xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 190cef3d52236565eb22e18b33e9e865ec634aa3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_quota.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/bio.h>
42 #include <sys/buf.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/proc.h>
46 #include <sys/racct.h>
47 #include <sys/random.h>
48 #include <sys/resourcevar.h>
49 #include <sys/rwlock.h>
50 #include <sys/stat.h>
51 #include <sys/vmmeter.h>
52 #include <sys/vnode.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_extern.h>
56 #include <vm/vm_object.h>
57 
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/ufsmount.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufs_extern.h>
63 
64 #include <ufs/ffs/fs.h>
65 #include <ufs/ffs/ffs_extern.h>
66 
67 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
68 	    ufs2_daddr_t, int, ufs2_daddr_t *);
69 
70 /*
71  * Update the access, modified, and inode change times as specified by the
72  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
73  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
74  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
75  * later if not now.  The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
76  * is currently being suspended (or is suspended) and vnode has been accessed.
77  * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
78  * reflect the presumably successful write, and if waitfor is set, then wait
79  * for the write to complete.
80  */
81 int
82 ffs_update(vp, waitfor)
83 	struct vnode *vp;
84 	int waitfor;
85 {
86 	struct fs *fs;
87 	struct buf *bp;
88 	struct inode *ip;
89 	int flags, error;
90 
91 	ASSERT_VOP_ELOCKED(vp, "ffs_update");
92 	ufs_itimes(vp);
93 	ip = VTOI(vp);
94 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
95 		return (0);
96 	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
97 	fs = ITOFS(ip);
98 	if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0)
99 		return (0);
100 	/*
101 	 * If we are updating a snapshot and another process is currently
102 	 * writing the buffer containing the inode for this snapshot then
103 	 * a deadlock can occur when it tries to check the snapshot to see
104 	 * if that block needs to be copied. Thus when updating a snapshot
105 	 * we check to see if the buffer is already locked, and if it is
106 	 * we drop the snapshot lock until the buffer has been written
107 	 * and is available to us. We have to grab a reference to the
108 	 * snapshot vnode to prevent it from being removed while we are
109 	 * waiting for the buffer.
110 	 */
111 	flags = 0;
112 	if (IS_SNAPSHOT(ip))
113 		flags = GB_LOCK_NOWAIT;
114 loop:
115 	error = bread_gb(ITODEVVP(ip),
116 	     fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
117 	     (int) fs->fs_bsize, NOCRED, flags, &bp);
118 	if (error != 0) {
119 		if (error != EBUSY)
120 			return (error);
121 		KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
122 		/*
123 		 * Wait for our inode block to become available.
124 		 *
125 		 * Hold a reference to the vnode to protect against
126 		 * ffs_snapgone(). Since we hold a reference, it can only
127 		 * get reclaimed (VI_DOOMED flag) in a forcible downgrade
128 		 * or unmount. For an unmount, the entire filesystem will be
129 		 * gone, so we cannot attempt to touch anything associated
130 		 * with it while the vnode is unlocked; all we can do is
131 		 * pause briefly and try again. If when we relock the vnode
132 		 * we discover that it has been reclaimed, updating it is no
133 		 * longer necessary and we can just return an error.
134 		 */
135 		vref(vp);
136 		VOP_UNLOCK(vp, 0);
137 		pause("ffsupd", 1);
138 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
139 		vrele(vp);
140 		if ((vp->v_iflag & VI_DOOMED) != 0)
141 			return (ENOENT);
142 		goto loop;
143 	}
144 	if (DOINGSOFTDEP(vp))
145 		softdep_update_inodeblock(ip, bp, waitfor);
146 	else if (ip->i_effnlink != ip->i_nlink)
147 		panic("ffs_update: bad link cnt");
148 	if (I_IS_UFS1(ip)) {
149 		*((struct ufs1_dinode *)bp->b_data +
150 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
151 		/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
152 		random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), RANDOM_FS_ATIME);
153 	} else {
154 		*((struct ufs2_dinode *)bp->b_data +
155 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
156 		/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
157 		random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), RANDOM_FS_ATIME);
158 	}
159 	if (waitfor)
160 		error = bwrite(bp);
161 	else if (vm_page_count_severe() || buf_dirty_count_severe()) {
162 		bawrite(bp);
163 		error = 0;
164 	} else {
165 		if (bp->b_bufsize == fs->fs_bsize)
166 			bp->b_flags |= B_CLUSTEROK;
167 		bdwrite(bp);
168 		error = 0;
169 	}
170 	return (error);
171 }
172 
173 #define	SINGLE	0	/* index of single indirect block */
174 #define	DOUBLE	1	/* index of double indirect block */
175 #define	TRIPLE	2	/* index of triple indirect block */
176 /*
177  * Truncate the inode ip to at most length size, freeing the
178  * disk blocks.
179  */
180 int
181 ffs_truncate(vp, length, flags, cred)
182 	struct vnode *vp;
183 	off_t length;
184 	int flags;
185 	struct ucred *cred;
186 {
187 	struct inode *ip;
188 	ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR];
189 	ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR];
190 	ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR];
191 	ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno;
192 	struct bufobj *bo;
193 	struct fs *fs;
194 	struct buf *bp;
195 	struct ufsmount *ump;
196 	int softdeptrunc, journaltrunc;
197 	int needextclean, extblocks;
198 	int offset, size, level, nblocks;
199 	int i, error, allerror, indiroff, waitforupdate;
200 	u_long key;
201 	off_t osize;
202 
203 	ip = VTOI(vp);
204 	ump = VFSTOUFS(vp->v_mount);
205 	fs = ump->um_fs;
206 	bo = &vp->v_bufobj;
207 
208 	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
209 
210 	if (length < 0)
211 		return (EINVAL);
212 	if (length > fs->fs_maxfilesize)
213 		return (EFBIG);
214 #ifdef QUOTA
215 	error = getinoquota(ip);
216 	if (error)
217 		return (error);
218 #endif
219 	/*
220 	 * Historically clients did not have to specify which data
221 	 * they were truncating. So, if not specified, we assume
222 	 * traditional behavior, e.g., just the normal data.
223 	 */
224 	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
225 		flags |= IO_NORMAL;
226 	if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
227 		flags |= IO_SYNC;
228 	waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp);
229 	/*
230 	 * If we are truncating the extended-attributes, and cannot
231 	 * do it with soft updates, then do it slowly here. If we are
232 	 * truncating both the extended attributes and the file contents
233 	 * (e.g., the file is being unlinked), then pick it off with
234 	 * soft updates below.
235 	 */
236 	allerror = 0;
237 	needextclean = 0;
238 	softdeptrunc = 0;
239 	journaltrunc = DOINGSUJ(vp);
240 	if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
241 		softdeptrunc = !softdep_slowdown(vp);
242 	extblocks = 0;
243 	datablocks = DIP(ip, i_blocks);
244 	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
245 		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
246 		datablocks -= extblocks;
247 	}
248 	if ((flags & IO_EXT) && extblocks > 0) {
249 		if (length != 0)
250 			panic("ffs_truncate: partial trunc of extdata");
251 		if (softdeptrunc || journaltrunc) {
252 			if ((flags & IO_NORMAL) == 0)
253 				goto extclean;
254 			needextclean = 1;
255 		} else {
256 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
257 				return (error);
258 #ifdef QUOTA
259 			(void) chkdq(ip, -extblocks, NOCRED, 0);
260 #endif
261 			vinvalbuf(vp, V_ALT, 0, 0);
262 			vn_pages_remove(vp,
263 			    OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
264 			osize = ip->i_din2->di_extsize;
265 			ip->i_din2->di_blocks -= extblocks;
266 			ip->i_din2->di_extsize = 0;
267 			for (i = 0; i < UFS_NXADDR; i++) {
268 				oldblks[i] = ip->i_din2->di_extb[i];
269 				ip->i_din2->di_extb[i] = 0;
270 			}
271 			ip->i_flag |= IN_CHANGE;
272 			if ((error = ffs_update(vp, waitforupdate)))
273 				return (error);
274 			for (i = 0; i < UFS_NXADDR; i++) {
275 				if (oldblks[i] == 0)
276 					continue;
277 				ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i],
278 				    sblksize(fs, osize, i), ip->i_number,
279 				    vp->v_type, NULL, SINGLETON_KEY);
280 			}
281 		}
282 	}
283 	if ((flags & IO_NORMAL) == 0)
284 		return (0);
285 	if (vp->v_type == VLNK &&
286 	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
287 	     datablocks == 0)) {
288 #ifdef INVARIANTS
289 		if (length != 0)
290 			panic("ffs_truncate: partial truncate of symlink");
291 #endif
292 		bzero(SHORTLINK(ip), (u_int)ip->i_size);
293 		ip->i_size = 0;
294 		DIP_SET(ip, i_size, 0);
295 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
296 		if (needextclean)
297 			goto extclean;
298 		return (ffs_update(vp, waitforupdate));
299 	}
300 	if (ip->i_size == length) {
301 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
302 		if (needextclean)
303 			goto extclean;
304 		return (ffs_update(vp, 0));
305 	}
306 	if (fs->fs_ronly)
307 		panic("ffs_truncate: read-only filesystem");
308 	if (IS_SNAPSHOT(ip))
309 		ffs_snapremove(vp);
310 	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
311 	osize = ip->i_size;
312 	/*
313 	 * Lengthen the size of the file. We must ensure that the
314 	 * last byte of the file is allocated. Since the smallest
315 	 * value of osize is 0, length will be at least 1.
316 	 */
317 	if (osize < length) {
318 		vnode_pager_setsize(vp, length);
319 		flags |= BA_CLRBUF;
320 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
321 		if (error) {
322 			vnode_pager_setsize(vp, osize);
323 			return (error);
324 		}
325 		ip->i_size = length;
326 		DIP_SET(ip, i_size, length);
327 		if (bp->b_bufsize == fs->fs_bsize)
328 			bp->b_flags |= B_CLUSTEROK;
329 		if (flags & IO_SYNC)
330 			bwrite(bp);
331 		else if (DOINGASYNC(vp))
332 			bdwrite(bp);
333 		else
334 			bawrite(bp);
335 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
336 		return (ffs_update(vp, waitforupdate));
337 	}
338 	/*
339 	 * Lookup block number for a given offset. Zero length files
340 	 * have no blocks, so return a blkno of -1.
341 	 */
342 	lbn = lblkno(fs, length - 1);
343 	if (length == 0) {
344 		blkno = -1;
345 	} else if (lbn < UFS_NDADDR) {
346 		blkno = DIP(ip, i_db[lbn]);
347 	} else {
348 		error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
349 		    cred, BA_METAONLY, &bp);
350 		if (error)
351 			return (error);
352 		indiroff = (lbn - UFS_NDADDR) % NINDIR(fs);
353 		if (I_IS_UFS1(ip))
354 			blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
355 		else
356 			blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
357 		/*
358 		 * If the block number is non-zero, then the indirect block
359 		 * must have been previously allocated and need not be written.
360 		 * If the block number is zero, then we may have allocated
361 		 * the indirect block and hence need to write it out.
362 		 */
363 		if (blkno != 0)
364 			brelse(bp);
365 		else if (flags & IO_SYNC)
366 			bwrite(bp);
367 		else
368 			bdwrite(bp);
369 	}
370 	/*
371 	 * If the block number at the new end of the file is zero,
372 	 * then we must allocate it to ensure that the last block of
373 	 * the file is allocated. Soft updates does not handle this
374 	 * case, so here we have to clean up the soft updates data
375 	 * structures describing the allocation past the truncation
376 	 * point. Finding and deallocating those structures is a lot of
377 	 * work. Since partial truncation with a hole at the end occurs
378 	 * rarely, we solve the problem by syncing the file so that it
379 	 * will have no soft updates data structures left.
380 	 */
381 	if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
382 		return (error);
383 	if (blkno != 0 && DOINGSOFTDEP(vp)) {
384 		if (softdeptrunc == 0 && journaltrunc == 0) {
385 			/*
386 			 * If soft updates cannot handle this truncation,
387 			 * clean up soft dependency data structures and
388 			 * fall through to the synchronous truncation.
389 			 */
390 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
391 				return (error);
392 		} else {
393 			flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
394 			if (journaltrunc)
395 				softdep_journal_freeblocks(ip, cred, length,
396 				    flags);
397 			else
398 				softdep_setup_freeblocks(ip, length, flags);
399 			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
400 			if (journaltrunc == 0) {
401 				ip->i_flag |= IN_CHANGE | IN_UPDATE;
402 				error = ffs_update(vp, 0);
403 			}
404 			return (error);
405 		}
406 	}
407 	/*
408 	 * Shorten the size of the file. If the last block of the
409 	 * shortened file is unallocated, we must allocate it.
410 	 * Additionally, if the file is not being truncated to a
411 	 * block boundary, the contents of the partial block
412 	 * following the end of the file must be zero'ed in
413 	 * case it ever becomes accessible again because of
414 	 * subsequent file growth. Directories however are not
415 	 * zero'ed as they should grow back initialized to empty.
416 	 */
417 	offset = blkoff(fs, length);
418 	if (blkno != 0 && offset == 0) {
419 		ip->i_size = length;
420 		DIP_SET(ip, i_size, length);
421 	} else {
422 		lbn = lblkno(fs, length);
423 		flags |= BA_CLRBUF;
424 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
425 		if (error)
426 			return (error);
427 		/*
428 		 * When we are doing soft updates and the UFS_BALLOC
429 		 * above fills in a direct block hole with a full sized
430 		 * block that will be truncated down to a fragment below,
431 		 * we must flush out the block dependency with an FSYNC
432 		 * so that we do not get a soft updates inconsistency
433 		 * when we create the fragment below.
434 		 */
435 		if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR &&
436 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
437 		    (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
438 			return (error);
439 		ip->i_size = length;
440 		DIP_SET(ip, i_size, length);
441 		size = blksize(fs, ip, lbn);
442 		if (vp->v_type != VDIR && offset != 0)
443 			bzero((char *)bp->b_data + offset,
444 			    (u_int)(size - offset));
445 		/* Kirk's code has reallocbuf(bp, size, 1) here */
446 		allocbuf(bp, size);
447 		if (bp->b_bufsize == fs->fs_bsize)
448 			bp->b_flags |= B_CLUSTEROK;
449 		if (flags & IO_SYNC)
450 			bwrite(bp);
451 		else if (DOINGASYNC(vp))
452 			bdwrite(bp);
453 		else
454 			bawrite(bp);
455 	}
456 	/*
457 	 * Calculate index into inode's block list of
458 	 * last direct and indirect blocks (if any)
459 	 * which we want to keep.  Lastblock is -1 when
460 	 * the file is truncated to 0.
461 	 */
462 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
463 	lastiblock[SINGLE] = lastblock - UFS_NDADDR;
464 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
465 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
466 	nblocks = btodb(fs->fs_bsize);
467 	/*
468 	 * Update file and block pointers on disk before we start freeing
469 	 * blocks.  If we crash before free'ing blocks below, the blocks
470 	 * will be returned to the free list.  lastiblock values are also
471 	 * normalized to -1 for calls to ffs_indirtrunc below.
472 	 */
473 	for (level = TRIPLE; level >= SINGLE; level--) {
474 		oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]);
475 		if (lastiblock[level] < 0) {
476 			DIP_SET(ip, i_ib[level], 0);
477 			lastiblock[level] = -1;
478 		}
479 	}
480 	for (i = 0; i < UFS_NDADDR; i++) {
481 		oldblks[i] = DIP(ip, i_db[i]);
482 		if (i > lastblock)
483 			DIP_SET(ip, i_db[i], 0);
484 	}
485 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
486 	allerror = ffs_update(vp, waitforupdate);
487 
488 	/*
489 	 * Having written the new inode to disk, save its new configuration
490 	 * and put back the old block pointers long enough to process them.
491 	 * Note that we save the new block configuration so we can check it
492 	 * when we are done.
493 	 */
494 	for (i = 0; i < UFS_NDADDR; i++) {
495 		newblks[i] = DIP(ip, i_db[i]);
496 		DIP_SET(ip, i_db[i], oldblks[i]);
497 	}
498 	for (i = 0; i < UFS_NIADDR; i++) {
499 		newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]);
500 		DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]);
501 	}
502 	ip->i_size = osize;
503 	DIP_SET(ip, i_size, osize);
504 
505 	error = vtruncbuf(vp, cred, length, fs->fs_bsize);
506 	if (error && (allerror == 0))
507 		allerror = error;
508 
509 	/*
510 	 * Indirect blocks first.
511 	 */
512 	indir_lbn[SINGLE] = -UFS_NDADDR;
513 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
514 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
515 	for (level = TRIPLE; level >= SINGLE; level--) {
516 		bn = DIP(ip, i_ib[level]);
517 		if (bn != 0) {
518 			error = ffs_indirtrunc(ip, indir_lbn[level],
519 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
520 			if (error)
521 				allerror = error;
522 			blocksreleased += count;
523 			if (lastiblock[level] < 0) {
524 				DIP_SET(ip, i_ib[level], 0);
525 				ffs_blkfree(ump, fs, ump->um_devvp, bn,
526 				    fs->fs_bsize, ip->i_number,
527 				    vp->v_type, NULL, SINGLETON_KEY);
528 				blocksreleased += nblocks;
529 			}
530 		}
531 		if (lastiblock[level] >= 0)
532 			goto done;
533 	}
534 
535 	/*
536 	 * All whole direct blocks or frags.
537 	 */
538 	key = ffs_blkrelease_start(ump, ump->um_devvp, ip->i_number);
539 	for (i = UFS_NDADDR - 1; i > lastblock; i--) {
540 		long bsize;
541 
542 		bn = DIP(ip, i_db[i]);
543 		if (bn == 0)
544 			continue;
545 		DIP_SET(ip, i_db[i], 0);
546 		bsize = blksize(fs, ip, i);
547 		ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number,
548 		    vp->v_type, NULL, key);
549 		blocksreleased += btodb(bsize);
550 	}
551 	ffs_blkrelease_finish(ump, key);
552 	if (lastblock < 0)
553 		goto done;
554 
555 	/*
556 	 * Finally, look for a change in size of the
557 	 * last direct block; release any frags.
558 	 */
559 	bn = DIP(ip, i_db[lastblock]);
560 	if (bn != 0) {
561 		long oldspace, newspace;
562 
563 		/*
564 		 * Calculate amount of space we're giving
565 		 * back as old block size minus new block size.
566 		 */
567 		oldspace = blksize(fs, ip, lastblock);
568 		ip->i_size = length;
569 		DIP_SET(ip, i_size, length);
570 		newspace = blksize(fs, ip, lastblock);
571 		if (newspace == 0)
572 			panic("ffs_truncate: newspace");
573 		if (oldspace - newspace > 0) {
574 			/*
575 			 * Block number of space to be free'd is
576 			 * the old block # plus the number of frags
577 			 * required for the storage we're keeping.
578 			 */
579 			bn += numfrags(fs, newspace);
580 			ffs_blkfree(ump, fs, ump->um_devvp, bn,
581 			   oldspace - newspace, ip->i_number, vp->v_type,
582 			   NULL, SINGLETON_KEY);
583 			blocksreleased += btodb(oldspace - newspace);
584 		}
585 	}
586 done:
587 #ifdef INVARIANTS
588 	for (level = SINGLE; level <= TRIPLE; level++)
589 		if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level]))
590 			panic("ffs_truncate1");
591 	for (i = 0; i < UFS_NDADDR; i++)
592 		if (newblks[i] != DIP(ip, i_db[i]))
593 			panic("ffs_truncate2");
594 	BO_LOCK(bo);
595 	if (length == 0 &&
596 	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
597 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
598 		panic("ffs_truncate3");
599 	BO_UNLOCK(bo);
600 #endif /* INVARIANTS */
601 	/*
602 	 * Put back the real size.
603 	 */
604 	ip->i_size = length;
605 	DIP_SET(ip, i_size, length);
606 	if (DIP(ip, i_blocks) >= blocksreleased)
607 		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
608 	else	/* sanity */
609 		DIP_SET(ip, i_blocks, 0);
610 	ip->i_flag |= IN_CHANGE;
611 #ifdef QUOTA
612 	(void) chkdq(ip, -blocksreleased, NOCRED, 0);
613 #endif
614 	return (allerror);
615 
616 extclean:
617 	if (journaltrunc)
618 		softdep_journal_freeblocks(ip, cred, length, IO_EXT);
619 	else
620 		softdep_setup_freeblocks(ip, length, IO_EXT);
621 	return (ffs_update(vp, waitforupdate));
622 }
623 
624 /*
625  * Release blocks associated with the inode ip and stored in the indirect
626  * block bn.  Blocks are free'd in LIFO order up to (but not including)
627  * lastbn.  If level is greater than SINGLE, the block is an indirect block
628  * and recursive calls to indirtrunc must be used to cleanse other indirect
629  * blocks.
630  */
631 static int
632 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
633 	struct inode *ip;
634 	ufs2_daddr_t lbn, lastbn;
635 	ufs2_daddr_t dbn;
636 	int level;
637 	ufs2_daddr_t *countp;
638 {
639 	struct buf *bp;
640 	struct fs *fs;
641 	struct ufsmount *ump;
642 	struct vnode *vp;
643 	caddr_t copy = NULL;
644 	u_long key;
645 	int i, nblocks, error = 0, allerror = 0;
646 	ufs2_daddr_t nb, nlbn, last;
647 	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
648 	ufs1_daddr_t *bap1 = NULL;
649 	ufs2_daddr_t *bap2 = NULL;
650 #define BAP(ip, i) (I_IS_UFS1(ip) ? bap1[i] : bap2[i])
651 
652 	fs = ITOFS(ip);
653 	ump = ITOUMP(ip);
654 
655 	/*
656 	 * Calculate index in current block of last
657 	 * block to be kept.  -1 indicates the entire
658 	 * block so we need not calculate the index.
659 	 */
660 	factor = lbn_offset(fs, level);
661 	last = lastbn;
662 	if (lastbn > 0)
663 		last /= factor;
664 	nblocks = btodb(fs->fs_bsize);
665 	/*
666 	 * Get buffer of block pointers, zero those entries corresponding
667 	 * to blocks to be free'd, and update on disk copy first.  Since
668 	 * double(triple) indirect before single(double) indirect, calls
669 	 * to bmap on these blocks will fail.  However, we already have
670 	 * the on disk address, so we have to set the b_blkno field
671 	 * explicitly instead of letting bread do everything for us.
672 	 */
673 	vp = ITOV(ip);
674 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
675 	if ((bp->b_flags & B_CACHE) == 0) {
676 #ifdef RACCT
677 		if (racct_enable) {
678 			PROC_LOCK(curproc);
679 			racct_add_buf(curproc, bp, 0);
680 			PROC_UNLOCK(curproc);
681 		}
682 #endif /* RACCT */
683 		curthread->td_ru.ru_inblock++;	/* pay for read */
684 		bp->b_iocmd = BIO_READ;
685 		bp->b_flags &= ~B_INVAL;
686 		bp->b_ioflags &= ~BIO_ERROR;
687 		if (bp->b_bcount > bp->b_bufsize)
688 			panic("ffs_indirtrunc: bad buffer size");
689 		bp->b_blkno = dbn;
690 		vfs_busy_pages(bp, 0);
691 		bp->b_iooffset = dbtob(bp->b_blkno);
692 		bstrategy(bp);
693 		error = bufwait(bp);
694 	}
695 	if (error) {
696 		brelse(bp);
697 		*countp = 0;
698 		return (error);
699 	}
700 
701 	if (I_IS_UFS1(ip))
702 		bap1 = (ufs1_daddr_t *)bp->b_data;
703 	else
704 		bap2 = (ufs2_daddr_t *)bp->b_data;
705 	if (lastbn != -1) {
706 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
707 		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
708 		for (i = last + 1; i < NINDIR(fs); i++)
709 			if (I_IS_UFS1(ip))
710 				bap1[i] = 0;
711 			else
712 				bap2[i] = 0;
713 		if (DOINGASYNC(vp)) {
714 			bdwrite(bp);
715 		} else {
716 			error = bwrite(bp);
717 			if (error)
718 				allerror = error;
719 		}
720 		if (I_IS_UFS1(ip))
721 			bap1 = (ufs1_daddr_t *)copy;
722 		else
723 			bap2 = (ufs2_daddr_t *)copy;
724 	}
725 
726 	/*
727 	 * Recursively free totally unused blocks.
728 	 */
729 	key = ffs_blkrelease_start(ump, ITODEVVP(ip), ip->i_number);
730 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
731 	    i--, nlbn += factor) {
732 		nb = BAP(ip, i);
733 		if (nb == 0)
734 			continue;
735 		if (level > SINGLE) {
736 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
737 			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
738 				allerror = error;
739 			blocksreleased += blkcount;
740 		}
741 		ffs_blkfree(ump, fs, ITODEVVP(ip), nb, fs->fs_bsize,
742 		    ip->i_number, vp->v_type, NULL, key);
743 		blocksreleased += nblocks;
744 	}
745 	ffs_blkrelease_finish(ump, key);
746 
747 	/*
748 	 * Recursively free last partial block.
749 	 */
750 	if (level > SINGLE && lastbn >= 0) {
751 		last = lastbn % factor;
752 		nb = BAP(ip, i);
753 		if (nb != 0) {
754 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
755 			    last, level - 1, &blkcount);
756 			if (error)
757 				allerror = error;
758 			blocksreleased += blkcount;
759 		}
760 	}
761 	if (copy != NULL) {
762 		free(copy, M_TEMP);
763 	} else {
764 		bp->b_flags |= B_INVAL | B_NOCACHE;
765 		brelse(bp);
766 	}
767 
768 	*countp = blocksreleased;
769 	return (allerror);
770 }
771 
772 int
773 ffs_rdonly(struct inode *ip)
774 {
775 
776 	return (ITOFS(ip)->fs_ronly != 0);
777 }
778 
779