xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: lfs_inode.c,v 1.106 2006/10/14 09:17:26 yamt 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 1.106 2006/10/14 09:17:26 yamt 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/malloc.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/trace.h>
86 #include <sys/resourcevar.h>
87 #include <sys/kauth.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 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t);
98 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t,
99 			   daddr_t, int, long *, long *, long *, size_t *,
100 			   struct lwp *);
101 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *);
102 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
103 
104 /* Search a block for a specific dinode. */
105 struct ufs1_dinode *
106 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
107 {
108 	struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data;
109 	struct ufs1_dinode *ldip, *fin;
110 
111 	ASSERT_NO_SEGLOCK(fs);
112 	/*
113 	 * Read the inode block backwards, since later versions of the
114 	 * inode will supercede earlier ones.  Though it is unlikely, it is
115 	 * possible that the same inode will appear in the same inode block.
116 	 */
117 	fin = dip + INOPB(fs);
118 	for (ldip = fin - 1; ldip >= dip; --ldip)
119 		if (ldip->di_inumber == ino)
120 			return (ldip);
121 
122 	printf("searched %d entries\n", (int)(fin - dip));
123 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
124 	       dtosn(fs, fs->lfs_offset));
125 	printf("block is 0x%llx (seg %lld)\n",
126 	       (unsigned long long)dbtofsb(fs, bp->b_blkno),
127 	       (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno)));
128 
129 	return NULL;
130 }
131 
132 int
133 lfs_update(struct vnode *vp, const struct timespec *acc,
134     const struct timespec *mod, int updflags)
135 {
136 	struct inode *ip;
137 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
138 	int s;
139 	int flags;
140 
141 	ASSERT_NO_SEGLOCK(fs);
142 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
143 		return (0);
144 	ip = VTOI(vp);
145 
146 	/*
147 	 * If we are called from vinvalbuf, and the file's blocks have
148 	 * already been scheduled for writing, but the writes have not
149 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
150 	 * will cause a panic.	So, we must wait until any pending write
151 	 * for our inode completes, if we are called with UPDATE_WAIT set.
152 	 */
153 	s = splbio();
154 	simple_lock(&vp->v_interlock);
155 	while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
156 	    WRITEINPROG(vp)) {
157 		DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d"
158 		      " (in progress)\n", ip->i_number));
159 		ltsleep(vp, (PRIBIO+1), "lfs_update", 0, &vp->v_interlock);
160 	}
161 	simple_unlock(&vp->v_interlock);
162 	splx(s);
163 	LFS_ITIMES(ip, acc, mod, NULL);
164 	if (updflags & UPDATE_CLOSE)
165 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING);
166 	else
167 		flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING);
168 	if (flags == 0)
169 		return (0);
170 
171 	/* If sync, push back the vnode and any dirty blocks it may have. */
172 	if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) {
173 		/* Avoid flushing VDIROP. */
174 		simple_lock(&fs->lfs_interlock);
175 		++fs->lfs_diropwait;
176 		while (vp->v_flag & VDIROP) {
177 			DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d"
178 			      " (dirops)\n", ip->i_number));
179 			DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags"
180 			      " 0x%x\n", vp->v_flag, ip->i_flag));
181 			if (fs->lfs_dirops == 0)
182 				lfs_flush_fs(fs, SEGM_SYNC);
183 			else
184 				ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
185 					0, &fs->lfs_interlock);
186 			/* XXX KS - by falling out here, are we writing the vn
187 			twice? */
188 		}
189 		--fs->lfs_diropwait;
190 		simple_unlock(&fs->lfs_interlock);
191 		return lfs_vflush(vp);
192 	}
193 	return 0;
194 }
195 
196 #define	SINGLE	0	/* index of single indirect block */
197 #define	DOUBLE	1	/* index of double indirect block */
198 #define	TRIPLE	2	/* index of triple indirect block */
199 /*
200  * Truncate the inode oip to at most length size, freeing the
201  * disk blocks.
202  */
203 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */
204 
205 int
206 lfs_truncate(struct vnode *ovp, off_t length, int ioflag,
207     kauth_cred_t cred, struct lwp *l)
208 {
209 	daddr_t lastblock;
210 	struct inode *oip = VTOI(ovp);
211 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
212 	/* XXX ondisk32 */
213 	int32_t newblks[NDADDR + NIADDR];
214 	struct lfs *fs;
215 	struct buf *bp;
216 	int offset, size, level;
217 	long count, rcount, blocksreleased = 0, real_released = 0;
218 	int i, nblocks;
219 	int aflags, error, allerror = 0;
220 	off_t osize;
221 	long lastseg;
222 	size_t bc;
223 	int obufsize, odb;
224 	int usepc;
225 	struct ufsmount *ump = oip->i_ump;
226 
227 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
228 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
229 		KASSERT(oip->i_size == 0);
230 		return 0;
231 	}
232 
233 	if (length < 0)
234 		return (EINVAL);
235 
236 	/*
237 	 * Just return and not update modification times.
238 	 */
239 	if (oip->i_size == length)
240 		return (0);
241 
242 	if (ovp->v_type == VLNK &&
243 	    (oip->i_size < ump->um_maxsymlinklen ||
244 	     (ump->um_maxsymlinklen == 0 &&
245 	      oip->i_ffs1_blocks == 0))) {
246 #ifdef DIAGNOSTIC
247 		if (length != 0)
248 			panic("lfs_truncate: partial truncate of symlink");
249 #endif
250 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
251 		oip->i_size = oip->i_ffs1_size = 0;
252 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
253 		return (lfs_update(ovp, NULL, NULL, 0));
254 	}
255 	if (oip->i_size == length) {
256 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
257 		return (lfs_update(ovp, NULL, NULL, 0));
258 	}
259 #ifdef QUOTA
260 	if ((error = getinoquota(oip)) != 0)
261 		return (error);
262 #endif
263 	fs = oip->i_lfs;
264 	lfs_imtime(fs);
265 	osize = oip->i_size;
266 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
267 
268 	ASSERT_NO_SEGLOCK(fs);
269 	/*
270 	 * Lengthen the size of the file. We must ensure that the
271 	 * last byte of the file is allocated. Since the smallest
272 	 * value of osize is 0, length will be at least 1.
273 	 */
274 	if (osize < length) {
275 		if (length > ump->um_maxfilesize)
276 			return (EFBIG);
277 		aflags = B_CLRBUF;
278 		if (ioflag & IO_SYNC)
279 			aflags |= B_SYNC;
280 		if (usepc) {
281 			if (lblkno(fs, osize) < NDADDR &&
282 			    lblkno(fs, osize) != lblkno(fs, length) &&
283 			    blkroundup(fs, osize) != osize) {
284 				off_t eob;
285 
286 				eob = blkroundup(fs, osize);
287 				error = ufs_balloc_range(ovp, osize,
288 				    eob - osize, cred, aflags);
289 				if (error)
290 					return error;
291 				if (ioflag & IO_SYNC) {
292 					ovp->v_size = eob;
293 					simple_lock(&ovp->v_interlock);
294 					VOP_PUTPAGES(ovp,
295 					    trunc_page(osize & fs->lfs_bmask),
296 					    round_page(eob),
297 					    PGO_CLEANIT | PGO_SYNCIO);
298 				}
299 			}
300 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
301 						 aflags);
302 			if (error) {
303 				(void) lfs_truncate(ovp, osize,
304 						    ioflag & IO_SYNC, cred, l);
305 				return error;
306 			}
307 			uvm_vnp_setsize(ovp, length);
308 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
309 			KASSERT(ovp->v_size == oip->i_size);
310 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
311 			return (lfs_update(ovp, NULL, NULL, 0));
312 		} else {
313 			error = lfs_reserve(fs, ovp, NULL,
314 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
315 			if (error)
316 				return (error);
317 			error = lfs_balloc(ovp, length - 1, 1, cred,
318 					   aflags, &bp);
319 			lfs_reserve(fs, ovp, NULL,
320 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
321 			if (error)
322 				return (error);
323 			oip->i_ffs1_size = oip->i_size = length;
324 			uvm_vnp_setsize(ovp, length);
325 			(void) VOP_BWRITE(bp);
326 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
327 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
328 			return (lfs_update(ovp, NULL, NULL, 0));
329 		}
330 	}
331 
332 	if ((error = lfs_reserve(fs, ovp, NULL,
333 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
334 		return (error);
335 
336 	/*
337 	 * Shorten the size of the file. If the file is not being
338 	 * truncated to a block boundary, the contents of the
339 	 * partial block following the end of the file must be
340 	 * zero'ed in case it ever becomes accessible again because
341 	 * of subsequent file growth. Directories however are not
342 	 * zero'ed as they should grow back initialized to empty.
343 	 */
344 	offset = blkoff(fs, length);
345 	lastseg = -1;
346 	bc = 0;
347 
348 	if (ovp != fs->lfs_ivnode)
349 		lfs_seglock(fs, SEGM_PROT);
350 	if (offset == 0) {
351 		oip->i_size = oip->i_ffs1_size = length;
352 	} else if (!usepc) {
353 		lbn = lblkno(fs, length);
354 		aflags = B_CLRBUF;
355 		if (ioflag & IO_SYNC)
356 			aflags |= B_SYNC;
357 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
358 		if (error) {
359 			lfs_reserve(fs, ovp, NULL,
360 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
361 			goto errout;
362 		}
363 		obufsize = bp->b_bufsize;
364 		odb = btofsb(fs, bp->b_bcount);
365 		oip->i_size = oip->i_ffs1_size = length;
366 		size = blksize(fs, oip, lbn);
367 		if (ovp->v_type != VDIR)
368 			memset((char *)bp->b_data + offset, 0,
369 			       (u_int)(size - offset));
370 		allocbuf(bp, size, 1);
371 		if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) {
372 			simple_lock(&lfs_subsys_lock);
373 			locked_queue_bytes -= obufsize - bp->b_bufsize;
374 			simple_unlock(&lfs_subsys_lock);
375 		}
376 		if (bp->b_flags & B_DELWRI)
377 			fs->lfs_avail += odb - btofsb(fs, size);
378 		(void) VOP_BWRITE(bp);
379 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
380 		/*
381 		 * When truncating a regular file down to a non-block-aligned
382 		 * size, we must zero the part of last block which is past
383 		 * the new EOF.  We must synchronously flush the zeroed pages
384 		 * to disk since the new pages will be invalidated as soon
385 		 * as we inform the VM system of the new, smaller size.
386 		 * We must do this before acquiring the GLOCK, since fetching
387 		 * the pages will acquire the GLOCK internally.
388 		 * So there is a window where another thread could see a whole
389 		 * zeroed page past EOF, but that's life.
390 		 */
391 		daddr_t xlbn;
392 		voff_t eoz;
393 
394 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
395 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags);
396 		if (error) {
397 			lfs_reserve(fs, ovp, NULL,
398 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
399 			goto errout;
400 		}
401 		xlbn = lblkno(fs, length);
402 		size = blksize(fs, oip, xlbn);
403 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
404 		uvm_vnp_zerorange(ovp, length, eoz - length);
405 		if (round_page(eoz) > round_page(length)) {
406 			simple_lock(&ovp->v_interlock);
407 			error = VOP_PUTPAGES(ovp, round_page(length),
408 			    round_page(eoz),
409 			    PGO_CLEANIT | PGO_DEACTIVATE |
410 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
411 			if (error) {
412 				lfs_reserve(fs, ovp, NULL,
413 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
414 				goto errout;
415 			}
416 		}
417 	}
418 
419 	genfs_node_wrlock(ovp);
420 
421 	oip->i_size = oip->i_ffs1_size = length;
422 	uvm_vnp_setsize(ovp, length);
423 	/*
424 	 * Calculate index into inode's block list of
425 	 * last direct and indirect blocks (if any)
426 	 * which we want to keep.  Lastblock is -1 when
427 	 * the file is truncated to 0.
428 	 */
429 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
430 	if (length > QUAD_MAX - fs->lfs_bsize)
431 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
432 	else
433 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
434 	lastiblock[SINGLE] = lastblock - NDADDR;
435 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
436 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
437 	nblocks = btofsb(fs, fs->lfs_bsize);
438 	/*
439 	 * Record changed file and block pointers before we start
440 	 * freeing blocks.  lastiblock values are also normalized to -1
441 	 * for calls to lfs_indirtrunc below.
442 	 */
443 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks);
444 	for (level = TRIPLE; level >= SINGLE; level--)
445 		if (lastiblock[level] < 0) {
446 			newblks[NDADDR+level] = 0;
447 			lastiblock[level] = -1;
448 		}
449 	for (i = NDADDR - 1; i > lastblock; i--)
450 		newblks[i] = 0;
451 
452 	oip->i_size = oip->i_ffs1_size = osize;
453 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
454 	if (error && !allerror)
455 		allerror = error;
456 
457 	/*
458 	 * Indirect blocks first.
459 	 */
460 	indir_lbn[SINGLE] = -NDADDR;
461 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
462 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
463 	for (level = TRIPLE; level >= SINGLE; level--) {
464 		bn = oip->i_ffs1_ib[level];
465 		if (bn != 0) {
466 			error = lfs_indirtrunc(oip, indir_lbn[level],
467 					       bn, lastiblock[level],
468 					       level, &count, &rcount,
469 					       &lastseg, &bc, l);
470 			if (error)
471 				allerror = error;
472 			real_released += rcount;
473 			blocksreleased += count;
474 			if (lastiblock[level] < 0) {
475 				if (oip->i_ffs1_ib[level] > 0)
476 					real_released += nblocks;
477 				blocksreleased += nblocks;
478 				oip->i_ffs1_ib[level] = 0;
479 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
480 					    &lastseg, &bc);
481         			lfs_deregister_block(ovp, bn);
482 			}
483 		}
484 		if (lastiblock[level] >= 0)
485 			goto done;
486 	}
487 
488 	/*
489 	 * All whole direct blocks or frags.
490 	 */
491 	for (i = NDADDR - 1; i > lastblock; i--) {
492 		long bsize, obsize;
493 
494 		bn = oip->i_ffs1_db[i];
495 		if (bn == 0)
496 			continue;
497 		bsize = blksize(fs, oip, i);
498 		if (oip->i_ffs1_db[i] > 0) {
499 			/* Check for fragment size changes */
500 			obsize = oip->i_lfs_fragsize[i];
501 			real_released += btofsb(fs, obsize);
502 			oip->i_lfs_fragsize[i] = 0;
503 		} else
504 			obsize = 0;
505 		blocksreleased += btofsb(fs, bsize);
506 		oip->i_ffs1_db[i] = 0;
507 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
508         	lfs_deregister_block(ovp, bn);
509 	}
510 	if (lastblock < 0)
511 		goto done;
512 
513 	/*
514 	 * Finally, look for a change in size of the
515 	 * last direct block; release any frags.
516 	 */
517 	bn = oip->i_ffs1_db[lastblock];
518 	if (bn != 0) {
519 		long oldspace, newspace;
520 #if 0
521 		long olddspace;
522 #endif
523 
524 		/*
525 		 * Calculate amount of space we're giving
526 		 * back as old block size minus new block size.
527 		 */
528 		oldspace = blksize(fs, oip, lastblock);
529 #if 0
530 		olddspace = oip->i_lfs_fragsize[lastblock];
531 #endif
532 
533 		oip->i_size = oip->i_ffs1_size = length;
534 		newspace = blksize(fs, oip, lastblock);
535 		if (newspace == 0)
536 			panic("itrunc: newspace");
537 		if (oldspace - newspace > 0) {
538 			blocksreleased += btofsb(fs, oldspace - newspace);
539 		}
540 #if 0
541 		if (bn > 0 && olddspace - newspace > 0) {
542 			/* No segment accounting here, just vnode */
543 			real_released += btofsb(fs, olddspace - newspace);
544 		}
545 #endif
546 	}
547 
548 done:
549 	/* Finish segment accounting corrections */
550 	lfs_update_seguse(fs, oip, lastseg, bc);
551 #ifdef DIAGNOSTIC
552 	for (level = SINGLE; level <= TRIPLE; level++)
553 		if ((newblks[NDADDR + level] == 0) !=
554 		    ((oip->i_ffs1_ib[level]) == 0)) {
555 			panic("lfs itrunc1");
556 		}
557 	for (i = 0; i < NDADDR; i++)
558 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
559 			panic("lfs itrunc2");
560 		}
561 	if (length == 0 &&
562 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
563 		panic("lfs itrunc3");
564 #endif /* DIAGNOSTIC */
565 	/*
566 	 * Put back the real size.
567 	 */
568 	oip->i_size = oip->i_ffs1_size = length;
569 	oip->i_lfs_effnblks -= blocksreleased;
570 	oip->i_ffs1_blocks -= real_released;
571 	simple_lock(&fs->lfs_interlock);
572 	fs->lfs_bfree += blocksreleased;
573 	simple_unlock(&fs->lfs_interlock);
574 #ifdef DIAGNOSTIC
575 	if (oip->i_size == 0 &&
576 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
577 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
578 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
579 		panic("lfs_truncate: persistent blocks");
580 	}
581 #endif
582 
583 	/*
584 	 * If we truncated to zero, take us off the paging queue.
585 	 */
586 	simple_lock(&fs->lfs_interlock);
587 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
588 		oip->i_flags &= ~IN_PAGING;
589 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
590 	}
591 	simple_unlock(&fs->lfs_interlock);
592 
593 	oip->i_flag |= IN_CHANGE;
594 #ifdef QUOTA
595 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
596 #endif
597 	lfs_reserve(fs, ovp, NULL,
598 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
599 	genfs_node_unlock(ovp);
600   errout:
601 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
602 	if (ovp != fs->lfs_ivnode)
603 		lfs_segunlock(fs);
604 	return (allerror ? allerror : error);
605 }
606 
607 /* Update segment and avail usage information when removing a block. */
608 static int
609 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
610 	    size_t bsize, long *lastseg, size_t *num)
611 {
612 	long seg;
613 	int error = 0;
614 
615 	ASSERT_SEGLOCK(fs);
616 	bsize = fragroundup(fs, bsize);
617 	if (daddr > 0) {
618 		if (*lastseg != (seg = dtosn(fs, daddr))) {
619 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
620 			*num = bsize;
621 			*lastseg = seg;
622 		} else
623 			*num += bsize;
624 	}
625 
626 	return error;
627 }
628 
629 /* Finish the accounting updates for a segment. */
630 static int
631 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
632 {
633 	struct segdelta *sd;
634 	struct vnode *vp;
635 
636 	ASSERT_SEGLOCK(fs);
637 	if (lastseg < 0 || num == 0)
638 		return 0;
639 
640 	vp = ITOV(ip);
641 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
642 		if (sd->segnum == lastseg)
643 			break;
644 	if (sd == NULL) {
645 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
646 		sd->segnum = lastseg;
647 		sd->num = 0;
648 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
649 	}
650 	sd->num += num;
651 
652 	return 0;
653 }
654 
655 static void
656 lfs_finalize_seguse(struct lfs *fs, void *v)
657 {
658 	SEGUSE *sup;
659 	struct buf *bp;
660 	struct segdelta *sd;
661 	LIST_HEAD(, segdelta) *hd = v;
662 
663 	ASSERT_SEGLOCK(fs);
664 	while((sd = LIST_FIRST(hd)) != NULL) {
665 		LIST_REMOVE(sd, list);
666 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
667 		if (sd->num > sup->su_nbytes) {
668 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
669 				sd->segnum, (long)(sd->num - sup->su_nbytes));
670 			panic("lfs_finalize_seguse: negative bytes");
671 			sup->su_nbytes = sd->num;
672 		}
673 		sup->su_nbytes -= sd->num;
674 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
675 		free(sd, M_SEGMENT);
676 	}
677 }
678 
679 /* Finish the accounting updates for a segment. */
680 void
681 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
682 {
683 	ASSERT_SEGLOCK(fs);
684 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
685 }
686 
687 /* Finish the accounting updates for a segment. */
688 void
689 lfs_finalize_fs_seguse(struct lfs *fs)
690 {
691 	ASSERT_SEGLOCK(fs);
692 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
693 }
694 
695 /*
696  * Release blocks associated with the inode ip and stored in the indirect
697  * block bn.  Blocks are free'd in LIFO order up to (but not including)
698  * lastbn.  If level is greater than SINGLE, the block is an indirect block
699  * and recursive calls to indirtrunc must be used to cleanse other indirect
700  * blocks.
701  *
702  * NB: triple indirect blocks are untested.
703  */
704 static int
705 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
706 	       daddr_t lastbn, int level, long *countp,
707 	       long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l)
708 {
709 	int i;
710 	struct buf *bp;
711 	struct lfs *fs = ip->i_lfs;
712 	int32_t *bap;	/* XXX ondisk32 */
713 	struct vnode *vp;
714 	daddr_t nb, nlbn, last;
715 	int32_t *copy = NULL;	/* XXX ondisk32 */
716 	long blkcount, rblkcount, factor;
717 	int nblocks, blocksreleased = 0, real_released = 0;
718 	int error = 0, allerror = 0;
719 
720 	ASSERT_SEGLOCK(fs);
721 	/*
722 	 * Calculate index in current block of last
723 	 * block to be kept.  -1 indicates the entire
724 	 * block so we need not calculate the index.
725 	 */
726 	factor = 1;
727 	for (i = SINGLE; i < level; i++)
728 		factor *= NINDIR(fs);
729 	last = lastbn;
730 	if (lastbn > 0)
731 		last /= factor;
732 	nblocks = btofsb(fs, fs->lfs_bsize);
733 	/*
734 	 * Get buffer of block pointers, zero those entries corresponding
735 	 * to blocks to be free'd, and update on disk copy first.  Since
736 	 * double(triple) indirect before single(double) indirect, calls
737 	 * to bmap on these blocks will fail.  However, we already have
738 	 * the on disk address, so we have to set the b_blkno field
739 	 * explicitly instead of letting bread do everything for us.
740 	 */
741 	vp = ITOV(ip);
742 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
743 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
744 		/* Braces must be here in case trace evaluates to nothing. */
745 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
746 	} else {
747 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
748 		l->l_proc->p_stats->p_ru.ru_inblock++;	/* pay for read */
749 		bp->b_flags |= B_READ;
750 		if (bp->b_bcount > bp->b_bufsize)
751 			panic("lfs_indirtrunc: bad buffer size");
752 		bp->b_blkno = fsbtodb(fs, dbn);
753 		VOP_STRATEGY(vp, bp);
754 		error = biowait(bp);
755 	}
756 	if (error) {
757 		brelse(bp);
758 		*countp = *rcountp = 0;
759 		return (error);
760 	}
761 
762 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
763 	if (lastbn >= 0) {
764 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
765 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
766 		memset((caddr_t)&bap[last + 1], 0,
767 		/* XXX ondisk32 */
768 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
769 		error = VOP_BWRITE(bp);
770 		if (error)
771 			allerror = error;
772 		bap = copy;
773 	}
774 
775 	/*
776 	 * Recursively free totally unused blocks.
777 	 */
778 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
779 	    i--, nlbn += factor) {
780 		nb = bap[i];
781 		if (nb == 0)
782 			continue;
783 		if (level > SINGLE) {
784 			error = lfs_indirtrunc(ip, nlbn, nb,
785 					       (daddr_t)-1, level - 1,
786 					       &blkcount, &rblkcount,
787 					       lastsegp, bcp, l);
788 			if (error)
789 				allerror = error;
790 			blocksreleased += blkcount;
791 			real_released += rblkcount;
792 		}
793 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
794 		if (bap[i] > 0)
795 			real_released += nblocks;
796 		blocksreleased += nblocks;
797 	}
798 
799 	/*
800 	 * Recursively free last partial block.
801 	 */
802 	if (level > SINGLE && lastbn >= 0) {
803 		last = lastbn % factor;
804 		nb = bap[i];
805 		if (nb != 0) {
806 			error = lfs_indirtrunc(ip, nlbn, nb,
807 					       last, level - 1, &blkcount,
808 					       &rblkcount, lastsegp, bcp, l);
809 			if (error)
810 				allerror = error;
811 			real_released += rblkcount;
812 			blocksreleased += blkcount;
813 		}
814 	}
815 
816 	if (copy != NULL) {
817 		lfs_free(fs, copy, LFS_NB_IBLOCK);
818 	} else {
819 		if (bp->b_flags & B_DELWRI) {
820 			LFS_UNLOCK_BUF(bp);
821 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
822 			wakeup(&fs->lfs_avail);
823 		}
824 		bp->b_flags |= B_INVAL;
825 		brelse(bp);
826 	}
827 
828 	*countp = blocksreleased;
829 	*rcountp = real_released;
830 	return (allerror);
831 }
832 
833 /*
834  * Destroy any in core blocks past the truncation length.
835  * Inlined from vtruncbuf, so that lfs_avail could be updated.
836  * We take the seglock to prevent cleaning from occurring while we are
837  * invalidating blocks.
838  */
839 static int
840 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo)
841 {
842 	struct buf *bp, *nbp;
843 	int s, error;
844 	struct lfs *fs;
845 	voff_t off;
846 
847 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
848 	simple_lock(&vp->v_interlock);
849 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
850 	if (error)
851 		return error;
852 
853 	fs = VTOI(vp)->i_lfs;
854 	s = splbio();
855 
856 	ASSERT_SEGLOCK(fs);
857 restart:
858 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
859 		nbp = LIST_NEXT(bp, b_vnbufs);
860 		if (bp->b_lblkno < lbn)
861 			continue;
862 		simple_lock(&bp->b_interlock);
863 		if (bp->b_flags & B_BUSY) {
864 			bp->b_flags |= B_WANTED;
865 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
866 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
867 			if (error) {
868 				splx(s);
869 				return (error);
870 			}
871 			goto restart;
872 		}
873 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
874 		if (bp->b_flags & B_DELWRI) {
875 			bp->b_flags &= ~B_DELWRI;
876 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
877 			wakeup(&fs->lfs_avail);
878 		}
879 		LFS_UNLOCK_BUF(bp);
880 		simple_unlock(&bp->b_interlock);
881 		brelse(bp);
882 	}
883 
884 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
885 		nbp = LIST_NEXT(bp, b_vnbufs);
886 		if (bp->b_lblkno < lbn)
887 			continue;
888 		simple_lock(&bp->b_interlock);
889 		if (bp->b_flags & B_BUSY) {
890 			bp->b_flags |= B_WANTED;
891 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
892 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
893 			if (error) {
894 				splx(s);
895 				return (error);
896 			}
897 			goto restart;
898 		}
899 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
900 		if (bp->b_flags & B_DELWRI) {
901 			bp->b_flags &= ~B_DELWRI;
902 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
903 			wakeup(&fs->lfs_avail);
904 		}
905 		LFS_UNLOCK_BUF(bp);
906 		simple_unlock(&bp->b_interlock);
907 		brelse(bp);
908 	}
909 
910 	splx(s);
911 
912 	return (0);
913 }
914 
915