xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision 17306b8fd0952c7489f93f0230818481e5a1e2c9)
1 /*	$NetBSD: lfs_inode.c,v 1.51 2001/05/30 11:57:18 mrg Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *      This product includes software developed by the NetBSD
21  *      Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*
39  * Copyright (c) 1986, 1989, 1991, 1993
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
71  */
72 
73 #if defined(_KERNEL_OPT)
74 #include "opt_quota.h"
75 #endif
76 
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/mount.h>
80 #include <sys/proc.h>
81 #include <sys/file.h>
82 #include <sys/buf.h>
83 #include <sys/vnode.h>
84 #include <sys/kernel.h>
85 #include <sys/malloc.h>
86 #include <sys/trace.h>
87 #include <sys/resourcevar.h>
88 
89 #include <ufs/ufs/quota.h>
90 #include <ufs/ufs/inode.h>
91 #include <ufs/ufs/ufsmount.h>
92 #include <ufs/ufs/ufs_extern.h>
93 
94 #include <ufs/lfs/lfs.h>
95 #include <ufs/lfs/lfs_extern.h>
96 
97 extern int locked_queue_count;
98 extern long locked_queue_bytes;
99 
100 static int lfs_update_seguse(struct lfs *, long, size_t);
101 static int lfs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t,
102 			   ufs_daddr_t, int, long *, long *, long *, size_t *,
103 			   struct proc *);
104 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *);
105 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
106 
107 /* Search a block for a specific dinode. */
108 struct dinode *
109 lfs_ifind(fs, ino, bp)
110 	struct lfs *fs;
111 	ino_t ino;
112 	struct buf *bp;
113 {
114 	int cnt;
115 	struct dinode *dip = (struct dinode *)bp->b_data;
116 	struct dinode *ldip;
117 
118 	for (cnt = INOPB(fs), ldip = dip + (cnt - 1); cnt--; --ldip)
119 		if (ldip->di_inumber == ino)
120 			return (ldip);
121 
122 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
123 	       datosn(fs,fs->lfs_offset));
124 	printf("block is 0x%x (seg %d)\n", bp->b_blkno,
125 	       datosn(fs,bp->b_blkno));
126 	panic("lfs_ifind: dinode %u not found", ino);
127 	/* NOTREACHED */
128 }
129 
130 int
131 lfs_update(v)
132 	void *v;
133 {
134 	struct vop_update_args /* {
135 				  struct vnode *a_vp;
136 				  struct timespec *a_access;
137 				  struct timespec *a_modify;
138 				  int a_flags;
139 				  } */ *ap = v;
140 	struct inode *ip;
141 	struct vnode *vp = ap->a_vp;
142 	int oflag;
143 	struct timespec ts;
144 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
145 
146 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
147 		return (0);
148 	ip = VTOI(vp);
149 
150 	/*
151 	 * If we are called from vinvalbuf, and the file's blocks have
152 	 * already been scheduled for writing, but the writes have not
153 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
154 	 * will cause a panic.  So, we must wait until any pending write
155 	 * for our inode completes, if we are called with UPDATE_WAIT set.
156 	 */
157 	while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
158 	    WRITEINPROG(vp)) {
159 #ifdef DEBUG_LFS
160 		printf("lfs_update: sleeping on inode %d (in-progress)\n",
161 		       ip->i_number);
162 #endif
163 		tsleep(vp, (PRIBIO+1), "lfs_update", 0);
164 	}
165 	oflag = ip->i_flag;
166 	TIMEVAL_TO_TIMESPEC(&time, &ts);
167 	LFS_ITIMES(ip,
168 		   ap->a_access ? ap->a_access : &ts,
169 		   ap->a_modify ? ap->a_modify : &ts, &ts);
170 	if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) {
171 		return (0);
172 	}
173 
174 	/* If sync, push back the vnode and any dirty blocks it may have. */
175 	if((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP))==UPDATE_WAIT) {
176 		/* Avoid flushing VDIROP. */
177 		++fs->lfs_diropwait;
178 		while(vp->v_flag & VDIROP) {
179 #ifdef DEBUG_LFS
180 			printf("lfs_update: sleeping on inode %d (dirops)\n",
181 			       ip->i_number);
182 			printf("lfs_update: vflags 0x%lx, iflags 0x%x\n",
183 				vp->v_flag, ip->i_flag);
184 #endif
185 			if(fs->lfs_dirops == 0)
186 				lfs_flush_fs(fs, SEGM_SYNC);
187 			else
188 				tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
189 				       0);
190 			/* XXX KS - by falling out here, are we writing the vn
191 			twice? */
192 		}
193 		--fs->lfs_diropwait;
194 		return lfs_vflush(vp);
195         }
196 	return 0;
197 }
198 
199 #define	SINGLE	0	/* index of single indirect block */
200 #define	DOUBLE	1	/* index of double indirect block */
201 #define	TRIPLE	2	/* index of triple indirect block */
202 /*
203  * Truncate the inode oip to at most length size, freeing the
204  * disk blocks.
205  */
206 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */
207 int
208 lfs_truncate(v)
209 	void *v;
210 {
211 	struct vop_truncate_args /* {
212 		struct vnode *a_vp;
213 		off_t a_length;
214 		int a_flags;
215 		struct ucred *a_cred;
216 		struct proc *a_p;
217 	} */ *ap = v;
218 	struct vnode *ovp = ap->a_vp;
219 	ufs_daddr_t lastblock;
220 	struct inode *oip;
221 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
222 	ufs_daddr_t newblks[NDADDR + NIADDR];
223 	off_t length = ap->a_length;
224 	struct lfs *fs;
225 	struct buf *bp;
226 	int offset, size, level;
227 	long count, rcount, nblocks, blocksreleased = 0, real_released = 0;
228 	int i;
229 	int aflags, error, allerror = 0;
230 	off_t osize;
231 	long lastseg;
232 	size_t bc;
233 	int obufsize, odb;
234 
235 	if (length < 0)
236 		return (EINVAL);
237 	oip = VTOI(ovp);
238 
239 	/*
240 	 * Just return and not update modification times.
241 	 */
242 	if (oip->i_ffs_size == length)
243 		return (0);
244 
245 	if (ovp->v_type == VLNK &&
246 	    (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen ||
247 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
248 	      oip->i_din.ffs_din.di_blocks == 0))) {
249 #ifdef DIAGNOSTIC
250 		if (length != 0)
251 			panic("lfs_truncate: partial truncate of symlink");
252 #endif
253 		memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size);
254 		oip->i_ffs_size = 0;
255 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
256 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
257 	}
258 	if (oip->i_ffs_size == length) {
259 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
260 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
261 	}
262 #ifdef QUOTA
263 	if ((error = getinoquota(oip)) != 0)
264 		return (error);
265 #endif
266 	fs = oip->i_lfs;
267 	lfs_imtime(fs);
268 	osize = oip->i_ffs_size;
269 	ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
270 
271 	/*
272 	 * Lengthen the size of the file. We must ensure that the
273 	 * last byte of the file is allocated. Since the smallest
274 	 * value of osize is 0, length will be at least 1.
275 	 */
276 	if (osize < length) {
277 		if (length > fs->lfs_maxfilesize)
278 			return (EFBIG);
279 		aflags = B_CLRBUF;
280 		if (ap->a_flags & IO_SYNC)
281 			aflags |= B_SYNC;
282 		error = lfs_reserve(fs, ovp, fsbtodb(fs, NIADDR + 2));
283 		if (error)
284 			return (error);
285 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
286 		lfs_reserve(fs, ovp, -fsbtodb(fs, NIADDR + 2));
287 		if (error)
288 			return (error);
289 		oip->i_ffs_size = length;
290 		uvm_vnp_setsize(ovp, length);
291 		(void) VOP_BWRITE(bp);
292 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
293 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
294 	}
295 
296 	if ((error = lfs_reserve(fs, ovp, fsbtodb(fs, 2 * NIADDR + 3))) != 0)
297 		return (error);
298 	/*
299 	 * Make sure no writes to this inode can happen while we're
300 	 * truncating.  Otherwise, blocks which are accounted for on the
301 	 * inode *and* which have been created for cleaning can coexist,
302 	 * and cause an overcounting.
303 	 *
304 	 * (We don't need to *hold* the seglock, though, because we already
305 	 * hold the inode lock; draining the seglock is sufficient.)
306 	 */
307 	if (ovp != fs->lfs_unlockvp) {
308 		while(fs->lfs_seglock) {
309 			tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0);
310 		}
311 	}
312 
313 	/*
314 	 * Shorten the size of the file. If the file is not being
315 	 * truncated to a block boundary, the contents of the
316 	 * partial block following the end of the file must be
317 	 * zero'ed in case it ever becomes accessible again because
318 	 * of subsequent file growth. Directories however are not
319 	 * zero'ed as they should grow back initialized to empty.
320 	 */
321 	offset = blkoff(fs, length);
322 	lastseg = -1;
323 	bc = 0;
324 	if (offset == 0) {
325 		oip->i_ffs_size = length;
326 	} else {
327 		lbn = lblkno(fs, length);
328 		aflags = B_CLRBUF;
329 		if (ap->a_flags & IO_SYNC)
330 			aflags |= B_SYNC;
331 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
332 		if (error) {
333 			lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3));
334 			return (error);
335 		}
336 		obufsize = bp->b_bufsize;
337 		odb = btodb(bp->b_bcount);
338 		oip->i_ffs_size = length;
339 		size = blksize(fs, oip, lbn);
340 		if (ovp->v_type != VDIR)
341 			memset((char *)bp->b_data + offset, 0,
342 			       (u_int)(size - offset));
343 		allocbuf(bp, size);
344 		if (bp->b_flags & B_DELWRI) {
345 			if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED)
346 				locked_queue_bytes -= obufsize - bp->b_bufsize;
347 			fs->lfs_avail += odb - btodb(size);
348 		}
349 		(void) VOP_BWRITE(bp);
350 	}
351 	uvm_vnp_setsize(ovp, length);
352 	/*
353 	 * Calculate index into inode's block list of
354 	 * last direct and indirect blocks (if any)
355 	 * which we want to keep.  Lastblock is -1 when
356 	 * the file is truncated to 0.
357 	 */
358 	lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
359 	lastiblock[SINGLE] = lastblock - NDADDR;
360 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
361 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
362 	nblocks = btodb(fs->lfs_bsize);
363 	/*
364 	 * Record changed file and block pointers before we start
365 	 * freeing blocks.  lastiblock values are also normalized to -1
366 	 * for calls to lfs_indirtrunc below.
367 	 */
368 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks);
369 	for (level = TRIPLE; level >= SINGLE; level--)
370 		if (lastiblock[level] < 0) {
371 			newblks[NDADDR+level] = 0;
372 			lastiblock[level] = -1;
373 		}
374 	for (i = NDADDR - 1; i > lastblock; i--)
375 		newblks[i] = 0;
376 
377 	oip->i_ffs_size = osize;
378 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
379 	if (error && !allerror)
380 		allerror = error;
381 
382 	/*
383 	 * Indirect blocks first.
384 	 */
385 	indir_lbn[SINGLE] = -NDADDR;
386 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
387 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
388 	for (level = TRIPLE; level >= SINGLE; level--) {
389 		bn = oip->i_ffs_ib[level];
390 		if (bn != 0) {
391 			error = lfs_indirtrunc(oip, indir_lbn[level],
392 					       bn, lastiblock[level],
393 					       level, &count, &rcount,
394 					       &lastseg, &bc, ap->a_p);
395 			if (error)
396 				allerror = error;
397 			real_released += rcount;
398 			blocksreleased += count;
399 			if (lastiblock[level] < 0) {
400 				if (oip->i_ffs_ib[level] > 0)
401 					real_released += nblocks;
402 				blocksreleased += nblocks;
403 				oip->i_ffs_ib[level] = 0;
404 				lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc);
405 			}
406 		}
407 		if (lastiblock[level] >= 0)
408 			goto done;
409 	}
410 
411 	/*
412 	 * All whole direct blocks or frags.
413 	 */
414 	for (i = NDADDR - 1; i > lastblock; i--) {
415 		long bsize;
416 
417 		bn = oip->i_ffs_db[i];
418 		if (bn == 0)
419 			continue;
420 		bsize = blksize(fs, oip, i);
421 		if (oip->i_ffs_db[i] > 0)
422 			real_released += btodb(bsize);
423 		blocksreleased += btodb(bsize);
424 		oip->i_ffs_db[i] = 0;
425 		lfs_blkfree(fs, bn, bsize, &lastseg, &bc);
426 	}
427 	if (lastblock < 0)
428 		goto done;
429 
430 	/*
431 	 * Finally, look for a change in size of the
432 	 * last direct block; release any frags.
433 	 */
434 	bn = oip->i_ffs_db[lastblock];
435 	if (bn != 0) {
436 		long oldspace, newspace;
437 
438 		/*
439 		 * Calculate amount of space we're giving
440 		 * back as old block size minus new block size.
441 		 */
442 		oldspace = blksize(fs, oip, lastblock);
443 		oip->i_ffs_size = length;
444 		newspace = blksize(fs, oip, lastblock);
445 		if (newspace == 0)
446 			panic("itrunc: newspace");
447 		if (oldspace - newspace > 0) {
448 			lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc);
449 			if (bn > 0)
450 				real_released += btodb(oldspace - newspace);
451 			blocksreleased += btodb(oldspace - newspace);
452 		}
453 	}
454 
455 done:
456 	/* Finish segment accounting corrections */
457 	lfs_update_seguse(fs, lastseg, bc);
458 #ifdef DIAGNOSTIC
459 	for (level = SINGLE; level <= TRIPLE; level++)
460 		if (newblks[NDADDR + level] != oip->i_ffs_ib[level])
461 			panic("lfs itrunc1");
462 	for (i = 0; i < NDADDR; i++)
463 		if (newblks[i] != oip->i_ffs_db[i])
464 			panic("lfs itrunc2");
465 	if (length == 0 &&
466 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
467 		panic("lfs itrunc3");
468 #endif /* DIAGNOSTIC */
469 	/*
470 	 * Put back the real size.
471 	 */
472 	oip->i_ffs_size = length;
473 	oip->i_lfs_effnblks -= blocksreleased;
474 	oip->i_ffs_blocks -= real_released;
475 	fs->lfs_bfree += blocksreleased;
476 #ifdef DIAGNOSTIC
477 	if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) {
478 		printf("lfs_truncate: truncate to 0 but %d blocks on inode\n",
479 		       oip->i_ffs_blocks);
480 		panic("lfs_truncate: persistent blocks\n");
481 	}
482 #endif
483 	oip->i_flag |= IN_CHANGE;
484 #ifdef QUOTA
485 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
486 #endif
487 	lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3));
488 	return (allerror);
489 }
490 
491 /* Update segment usage information when removing a block. */
492 static int
493 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg,
494 	    size_t *num)
495 {
496 	long seg;
497 	int error = 0;
498 
499 	bsize = fragroundup(fs, bsize);
500 	if (daddr > 0) {
501 		if (*lastseg != (seg = datosn(fs, daddr))) {
502 			error = lfs_update_seguse(fs, *lastseg, *num);
503 			*num = bsize;
504 			*lastseg = seg;
505 		} else
506 			*num += bsize;
507 	}
508 	return error;
509 }
510 
511 /* Finish the accounting updates for a segment. */
512 static int
513 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num)
514 {
515 	SEGUSE *sup;
516 	struct buf *bp;
517 
518 	if (lastseg < 0 || num == 0)
519 		return 0;
520 
521 
522 	LFS_SEGENTRY(sup, fs, lastseg, bp);
523 	if (num > sup->su_nbytes) {
524 		printf("lfs_truncate: segment %ld short by %ld\n",
525 		       lastseg, (long)num - sup->su_nbytes);
526 		panic("lfs_truncate: negative bytes");
527 		sup->su_nbytes = num;
528 	}
529 	sup->su_nbytes -= num;
530 	return (VOP_BWRITE(bp)); /* Ifile */
531 }
532 
533 /*
534  * Release blocks associated with the inode ip and stored in the indirect
535  * block bn.  Blocks are free'd in LIFO order up to (but not including)
536  * lastbn.  If level is greater than SINGLE, the block is an indirect block
537  * and recursive calls to indirtrunc must be used to cleanse other indirect
538  * blocks.
539  *
540  * NB: triple indirect blocks are untested.
541  */
542 static int
543 lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn,
544 	       ufs_daddr_t lastbn, int level, long *countp,
545 	       long *rcountp, long *lastsegp, size_t *bcp, struct proc *p)
546 {
547 	int i;
548 	struct buf *bp;
549 	struct lfs *fs = ip->i_lfs;
550 	ufs_daddr_t *bap;
551 	struct vnode *vp;
552 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
553 	long blkcount, rblkcount, factor;
554 	int nblocks, blocksreleased = 0, real_released = 0;
555 	int error = 0, allerror = 0;
556 
557 	/*
558 	 * Calculate index in current block of last
559 	 * block to be kept.  -1 indicates the entire
560 	 * block so we need not calculate the index.
561 	 */
562 	factor = 1;
563 	for (i = SINGLE; i < level; i++)
564 		factor *= NINDIR(fs);
565 	last = lastbn;
566 	if (lastbn > 0)
567 		last /= factor;
568 	nblocks = btodb(fs->lfs_bsize);
569 	/*
570 	 * Get buffer of block pointers, zero those entries corresponding
571 	 * to blocks to be free'd, and update on disk copy first.  Since
572 	 * double(triple) indirect before single(double) indirect, calls
573 	 * to bmap on these blocks will fail.  However, we already have
574 	 * the on disk address, so we have to set the b_blkno field
575 	 * explicitly instead of letting bread do everything for us.
576 	 */
577 	vp = ITOV(ip);
578 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
579 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
580 		/* Braces must be here in case trace evaluates to nothing. */
581 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
582 	} else {
583 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
584 		p->p_stats->p_ru.ru_inblock++;	/* pay for read */
585 		bp->b_flags |= B_READ;
586 		if (bp->b_bcount > bp->b_bufsize)
587 			panic("lfs_indirtrunc: bad buffer size");
588 		bp->b_blkno = dbn;
589 		VOP_STRATEGY(bp);
590 		error = biowait(bp);
591 	}
592 	if (error) {
593 		brelse(bp);
594 		*countp = *rcountp = 0;
595 		return (error);
596 	}
597 
598 	bap = (ufs_daddr_t *)bp->b_data;
599 	if (lastbn >= 0) {
600 		MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK);
601 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
602 		memset((caddr_t)&bap[last + 1], 0,
603 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
604 		error = VOP_BWRITE(bp);
605 		if (error)
606 			allerror = error;
607 		bap = copy;
608 	}
609 
610 	/*
611 	 * Recursively free totally unused blocks.
612 	 */
613 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
614 	    i--, nlbn += factor) {
615 		nb = bap[i];
616 		if (nb == 0)
617 			continue;
618 		if (level > SINGLE) {
619 			error = lfs_indirtrunc(ip, nlbn, nb,
620 					       (ufs_daddr_t)-1, level - 1,
621 					       &blkcount, &rblkcount,
622 					       lastsegp, bcp, p);
623 			if (error)
624 				allerror = error;
625 			blocksreleased += blkcount;
626 			real_released += rblkcount;
627 		}
628 		lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp);
629 		if (bap[i] > 0)
630 			real_released += nblocks;
631 		blocksreleased += nblocks;
632 	}
633 
634 	/*
635 	 * Recursively free last partial block.
636 	 */
637 	if (level > SINGLE && lastbn >= 0) {
638 		last = lastbn % factor;
639 		nb = bap[i];
640 		if (nb != 0) {
641 			error = lfs_indirtrunc(ip, nlbn, nb,
642 					       last, level - 1, &blkcount,
643 					       &rblkcount, lastsegp, bcp, p);
644 			if (error)
645 				allerror = error;
646 			real_released += rblkcount;
647 			blocksreleased += blkcount;
648 		}
649 	}
650 
651 	if (copy != NULL) {
652 		FREE(copy, M_TEMP);
653 	} else {
654 		if (bp->b_flags & B_DELWRI) {
655 			LFS_UNLOCK_BUF(bp);
656 			fs->lfs_avail += btodb(bp->b_bcount);
657 			wakeup(&fs->lfs_avail);
658 		}
659 		bp->b_flags |= B_INVAL;
660 		brelse(bp);
661 	}
662 
663 	*countp = blocksreleased;
664 	*rcountp = real_released;
665 	return (allerror);
666 }
667 
668 /*
669  * Destroy any in core blocks past the truncation length.
670  * Inlined from vtruncbuf, so that lfs_avail could be updated.
671  */
672 static int
673 lfs_vtruncbuf(vp, lbn, slpflag, slptimeo)
674 	struct vnode *vp;
675 	daddr_t lbn;
676 	int slpflag, slptimeo;
677 {
678 	struct buf *bp, *nbp;
679 	int s, error;
680 	struct lfs *fs;
681 
682 	fs = VTOI(vp)->i_lfs;
683 	s = splbio();
684 
685 restart:
686 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
687 		nbp = LIST_NEXT(bp, b_vnbufs);
688 		if (bp->b_lblkno < lbn)
689 			continue;
690 		if (bp->b_flags & B_BUSY) {
691 			bp->b_flags |= B_WANTED;
692 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
693 			    "lfs_vtruncbuf", slptimeo);
694 			if (error) {
695 				splx(s);
696 				return (error);
697 			}
698 			goto restart;
699 		}
700 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
701 		if (bp->b_flags & B_DELWRI) {
702 			bp->b_flags &= ~B_DELWRI;
703 			fs->lfs_avail += btodb(bp->b_bcount);
704 			wakeup(&fs->lfs_avail);
705 		}
706 		LFS_UNLOCK_BUF(bp);
707 		brelse(bp);
708 	}
709 
710 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
711 		nbp = LIST_NEXT(bp, b_vnbufs);
712 		if (bp->b_lblkno < lbn)
713 			continue;
714 		if (bp->b_flags & B_BUSY) {
715 			bp->b_flags |= B_WANTED;
716 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
717 			    "lfs_vtruncbuf", slptimeo);
718 			if (error) {
719 				splx(s);
720 				return (error);
721 			}
722 			goto restart;
723 		}
724 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
725 		if (bp->b_flags & B_DELWRI) {
726 			bp->b_flags &= ~B_DELWRI;
727 			fs->lfs_avail += btodb(bp->b_bcount);
728 			wakeup(&fs->lfs_avail);
729 		}
730 		LFS_UNLOCK_BUF(bp);
731 		brelse(bp);
732 	}
733 
734 	splx(s);
735 
736 	return (0);
737 }
738 
739