xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision bf1e9b32e27832f0c493206710fb8b58a980838a)
1 /*	$NetBSD: lfs_inode.c,v 1.96 2005/05/29 21:25:24 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 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. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.96 2005/05/29 21:25:24 christos Exp $");
71 
72 #if defined(_KERNEL_OPT)
73 #include "opt_quota.h"
74 #endif
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/mount.h>
79 #include <sys/proc.h>
80 #include <sys/file.h>
81 #include <sys/buf.h>
82 #include <sys/vnode.h>
83 #include <sys/kernel.h>
84 #include <sys/trace.h>
85 #include <sys/resourcevar.h>
86 
87 #include <ufs/ufs/quota.h>
88 #include <ufs/ufs/inode.h>
89 #include <ufs/ufs/ufsmount.h>
90 #include <ufs/ufs/ufs_extern.h>
91 
92 #include <ufs/lfs/lfs.h>
93 #include <ufs/lfs/lfs_extern.h>
94 
95 static int lfs_update_seguse(struct lfs *, long, size_t);
96 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t,
97 			   daddr_t, int, long *, long *, long *, size_t *,
98 			   struct proc *);
99 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *);
100 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
101 
102 /* Search a block for a specific dinode. */
103 struct ufs1_dinode *
104 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
105 {
106 	struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data;
107 	struct ufs1_dinode *ldip, *fin;
108 
109 	ASSERT_NO_SEGLOCK(fs);
110 	/*
111 	 * Read the inode block backwards, since later versions of the
112 	 * inode will supercede earlier ones.  Though it is unlikely, it is
113 	 * possible that the same inode will appear in the same inode block.
114 	 */
115 	fin = dip + INOPB(fs);
116 	for (ldip = fin - 1; ldip >= dip; --ldip)
117 		if (ldip->di_inumber == ino)
118 			return (ldip);
119 
120 	printf("searched %d entries\n", (int)(fin - dip));
121 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
122 	       dtosn(fs, fs->lfs_offset));
123 	printf("block is 0x%llx (seg %lld)\n",
124 	       (unsigned long long)dbtofsb(fs, bp->b_blkno),
125 	       (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno)));
126 
127 	return NULL;
128 }
129 
130 int
131 lfs_update(void *v)
132 {
133 	struct vop_update_args /* {
134 				  struct vnode *a_vp;
135 				  struct timespec *a_access;
136 				  struct timespec *a_modify;
137 				  int a_flags;
138 				  } */ *ap = v;
139 	struct inode *ip;
140 	struct vnode *vp = ap->a_vp;
141 	struct timespec ts;
142 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
143 	int s;
144 	int flags;
145 
146 	ASSERT_NO_SEGLOCK(fs);
147 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
148 		return (0);
149 	ip = VTOI(vp);
150 
151 	/*
152 	 * If we are called from vinvalbuf, and the file's blocks have
153 	 * already been scheduled for writing, but the writes have not
154 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
155 	 * will cause a panic.	So, we must wait until any pending write
156 	 * for our inode completes, if we are called with UPDATE_WAIT set.
157 	 */
158 	s = splbio();
159 	simple_lock(&vp->v_interlock);
160 	while ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
161 	    WRITEINPROG(vp)) {
162 		DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d"
163 		      " (in progress)\n", ip->i_number));
164 		ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock);
165 	}
166 	simple_unlock(&vp->v_interlock);
167 	splx(s);
168 	TIMEVAL_TO_TIMESPEC(&time, &ts);
169 	LFS_ITIMES(ip,
170 		   ap->a_access ? ap->a_access : &ts,
171 		   ap->a_modify ? ap->a_modify : &ts, &ts);
172 	if (ap->a_flags & UPDATE_CLOSE)
173 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING);
174 	else
175 		flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING);
176 	if (flags == 0)
177 		return (0);
178 
179 	/* If sync, push back the vnode and any dirty blocks it may have. */
180 	if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) {
181 		/* Avoid flushing VDIROP. */
182 		simple_lock(&fs->lfs_interlock);
183 		++fs->lfs_diropwait;
184 		while (vp->v_flag & VDIROP) {
185 			DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d"
186 			      " (dirops)\n", ip->i_number));
187 			DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags"
188 			      " 0x%x\n", vp->v_flag, ip->i_flag));
189 			if (fs->lfs_dirops == 0)
190 				lfs_flush_fs(fs, SEGM_SYNC);
191 			else
192 				ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
193 					0, &fs->lfs_interlock);
194 			/* XXX KS - by falling out here, are we writing the vn
195 			twice? */
196 		}
197 		--fs->lfs_diropwait;
198 		simple_unlock(&fs->lfs_interlock);
199 		return lfs_vflush(vp);
200 	}
201 	return 0;
202 }
203 
204 #define	SINGLE	0	/* index of single indirect block */
205 #define	DOUBLE	1	/* index of double indirect block */
206 #define	TRIPLE	2	/* index of triple indirect block */
207 /*
208  * Truncate the inode oip to at most length size, freeing the
209  * disk blocks.
210  */
211 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */
212 
213 int
214 lfs_truncate(void *v)
215 {
216 	struct vop_truncate_args /* {
217 		struct vnode *a_vp;
218 		off_t a_length;
219 		int a_flags;
220 		struct ucred *a_cred;
221 		struct proc *a_p;
222 	} */ *ap = v;
223 	struct vnode *ovp = ap->a_vp;
224 	struct genfs_node *gp = VTOG(ovp);
225 	daddr_t lastblock;
226 	struct inode *oip = VTOI(ovp);
227 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
228 	/* XXX ondisk32 */
229 	int32_t newblks[NDADDR + NIADDR];
230 	off_t length = ap->a_length;
231 	struct lfs *fs;
232 	struct buf *bp;
233 	int offset, size, level;
234 	long count, rcount, blocksreleased = 0, real_released = 0;
235 	int i, ioflag, nblocks;
236 	int aflags, error, allerror = 0;
237 	off_t osize;
238 	long lastseg;
239 	size_t bc;
240 	int obufsize, odb;
241 	int usepc;
242 	struct ufsmount *ump = oip->i_ump;
243 
244 	if (length < 0)
245 		return (EINVAL);
246 
247 	/*
248 	 * Just return and not update modification times.
249 	 */
250 	if (oip->i_size == length)
251 		return (0);
252 
253 	if (ovp->v_type == VLNK &&
254 	    (oip->i_size < ump->um_maxsymlinklen ||
255 	     (ump->um_maxsymlinklen == 0 &&
256 	      oip->i_ffs1_blocks == 0))) {
257 #ifdef DIAGNOSTIC
258 		if (length != 0)
259 			panic("lfs_truncate: partial truncate of symlink");
260 #endif
261 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
262 		oip->i_size = oip->i_ffs1_size = 0;
263 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
264 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
265 	}
266 	if (oip->i_size == length) {
267 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
268 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
269 	}
270 #ifdef QUOTA
271 	if ((error = getinoquota(oip)) != 0)
272 		return (error);
273 #endif
274 	fs = oip->i_lfs;
275 	lfs_imtime(fs);
276 	osize = oip->i_size;
277 	ioflag = ap->a_flags;
278 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
279 
280 	ASSERT_NO_SEGLOCK(fs);
281 	/*
282 	 * Lengthen the size of the file. We must ensure that the
283 	 * last byte of the file is allocated. Since the smallest
284 	 * value of osize is 0, length will be at least 1.
285 	 */
286 	if (osize < length) {
287 		if (length > ump->um_maxfilesize)
288 			return (EFBIG);
289 		aflags = B_CLRBUF;
290 		if (ioflag & IO_SYNC)
291 			aflags |= B_SYNC;
292 		if (usepc) {
293 			if (lblkno(fs, osize) < NDADDR &&
294 			    lblkno(fs, osize) != lblkno(fs, length) &&
295 			    blkroundup(fs, osize) != osize) {
296 				off_t eob;
297 
298 				eob = blkroundup(fs, osize);
299 				error = ufs_balloc_range(ovp, osize,
300 				    eob - osize, ap->a_cred, aflags);
301 				if (error)
302 					return error;
303 				if (ioflag & IO_SYNC) {
304 					ovp->v_size = eob;
305 					simple_lock(&ovp->v_interlock);
306 					VOP_PUTPAGES(ovp,
307 					    trunc_page(osize & fs->lfs_bmask),
308 					    round_page(eob),
309 					    PGO_CLEANIT | PGO_SYNCIO);
310 				}
311 			}
312 			error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
313 						 aflags);
314 			if (error) {
315 				(void) VOP_TRUNCATE(ovp, osize,
316 						    ioflag & IO_SYNC,
317 				    		    ap->a_cred, ap->a_p);
318 				return error;
319 			}
320 			uvm_vnp_setsize(ovp, length);
321 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
322 			KASSERT(ovp->v_size == oip->i_size);
323 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
324 			return (VOP_UPDATE(ovp, NULL, NULL, 0));
325 		} else {
326 			error = lfs_reserve(fs, ovp, NULL,
327 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
328 			if (error)
329 				return (error);
330 			error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred,
331 					   aflags, &bp);
332 			lfs_reserve(fs, ovp, NULL,
333 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
334 			if (error)
335 				return (error);
336 			oip->i_ffs1_size = oip->i_size = length;
337 			uvm_vnp_setsize(ovp, length);
338 			(void) VOP_BWRITE(bp);
339 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
340 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
341 			return (VOP_UPDATE(ovp, NULL, NULL, 0));
342 		}
343 	}
344 
345 	if ((error = lfs_reserve(fs, ovp, NULL,
346 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
347 		return (error);
348 
349 	/*
350 	 * Shorten the size of the file. If the file is not being
351 	 * truncated to a block boundary, the contents of the
352 	 * partial block following the end of the file must be
353 	 * zero'ed in case it ever becomes accessible again because
354 	 * of subsequent file growth. Directories however are not
355 	 * zero'ed as they should grow back initialized to empty.
356 	 */
357 	offset = blkoff(fs, length);
358 	lastseg = -1;
359 	bc = 0;
360 
361 	if (ovp != fs->lfs_ivnode)
362 		lfs_seglock(fs, SEGM_PROT);
363 	if (offset == 0) {
364 		oip->i_size = oip->i_ffs1_size = length;
365 	} else if (!usepc) {
366 		lbn = lblkno(fs, length);
367 		aflags = B_CLRBUF;
368 		if (ioflag & IO_SYNC)
369 			aflags |= B_SYNC;
370 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
371 		if (error) {
372 			lfs_reserve(fs, ovp, NULL,
373 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
374 			goto errout;
375 		}
376 		obufsize = bp->b_bufsize;
377 		odb = btofsb(fs, bp->b_bcount);
378 		oip->i_size = oip->i_ffs1_size = length;
379 		size = blksize(fs, oip, lbn);
380 		if (ovp->v_type != VDIR)
381 			memset((char *)bp->b_data + offset, 0,
382 			       (u_int)(size - offset));
383 		allocbuf(bp, size, 1);
384 		if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) {
385 			simple_lock(&lfs_subsys_lock);
386 			locked_queue_bytes -= obufsize - bp->b_bufsize;
387 			simple_unlock(&lfs_subsys_lock);
388 		}
389 		if (bp->b_flags & B_DELWRI)
390 			fs->lfs_avail += odb - btofsb(fs, size);
391 		(void) VOP_BWRITE(bp);
392 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
393 		/*
394 		 * When truncating a regular file down to a non-block-aligned
395 		 * size, we must zero the part of last block which is past
396 		 * the new EOF.  We must synchronously flush the zeroed pages
397 		 * to disk since the new pages will be invalidated as soon
398 		 * as we inform the VM system of the new, smaller size.
399 		 * We must do this before acquiring the GLOCK, since fetching
400 		 * the pages will acquire the GLOCK internally.
401 		 * So there is a window where another thread could see a whole
402 		 * zeroed page past EOF, but that's life.
403 		 */
404 		daddr_t xlbn;
405 		voff_t eoz;
406 
407 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
408 		error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
409 		    aflags);
410 		if (error) {
411 			lfs_reserve(fs, ovp, NULL,
412 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
413 			goto errout;
414 		}
415 		xlbn = lblkno(fs, length);
416 		size = blksize(fs, oip, xlbn);
417 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
418 		uvm_vnp_zerorange(ovp, length, eoz - length);
419 		if (round_page(eoz) > round_page(length)) {
420 			simple_lock(&ovp->v_interlock);
421 			error = VOP_PUTPAGES(ovp, round_page(length),
422 			    round_page(eoz),
423 			    PGO_CLEANIT | PGO_DEACTIVATE |
424 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
425 			if (error) {
426 				lfs_reserve(fs, ovp, NULL,
427 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
428 				goto errout;
429 			}
430 		}
431 	}
432 
433 	lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
434 
435 	oip->i_size = oip->i_ffs1_size = length;
436 	uvm_vnp_setsize(ovp, length);
437 	/*
438 	 * Calculate index into inode's block list of
439 	 * last direct and indirect blocks (if any)
440 	 * which we want to keep.  Lastblock is -1 when
441 	 * the file is truncated to 0.
442 	 */
443 	lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
444 	lastiblock[SINGLE] = lastblock - NDADDR;
445 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
446 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
447 	nblocks = btofsb(fs, fs->lfs_bsize);
448 	/*
449 	 * Record changed file and block pointers before we start
450 	 * freeing blocks.  lastiblock values are also normalized to -1
451 	 * for calls to lfs_indirtrunc below.
452 	 */
453 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks);
454 	for (level = TRIPLE; level >= SINGLE; level--)
455 		if (lastiblock[level] < 0) {
456 			newblks[NDADDR+level] = 0;
457 			lastiblock[level] = -1;
458 		}
459 	for (i = NDADDR - 1; i > lastblock; i--)
460 		newblks[i] = 0;
461 
462 	oip->i_size = oip->i_ffs1_size = osize;
463 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
464 	if (error && !allerror)
465 		allerror = error;
466 
467 	/*
468 	 * Indirect blocks first.
469 	 */
470 	indir_lbn[SINGLE] = -NDADDR;
471 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
472 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
473 	for (level = TRIPLE; level >= SINGLE; level--) {
474 		bn = oip->i_ffs1_ib[level];
475 		if (bn != 0) {
476 			error = lfs_indirtrunc(oip, indir_lbn[level],
477 					       bn, lastiblock[level],
478 					       level, &count, &rcount,
479 					       &lastseg, &bc, ap->a_p);
480 			if (error)
481 				allerror = error;
482 			real_released += rcount;
483 			blocksreleased += count;
484 			if (lastiblock[level] < 0) {
485 				if (oip->i_ffs1_ib[level] > 0)
486 					real_released += nblocks;
487 				blocksreleased += nblocks;
488 				oip->i_ffs1_ib[level] = 0;
489 				lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc);
490         			lfs_deregister_block(ovp, bn);
491 			}
492 		}
493 		if (lastiblock[level] >= 0)
494 			goto done;
495 	}
496 
497 	/*
498 	 * All whole direct blocks or frags.
499 	 */
500 	for (i = NDADDR - 1; i > lastblock; i--) {
501 		long bsize, obsize;
502 
503 		bn = oip->i_ffs1_db[i];
504 		if (bn == 0)
505 			continue;
506 		bsize = blksize(fs, oip, i);
507 		if (oip->i_ffs1_db[i] > 0) {
508 			/* Check for fragment size changes */
509 			obsize = oip->i_lfs_fragsize[i];
510 			real_released += btofsb(fs, obsize);
511 			oip->i_lfs_fragsize[i] = 0;
512 		} else
513 			obsize = 0;
514 		blocksreleased += btofsb(fs, bsize);
515 		oip->i_ffs1_db[i] = 0;
516 		lfs_blkfree(fs, bn, obsize, &lastseg, &bc);
517         	lfs_deregister_block(ovp, bn);
518 	}
519 	if (lastblock < 0)
520 		goto done;
521 
522 	/*
523 	 * Finally, look for a change in size of the
524 	 * last direct block; release any frags.
525 	 */
526 	bn = oip->i_ffs1_db[lastblock];
527 	if (bn != 0) {
528 		long oldspace, newspace;
529 #if 0
530 		long olddspace;
531 #endif
532 
533 		/*
534 		 * Calculate amount of space we're giving
535 		 * back as old block size minus new block size.
536 		 */
537 		oldspace = blksize(fs, oip, lastblock);
538 #if 0
539 		olddspace = oip->i_lfs_fragsize[lastblock];
540 #endif
541 
542 		oip->i_size = oip->i_ffs1_size = length;
543 		newspace = blksize(fs, oip, lastblock);
544 		if (newspace == 0)
545 			panic("itrunc: newspace");
546 		if (oldspace - newspace > 0) {
547 			blocksreleased += btofsb(fs, oldspace - newspace);
548 		}
549 #if 0
550 		if (bn > 0 && olddspace - newspace > 0) {
551 			/* No segment accounting here, just vnode */
552 			real_released += btofsb(fs, olddspace - newspace);
553 		}
554 #endif
555 	}
556 
557 done:
558 	/* Finish segment accounting corrections */
559 	lfs_update_seguse(fs, lastseg, bc);
560 #ifdef DIAGNOSTIC
561 	for (level = SINGLE; level <= TRIPLE; level++)
562 		if ((newblks[NDADDR + level] == 0) !=
563 		    (oip->i_ffs1_ib[level]) == 0) {
564 			panic("lfs itrunc1");
565 		}
566 	for (i = 0; i < NDADDR; i++)
567 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
568 			panic("lfs itrunc2");
569 		}
570 	if (length == 0 &&
571 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
572 		panic("lfs itrunc3");
573 #endif /* DIAGNOSTIC */
574 	/*
575 	 * Put back the real size.
576 	 */
577 	oip->i_size = oip->i_ffs1_size = length;
578 	oip->i_lfs_effnblks -= blocksreleased;
579 	oip->i_ffs1_blocks -= real_released;
580 	simple_lock(&fs->lfs_interlock);
581 	fs->lfs_bfree += blocksreleased;
582 	simple_unlock(&fs->lfs_interlock);
583 #ifdef DIAGNOSTIC
584 	if (oip->i_size == 0 &&
585 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
586 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
587 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
588 		panic("lfs_truncate: persistent blocks");
589 	}
590 #endif
591 	oip->i_flag |= IN_CHANGE;
592 #ifdef QUOTA
593 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
594 #endif
595 	lfs_reserve(fs, ovp, NULL,
596 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
597 	lockmgr(&gp->g_glock, LK_RELEASE, NULL);
598   errout:
599 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
600 	if (ovp != fs->lfs_ivnode)
601 		lfs_segunlock(fs);
602 	return (allerror ? allerror : error);
603 }
604 
605 /* Update segment and avail usage information when removing a block. */
606 static int
607 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg,
608 	    size_t *num)
609 {
610 	long seg;
611 	int error = 0;
612 
613 	ASSERT_SEGLOCK(fs);
614 	bsize = fragroundup(fs, bsize);
615 	if (daddr > 0) {
616 		if (*lastseg != (seg = dtosn(fs, daddr))) {
617 			error = lfs_update_seguse(fs, *lastseg, *num);
618 			*num = bsize;
619 			*lastseg = seg;
620 		} else
621 			*num += bsize;
622 	}
623 
624 	return error;
625 }
626 
627 /* Finish the accounting updates for a segment. */
628 static int
629 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num)
630 {
631 	SEGUSE *sup;
632 	struct buf *bp;
633 
634 	ASSERT_SEGLOCK(fs);
635 	if (lastseg < 0 || num == 0)
636 		return 0;
637 
638 	LFS_SEGENTRY(sup, fs, lastseg, bp);
639 	if (num > sup->su_nbytes) {
640 		printf("lfs_truncate: segment %ld short by %ld\n",
641 		       lastseg, (long)num - sup->su_nbytes);
642 		panic("lfs_truncate: negative bytes");
643 		sup->su_nbytes = num;
644 	}
645 	sup->su_nbytes -= num;
646 	LFS_WRITESEGENTRY(sup, fs, lastseg, bp);
647 
648 	return 0;
649 }
650 
651 /*
652  * Release blocks associated with the inode ip and stored in the indirect
653  * block bn.  Blocks are free'd in LIFO order up to (but not including)
654  * lastbn.  If level is greater than SINGLE, the block is an indirect block
655  * and recursive calls to indirtrunc must be used to cleanse other indirect
656  * blocks.
657  *
658  * NB: triple indirect blocks are untested.
659  */
660 static int
661 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
662 	       daddr_t lastbn, int level, long *countp,
663 	       long *rcountp, long *lastsegp, size_t *bcp, struct proc *p)
664 {
665 	int i;
666 	struct buf *bp;
667 	struct lfs *fs = ip->i_lfs;
668 	int32_t *bap;	/* XXX ondisk32 */
669 	struct vnode *vp;
670 	daddr_t nb, nlbn, last;
671 	int32_t *copy = NULL;	/* XXX ondisk32 */
672 	long blkcount, rblkcount, factor;
673 	int nblocks, blocksreleased = 0, real_released = 0;
674 	int error = 0, allerror = 0;
675 
676 	ASSERT_SEGLOCK(fs);
677 	/*
678 	 * Calculate index in current block of last
679 	 * block to be kept.  -1 indicates the entire
680 	 * block so we need not calculate the index.
681 	 */
682 	factor = 1;
683 	for (i = SINGLE; i < level; i++)
684 		factor *= NINDIR(fs);
685 	last = lastbn;
686 	if (lastbn > 0)
687 		last /= factor;
688 	nblocks = btofsb(fs, fs->lfs_bsize);
689 	/*
690 	 * Get buffer of block pointers, zero those entries corresponding
691 	 * to blocks to be free'd, and update on disk copy first.  Since
692 	 * double(triple) indirect before single(double) indirect, calls
693 	 * to bmap on these blocks will fail.  However, we already have
694 	 * the on disk address, so we have to set the b_blkno field
695 	 * explicitly instead of letting bread do everything for us.
696 	 */
697 	vp = ITOV(ip);
698 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
699 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
700 		/* Braces must be here in case trace evaluates to nothing. */
701 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
702 	} else {
703 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
704 		p->p_stats->p_ru.ru_inblock++;	/* pay for read */
705 		bp->b_flags |= B_READ;
706 		if (bp->b_bcount > bp->b_bufsize)
707 			panic("lfs_indirtrunc: bad buffer size");
708 		bp->b_blkno = fsbtodb(fs, dbn);
709 		VOP_STRATEGY(vp, bp);
710 		error = biowait(bp);
711 	}
712 	if (error) {
713 		brelse(bp);
714 		*countp = *rcountp = 0;
715 		return (error);
716 	}
717 
718 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
719 	if (lastbn >= 0) {
720 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
721 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
722 		memset((caddr_t)&bap[last + 1], 0,
723 		/* XXX ondisk32 */
724 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
725 		error = VOP_BWRITE(bp);
726 		if (error)
727 			allerror = error;
728 		bap = copy;
729 	}
730 
731 	/*
732 	 * Recursively free totally unused blocks.
733 	 */
734 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
735 	    i--, nlbn += factor) {
736 		nb = bap[i];
737 		if (nb == 0)
738 			continue;
739 		if (level > SINGLE) {
740 			error = lfs_indirtrunc(ip, nlbn, nb,
741 					       (daddr_t)-1, level - 1,
742 					       &blkcount, &rblkcount,
743 					       lastsegp, bcp, p);
744 			if (error)
745 				allerror = error;
746 			blocksreleased += blkcount;
747 			real_released += rblkcount;
748 		}
749 		lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp);
750 		if (bap[i] > 0)
751 			real_released += nblocks;
752 		blocksreleased += nblocks;
753 	}
754 
755 	/*
756 	 * Recursively free last partial block.
757 	 */
758 	if (level > SINGLE && lastbn >= 0) {
759 		last = lastbn % factor;
760 		nb = bap[i];
761 		if (nb != 0) {
762 			error = lfs_indirtrunc(ip, nlbn, nb,
763 					       last, level - 1, &blkcount,
764 					       &rblkcount, lastsegp, bcp, p);
765 			if (error)
766 				allerror = error;
767 			real_released += rblkcount;
768 			blocksreleased += blkcount;
769 		}
770 	}
771 
772 	if (copy != NULL) {
773 		lfs_free(fs, copy, LFS_NB_IBLOCK);
774 	} else {
775 		if (bp->b_flags & B_DELWRI) {
776 			LFS_UNLOCK_BUF(bp);
777 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
778 			wakeup(&fs->lfs_avail);
779 		}
780 		bp->b_flags |= B_INVAL;
781 		brelse(bp);
782 	}
783 
784 	*countp = blocksreleased;
785 	*rcountp = real_released;
786 	return (allerror);
787 }
788 
789 /*
790  * Destroy any in core blocks past the truncation length.
791  * Inlined from vtruncbuf, so that lfs_avail could be updated.
792  * We take the seglock to prevent cleaning from occurring while we are
793  * invalidating blocks.
794  */
795 static int
796 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo)
797 {
798 	struct buf *bp, *nbp;
799 	int s, error;
800 	struct lfs *fs;
801 	voff_t off;
802 
803 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
804 	simple_lock(&vp->v_interlock);
805 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
806 	if (error)
807 		return error;
808 
809 	fs = VTOI(vp)->i_lfs;
810 	s = splbio();
811 
812 	ASSERT_SEGLOCK(fs);
813 restart:
814 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
815 		nbp = LIST_NEXT(bp, b_vnbufs);
816 		if (bp->b_lblkno < lbn)
817 			continue;
818 		simple_lock(&bp->b_interlock);
819 		if (bp->b_flags & B_BUSY) {
820 			bp->b_flags |= B_WANTED;
821 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
822 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
823 			if (error) {
824 				splx(s);
825 				return (error);
826 			}
827 			goto restart;
828 		}
829 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
830 		if (bp->b_flags & B_DELWRI) {
831 			bp->b_flags &= ~B_DELWRI;
832 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
833 			wakeup(&fs->lfs_avail);
834 		}
835 		LFS_UNLOCK_BUF(bp);
836 		simple_unlock(&bp->b_interlock);
837 		brelse(bp);
838 	}
839 
840 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
841 		nbp = LIST_NEXT(bp, b_vnbufs);
842 		if (bp->b_lblkno < lbn)
843 			continue;
844 		simple_lock(&bp->b_interlock);
845 		if (bp->b_flags & B_BUSY) {
846 			bp->b_flags |= B_WANTED;
847 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
848 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
849 			if (error) {
850 				splx(s);
851 				return (error);
852 			}
853 			goto restart;
854 		}
855 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
856 		if (bp->b_flags & B_DELWRI) {
857 			bp->b_flags &= ~B_DELWRI;
858 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
859 			wakeup(&fs->lfs_avail);
860 		}
861 		LFS_UNLOCK_BUF(bp);
862 		simple_unlock(&bp->b_interlock);
863 		brelse(bp);
864 	}
865 
866 	splx(s);
867 
868 	return (0);
869 }
870 
871