xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision eac7052fdebb90caf2f653e06187bdbca837b9c7)
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 	daddr_t bn;
90 	int flags, error;
91 
92 	ASSERT_VOP_ELOCKED(vp, "ffs_update");
93 	ufs_itimes(vp);
94 	ip = VTOI(vp);
95 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
96 		return (0);
97 	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
98 	/*
99 	 * The IN_SIZEMOD and IN_IBLKDATA flags indicate changes to the
100 	 * file size and block pointer fields in the inode. When these
101 	 * fields have been changed, the fsync() and fsyncdata() system
102 	 * calls must write the inode to ensure their semantics that the
103 	 * file is on stable store.
104 	 *
105 	 * The IN_SIZEMOD and IN_IBLKDATA flags cannot be cleared until
106 	 * a synchronous write of the inode is done. If they are cleared
107 	 * on an asynchronous write, then the inode may not yet have been
108 	 * written to the disk when an fsync() or fsyncdata() call is done.
109 	 * Absent these flags, these calls would not know that they needed
110 	 * to write the inode. Thus, these flags only can be cleared on
111 	 * synchronous writes of the inode. Since the inode will be locked
112 	 * for the duration of the I/O that writes it to disk, no fsync()
113 	 * or fsyncdata() will be able to run before the on-disk inode
114 	 * is complete.
115 	 */
116 	if (waitfor)
117 		ip->i_flag &= ~(IN_SIZEMOD | IN_IBLKDATA);
118 	fs = ITOFS(ip);
119 	if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0)
120 		return (0);
121 	/*
122 	 * If we are updating a snapshot and another process is currently
123 	 * writing the buffer containing the inode for this snapshot then
124 	 * a deadlock can occur when it tries to check the snapshot to see
125 	 * if that block needs to be copied. Thus when updating a snapshot
126 	 * we check to see if the buffer is already locked, and if it is
127 	 * we drop the snapshot lock until the buffer has been written
128 	 * and is available to us. We have to grab a reference to the
129 	 * snapshot vnode to prevent it from being removed while we are
130 	 * waiting for the buffer.
131 	 */
132 	flags = 0;
133 	if (IS_SNAPSHOT(ip))
134 		flags = GB_LOCK_NOWAIT;
135 loop:
136 	bn = fsbtodb(fs, ino_to_fsba(fs, ip->i_number));
137 	error = ffs_breadz(VFSTOUFS(vp->v_mount), ITODEVVP(ip), bn, bn,
138 	     (int) fs->fs_bsize, NULL, NULL, 0, NOCRED, flags, NULL, &bp);
139 	if (error != 0) {
140 		if (error != EBUSY)
141 			return (error);
142 		KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
143 		/*
144 		 * Wait for our inode block to become available.
145 		 *
146 		 * Hold a reference to the vnode to protect against
147 		 * ffs_snapgone(). Since we hold a reference, it can only
148 		 * get reclaimed (VIRF_DOOMED flag) in a forcible downgrade
149 		 * or unmount. For an unmount, the entire filesystem will be
150 		 * gone, so we cannot attempt to touch anything associated
151 		 * with it while the vnode is unlocked; all we can do is
152 		 * pause briefly and try again. If when we relock the vnode
153 		 * we discover that it has been reclaimed, updating it is no
154 		 * longer necessary and we can just return an error.
155 		 */
156 		vref(vp);
157 		VOP_UNLOCK(vp);
158 		pause("ffsupd", 1);
159 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
160 		vrele(vp);
161 		if (VN_IS_DOOMED(vp))
162 			return (ENOENT);
163 		goto loop;
164 	}
165 	if (DOINGSOFTDEP(vp))
166 		softdep_update_inodeblock(ip, bp, waitfor);
167 	else if (ip->i_effnlink != ip->i_nlink)
168 		panic("ffs_update: bad link cnt");
169 	if (I_IS_UFS1(ip)) {
170 		*((struct ufs1_dinode *)bp->b_data +
171 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
172 		/*
173 		 * XXX: FIX? The entropy here is desirable,
174 		 * but the harvesting may be expensive
175 		 */
176 		random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), RANDOM_FS_ATIME);
177 	} else {
178 		ffs_update_dinode_ckhash(fs, ip->i_din2);
179 		*((struct ufs2_dinode *)bp->b_data +
180 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
181 		/*
182 		 * XXX: FIX? The entropy here is desirable,
183 		 * but the harvesting may be expensive
184 		 */
185 		random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), RANDOM_FS_ATIME);
186 	}
187 	if (waitfor) {
188 		error = bwrite(bp);
189 		if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error))
190 			error = 0;
191 	} else if (vm_page_count_severe() || buf_dirty_count_severe()) {
192 		bawrite(bp);
193 		error = 0;
194 	} else {
195 		if (bp->b_bufsize == fs->fs_bsize)
196 			bp->b_flags |= B_CLUSTEROK;
197 		bdwrite(bp);
198 		error = 0;
199 	}
200 	return (error);
201 }
202 
203 #define	SINGLE	0	/* index of single indirect block */
204 #define	DOUBLE	1	/* index of double indirect block */
205 #define	TRIPLE	2	/* index of triple indirect block */
206 /*
207  * Truncate the inode ip to at most length size, freeing the
208  * disk blocks.
209  */
210 int
211 ffs_truncate(vp, length, flags, cred)
212 	struct vnode *vp;
213 	off_t length;
214 	int flags;
215 	struct ucred *cred;
216 {
217 	struct inode *ip;
218 	ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR];
219 	ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR];
220 	ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR];
221 	ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno;
222 	struct bufobj *bo;
223 	struct fs *fs;
224 	struct buf *bp;
225 	struct ufsmount *ump;
226 	int softdeptrunc, journaltrunc;
227 	int needextclean, extblocks;
228 	int offset, size, level, nblocks;
229 	int i, error, allerror, indiroff, waitforupdate;
230 	u_long key;
231 	off_t osize;
232 
233 	ip = VTOI(vp);
234 	ump = VFSTOUFS(vp->v_mount);
235 	fs = ump->um_fs;
236 	bo = &vp->v_bufobj;
237 
238 	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
239 
240 	if (length < 0)
241 		return (EINVAL);
242 	if (length > fs->fs_maxfilesize)
243 		return (EFBIG);
244 #ifdef QUOTA
245 	error = getinoquota(ip);
246 	if (error)
247 		return (error);
248 #endif
249 	/*
250 	 * Historically clients did not have to specify which data
251 	 * they were truncating. So, if not specified, we assume
252 	 * traditional behavior, e.g., just the normal data.
253 	 */
254 	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
255 		flags |= IO_NORMAL;
256 	if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
257 		flags |= IO_SYNC;
258 	waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp);
259 	/*
260 	 * If we are truncating the extended-attributes, and cannot
261 	 * do it with soft updates, then do it slowly here. If we are
262 	 * truncating both the extended attributes and the file contents
263 	 * (e.g., the file is being unlinked), then pick it off with
264 	 * soft updates below.
265 	 */
266 	allerror = 0;
267 	needextclean = 0;
268 	softdeptrunc = 0;
269 	journaltrunc = DOINGSUJ(vp);
270 	journaltrunc = 0;	/* XXX temp patch until bug found */
271 	if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
272 		softdeptrunc = !softdep_slowdown(vp);
273 	extblocks = 0;
274 	datablocks = DIP(ip, i_blocks);
275 	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
276 		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
277 		datablocks -= extblocks;
278 	}
279 	if ((flags & IO_EXT) && extblocks > 0) {
280 		if (length != 0)
281 			panic("ffs_truncate: partial trunc of extdata");
282 		if (softdeptrunc || journaltrunc) {
283 			if ((flags & IO_NORMAL) == 0)
284 				goto extclean;
285 			needextclean = 1;
286 		} else {
287 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
288 				return (error);
289 #ifdef QUOTA
290 			(void) chkdq(ip, -extblocks, NOCRED, FORCE);
291 #endif
292 			vinvalbuf(vp, V_ALT, 0, 0);
293 			vn_pages_remove(vp,
294 			    OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
295 			osize = ip->i_din2->di_extsize;
296 			ip->i_din2->di_blocks -= extblocks;
297 			ip->i_din2->di_extsize = 0;
298 			for (i = 0; i < UFS_NXADDR; i++) {
299 				oldblks[i] = ip->i_din2->di_extb[i];
300 				ip->i_din2->di_extb[i] = 0;
301 			}
302 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
303 			if ((error = ffs_update(vp, waitforupdate)))
304 				return (error);
305 			for (i = 0; i < UFS_NXADDR; i++) {
306 				if (oldblks[i] == 0)
307 					continue;
308 				ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i],
309 				    sblksize(fs, osize, i), ip->i_number,
310 				    vp->v_type, NULL, SINGLETON_KEY);
311 			}
312 		}
313 	}
314 	if ((flags & IO_NORMAL) == 0)
315 		return (0);
316 	if (vp->v_type == VLNK &&
317 	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
318 	     datablocks == 0)) {
319 #ifdef INVARIANTS
320 		if (length != 0)
321 			panic("ffs_truncate: partial truncate of symlink");
322 #endif
323 		bzero(SHORTLINK(ip), (u_int)ip->i_size);
324 		ip->i_size = 0;
325 		DIP_SET(ip, i_size, 0);
326 		UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
327 		if (needextclean)
328 			goto extclean;
329 		return (ffs_update(vp, waitforupdate));
330 	}
331 	if (ip->i_size == length) {
332 		UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
333 		if (needextclean)
334 			goto extclean;
335 		return (ffs_update(vp, 0));
336 	}
337 	if (fs->fs_ronly)
338 		panic("ffs_truncate: read-only filesystem");
339 	if (IS_SNAPSHOT(ip))
340 		ffs_snapremove(vp);
341 	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
342 	osize = ip->i_size;
343 	/*
344 	 * Lengthen the size of the file. We must ensure that the
345 	 * last byte of the file is allocated. Since the smallest
346 	 * value of osize is 0, length will be at least 1.
347 	 */
348 	if (osize < length) {
349 		vnode_pager_setsize(vp, length);
350 		flags |= BA_CLRBUF;
351 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
352 		if (error) {
353 			vnode_pager_setsize(vp, osize);
354 			return (error);
355 		}
356 		ip->i_size = length;
357 		DIP_SET(ip, i_size, length);
358 		if (bp->b_bufsize == fs->fs_bsize)
359 			bp->b_flags |= B_CLUSTEROK;
360 		if (flags & IO_SYNC)
361 			bwrite(bp);
362 		else if (DOINGASYNC(vp))
363 			bdwrite(bp);
364 		else
365 			bawrite(bp);
366 		UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
367 		return (ffs_update(vp, waitforupdate));
368 	}
369 	/*
370 	 * Lookup block number for a given offset. Zero length files
371 	 * have no blocks, so return a blkno of -1.
372 	 */
373 	lbn = lblkno(fs, length - 1);
374 	if (length == 0) {
375 		blkno = -1;
376 	} else if (lbn < UFS_NDADDR) {
377 		blkno = DIP(ip, i_db[lbn]);
378 	} else {
379 		error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
380 		    cred, BA_METAONLY, &bp);
381 		if (error)
382 			return (error);
383 		indiroff = (lbn - UFS_NDADDR) % NINDIR(fs);
384 		if (I_IS_UFS1(ip))
385 			blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
386 		else
387 			blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
388 		/*
389 		 * If the block number is non-zero, then the indirect block
390 		 * must have been previously allocated and need not be written.
391 		 * If the block number is zero, then we may have allocated
392 		 * the indirect block and hence need to write it out.
393 		 */
394 		if (blkno != 0)
395 			brelse(bp);
396 		else if (flags & IO_SYNC)
397 			bwrite(bp);
398 		else
399 			bdwrite(bp);
400 	}
401 	/*
402 	 * If the block number at the new end of the file is zero,
403 	 * then we must allocate it to ensure that the last block of
404 	 * the file is allocated. Soft updates does not handle this
405 	 * case, so here we have to clean up the soft updates data
406 	 * structures describing the allocation past the truncation
407 	 * point. Finding and deallocating those structures is a lot of
408 	 * work. Since partial truncation with a hole at the end occurs
409 	 * rarely, we solve the problem by syncing the file so that it
410 	 * will have no soft updates data structures left.
411 	 */
412 	if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
413 		return (error);
414 	if (blkno != 0 && DOINGSOFTDEP(vp)) {
415 		if (softdeptrunc == 0 && journaltrunc == 0) {
416 			/*
417 			 * If soft updates cannot handle this truncation,
418 			 * clean up soft dependency data structures and
419 			 * fall through to the synchronous truncation.
420 			 */
421 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
422 				return (error);
423 		} else {
424 			flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
425 			if (journaltrunc)
426 				softdep_journal_freeblocks(ip, cred, length,
427 				    flags);
428 			else
429 				softdep_setup_freeblocks(ip, length, flags);
430 			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
431 			if (journaltrunc == 0) {
432 				UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
433 				error = ffs_update(vp, 0);
434 			}
435 			return (error);
436 		}
437 	}
438 	/*
439 	 * Shorten the size of the file. If the last block of the
440 	 * shortened file is unallocated, we must allocate it.
441 	 * Additionally, if the file is not being truncated to a
442 	 * block boundary, the contents of the partial block
443 	 * following the end of the file must be zero'ed in
444 	 * case it ever becomes accessible again because of
445 	 * subsequent file growth. Directories however are not
446 	 * zero'ed as they should grow back initialized to empty.
447 	 */
448 	offset = blkoff(fs, length);
449 	if (blkno != 0 && offset == 0) {
450 		ip->i_size = length;
451 		DIP_SET(ip, i_size, length);
452 		UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
453 	} else {
454 		lbn = lblkno(fs, length);
455 		flags |= BA_CLRBUF;
456 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
457 		if (error)
458 			return (error);
459 		/*
460 		 * When we are doing soft updates and the UFS_BALLOC
461 		 * above fills in a direct block hole with a full sized
462 		 * block that will be truncated down to a fragment below,
463 		 * we must flush out the block dependency with an FSYNC
464 		 * so that we do not get a soft updates inconsistency
465 		 * when we create the fragment below.
466 		 */
467 		if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR &&
468 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
469 		    (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
470 			return (error);
471 		ip->i_size = length;
472 		DIP_SET(ip, i_size, length);
473 		size = blksize(fs, ip, lbn);
474 		if (vp->v_type != VDIR && offset != 0)
475 			bzero((char *)bp->b_data + offset,
476 			    (u_int)(size - offset));
477 		/* Kirk's code has reallocbuf(bp, size, 1) here */
478 		allocbuf(bp, size);
479 		if (bp->b_bufsize == fs->fs_bsize)
480 			bp->b_flags |= B_CLUSTEROK;
481 		if (flags & IO_SYNC)
482 			bwrite(bp);
483 		else if (DOINGASYNC(vp))
484 			bdwrite(bp);
485 		else
486 			bawrite(bp);
487 		UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
488 	}
489 	/*
490 	 * Calculate index into inode's block list of
491 	 * last direct and indirect blocks (if any)
492 	 * which we want to keep.  Lastblock is -1 when
493 	 * the file is truncated to 0.
494 	 */
495 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
496 	lastiblock[SINGLE] = lastblock - UFS_NDADDR;
497 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
498 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
499 	nblocks = btodb(fs->fs_bsize);
500 	/*
501 	 * Update file and block pointers on disk before we start freeing
502 	 * blocks.  If we crash before free'ing blocks below, the blocks
503 	 * will be returned to the free list.  lastiblock values are also
504 	 * normalized to -1 for calls to ffs_indirtrunc below.
505 	 */
506 	for (level = TRIPLE; level >= SINGLE; level--) {
507 		oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]);
508 		if (lastiblock[level] < 0) {
509 			DIP_SET(ip, i_ib[level], 0);
510 			lastiblock[level] = -1;
511 		}
512 	}
513 	for (i = 0; i < UFS_NDADDR; i++) {
514 		oldblks[i] = DIP(ip, i_db[i]);
515 		if (i > lastblock)
516 			DIP_SET(ip, i_db[i], 0);
517 	}
518 	UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
519 	allerror = ffs_update(vp, waitforupdate);
520 
521 	/*
522 	 * Having written the new inode to disk, save its new configuration
523 	 * and put back the old block pointers long enough to process them.
524 	 * Note that we save the new block configuration so we can check it
525 	 * when we are done.
526 	 */
527 	for (i = 0; i < UFS_NDADDR; i++) {
528 		newblks[i] = DIP(ip, i_db[i]);
529 		DIP_SET(ip, i_db[i], oldblks[i]);
530 	}
531 	for (i = 0; i < UFS_NIADDR; i++) {
532 		newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]);
533 		DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]);
534 	}
535 	ip->i_size = osize;
536 	DIP_SET(ip, i_size, osize);
537 	UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
538 
539 	error = vtruncbuf(vp, length, fs->fs_bsize);
540 	if (error && (allerror == 0))
541 		allerror = error;
542 
543 	/*
544 	 * Indirect blocks first.
545 	 */
546 	indir_lbn[SINGLE] = -UFS_NDADDR;
547 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
548 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
549 	for (level = TRIPLE; level >= SINGLE; level--) {
550 		bn = DIP(ip, i_ib[level]);
551 		if (bn != 0) {
552 			error = ffs_indirtrunc(ip, indir_lbn[level],
553 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
554 			if (error)
555 				allerror = error;
556 			blocksreleased += count;
557 			if (lastiblock[level] < 0) {
558 				DIP_SET(ip, i_ib[level], 0);
559 				ffs_blkfree(ump, fs, ump->um_devvp, bn,
560 				    fs->fs_bsize, ip->i_number,
561 				    vp->v_type, NULL, SINGLETON_KEY);
562 				blocksreleased += nblocks;
563 			}
564 		}
565 		if (lastiblock[level] >= 0)
566 			goto done;
567 	}
568 
569 	/*
570 	 * All whole direct blocks or frags.
571 	 */
572 	key = ffs_blkrelease_start(ump, ump->um_devvp, ip->i_number);
573 	for (i = UFS_NDADDR - 1; i > lastblock; i--) {
574 		long bsize;
575 
576 		bn = DIP(ip, i_db[i]);
577 		if (bn == 0)
578 			continue;
579 		DIP_SET(ip, i_db[i], 0);
580 		bsize = blksize(fs, ip, i);
581 		ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number,
582 		    vp->v_type, NULL, key);
583 		blocksreleased += btodb(bsize);
584 	}
585 	ffs_blkrelease_finish(ump, key);
586 	if (lastblock < 0)
587 		goto done;
588 
589 	/*
590 	 * Finally, look for a change in size of the
591 	 * last direct block; release any frags.
592 	 */
593 	bn = DIP(ip, i_db[lastblock]);
594 	if (bn != 0) {
595 		long oldspace, newspace;
596 
597 		/*
598 		 * Calculate amount of space we're giving
599 		 * back as old block size minus new block size.
600 		 */
601 		oldspace = blksize(fs, ip, lastblock);
602 		ip->i_size = length;
603 		DIP_SET(ip, i_size, length);
604 		UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
605 		newspace = blksize(fs, ip, lastblock);
606 		if (newspace == 0)
607 			panic("ffs_truncate: newspace");
608 		if (oldspace - newspace > 0) {
609 			/*
610 			 * Block number of space to be free'd is
611 			 * the old block # plus the number of frags
612 			 * required for the storage we're keeping.
613 			 */
614 			bn += numfrags(fs, newspace);
615 			ffs_blkfree(ump, fs, ump->um_devvp, bn,
616 			   oldspace - newspace, ip->i_number, vp->v_type,
617 			   NULL, SINGLETON_KEY);
618 			blocksreleased += btodb(oldspace - newspace);
619 		}
620 	}
621 done:
622 #ifdef INVARIANTS
623 	for (level = SINGLE; level <= TRIPLE; level++)
624 		if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level]))
625 			panic("ffs_truncate1: level %d newblks %jd != i_ib %jd",
626 			    level, (intmax_t)newblks[UFS_NDADDR + level],
627 			    (intmax_t)DIP(ip, i_ib[level]));
628 	for (i = 0; i < UFS_NDADDR; i++)
629 		if (newblks[i] != DIP(ip, i_db[i]))
630 			panic("ffs_truncate2: blkno %d newblks %jd != i_db %jd",
631 			    i, (intmax_t)newblks[UFS_NDADDR + level],
632 			    (intmax_t)DIP(ip, i_ib[level]));
633 	BO_LOCK(bo);
634 	if (length == 0 &&
635 	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
636 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
637 		panic("ffs_truncate3: vp = %p, buffers: dirty = %d, clean = %d",
638 			vp, bo->bo_dirty.bv_cnt, bo->bo_clean.bv_cnt);
639 	BO_UNLOCK(bo);
640 #endif /* INVARIANTS */
641 	/*
642 	 * Put back the real size.
643 	 */
644 	ip->i_size = length;
645 	DIP_SET(ip, i_size, length);
646 	if (DIP(ip, i_blocks) >= blocksreleased)
647 		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
648 	else	/* sanity */
649 		DIP_SET(ip, i_blocks, 0);
650 	UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
651 #ifdef QUOTA
652 	(void) chkdq(ip, -blocksreleased, NOCRED, FORCE);
653 #endif
654 	return (allerror);
655 
656 extclean:
657 	if (journaltrunc)
658 		softdep_journal_freeblocks(ip, cred, length, IO_EXT);
659 	else
660 		softdep_setup_freeblocks(ip, length, IO_EXT);
661 	return (ffs_update(vp, waitforupdate));
662 }
663 
664 /*
665  * Release blocks associated with the inode ip and stored in the indirect
666  * block bn.  Blocks are free'd in LIFO order up to (but not including)
667  * lastbn.  If level is greater than SINGLE, the block is an indirect block
668  * and recursive calls to indirtrunc must be used to cleanse other indirect
669  * blocks.
670  */
671 static int
672 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
673 	struct inode *ip;
674 	ufs2_daddr_t lbn, lastbn;
675 	ufs2_daddr_t dbn;
676 	int level;
677 	ufs2_daddr_t *countp;
678 {
679 	struct buf *bp;
680 	struct fs *fs;
681 	struct ufsmount *ump;
682 	struct vnode *vp;
683 	caddr_t copy = NULL;
684 	u_long key;
685 	int i, nblocks, error = 0, allerror = 0;
686 	ufs2_daddr_t nb, nlbn, last;
687 	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
688 	ufs1_daddr_t *bap1 = NULL;
689 	ufs2_daddr_t *bap2 = NULL;
690 #define BAP(ip, i) (I_IS_UFS1(ip) ? bap1[i] : bap2[i])
691 
692 	fs = ITOFS(ip);
693 	ump = ITOUMP(ip);
694 
695 	/*
696 	 * Calculate index in current block of last
697 	 * block to be kept.  -1 indicates the entire
698 	 * block so we need not calculate the index.
699 	 */
700 	factor = lbn_offset(fs, level);
701 	last = lastbn;
702 	if (lastbn > 0)
703 		last /= factor;
704 	nblocks = btodb(fs->fs_bsize);
705 	/*
706 	 * Get buffer of block pointers, zero those entries corresponding
707 	 * to blocks to be free'd, and update on disk copy first.  Since
708 	 * double(triple) indirect before single(double) indirect, calls
709 	 * to VOP_BMAP() on these blocks will fail.  However, we already
710 	 * have the on-disk address, so we just pass it to bread() instead
711 	 * of having bread() attempt to calculate it using VOP_BMAP().
712 	 */
713 	vp = ITOV(ip);
714 	error = ffs_breadz(ump, vp, lbn, dbn, (int)fs->fs_bsize, NULL, NULL, 0,
715 	    NOCRED, 0, NULL, &bp);
716 	if (error) {
717 		*countp = 0;
718 		return (error);
719 	}
720 
721 	if (I_IS_UFS1(ip))
722 		bap1 = (ufs1_daddr_t *)bp->b_data;
723 	else
724 		bap2 = (ufs2_daddr_t *)bp->b_data;
725 	if (lastbn != -1) {
726 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
727 		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
728 		for (i = last + 1; i < NINDIR(fs); i++)
729 			if (I_IS_UFS1(ip))
730 				bap1[i] = 0;
731 			else
732 				bap2[i] = 0;
733 		if (DOINGASYNC(vp)) {
734 			bdwrite(bp);
735 		} else {
736 			error = bwrite(bp);
737 			if (error)
738 				allerror = error;
739 		}
740 		if (I_IS_UFS1(ip))
741 			bap1 = (ufs1_daddr_t *)copy;
742 		else
743 			bap2 = (ufs2_daddr_t *)copy;
744 	}
745 
746 	/*
747 	 * Recursively free totally unused blocks.
748 	 */
749 	key = ffs_blkrelease_start(ump, ITODEVVP(ip), ip->i_number);
750 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
751 	    i--, nlbn += factor) {
752 		nb = BAP(ip, i);
753 		if (nb == 0)
754 			continue;
755 		if (level > SINGLE) {
756 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
757 			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
758 				allerror = error;
759 			blocksreleased += blkcount;
760 		}
761 		ffs_blkfree(ump, fs, ITODEVVP(ip), nb, fs->fs_bsize,
762 		    ip->i_number, vp->v_type, NULL, key);
763 		blocksreleased += nblocks;
764 	}
765 	ffs_blkrelease_finish(ump, key);
766 
767 	/*
768 	 * Recursively free last partial block.
769 	 */
770 	if (level > SINGLE && lastbn >= 0) {
771 		last = lastbn % factor;
772 		nb = BAP(ip, i);
773 		if (nb != 0) {
774 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
775 			    last, level - 1, &blkcount);
776 			if (error)
777 				allerror = error;
778 			blocksreleased += blkcount;
779 		}
780 	}
781 	if (copy != NULL) {
782 		free(copy, M_TEMP);
783 	} else {
784 		bp->b_flags |= B_INVAL | B_NOCACHE;
785 		brelse(bp);
786 	}
787 
788 	*countp = blocksreleased;
789 	return (allerror);
790 }
791 
792 int
793 ffs_rdonly(struct inode *ip)
794 {
795 
796 	return (ITOFS(ip)->fs_ronly != 0);
797 }
798