (iip != NULL) {
3589 			ASSERT(iip->ili_logged == 0);
3590 			ASSERT(iip->ili_last_fields == 0);
3591 			ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3592 		}
3593 		xfs_ifunlock(ip);
3594 	}
3595 
3596 	return 0;
3597 
3598 corrupt_out:
3599 	return XFS_ERROR(EFSCORRUPTED);
3600 }
3601 
3602 
3603 /*
3604  * Flush all inactive inodes in mp.  Return true if no user references
3605  * were found, false otherwise.
3606  */
3607 int
3608 xfs_iflush_all(
3609 	xfs_mount_t	*mp,
3610 	int		flag)
3611 {
3612 	int		busy;
3613 	int		done;
3614 	int		purged;
3615 	xfs_inode_t	*ip;
3616 	vmap_t		vmap;
3617 	vnode_t		*vp;
3618 
3619 	busy = done = 0;
3620 	while (!done) {
3621 		purged = 0;
3622 		XFS_MOUNT_ILOCK(mp);
3623 		ip = mp->m_inodes;
3624 		if (ip == NULL) {
3625 			break;
3626 		}
3627 		do {
3628 			/* Make sure we skip markers inserted by sync */
3629 			if (ip->i_mount == NULL) {
3630 				ip = ip->i_mnext;
3631 				continue;
3632 			}
3633 
3634 			/*
3635 			 * It's up to our caller to purge the root
3636 			 * and quota vnodes later.
3637 			 */
3638 			vp = XFS_ITOV_NULL(ip);
3639 
3640 			if (!vp) {
3641 				XFS_MOUNT_IUNLOCK(mp);
3642 				xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3643 				purged = 1;
3644 				break;
3645 			}
3646 
3647 			if (vn_count(vp) != 0) {
3648 				if (vn_count(vp) == 1 &&
3649 				    (ip == mp->m_rootip ||
3650 				     (mp->m_quotainfo &&
3651 				      (ip->i_ino == mp->m_sb.sb_uquotino ||
3652 				       ip->i_ino == mp->m_sb.sb_gquotino)))) {
3653 
3654 					ip = ip->i_mnext;
3655 					continue;
3656 				}
3657 				if (!(flag & XFS_FLUSH_ALL)) {
3658 					busy = 1;
3659 					done = 1;
3660 					break;
3661 				}
3662 				/*
3663 				 * Ignore busy inodes but continue flushing
3664 				 * others.
3665 				 */
3666 				ip = ip->i_mnext;
3667 				continue;
3668 			}
3669 			/*
3670 			 * Sample vp mapping while holding mp locked on MP
3671 			 * systems, so we don't purge a reclaimed or
3672 			 * nonexistent vnode.  We break from the loop
3673 			 * since we know that we modify
3674 			 * it by pulling ourselves from it in xfs_reclaim()
3675 			 * called via vn_purge() below.  Set ip to the next
3676 			 * entry in the list anyway so we'll know below
3677 			 * whether we reached the end or not.
3678 			 */
3679 			VMAP(vp, vmap);
3680 			XFS_MOUNT_IUNLOCK(mp);
3681 
3682 			vn_purge(vp, &vmap);
3683 
3684 			purged = 1;
3685 			break;
3686 		} while (ip != mp->m_inodes);
3687 		/*
3688 		 * We need to distinguish between when we exit the loop
3689 		 * after a purge and when we simply hit the end of the
3690 		 * list.  We can't use the (ip == mp->m_inodes) test,
3691 		 * because when we purge an inode at the start of the list
3692 		 * the next inode on the list becomes mp->m_inodes.  That
3693 		 * would cause such a test to bail out early.  The purged
3694 		 * variable tells us how we got out of the loop.
3695 		 */
3696 		if (!purged) {
3697 			done = 1;
3698 		}
3699 	}
3700 	XFS_MOUNT_IUNLOCK(mp);
3701 	return !busy;
3702 }
3703 
3704 
3705 /*
3706  * xfs_iaccess: check accessibility of inode for mode.
3707  */
3708 int
3709 xfs_iaccess(
3710 	xfs_inode_t	*ip,
3711 	mode_t		mode,
3712 	cred_t		*cr)
3713 {
3714 	int		error;
3715 	mode_t		orgmode = mode;
3716 	struct inode	*inode = LINVFS_GET_IP(XFS_ITOV(ip));
3717 
3718 	if (mode & S_IWUSR) {
3719 		umode_t		imode = inode->i_mode;
3720 
3721 		if (IS_RDONLY(inode) &&
3722 		    (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3723 			return XFS_ERROR(EROFS);
3724 
3725 		if (IS_IMMUTABLE(inode))
3726 			return XFS_ERROR(EACCES);
3727 	}
3728 
3729 	/*
3730 	 * If there's an Access Control List it's used instead of
3731 	 * the mode bits.
3732 	 */
3733 	if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3734 		return error ? XFS_ERROR(error) : 0;
3735 
3736 	if (current_fsuid(cr) != ip->i_d.di_uid) {
3737 		mode >>= 3;
3738 		if (!in_group_p((gid_t)ip->i_d.di_gid))
3739 			mode >>= 3;
3740 	}
3741 
3742 	/*
3743 	 * If the DACs are ok we don't need any capability check.
3744 	 */
3745 	if ((ip->i_d.di_mode & mode) == mode)
3746 		return 0;
3747 	/*
3748 	 * Read/write DACs are always overridable.
3749 	 * Executable DACs are overridable if at least one exec bit is set.
3750 	 */
3751 	if (!(orgmode & S_IXUSR) ||
3752 	    (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3753 		if (capable_cred(cr, CAP_DAC_OVERRIDE))
3754 			return 0;
3755 
3756 	if ((orgmode == S_IRUSR) ||
3757 	    (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3758 		if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3759 			return 0;
3760 #ifdef	NOISE
3761 		cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3762 #endif	/* NOISE */
3763 		return XFS_ERROR(EACCES);
3764 	}
3765 	return XFS_ERROR(EACCES);
3766 }
3767 
3768 /*
3769  * xfs_iroundup: round up argument to next power of two
3770  */
3771 uint
3772 xfs_iroundup(
3773 	uint	v)
3774 {
3775 	int i;
3776 	uint m;
3777 
3778 	if ((v & (v - 1)) == 0)
3779 		return v;
3780 	ASSERT((v & 0x80000000) == 0);
3781 	if ((v & (v + 1)) == 0)
3782 		return v + 1;
3783 	for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3784 		if (v & m)
3785 			continue;
3786 		v |= m;
3787 		if ((v & (v + 1)) == 0)
3788 			return v + 1;
3789 	}
3790 	ASSERT(0);
3791 	return( 0 );
3792 }
3793 
3794 /*
3795  * Change the requested timestamp in the given inode.
3796  * We don't lock across timestamp updates, and we don't log them but
3797  * we do record the fact that there is dirty information in core.
3798  *
3799  * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
3800  *		with XFS_ICHGTIME_ACC to be sure that access time
3801  *		update will take.  Calling first with XFS_ICHGTIME_ACC
3802  *		and then XFS_ICHGTIME_MOD may fail to modify the access
3803  *		timestamp if the filesystem is mounted noacctm.
3804  */
3805 void
3806 xfs_ichgtime(xfs_inode_t *ip,
3807 	     int flags)
3808 {
3809 	timespec_t	tv;
3810 	vnode_t		*vp = XFS_ITOV(ip);
3811 	struct inode	*inode = LINVFS_GET_IP(vp);
3812 
3813 	/*
3814 	 * We're not supposed to change timestamps in readonly-mounted
3815 	 * filesystems.  Throw it away if anyone asks us.
3816 	 */
3817 	if (unlikely(vp->v_vfsp->vfs_flag & VFS_RDONLY))
3818 		return;
3819 
3820 	/*
3821 	 * Don't update access timestamps on reads if mounted "noatime"
3822 	 * Throw it away if anyone asks us.
3823 	 */
3824 	if ((ip->i_mount->m_flags & XFS_MOUNT_NOATIME || IS_NOATIME(inode)) &&
3825 	    ((flags & (XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD|XFS_ICHGTIME_CHG))
3826 			== XFS_ICHGTIME_ACC))
3827 		return;
3828 
3829 	nanotime(&tv);
3830 	if (flags & XFS_ICHGTIME_MOD) {
3831 		VN_MTIMESET(vp, &tv);
3832 		ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
3833 		ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
3834 	}
3835 	if (flags & XFS_ICHGTIME_ACC) {
3836 		VN_ATIMESET(vp, &tv);
3837 		ip->i_d.di_atime.t_sec = (__int32_t)tv.tv_sec;
3838 		ip->i_d.di_atime.t_nsec = (__int32_t)tv.tv_nsec;
3839 	}
3840 	if (flags & XFS_ICHGTIME_CHG) {
3841 		VN_CTIMESET(vp, &tv);
3842 		ip->i_d.di_ctime.t_sec = (__int32_t)tv.tv_sec;
3843 		ip->i_d.di_ctime.t_nsec = (__int32_t)tv.tv_nsec;
3844 	}
3845 
3846 	/*
3847 	 * We update the i_update_core field _after_ changing
3848 	 * the timestamps in order to coordinate properly with
3849 	 * xfs_iflush() so that we don't lose timestamp updates.
3850 	 * This keeps us from having to hold the inode lock
3851 	 * while doing this.  We use the SYNCHRONIZE macro to
3852 	 * ensure that the compiler does not reorder the update
3853 	 * of i_update_core above the timestamp updates above.
3854 	 */
3855 	SYNCHRONIZE();
3856 	ip->i_update_core = 1;
3857 	if (!(inode->i_state & I_LOCK))
3858 		mark_inode_dirty_sync(inode);
3859 }
3860 
3861 #ifdef XFS_ILOCK_TRACE
3862 ktrace_t	*xfs_ilock_trace_buf;
3863 
3864 void
3865 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3866 {
3867 	ktrace_enter(ip->i_lock_trace,
3868 		     (void *)ip,
3869 		     (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3870 		     (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3871 		     (void *)ra,		/* caller of ilock */
3872 		     (void *)(unsigned long)current_cpu(),
3873 		     (void *)(unsigned long)current_pid(),
3874 		     NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3875 }
3876 #endif
3877