xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 87c1627502a5dde91e5284118eec8682b60f27a2)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_quota.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mount.h>
40 #include <sys/proc.h>
41 #include <sys/bio.h>
42 #include <sys/buf.h>
43 #include <sys/vnode.h>
44 #include <sys/malloc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/vmmeter.h>
47 #include <sys/stat.h>
48 
49 #include <vm/vm.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_object.h>
52 
53 #include <ufs/ufs/extattr.h>
54 #include <ufs/ufs/quota.h>
55 #include <ufs/ufs/ufsmount.h>
56 #include <ufs/ufs/inode.h>
57 #include <ufs/ufs/ufs_extern.h>
58 
59 #include <ufs/ffs/fs.h>
60 #include <ufs/ffs/ffs_extern.h>
61 
62 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
63 	    ufs2_daddr_t, int, ufs2_daddr_t *);
64 
65 /*
66  * Update the access, modified, and inode change times as specified by the
67  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
68  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
69  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
70  * later if not now.  The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
71  * is currently being suspended (or is suspended) and vnode has been accessed.
72  * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
73  * reflect the presumably successful write, and if waitfor is set, then wait
74  * for the write to complete.
75  */
76 int
77 ffs_update(vp, waitfor)
78 	struct vnode *vp;
79 	int waitfor;
80 {
81 	struct fs *fs;
82 	struct buf *bp;
83 	struct inode *ip;
84 	int flags, error;
85 
86 	ASSERT_VOP_ELOCKED(vp, "ffs_update");
87 	ufs_itimes(vp);
88 	ip = VTOI(vp);
89 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
90 		return (0);
91 	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
92 	fs = ip->i_fs;
93 	if (fs->fs_ronly && ip->i_ump->um_fsckpid == 0)
94 		return (0);
95 	/*
96 	 * If we are updating a snapshot and another process is currently
97 	 * writing the buffer containing the inode for this snapshot then
98 	 * a deadlock can occur when it tries to check the snapshot to see
99 	 * if that block needs to be copied. Thus when updating a snapshot
100 	 * we check to see if the buffer is already locked, and if it is
101 	 * we drop the snapshot lock until the buffer has been written
102 	 * and is available to us. We have to grab a reference to the
103 	 * snapshot vnode to prevent it from being removed while we are
104 	 * waiting for the buffer.
105 	 */
106 	flags = 0;
107 	if (IS_SNAPSHOT(ip))
108 		flags = GB_LOCK_NOWAIT;
109 loop:
110 	error = breadn_flags(ip->i_devvp,
111 	     fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
112 	     (int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
113 	if (error != 0) {
114 		if (error != EBUSY) {
115 			brelse(bp);
116 			return (error);
117 		}
118 		KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
119 		/*
120 		 * Wait for our inode block to become available.
121 		 *
122 		 * Hold a reference to the vnode to protect against
123 		 * ffs_snapgone(). Since we hold a reference, it can only
124 		 * get reclaimed (VI_DOOMED flag) in a forcible downgrade
125 		 * or unmount. For an unmount, the entire filesystem will be
126 		 * gone, so we cannot attempt to touch anything associated
127 		 * with it while the vnode is unlocked; all we can do is
128 		 * pause briefly and try again. If when we relock the vnode
129 		 * we discover that it has been reclaimed, updating it is no
130 		 * longer necessary and we can just return an error.
131 		 */
132 		vref(vp);
133 		VOP_UNLOCK(vp, 0);
134 		pause("ffsupd", 1);
135 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
136 		vrele(vp);
137 		if ((vp->v_iflag & VI_DOOMED) != 0)
138 			return (ENOENT);
139 		goto loop;
140 	}
141 	if (DOINGSOFTDEP(vp))
142 		softdep_update_inodeblock(ip, bp, waitfor);
143 	else if (ip->i_effnlink != ip->i_nlink)
144 		panic("ffs_update: bad link cnt");
145 	if (ip->i_ump->um_fstype == UFS1)
146 		*((struct ufs1_dinode *)bp->b_data +
147 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
148 	else
149 		*((struct ufs2_dinode *)bp->b_data +
150 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
151 	if (waitfor && !DOINGASYNC(vp))
152 		error = bwrite(bp);
153 	else if (vm_page_count_severe() || buf_dirty_count_severe()) {
154 		bawrite(bp);
155 		error = 0;
156 	} else {
157 		if (bp->b_bufsize == fs->fs_bsize)
158 			bp->b_flags |= B_CLUSTEROK;
159 		bdwrite(bp);
160 		error = 0;
161 	}
162 	return (error);
163 }
164 
165 #define	SINGLE	0	/* index of single indirect block */
166 #define	DOUBLE	1	/* index of double indirect block */
167 #define	TRIPLE	2	/* index of triple indirect block */
168 /*
169  * Truncate the inode ip to at most length size, freeing the
170  * disk blocks.
171  */
172 int
173 ffs_truncate(vp, length, flags, cred)
174 	struct vnode *vp;
175 	off_t length;
176 	int flags;
177 	struct ucred *cred;
178 {
179 	struct inode *ip;
180 	ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
181 	ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
182 	ufs2_daddr_t count, blocksreleased = 0, datablocks;
183 	struct bufobj *bo;
184 	struct fs *fs;
185 	struct buf *bp;
186 	struct ufsmount *ump;
187 	int softdeptrunc, journaltrunc;
188 	int needextclean, extblocks;
189 	int offset, size, level, nblocks;
190 	int i, error, allerror;
191 	off_t osize;
192 
193 	ip = VTOI(vp);
194 	fs = ip->i_fs;
195 	ump = ip->i_ump;
196 	bo = &vp->v_bufobj;
197 
198 	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
199 
200 	if (length < 0)
201 		return (EINVAL);
202 	if (length > fs->fs_maxfilesize)
203 		return (EFBIG);
204 #ifdef QUOTA
205 	error = getinoquota(ip);
206 	if (error)
207 		return (error);
208 #endif
209 	/*
210 	 * Historically clients did not have to specify which data
211 	 * they were truncating. So, if not specified, we assume
212 	 * traditional behavior, e.g., just the normal data.
213 	 */
214 	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
215 		flags |= IO_NORMAL;
216 	if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
217 		flags |= IO_SYNC;
218 	/*
219 	 * If we are truncating the extended-attributes, and cannot
220 	 * do it with soft updates, then do it slowly here. If we are
221 	 * truncating both the extended attributes and the file contents
222 	 * (e.g., the file is being unlinked), then pick it off with
223 	 * soft updates below.
224 	 */
225 	allerror = 0;
226 	needextclean = 0;
227 	softdeptrunc = 0;
228 	journaltrunc = DOINGSUJ(vp);
229 	if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
230 		softdeptrunc = !softdep_slowdown(vp);
231 	extblocks = 0;
232 	datablocks = DIP(ip, i_blocks);
233 	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
234 		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
235 		datablocks -= extblocks;
236 	}
237 	if ((flags & IO_EXT) && extblocks > 0) {
238 		if (length != 0)
239 			panic("ffs_truncate: partial trunc of extdata");
240 		if (softdeptrunc || journaltrunc) {
241 			if ((flags & IO_NORMAL) == 0)
242 				goto extclean;
243 			needextclean = 1;
244 		} else {
245 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
246 				return (error);
247 #ifdef QUOTA
248 			(void) chkdq(ip, -extblocks, NOCRED, 0);
249 #endif
250 			vinvalbuf(vp, V_ALT, 0, 0);
251 			vn_pages_remove(vp,
252 			    OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
253 			osize = ip->i_din2->di_extsize;
254 			ip->i_din2->di_blocks -= extblocks;
255 			ip->i_din2->di_extsize = 0;
256 			for (i = 0; i < NXADDR; i++) {
257 				oldblks[i] = ip->i_din2->di_extb[i];
258 				ip->i_din2->di_extb[i] = 0;
259 			}
260 			ip->i_flag |= IN_CHANGE;
261 			if ((error = ffs_update(vp, !DOINGASYNC(vp))))
262 				return (error);
263 			for (i = 0; i < NXADDR; i++) {
264 				if (oldblks[i] == 0)
265 					continue;
266 				ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
267 				    sblksize(fs, osize, i), ip->i_number,
268 				    vp->v_type, NULL);
269 			}
270 		}
271 	}
272 	if ((flags & IO_NORMAL) == 0)
273 		return (0);
274 	if (vp->v_type == VLNK &&
275 	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
276 	     datablocks == 0)) {
277 #ifdef INVARIANTS
278 		if (length != 0)
279 			panic("ffs_truncate: partial truncate of symlink");
280 #endif
281 		bzero(SHORTLINK(ip), (u_int)ip->i_size);
282 		ip->i_size = 0;
283 		DIP_SET(ip, i_size, 0);
284 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
285 		if (needextclean)
286 			goto extclean;
287 		return (ffs_update(vp, !DOINGASYNC(vp)));
288 	}
289 	if (ip->i_size == length) {
290 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
291 		if (needextclean)
292 			goto extclean;
293 		return (ffs_update(vp, 0));
294 	}
295 	if (fs->fs_ronly)
296 		panic("ffs_truncate: read-only filesystem");
297 	if (IS_SNAPSHOT(ip))
298 		ffs_snapremove(vp);
299 	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
300 	osize = ip->i_size;
301 	/*
302 	 * Lengthen the size of the file. We must ensure that the
303 	 * last byte of the file is allocated. Since the smallest
304 	 * value of osize is 0, length will be at least 1.
305 	 */
306 	if (osize < length) {
307 		vnode_pager_setsize(vp, length);
308 		flags |= BA_CLRBUF;
309 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
310 		if (error) {
311 			vnode_pager_setsize(vp, osize);
312 			return (error);
313 		}
314 		ip->i_size = length;
315 		DIP_SET(ip, i_size, length);
316 		if (bp->b_bufsize == fs->fs_bsize)
317 			bp->b_flags |= B_CLUSTEROK;
318 		if (flags & IO_SYNC)
319 			bwrite(bp);
320 		else if (DOINGASYNC(vp))
321 			bdwrite(bp);
322 		else
323 			bawrite(bp);
324 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
325 		return (ffs_update(vp, !DOINGASYNC(vp)));
326 	}
327 	if (DOINGSOFTDEP(vp)) {
328 		if (softdeptrunc == 0 && journaltrunc == 0) {
329 			/*
330 			 * If a file is only partially truncated, then
331 			 * we have to clean up the data structures
332 			 * describing the allocation past the truncation
333 			 * point. Finding and deallocating those structures
334 			 * is a lot of work. Since partial truncation occurs
335 			 * rarely, we solve the problem by syncing the file
336 			 * so that it will have no data structures left.
337 			 */
338 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
339 				return (error);
340 		} else {
341 			flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
342 			if (journaltrunc)
343 				softdep_journal_freeblocks(ip, cred, length,
344 				    flags);
345 			else
346 				softdep_setup_freeblocks(ip, length, flags);
347 			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
348 			if (journaltrunc == 0) {
349 				ip->i_flag |= IN_CHANGE | IN_UPDATE;
350 				error = ffs_update(vp, 0);
351 			}
352 			return (error);
353 		}
354 	}
355 	/*
356 	 * Shorten the size of the file. If the file is not being
357 	 * truncated to a block boundary, the contents of the
358 	 * partial block following the end of the file must be
359 	 * zero'ed in case it ever becomes accessible again because
360 	 * of subsequent file growth. Directories however are not
361 	 * zero'ed as they should grow back initialized to empty.
362 	 */
363 	offset = blkoff(fs, length);
364 	if (offset == 0) {
365 		ip->i_size = length;
366 		DIP_SET(ip, i_size, length);
367 	} else {
368 		lbn = lblkno(fs, length);
369 		flags |= BA_CLRBUF;
370 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
371 		if (error)
372 			return (error);
373 		/*
374 		 * When we are doing soft updates and the UFS_BALLOC
375 		 * above fills in a direct block hole with a full sized
376 		 * block that will be truncated down to a fragment below,
377 		 * we must flush out the block dependency with an FSYNC
378 		 * so that we do not get a soft updates inconsistency
379 		 * when we create the fragment below.
380 		 */
381 		if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
382 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
383 		    (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
384 			return (error);
385 		ip->i_size = length;
386 		DIP_SET(ip, i_size, length);
387 		size = blksize(fs, ip, lbn);
388 		if (vp->v_type != VDIR)
389 			bzero((char *)bp->b_data + offset,
390 			    (u_int)(size - offset));
391 		/* Kirk's code has reallocbuf(bp, size, 1) here */
392 		allocbuf(bp, size);
393 		if (bp->b_bufsize == fs->fs_bsize)
394 			bp->b_flags |= B_CLUSTEROK;
395 		if (flags & IO_SYNC)
396 			bwrite(bp);
397 		else if (DOINGASYNC(vp))
398 			bdwrite(bp);
399 		else
400 			bawrite(bp);
401 	}
402 	/*
403 	 * Calculate index into inode's block list of
404 	 * last direct and indirect blocks (if any)
405 	 * which we want to keep.  Lastblock is -1 when
406 	 * the file is truncated to 0.
407 	 */
408 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
409 	lastiblock[SINGLE] = lastblock - NDADDR;
410 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
411 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
412 	nblocks = btodb(fs->fs_bsize);
413 	/*
414 	 * Update file and block pointers on disk before we start freeing
415 	 * blocks.  If we crash before free'ing blocks below, the blocks
416 	 * will be returned to the free list.  lastiblock values are also
417 	 * normalized to -1 for calls to ffs_indirtrunc below.
418 	 */
419 	for (level = TRIPLE; level >= SINGLE; level--) {
420 		oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
421 		if (lastiblock[level] < 0) {
422 			DIP_SET(ip, i_ib[level], 0);
423 			lastiblock[level] = -1;
424 		}
425 	}
426 	for (i = 0; i < NDADDR; i++) {
427 		oldblks[i] = DIP(ip, i_db[i]);
428 		if (i > lastblock)
429 			DIP_SET(ip, i_db[i], 0);
430 	}
431 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
432 	allerror = ffs_update(vp, !DOINGASYNC(vp));
433 
434 	/*
435 	 * Having written the new inode to disk, save its new configuration
436 	 * and put back the old block pointers long enough to process them.
437 	 * Note that we save the new block configuration so we can check it
438 	 * when we are done.
439 	 */
440 	for (i = 0; i < NDADDR; i++) {
441 		newblks[i] = DIP(ip, i_db[i]);
442 		DIP_SET(ip, i_db[i], oldblks[i]);
443 	}
444 	for (i = 0; i < NIADDR; i++) {
445 		newblks[NDADDR + i] = DIP(ip, i_ib[i]);
446 		DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
447 	}
448 	ip->i_size = osize;
449 	DIP_SET(ip, i_size, osize);
450 
451 	error = vtruncbuf(vp, cred, length, fs->fs_bsize);
452 	if (error && (allerror == 0))
453 		allerror = error;
454 
455 	/*
456 	 * Indirect blocks first.
457 	 */
458 	indir_lbn[SINGLE] = -NDADDR;
459 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
460 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
461 	for (level = TRIPLE; level >= SINGLE; level--) {
462 		bn = DIP(ip, i_ib[level]);
463 		if (bn != 0) {
464 			error = ffs_indirtrunc(ip, indir_lbn[level],
465 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
466 			if (error)
467 				allerror = error;
468 			blocksreleased += count;
469 			if (lastiblock[level] < 0) {
470 				DIP_SET(ip, i_ib[level], 0);
471 				ffs_blkfree(ump, fs, ip->i_devvp, bn,
472 				    fs->fs_bsize, ip->i_number,
473 				    vp->v_type, NULL);
474 				blocksreleased += nblocks;
475 			}
476 		}
477 		if (lastiblock[level] >= 0)
478 			goto done;
479 	}
480 
481 	/*
482 	 * All whole direct blocks or frags.
483 	 */
484 	for (i = NDADDR - 1; i > lastblock; i--) {
485 		long bsize;
486 
487 		bn = DIP(ip, i_db[i]);
488 		if (bn == 0)
489 			continue;
490 		DIP_SET(ip, i_db[i], 0);
491 		bsize = blksize(fs, ip, i);
492 		ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number,
493 		    vp->v_type, NULL);
494 		blocksreleased += btodb(bsize);
495 	}
496 	if (lastblock < 0)
497 		goto done;
498 
499 	/*
500 	 * Finally, look for a change in size of the
501 	 * last direct block; release any frags.
502 	 */
503 	bn = DIP(ip, i_db[lastblock]);
504 	if (bn != 0) {
505 		long oldspace, newspace;
506 
507 		/*
508 		 * Calculate amount of space we're giving
509 		 * back as old block size minus new block size.
510 		 */
511 		oldspace = blksize(fs, ip, lastblock);
512 		ip->i_size = length;
513 		DIP_SET(ip, i_size, length);
514 		newspace = blksize(fs, ip, lastblock);
515 		if (newspace == 0)
516 			panic("ffs_truncate: newspace");
517 		if (oldspace - newspace > 0) {
518 			/*
519 			 * Block number of space to be free'd is
520 			 * the old block # plus the number of frags
521 			 * required for the storage we're keeping.
522 			 */
523 			bn += numfrags(fs, newspace);
524 			ffs_blkfree(ump, fs, ip->i_devvp, bn,
525 			   oldspace - newspace, ip->i_number, vp->v_type, NULL);
526 			blocksreleased += btodb(oldspace - newspace);
527 		}
528 	}
529 done:
530 #ifdef INVARIANTS
531 	for (level = SINGLE; level <= TRIPLE; level++)
532 		if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
533 			panic("ffs_truncate1");
534 	for (i = 0; i < NDADDR; i++)
535 		if (newblks[i] != DIP(ip, i_db[i]))
536 			panic("ffs_truncate2");
537 	BO_LOCK(bo);
538 	if (length == 0 &&
539 	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
540 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
541 		panic("ffs_truncate3");
542 	BO_UNLOCK(bo);
543 #endif /* INVARIANTS */
544 	/*
545 	 * Put back the real size.
546 	 */
547 	ip->i_size = length;
548 	DIP_SET(ip, i_size, length);
549 	if (DIP(ip, i_blocks) >= blocksreleased)
550 		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
551 	else	/* sanity */
552 		DIP_SET(ip, i_blocks, 0);
553 	ip->i_flag |= IN_CHANGE;
554 #ifdef QUOTA
555 	(void) chkdq(ip, -blocksreleased, NOCRED, 0);
556 #endif
557 	return (allerror);
558 
559 extclean:
560 	if (journaltrunc)
561 		softdep_journal_freeblocks(ip, cred, length, IO_EXT);
562 	else
563 		softdep_setup_freeblocks(ip, length, IO_EXT);
564 	return (ffs_update(vp, !DOINGASYNC(vp)));
565 }
566 
567 /*
568  * Release blocks associated with the inode ip and stored in the indirect
569  * block bn.  Blocks are free'd in LIFO order up to (but not including)
570  * lastbn.  If level is greater than SINGLE, the block is an indirect block
571  * and recursive calls to indirtrunc must be used to cleanse other indirect
572  * blocks.
573  */
574 static int
575 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
576 	struct inode *ip;
577 	ufs2_daddr_t lbn, lastbn;
578 	ufs2_daddr_t dbn;
579 	int level;
580 	ufs2_daddr_t *countp;
581 {
582 	struct buf *bp;
583 	struct fs *fs = ip->i_fs;
584 	struct vnode *vp;
585 	caddr_t copy = NULL;
586 	int i, nblocks, error = 0, allerror = 0;
587 	ufs2_daddr_t nb, nlbn, last;
588 	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
589 	ufs1_daddr_t *bap1 = NULL;
590 	ufs2_daddr_t *bap2 = NULL;
591 #	define BAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? bap1[i] : bap2[i])
592 
593 	/*
594 	 * Calculate index in current block of last
595 	 * block to be kept.  -1 indicates the entire
596 	 * block so we need not calculate the index.
597 	 */
598 	factor = lbn_offset(fs, level);
599 	last = lastbn;
600 	if (lastbn > 0)
601 		last /= factor;
602 	nblocks = btodb(fs->fs_bsize);
603 	/*
604 	 * Get buffer of block pointers, zero those entries corresponding
605 	 * to blocks to be free'd, and update on disk copy first.  Since
606 	 * double(triple) indirect before single(double) indirect, calls
607 	 * to bmap on these blocks will fail.  However, we already have
608 	 * the on disk address, so we have to set the b_blkno field
609 	 * explicitly instead of letting bread do everything for us.
610 	 */
611 	vp = ITOV(ip);
612 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
613 	if ((bp->b_flags & B_CACHE) == 0) {
614 		curthread->td_ru.ru_inblock++;	/* pay for read */
615 		bp->b_iocmd = BIO_READ;
616 		bp->b_flags &= ~B_INVAL;
617 		bp->b_ioflags &= ~BIO_ERROR;
618 		if (bp->b_bcount > bp->b_bufsize)
619 			panic("ffs_indirtrunc: bad buffer size");
620 		bp->b_blkno = dbn;
621 		vfs_busy_pages(bp, 0);
622 		bp->b_iooffset = dbtob(bp->b_blkno);
623 		bstrategy(bp);
624 		error = bufwait(bp);
625 	}
626 	if (error) {
627 		brelse(bp);
628 		*countp = 0;
629 		return (error);
630 	}
631 
632 	if (ip->i_ump->um_fstype == UFS1)
633 		bap1 = (ufs1_daddr_t *)bp->b_data;
634 	else
635 		bap2 = (ufs2_daddr_t *)bp->b_data;
636 	if (lastbn != -1) {
637 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
638 		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
639 		for (i = last + 1; i < NINDIR(fs); i++)
640 			if (ip->i_ump->um_fstype == UFS1)
641 				bap1[i] = 0;
642 			else
643 				bap2[i] = 0;
644 		if (DOINGASYNC(vp)) {
645 			bdwrite(bp);
646 		} else {
647 			error = bwrite(bp);
648 			if (error)
649 				allerror = error;
650 		}
651 		if (ip->i_ump->um_fstype == UFS1)
652 			bap1 = (ufs1_daddr_t *)copy;
653 		else
654 			bap2 = (ufs2_daddr_t *)copy;
655 	}
656 
657 	/*
658 	 * Recursively free totally unused blocks.
659 	 */
660 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
661 	    i--, nlbn += factor) {
662 		nb = BAP(ip, i);
663 		if (nb == 0)
664 			continue;
665 		if (level > SINGLE) {
666 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
667 			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
668 				allerror = error;
669 			blocksreleased += blkcount;
670 		}
671 		ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
672 		    ip->i_number, vp->v_type, NULL);
673 		blocksreleased += nblocks;
674 	}
675 
676 	/*
677 	 * Recursively free last partial block.
678 	 */
679 	if (level > SINGLE && lastbn >= 0) {
680 		last = lastbn % factor;
681 		nb = BAP(ip, i);
682 		if (nb != 0) {
683 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
684 			    last, level - 1, &blkcount);
685 			if (error)
686 				allerror = error;
687 			blocksreleased += blkcount;
688 		}
689 	}
690 	if (copy != NULL) {
691 		free(copy, M_TEMP);
692 	} else {
693 		bp->b_flags |= B_INVAL | B_NOCACHE;
694 		brelse(bp);
695 	}
696 
697 	*countp = blocksreleased;
698 	return (allerror);
699 }
700 
701 int
702 ffs_rdonly(struct inode *ip)
703 {
704 
705 	return (ip->i_ump->um_fs->fs_ronly != 0);
706 }
707 
708