xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
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 
52 #include <ufs/ufs/extattr.h>
53 #include <ufs/ufs/quota.h>
54 #include <ufs/ufs/ufsmount.h>
55 #include <ufs/ufs/inode.h>
56 #include <ufs/ufs/ufs_extern.h>
57 
58 #include <ufs/ffs/fs.h>
59 #include <ufs/ffs/ffs_extern.h>
60 
61 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
62 	    ufs2_daddr_t, int, ufs2_daddr_t *);
63 
64 /*
65  * Update the access, modified, and inode change times as specified by the
66  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
67  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
68  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
69  * later if not now.  The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
70  * is currently being suspended (or is suspended) and vnode has been accessed.
71  * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
72  * reflect the presumably successful write, and if waitfor is set, then wait
73  * for the write to complete.
74  */
75 int
76 ffs_update(vp, waitfor)
77 	struct vnode *vp;
78 	int waitfor;
79 {
80 	struct fs *fs;
81 	struct buf *bp;
82 	struct inode *ip;
83 	int error;
84 
85 	ASSERT_VOP_ELOCKED(vp, "ffs_update");
86 	ufs_itimes(vp);
87 	ip = VTOI(vp);
88 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
89 		return (0);
90 	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
91 	fs = ip->i_fs;
92 	if (fs->fs_ronly)
93 		return (0);
94 	/*
95 	 * Ensure that uid and gid are correct. This is a temporary
96 	 * fix until fsck has been changed to do the update.
97 	 */
98 	if (fs->fs_magic == FS_UFS1_MAGIC &&		/* XXX */
99 	    fs->fs_old_inodefmt < FS_44INODEFMT) {	/* XXX */
100 		ip->i_din1->di_ouid = ip->i_uid;	/* XXX */
101 		ip->i_din1->di_ogid = ip->i_gid;	/* XXX */
102 	}						/* XXX */
103 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
104 		(int)fs->fs_bsize, NOCRED, &bp);
105 	if (error) {
106 		brelse(bp);
107 		return (error);
108 	}
109 	if (DOINGSOFTDEP(vp))
110 		softdep_update_inodeblock(ip, bp, waitfor);
111 	else if (ip->i_effnlink != ip->i_nlink)
112 		panic("ffs_update: bad link cnt");
113 	if (ip->i_ump->um_fstype == UFS1)
114 		*((struct ufs1_dinode *)bp->b_data +
115 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
116 	else
117 		*((struct ufs2_dinode *)bp->b_data +
118 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
119 	if (waitfor && !DOINGASYNC(vp)) {
120 		return (bwrite(bp));
121 	} else if (vm_page_count_severe() || buf_dirty_count_severe()) {
122 		return (bwrite(bp));
123 	} else {
124 		if (bp->b_bufsize == fs->fs_bsize)
125 			bp->b_flags |= B_CLUSTEROK;
126 		bdwrite(bp);
127 		return (0);
128 	}
129 }
130 
131 #define	SINGLE	0	/* index of single indirect block */
132 #define	DOUBLE	1	/* index of double indirect block */
133 #define	TRIPLE	2	/* index of triple indirect block */
134 /*
135  * Truncate the inode ip to at most length size, freeing the
136  * disk blocks.
137  */
138 int
139 ffs_truncate(vp, length, flags, cred, td)
140 	struct vnode *vp;
141 	off_t length;
142 	int flags;
143 	struct ucred *cred;
144 	struct thread *td;
145 {
146 	struct inode *ip;
147 	ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
148 	ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
149 	ufs2_daddr_t count, blocksreleased = 0, datablocks;
150 	struct fs *fs;
151 	struct buf *bp;
152 	struct ufsmount *ump;
153 	int needextclean, softdepslowdown, extblocks;
154 	int offset, size, level, nblocks;
155 	int i, error, allerror;
156 	off_t osize;
157 
158 	ip = VTOI(vp);
159 	fs = ip->i_fs;
160 	ump = ip->i_ump;
161 
162 	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
163 
164 	if (length < 0)
165 		return (EINVAL);
166 	/*
167 	 * Historically clients did not have to specify which data
168 	 * they were truncating. So, if not specified, we assume
169 	 * traditional behavior, e.g., just the normal data.
170 	 */
171 	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
172 		flags |= IO_NORMAL;
173 	/*
174 	 * If we are truncating the extended-attributes, and cannot
175 	 * do it with soft updates, then do it slowly here. If we are
176 	 * truncating both the extended attributes and the file contents
177 	 * (e.g., the file is being unlinked), then pick it off with
178 	 * soft updates below.
179 	 */
180 	needextclean = 0;
181 	softdepslowdown = DOINGSOFTDEP(vp) && softdep_slowdown(vp);
182 	extblocks = 0;
183 	datablocks = DIP(ip, i_blocks);
184 	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
185 		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
186 		datablocks -= extblocks;
187 	}
188 	if ((flags & IO_EXT) && extblocks > 0) {
189 		if (DOINGSOFTDEP(vp) && softdepslowdown == 0 && length == 0) {
190 			if ((flags & IO_NORMAL) == 0) {
191 				softdep_setup_freeblocks(ip, length, IO_EXT);
192 				return (0);
193 			}
194 			needextclean = 1;
195 		} else {
196 			if (length != 0)
197 				panic("ffs_truncate: partial trunc of extdata");
198 			if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
199 				return (error);
200 			osize = ip->i_din2->di_extsize;
201 			ip->i_din2->di_blocks -= extblocks;
202 #ifdef QUOTA
203 			(void) chkdq(ip, -extblocks, NOCRED, 0);
204 #endif
205 			vinvalbuf(vp, V_ALT, td, 0, 0);
206 			ip->i_din2->di_extsize = 0;
207 			for (i = 0; i < NXADDR; i++) {
208 				oldblks[i] = ip->i_din2->di_extb[i];
209 				ip->i_din2->di_extb[i] = 0;
210 			}
211 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
212 			if ((error = ffs_update(vp, 1)))
213 				return (error);
214 			for (i = 0; i < NXADDR; i++) {
215 				if (oldblks[i] == 0)
216 					continue;
217 				ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
218 				    sblksize(fs, osize, i), ip->i_number);
219 			}
220 		}
221 	}
222 	if ((flags & IO_NORMAL) == 0)
223 		return (0);
224 	if (length > fs->fs_maxfilesize)
225 		return (EFBIG);
226 	if (vp->v_type == VLNK &&
227 	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
228 	     datablocks == 0)) {
229 #ifdef INVARIANTS
230 		if (length != 0)
231 			panic("ffs_truncate: partial truncate of symlink");
232 #endif
233 		bzero(SHORTLINK(ip), (u_int)ip->i_size);
234 		ip->i_size = 0;
235 		DIP_SET(ip, i_size, 0);
236 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
237 		if (needextclean)
238 			softdep_setup_freeblocks(ip, length, IO_EXT);
239 		return (ffs_update(vp, 1));
240 	}
241 	if (ip->i_size == length) {
242 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
243 		if (needextclean)
244 			softdep_setup_freeblocks(ip, length, IO_EXT);
245 		return (ffs_update(vp, 0));
246 	}
247 	if (fs->fs_ronly)
248 		panic("ffs_truncate: read-only filesystem");
249 #ifdef QUOTA
250 	error = getinoquota(ip);
251 	if (error)
252 		return (error);
253 #endif
254 	if ((ip->i_flags & SF_SNAPSHOT) != 0)
255 		ffs_snapremove(vp);
256 	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
257 	if (DOINGSOFTDEP(vp)) {
258 		if (length > 0 || softdepslowdown) {
259 			/*
260 			 * If a file is only partially truncated, then
261 			 * we have to clean up the data structures
262 			 * describing the allocation past the truncation
263 			 * point. Finding and deallocating those structures
264 			 * is a lot of work. Since partial truncation occurs
265 			 * rarely, we solve the problem by syncing the file
266 			 * so that it will have no data structures left.
267 			 */
268 			if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
269 				return (error);
270 			UFS_LOCK(ump);
271 			if (ip->i_flag & IN_SPACECOUNTED)
272 				fs->fs_pendingblocks -= datablocks;
273 			UFS_UNLOCK(ump);
274 		} else {
275 #ifdef QUOTA
276 			(void) chkdq(ip, -datablocks, NOCRED, 0);
277 #endif
278 			softdep_setup_freeblocks(ip, length, needextclean ?
279 			    IO_EXT | IO_NORMAL : IO_NORMAL);
280 			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
281 			vinvalbuf(vp, needextclean ? 0 : V_NORMAL, td, 0, 0);
282 			vnode_pager_setsize(vp, 0);
283 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
284 			return (ffs_update(vp, 0));
285 		}
286 	}
287 	osize = ip->i_size;
288 	/*
289 	 * Lengthen the size of the file. We must ensure that the
290 	 * last byte of the file is allocated. Since the smallest
291 	 * value of osize is 0, length will be at least 1.
292 	 */
293 	if (osize < length) {
294 		vnode_pager_setsize(vp, length);
295 		flags |= BA_CLRBUF;
296 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
297 		if (error)
298 			return (error);
299 		ip->i_size = length;
300 		DIP_SET(ip, i_size, length);
301 		if (bp->b_bufsize == fs->fs_bsize)
302 			bp->b_flags |= B_CLUSTEROK;
303 		if (flags & IO_SYNC)
304 			bwrite(bp);
305 		else
306 			bawrite(bp);
307 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
308 		return (ffs_update(vp, 1));
309 	}
310 	/*
311 	 * Shorten the size of the file. If the file is not being
312 	 * truncated to a block boundary, the contents of the
313 	 * partial block following the end of the file must be
314 	 * zero'ed in case it ever becomes accessible again because
315 	 * of subsequent file growth. Directories however are not
316 	 * zero'ed as they should grow back initialized to empty.
317 	 */
318 	offset = blkoff(fs, length);
319 	if (offset == 0) {
320 		ip->i_size = length;
321 		DIP_SET(ip, i_size, length);
322 	} else {
323 		lbn = lblkno(fs, length);
324 		flags |= BA_CLRBUF;
325 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
326 		if (error) {
327 			return (error);
328 		}
329 		/*
330 		 * When we are doing soft updates and the UFS_BALLOC
331 		 * above fills in a direct block hole with a full sized
332 		 * block that will be truncated down to a fragment below,
333 		 * we must flush out the block dependency with an FSYNC
334 		 * so that we do not get a soft updates inconsistency
335 		 * when we create the fragment below.
336 		 */
337 		if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
338 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
339 		    (error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
340 			return (error);
341 		ip->i_size = length;
342 		DIP_SET(ip, i_size, length);
343 		size = blksize(fs, ip, lbn);
344 		if (vp->v_type != VDIR)
345 			bzero((char *)bp->b_data + offset,
346 			    (u_int)(size - offset));
347 		/* Kirk's code has reallocbuf(bp, size, 1) here */
348 		allocbuf(bp, size);
349 		if (bp->b_bufsize == fs->fs_bsize)
350 			bp->b_flags |= B_CLUSTEROK;
351 		if (flags & IO_SYNC)
352 			bwrite(bp);
353 		else
354 			bawrite(bp);
355 	}
356 	/*
357 	 * Calculate index into inode's block list of
358 	 * last direct and indirect blocks (if any)
359 	 * which we want to keep.  Lastblock is -1 when
360 	 * the file is truncated to 0.
361 	 */
362 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
363 	lastiblock[SINGLE] = lastblock - NDADDR;
364 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
365 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
366 	nblocks = btodb(fs->fs_bsize);
367 	/*
368 	 * Update file and block pointers on disk before we start freeing
369 	 * blocks.  If we crash before free'ing blocks below, the blocks
370 	 * will be returned to the free list.  lastiblock values are also
371 	 * normalized to -1 for calls to ffs_indirtrunc below.
372 	 */
373 	for (level = TRIPLE; level >= SINGLE; level--) {
374 		oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
375 		if (lastiblock[level] < 0) {
376 			DIP_SET(ip, i_ib[level], 0);
377 			lastiblock[level] = -1;
378 		}
379 	}
380 	for (i = 0; i < NDADDR; i++) {
381 		oldblks[i] = DIP(ip, i_db[i]);
382 		if (i > lastblock)
383 			DIP_SET(ip, i_db[i], 0);
384 	}
385 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
386 	allerror = ffs_update(vp, 1);
387 
388 	/*
389 	 * Having written the new inode to disk, save its new configuration
390 	 * and put back the old block pointers long enough to process them.
391 	 * Note that we save the new block configuration so we can check it
392 	 * when we are done.
393 	 */
394 	for (i = 0; i < NDADDR; i++) {
395 		newblks[i] = DIP(ip, i_db[i]);
396 		DIP_SET(ip, i_db[i], oldblks[i]);
397 	}
398 	for (i = 0; i < NIADDR; i++) {
399 		newblks[NDADDR + i] = DIP(ip, i_ib[i]);
400 		DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
401 	}
402 	ip->i_size = osize;
403 	DIP_SET(ip, i_size, osize);
404 
405 	error = vtruncbuf(vp, cred, td, length, fs->fs_bsize);
406 	if (error && (allerror == 0))
407 		allerror = error;
408 
409 	/*
410 	 * Indirect blocks first.
411 	 */
412 	indir_lbn[SINGLE] = -NDADDR;
413 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
414 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
415 	for (level = TRIPLE; level >= SINGLE; level--) {
416 		bn = DIP(ip, i_ib[level]);
417 		if (bn != 0) {
418 			error = ffs_indirtrunc(ip, indir_lbn[level],
419 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
420 			if (error)
421 				allerror = error;
422 			blocksreleased += count;
423 			if (lastiblock[level] < 0) {
424 				DIP_SET(ip, i_ib[level], 0);
425 				ffs_blkfree(ump, fs, ip->i_devvp, bn,
426 				    fs->fs_bsize, ip->i_number);
427 				blocksreleased += nblocks;
428 			}
429 		}
430 		if (lastiblock[level] >= 0)
431 			goto done;
432 	}
433 
434 	/*
435 	 * All whole direct blocks or frags.
436 	 */
437 	for (i = NDADDR - 1; i > lastblock; i--) {
438 		long bsize;
439 
440 		bn = DIP(ip, i_db[i]);
441 		if (bn == 0)
442 			continue;
443 		DIP_SET(ip, i_db[i], 0);
444 		bsize = blksize(fs, ip, i);
445 		ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number);
446 		blocksreleased += btodb(bsize);
447 	}
448 	if (lastblock < 0)
449 		goto done;
450 
451 	/*
452 	 * Finally, look for a change in size of the
453 	 * last direct block; release any frags.
454 	 */
455 	bn = DIP(ip, i_db[lastblock]);
456 	if (bn != 0) {
457 		long oldspace, newspace;
458 
459 		/*
460 		 * Calculate amount of space we're giving
461 		 * back as old block size minus new block size.
462 		 */
463 		oldspace = blksize(fs, ip, lastblock);
464 		ip->i_size = length;
465 		DIP_SET(ip, i_size, length);
466 		newspace = blksize(fs, ip, lastblock);
467 		if (newspace == 0)
468 			panic("ffs_truncate: newspace");
469 		if (oldspace - newspace > 0) {
470 			/*
471 			 * Block number of space to be free'd is
472 			 * the old block # plus the number of frags
473 			 * required for the storage we're keeping.
474 			 */
475 			bn += numfrags(fs, newspace);
476 			ffs_blkfree(ump, fs, ip->i_devvp, bn,
477 			    oldspace - newspace, ip->i_number);
478 			blocksreleased += btodb(oldspace - newspace);
479 		}
480 	}
481 done:
482 #ifdef INVARIANTS
483 	for (level = SINGLE; level <= TRIPLE; level++)
484 		if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
485 			panic("ffs_truncate1");
486 	for (i = 0; i < NDADDR; i++)
487 		if (newblks[i] != DIP(ip, i_db[i]))
488 			panic("ffs_truncate2");
489 	VI_LOCK(vp);
490 	if (length == 0 &&
491 	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
492 	    (vp->v_bufobj.bo_dirty.bv_cnt > 0 ||
493 	     vp->v_bufobj.bo_clean.bv_cnt > 0))
494 		panic("ffs_truncate3");
495 	VI_UNLOCK(vp);
496 #endif /* INVARIANTS */
497 	/*
498 	 * Put back the real size.
499 	 */
500 	ip->i_size = length;
501 	DIP_SET(ip, i_size, length);
502 	DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
503 
504 	if (DIP(ip, i_blocks) < 0)			/* sanity */
505 		DIP_SET(ip, i_blocks, 0);
506 	ip->i_flag |= IN_CHANGE;
507 #ifdef QUOTA
508 	(void) chkdq(ip, -blocksreleased, NOCRED, 0);
509 #endif
510 	return (allerror);
511 }
512 
513 /*
514  * Release blocks associated with the inode ip and stored in the indirect
515  * block bn.  Blocks are free'd in LIFO order up to (but not including)
516  * lastbn.  If level is greater than SINGLE, the block is an indirect block
517  * and recursive calls to indirtrunc must be used to cleanse other indirect
518  * blocks.
519  */
520 static int
521 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
522 	struct inode *ip;
523 	ufs2_daddr_t lbn, lastbn;
524 	ufs2_daddr_t dbn;
525 	int level;
526 	ufs2_daddr_t *countp;
527 {
528 	struct buf *bp;
529 	struct fs *fs = ip->i_fs;
530 	struct vnode *vp;
531 	caddr_t copy = NULL;
532 	int i, nblocks, error = 0, allerror = 0;
533 	ufs2_daddr_t nb, nlbn, last;
534 	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
535 	ufs1_daddr_t *bap1 = NULL;
536 	ufs2_daddr_t *bap2 = NULL;
537 #	define BAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? bap1[i] : bap2[i])
538 
539 	/*
540 	 * Calculate index in current block of last
541 	 * block to be kept.  -1 indicates the entire
542 	 * block so we need not calculate the index.
543 	 */
544 	factor = 1;
545 	for (i = SINGLE; i < level; i++)
546 		factor *= NINDIR(fs);
547 	last = lastbn;
548 	if (lastbn > 0)
549 		last /= factor;
550 	nblocks = btodb(fs->fs_bsize);
551 	/*
552 	 * Get buffer of block pointers, zero those entries corresponding
553 	 * to blocks to be free'd, and update on disk copy first.  Since
554 	 * double(triple) indirect before single(double) indirect, calls
555 	 * to bmap on these blocks will fail.  However, we already have
556 	 * the on disk address, so we have to set the b_blkno field
557 	 * explicitly instead of letting bread do everything for us.
558 	 */
559 	vp = ITOV(ip);
560 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
561 	if ((bp->b_flags & B_CACHE) == 0) {
562 		curthread->td_ru.ru_inblock++;	/* pay for read */
563 		bp->b_iocmd = BIO_READ;
564 		bp->b_flags &= ~B_INVAL;
565 		bp->b_ioflags &= ~BIO_ERROR;
566 		if (bp->b_bcount > bp->b_bufsize)
567 			panic("ffs_indirtrunc: bad buffer size");
568 		bp->b_blkno = dbn;
569 		vfs_busy_pages(bp, 0);
570 		bp->b_iooffset = dbtob(bp->b_blkno);
571 		bstrategy(bp);
572 		error = bufwait(bp);
573 	}
574 	if (error) {
575 		brelse(bp);
576 		*countp = 0;
577 		return (error);
578 	}
579 
580 	if (ip->i_ump->um_fstype == UFS1)
581 		bap1 = (ufs1_daddr_t *)bp->b_data;
582 	else
583 		bap2 = (ufs2_daddr_t *)bp->b_data;
584 	if (lastbn != -1) {
585 		MALLOC(copy, caddr_t, fs->fs_bsize, M_TEMP, M_WAITOK);
586 		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
587 		for (i = last + 1; i < NINDIR(fs); i++)
588 			if (ip->i_ump->um_fstype == UFS1)
589 				bap1[i] = 0;
590 			else
591 				bap2[i] = 0;
592 		if (DOINGASYNC(vp)) {
593 			bawrite(bp);
594 		} else {
595 			error = bwrite(bp);
596 			if (error)
597 				allerror = error;
598 		}
599 		if (ip->i_ump->um_fstype == UFS1)
600 			bap1 = (ufs1_daddr_t *)copy;
601 		else
602 			bap2 = (ufs2_daddr_t *)copy;
603 	}
604 
605 	/*
606 	 * Recursively free totally unused blocks.
607 	 */
608 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
609 	    i--, nlbn += factor) {
610 		nb = BAP(ip, i);
611 		if (nb == 0)
612 			continue;
613 		if (level > SINGLE) {
614 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
615 			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
616 				allerror = error;
617 			blocksreleased += blkcount;
618 		}
619 		ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
620 		    ip->i_number);
621 		blocksreleased += nblocks;
622 	}
623 
624 	/*
625 	 * Recursively free last partial block.
626 	 */
627 	if (level > SINGLE && lastbn >= 0) {
628 		last = lastbn % factor;
629 		nb = BAP(ip, i);
630 		if (nb != 0) {
631 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
632 			    last, level - 1, &blkcount);
633 			if (error)
634 				allerror = error;
635 			blocksreleased += blkcount;
636 		}
637 	}
638 	if (copy != NULL) {
639 		FREE(copy, M_TEMP);
640 	} else {
641 		bp->b_flags |= B_INVAL | B_NOCACHE;
642 		brelse(bp);
643 	}
644 
645 	*countp = blocksreleased;
646 	return (allerror);
647 }
648