xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
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/bio.h>
40 #include <sys/buf.h>
41 #include <sys/malloc.h>
42 #include <sys/mount.h>
43 #include <sys/proc.h>
44 #include <sys/random.h>
45 #include <sys/resourcevar.h>
46 #include <sys/rwlock.h>
47 #include <sys/stat.h>
48 #include <sys/vmmeter.h>
49 #include <sys/vnode.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_object.h>
54 
55 #include <ufs/ufs/extattr.h>
56 #include <ufs/ufs/quota.h>
57 #include <ufs/ufs/ufsmount.h>
58 #include <ufs/ufs/inode.h>
59 #include <ufs/ufs/ufs_extern.h>
60 
61 #include <ufs/ffs/fs.h>
62 #include <ufs/ffs/ffs_extern.h>
63 
64 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
65 	    ufs2_daddr_t, int, ufs2_daddr_t *);
66 
67 /*
68  * Update the access, modified, and inode change times as specified by the
69  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
70  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
71  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
72  * later if not now.  The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
73  * is currently being suspended (or is suspended) and vnode has been accessed.
74  * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
75  * reflect the presumably successful write, and if waitfor is set, then wait
76  * for the write to complete.
77  */
78 int
79 ffs_update(vp, waitfor)
80 	struct vnode *vp;
81 	int waitfor;
82 {
83 	struct fs *fs;
84 	struct buf *bp;
85 	struct inode *ip;
86 	int flags, error;
87 
88 	ASSERT_VOP_ELOCKED(vp, "ffs_update");
89 	ufs_itimes(vp);
90 	ip = VTOI(vp);
91 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
92 		return (0);
93 	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
94 	fs = ip->i_fs;
95 	if (fs->fs_ronly && ip->i_ump->um_fsckpid == 0)
96 		return (0);
97 	/*
98 	 * If we are updating a snapshot and another process is currently
99 	 * writing the buffer containing the inode for this snapshot then
100 	 * a deadlock can occur when it tries to check the snapshot to see
101 	 * if that block needs to be copied. Thus when updating a snapshot
102 	 * we check to see if the buffer is already locked, and if it is
103 	 * we drop the snapshot lock until the buffer has been written
104 	 * and is available to us. We have to grab a reference to the
105 	 * snapshot vnode to prevent it from being removed while we are
106 	 * waiting for the buffer.
107 	 */
108 	flags = 0;
109 	if (IS_SNAPSHOT(ip))
110 		flags = GB_LOCK_NOWAIT;
111 loop:
112 	error = breadn_flags(ip->i_devvp,
113 	     fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
114 	     (int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
115 	if (error != 0) {
116 		if (error != EBUSY)
117 			return (error);
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 		/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
149 		random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), 1, RANDOM_FS_ATIME);
150 	} else {
151 		*((struct ufs2_dinode *)bp->b_data +
152 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
153 		/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
154 		random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), 1, RANDOM_FS_ATIME);
155 	}
156 	if (waitfor && !DOINGASYNC(vp))
157 		error = bwrite(bp);
158 	else if (vm_page_count_severe() || buf_dirty_count_severe()) {
159 		bawrite(bp);
160 		error = 0;
161 	} else {
162 		if (bp->b_bufsize == fs->fs_bsize)
163 			bp->b_flags |= B_CLUSTEROK;
164 		bdwrite(bp);
165 		error = 0;
166 	}
167 	return (error);
168 }
169 
170 #define	SINGLE	0	/* index of single indirect block */
171 #define	DOUBLE	1	/* index of double indirect block */
172 #define	TRIPLE	2	/* index of triple indirect block */
173 /*
174  * Truncate the inode ip to at most length size, freeing the
175  * disk blocks.
176  */
177 int
178 ffs_truncate(vp, length, flags, cred)
179 	struct vnode *vp;
180 	off_t length;
181 	int flags;
182 	struct ucred *cred;
183 {
184 	struct inode *ip;
185 	ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
186 	ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
187 	ufs2_daddr_t count, blocksreleased = 0, datablocks;
188 	struct bufobj *bo;
189 	struct fs *fs;
190 	struct buf *bp;
191 	struct ufsmount *ump;
192 	int softdeptrunc, journaltrunc;
193 	int needextclean, extblocks;
194 	int offset, size, level, nblocks;
195 	int i, error, allerror;
196 	off_t osize;
197 
198 	ip = VTOI(vp);
199 	fs = ip->i_fs;
200 	ump = ip->i_ump;
201 	bo = &vp->v_bufobj;
202 
203 	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
204 
205 	if (length < 0)
206 		return (EINVAL);
207 	if (length > fs->fs_maxfilesize)
208 		return (EFBIG);
209 #ifdef QUOTA
210 	error = getinoquota(ip);
211 	if (error)
212 		return (error);
213 #endif
214 	/*
215 	 * Historically clients did not have to specify which data
216 	 * they were truncating. So, if not specified, we assume
217 	 * traditional behavior, e.g., just the normal data.
218 	 */
219 	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
220 		flags |= IO_NORMAL;
221 	if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
222 		flags |= IO_SYNC;
223 	/*
224 	 * If we are truncating the extended-attributes, and cannot
225 	 * do it with soft updates, then do it slowly here. If we are
226 	 * truncating both the extended attributes and the file contents
227 	 * (e.g., the file is being unlinked), then pick it off with
228 	 * soft updates below.
229 	 */
230 	allerror = 0;
231 	needextclean = 0;
232 	softdeptrunc = 0;
233 	journaltrunc = DOINGSUJ(vp);
234 	if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
235 		softdeptrunc = !softdep_slowdown(vp);
236 	extblocks = 0;
237 	datablocks = DIP(ip, i_blocks);
238 	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
239 		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
240 		datablocks -= extblocks;
241 	}
242 	if ((flags & IO_EXT) && extblocks > 0) {
243 		if (length != 0)
244 			panic("ffs_truncate: partial trunc of extdata");
245 		if (softdeptrunc || journaltrunc) {
246 			if ((flags & IO_NORMAL) == 0)
247 				goto extclean;
248 			needextclean = 1;
249 		} else {
250 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
251 				return (error);
252 #ifdef QUOTA
253 			(void) chkdq(ip, -extblocks, NOCRED, 0);
254 #endif
255 			vinvalbuf(vp, V_ALT, 0, 0);
256 			vn_pages_remove(vp,
257 			    OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
258 			osize = ip->i_din2->di_extsize;
259 			ip->i_din2->di_blocks -= extblocks;
260 			ip->i_din2->di_extsize = 0;
261 			for (i = 0; i < NXADDR; i++) {
262 				oldblks[i] = ip->i_din2->di_extb[i];
263 				ip->i_din2->di_extb[i] = 0;
264 			}
265 			ip->i_flag |= IN_CHANGE;
266 			if ((error = ffs_update(vp, !DOINGASYNC(vp))))
267 				return (error);
268 			for (i = 0; i < NXADDR; i++) {
269 				if (oldblks[i] == 0)
270 					continue;
271 				ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
272 				    sblksize(fs, osize, i), ip->i_number,
273 				    vp->v_type, NULL);
274 			}
275 		}
276 	}
277 	if ((flags & IO_NORMAL) == 0)
278 		return (0);
279 	if (vp->v_type == VLNK &&
280 	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
281 	     datablocks == 0)) {
282 #ifdef INVARIANTS
283 		if (length != 0)
284 			panic("ffs_truncate: partial truncate of symlink");
285 #endif
286 		bzero(SHORTLINK(ip), (u_int)ip->i_size);
287 		ip->i_size = 0;
288 		DIP_SET(ip, i_size, 0);
289 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
290 		if (needextclean)
291 			goto extclean;
292 		return (ffs_update(vp, !DOINGASYNC(vp)));
293 	}
294 	if (ip->i_size == length) {
295 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
296 		if (needextclean)
297 			goto extclean;
298 		return (ffs_update(vp, 0));
299 	}
300 	if (fs->fs_ronly)
301 		panic("ffs_truncate: read-only filesystem");
302 	if (IS_SNAPSHOT(ip))
303 		ffs_snapremove(vp);
304 	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
305 	osize = ip->i_size;
306 	/*
307 	 * Lengthen the size of the file. We must ensure that the
308 	 * last byte of the file is allocated. Since the smallest
309 	 * value of osize is 0, length will be at least 1.
310 	 */
311 	if (osize < length) {
312 		vnode_pager_setsize(vp, length);
313 		flags |= BA_CLRBUF;
314 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
315 		if (error) {
316 			vnode_pager_setsize(vp, osize);
317 			return (error);
318 		}
319 		ip->i_size = length;
320 		DIP_SET(ip, i_size, length);
321 		if (bp->b_bufsize == fs->fs_bsize)
322 			bp->b_flags |= B_CLUSTEROK;
323 		if (flags & IO_SYNC)
324 			bwrite(bp);
325 		else if (DOINGASYNC(vp))
326 			bdwrite(bp);
327 		else
328 			bawrite(bp);
329 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
330 		return (ffs_update(vp, !DOINGASYNC(vp)));
331 	}
332 	if (DOINGSOFTDEP(vp)) {
333 		if (softdeptrunc == 0 && journaltrunc == 0) {
334 			/*
335 			 * If a file is only partially truncated, then
336 			 * we have to clean up the data structures
337 			 * describing the allocation past the truncation
338 			 * point. Finding and deallocating those structures
339 			 * is a lot of work. Since partial truncation occurs
340 			 * rarely, we solve the problem by syncing the file
341 			 * so that it will have no data structures left.
342 			 */
343 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
344 				return (error);
345 		} else {
346 			flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
347 			if (journaltrunc)
348 				softdep_journal_freeblocks(ip, cred, length,
349 				    flags);
350 			else
351 				softdep_setup_freeblocks(ip, length, flags);
352 			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
353 			if (journaltrunc == 0) {
354 				ip->i_flag |= IN_CHANGE | IN_UPDATE;
355 				error = ffs_update(vp, 0);
356 			}
357 			return (error);
358 		}
359 	}
360 	/*
361 	 * Shorten the size of the file. If the file is not being
362 	 * truncated to a block boundary, the contents of the
363 	 * partial block following the end of the file must be
364 	 * zero'ed in case it ever becomes accessible again because
365 	 * of subsequent file growth. Directories however are not
366 	 * zero'ed as they should grow back initialized to empty.
367 	 */
368 	offset = blkoff(fs, length);
369 	if (offset == 0) {
370 		ip->i_size = length;
371 		DIP_SET(ip, i_size, length);
372 	} else {
373 		lbn = lblkno(fs, length);
374 		flags |= BA_CLRBUF;
375 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
376 		if (error)
377 			return (error);
378 		/*
379 		 * When we are doing soft updates and the UFS_BALLOC
380 		 * above fills in a direct block hole with a full sized
381 		 * block that will be truncated down to a fragment below,
382 		 * we must flush out the block dependency with an FSYNC
383 		 * so that we do not get a soft updates inconsistency
384 		 * when we create the fragment below.
385 		 */
386 		if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
387 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
388 		    (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
389 			return (error);
390 		ip->i_size = length;
391 		DIP_SET(ip, i_size, length);
392 		size = blksize(fs, ip, lbn);
393 		if (vp->v_type != VDIR)
394 			bzero((char *)bp->b_data + offset,
395 			    (u_int)(size - offset));
396 		/* Kirk's code has reallocbuf(bp, size, 1) here */
397 		allocbuf(bp, size);
398 		if (bp->b_bufsize == fs->fs_bsize)
399 			bp->b_flags |= B_CLUSTEROK;
400 		if (flags & IO_SYNC)
401 			bwrite(bp);
402 		else if (DOINGASYNC(vp))
403 			bdwrite(bp);
404 		else
405 			bawrite(bp);
406 	}
407 	/*
408 	 * Calculate index into inode's block list of
409 	 * last direct and indirect blocks (if any)
410 	 * which we want to keep.  Lastblock is -1 when
411 	 * the file is truncated to 0.
412 	 */
413 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
414 	lastiblock[SINGLE] = lastblock - NDADDR;
415 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
416 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
417 	nblocks = btodb(fs->fs_bsize);
418 	/*
419 	 * Update file and block pointers on disk before we start freeing
420 	 * blocks.  If we crash before free'ing blocks below, the blocks
421 	 * will be returned to the free list.  lastiblock values are also
422 	 * normalized to -1 for calls to ffs_indirtrunc below.
423 	 */
424 	for (level = TRIPLE; level >= SINGLE; level--) {
425 		oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
426 		if (lastiblock[level] < 0) {
427 			DIP_SET(ip, i_ib[level], 0);
428 			lastiblock[level] = -1;
429 		}
430 	}
431 	for (i = 0; i < NDADDR; i++) {
432 		oldblks[i] = DIP(ip, i_db[i]);
433 		if (i > lastblock)
434 			DIP_SET(ip, i_db[i], 0);
435 	}
436 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
437 	allerror = ffs_update(vp, !DOINGASYNC(vp));
438 
439 	/*
440 	 * Having written the new inode to disk, save its new configuration
441 	 * and put back the old block pointers long enough to process them.
442 	 * Note that we save the new block configuration so we can check it
443 	 * when we are done.
444 	 */
445 	for (i = 0; i < NDADDR; i++) {
446 		newblks[i] = DIP(ip, i_db[i]);
447 		DIP_SET(ip, i_db[i], oldblks[i]);
448 	}
449 	for (i = 0; i < NIADDR; i++) {
450 		newblks[NDADDR + i] = DIP(ip, i_ib[i]);
451 		DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
452 	}
453 	ip->i_size = osize;
454 	DIP_SET(ip, i_size, osize);
455 
456 	error = vtruncbuf(vp, cred, length, fs->fs_bsize);
457 	if (error && (allerror == 0))
458 		allerror = error;
459 
460 	/*
461 	 * Indirect blocks first.
462 	 */
463 	indir_lbn[SINGLE] = -NDADDR;
464 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
465 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
466 	for (level = TRIPLE; level >= SINGLE; level--) {
467 		bn = DIP(ip, i_ib[level]);
468 		if (bn != 0) {
469 			error = ffs_indirtrunc(ip, indir_lbn[level],
470 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
471 			if (error)
472 				allerror = error;
473 			blocksreleased += count;
474 			if (lastiblock[level] < 0) {
475 				DIP_SET(ip, i_ib[level], 0);
476 				ffs_blkfree(ump, fs, ip->i_devvp, bn,
477 				    fs->fs_bsize, ip->i_number,
478 				    vp->v_type, NULL);
479 				blocksreleased += nblocks;
480 			}
481 		}
482 		if (lastiblock[level] >= 0)
483 			goto done;
484 	}
485 
486 	/*
487 	 * All whole direct blocks or frags.
488 	 */
489 	for (i = NDADDR - 1; i > lastblock; i--) {
490 		long bsize;
491 
492 		bn = DIP(ip, i_db[i]);
493 		if (bn == 0)
494 			continue;
495 		DIP_SET(ip, i_db[i], 0);
496 		bsize = blksize(fs, ip, i);
497 		ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number,
498 		    vp->v_type, NULL);
499 		blocksreleased += btodb(bsize);
500 	}
501 	if (lastblock < 0)
502 		goto done;
503 
504 	/*
505 	 * Finally, look for a change in size of the
506 	 * last direct block; release any frags.
507 	 */
508 	bn = DIP(ip, i_db[lastblock]);
509 	if (bn != 0) {
510 		long oldspace, newspace;
511 
512 		/*
513 		 * Calculate amount of space we're giving
514 		 * back as old block size minus new block size.
515 		 */
516 		oldspace = blksize(fs, ip, lastblock);
517 		ip->i_size = length;
518 		DIP_SET(ip, i_size, length);
519 		newspace = blksize(fs, ip, lastblock);
520 		if (newspace == 0)
521 			panic("ffs_truncate: newspace");
522 		if (oldspace - newspace > 0) {
523 			/*
524 			 * Block number of space to be free'd is
525 			 * the old block # plus the number of frags
526 			 * required for the storage we're keeping.
527 			 */
528 			bn += numfrags(fs, newspace);
529 			ffs_blkfree(ump, fs, ip->i_devvp, bn,
530 			   oldspace - newspace, ip->i_number, vp->v_type, NULL);
531 			blocksreleased += btodb(oldspace - newspace);
532 		}
533 	}
534 done:
535 #ifdef INVARIANTS
536 	for (level = SINGLE; level <= TRIPLE; level++)
537 		if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
538 			panic("ffs_truncate1");
539 	for (i = 0; i < NDADDR; i++)
540 		if (newblks[i] != DIP(ip, i_db[i]))
541 			panic("ffs_truncate2");
542 	BO_LOCK(bo);
543 	if (length == 0 &&
544 	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
545 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
546 		panic("ffs_truncate3");
547 	BO_UNLOCK(bo);
548 #endif /* INVARIANTS */
549 	/*
550 	 * Put back the real size.
551 	 */
552 	ip->i_size = length;
553 	DIP_SET(ip, i_size, length);
554 	if (DIP(ip, i_blocks) >= blocksreleased)
555 		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
556 	else	/* sanity */
557 		DIP_SET(ip, i_blocks, 0);
558 	ip->i_flag |= IN_CHANGE;
559 #ifdef QUOTA
560 	(void) chkdq(ip, -blocksreleased, NOCRED, 0);
561 #endif
562 	return (allerror);
563 
564 extclean:
565 	if (journaltrunc)
566 		softdep_journal_freeblocks(ip, cred, length, IO_EXT);
567 	else
568 		softdep_setup_freeblocks(ip, length, IO_EXT);
569 	return (ffs_update(vp, !DOINGASYNC(vp)));
570 }
571 
572 /*
573  * Release blocks associated with the inode ip and stored in the indirect
574  * block bn.  Blocks are free'd in LIFO order up to (but not including)
575  * lastbn.  If level is greater than SINGLE, the block is an indirect block
576  * and recursive calls to indirtrunc must be used to cleanse other indirect
577  * blocks.
578  */
579 static int
580 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
581 	struct inode *ip;
582 	ufs2_daddr_t lbn, lastbn;
583 	ufs2_daddr_t dbn;
584 	int level;
585 	ufs2_daddr_t *countp;
586 {
587 	struct buf *bp;
588 	struct fs *fs = ip->i_fs;
589 	struct vnode *vp;
590 	caddr_t copy = NULL;
591 	int i, nblocks, error = 0, allerror = 0;
592 	ufs2_daddr_t nb, nlbn, last;
593 	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
594 	ufs1_daddr_t *bap1 = NULL;
595 	ufs2_daddr_t *bap2 = NULL;
596 #	define BAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? bap1[i] : bap2[i])
597 
598 	/*
599 	 * Calculate index in current block of last
600 	 * block to be kept.  -1 indicates the entire
601 	 * block so we need not calculate the index.
602 	 */
603 	factor = lbn_offset(fs, level);
604 	last = lastbn;
605 	if (lastbn > 0)
606 		last /= factor;
607 	nblocks = btodb(fs->fs_bsize);
608 	/*
609 	 * Get buffer of block pointers, zero those entries corresponding
610 	 * to blocks to be free'd, and update on disk copy first.  Since
611 	 * double(triple) indirect before single(double) indirect, calls
612 	 * to bmap on these blocks will fail.  However, we already have
613 	 * the on disk address, so we have to set the b_blkno field
614 	 * explicitly instead of letting bread do everything for us.
615 	 */
616 	vp = ITOV(ip);
617 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
618 	if ((bp->b_flags & B_CACHE) == 0) {
619 		curthread->td_ru.ru_inblock++;	/* pay for read */
620 		bp->b_iocmd = BIO_READ;
621 		bp->b_flags &= ~B_INVAL;
622 		bp->b_ioflags &= ~BIO_ERROR;
623 		if (bp->b_bcount > bp->b_bufsize)
624 			panic("ffs_indirtrunc: bad buffer size");
625 		bp->b_blkno = dbn;
626 		vfs_busy_pages(bp, 0);
627 		bp->b_iooffset = dbtob(bp->b_blkno);
628 		bstrategy(bp);
629 		error = bufwait(bp);
630 	}
631 	if (error) {
632 		brelse(bp);
633 		*countp = 0;
634 		return (error);
635 	}
636 
637 	if (ip->i_ump->um_fstype == UFS1)
638 		bap1 = (ufs1_daddr_t *)bp->b_data;
639 	else
640 		bap2 = (ufs2_daddr_t *)bp->b_data;
641 	if (lastbn != -1) {
642 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
643 		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
644 		for (i = last + 1; i < NINDIR(fs); i++)
645 			if (ip->i_ump->um_fstype == UFS1)
646 				bap1[i] = 0;
647 			else
648 				bap2[i] = 0;
649 		if (DOINGASYNC(vp)) {
650 			bdwrite(bp);
651 		} else {
652 			error = bwrite(bp);
653 			if (error)
654 				allerror = error;
655 		}
656 		if (ip->i_ump->um_fstype == UFS1)
657 			bap1 = (ufs1_daddr_t *)copy;
658 		else
659 			bap2 = (ufs2_daddr_t *)copy;
660 	}
661 
662 	/*
663 	 * Recursively free totally unused blocks.
664 	 */
665 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
666 	    i--, nlbn += factor) {
667 		nb = BAP(ip, i);
668 		if (nb == 0)
669 			continue;
670 		if (level > SINGLE) {
671 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
672 			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
673 				allerror = error;
674 			blocksreleased += blkcount;
675 		}
676 		ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
677 		    ip->i_number, vp->v_type, NULL);
678 		blocksreleased += nblocks;
679 	}
680 
681 	/*
682 	 * Recursively free last partial block.
683 	 */
684 	if (level > SINGLE && lastbn >= 0) {
685 		last = lastbn % factor;
686 		nb = BAP(ip, i);
687 		if (nb != 0) {
688 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
689 			    last, level - 1, &blkcount);
690 			if (error)
691 				allerror = error;
692 			blocksreleased += blkcount;
693 		}
694 	}
695 	if (copy != NULL) {
696 		free(copy, M_TEMP);
697 	} else {
698 		bp->b_flags |= B_INVAL | B_NOCACHE;
699 		brelse(bp);
700 	}
701 
702 	*countp = blocksreleased;
703 	return (allerror);
704 }
705 
706 int
707 ffs_rdonly(struct inode *ip)
708 {
709 
710 	return (ip->i_ump->um_fs->fs_ronly != 0);
711 }
712 
713