xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision 7fa608457b817eca6e0977b37f758ae064f3c99c)
1 /*	$NetBSD: lfs_inode.c,v 1.114 2007/10/10 22:38:00 ad 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 1.114 2007/10/10 22:38:00 ad 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 VU_DIROP. */
174 		simple_lock(&fs->lfs_interlock);
175 		++fs->lfs_diropwait;
176 		while (vp->v_uflag & VU_DIROP) {
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_uflag|vp->v_iflag|vp->v_vflag,
181 			      ip->i_flag));
182 			if (fs->lfs_dirops == 0)
183 				lfs_flush_fs(fs, SEGM_SYNC);
184 			else
185 				ltsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
186 					0, &fs->lfs_interlock);
187 			/* XXX KS - by falling out here, are we writing the vn
188 			twice? */
189 		}
190 		--fs->lfs_diropwait;
191 		simple_unlock(&fs->lfs_interlock);
192 		return lfs_vflush(vp);
193 	}
194 	return 0;
195 }
196 
197 #define	SINGLE	0	/* index of single indirect block */
198 #define	DOUBLE	1	/* index of double indirect block */
199 #define	TRIPLE	2	/* index of triple indirect block */
200 /*
201  * Truncate the inode oip to at most length size, freeing the
202  * disk blocks.
203  */
204 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */
205 
206 int
207 lfs_truncate(struct vnode *ovp, off_t length, int ioflag,
208     kauth_cred_t cred, struct lwp *l)
209 {
210 	daddr_t lastblock;
211 	struct inode *oip = VTOI(ovp);
212 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
213 	/* XXX ondisk32 */
214 	int32_t newblks[NDADDR + NIADDR];
215 	struct lfs *fs;
216 	struct buf *bp;
217 	int offset, size, level;
218 	long count, rcount, blocksreleased = 0, real_released = 0;
219 	int i, nblocks;
220 	int aflags, error, allerror = 0;
221 	off_t osize;
222 	long lastseg;
223 	size_t bc;
224 	int obufsize, odb;
225 	int usepc;
226 	struct ufsmount *ump = oip->i_ump;
227 
228 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
229 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
230 		KASSERT(oip->i_size == 0);
231 		return 0;
232 	}
233 
234 	if (length < 0)
235 		return (EINVAL);
236 
237 	/*
238 	 * Just return and not update modification times.
239 	 */
240 	if (oip->i_size == length)
241 		return (0);
242 
243 	if (ovp->v_type == VLNK &&
244 	    (oip->i_size < ump->um_maxsymlinklen ||
245 	     (ump->um_maxsymlinklen == 0 &&
246 	      oip->i_ffs1_blocks == 0))) {
247 #ifdef DIAGNOSTIC
248 		if (length != 0)
249 			panic("lfs_truncate: partial truncate of symlink");
250 #endif
251 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
252 		oip->i_size = oip->i_ffs1_size = 0;
253 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
254 		return (lfs_update(ovp, NULL, NULL, 0));
255 	}
256 	if (oip->i_size == length) {
257 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
258 		return (lfs_update(ovp, NULL, NULL, 0));
259 	}
260 	fs = oip->i_lfs;
261 	lfs_imtime(fs);
262 	osize = oip->i_size;
263 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
264 
265 	ASSERT_NO_SEGLOCK(fs);
266 	/*
267 	 * Lengthen the size of the file. We must ensure that the
268 	 * last byte of the file is allocated. Since the smallest
269 	 * value of osize is 0, length will be at least 1.
270 	 */
271 	if (osize < length) {
272 		if (length > ump->um_maxfilesize)
273 			return (EFBIG);
274 		aflags = B_CLRBUF;
275 		if (ioflag & IO_SYNC)
276 			aflags |= B_SYNC;
277 		if (usepc) {
278 			if (lblkno(fs, osize) < NDADDR &&
279 			    lblkno(fs, osize) != lblkno(fs, length) &&
280 			    blkroundup(fs, osize) != osize) {
281 				off_t eob;
282 
283 				eob = blkroundup(fs, osize);
284 				uvm_vnp_setwritesize(ovp, eob);
285 				error = ufs_balloc_range(ovp, osize,
286 				    eob - osize, cred, aflags);
287 				if (error)
288 					return error;
289 				if (ioflag & IO_SYNC) {
290 					simple_lock(&ovp->v_interlock);
291 					VOP_PUTPAGES(ovp,
292 					    trunc_page(osize & fs->lfs_bmask),
293 					    round_page(eob),
294 					    PGO_CLEANIT | PGO_SYNCIO);
295 				}
296 			}
297 			uvm_vnp_setwritesize(ovp, length);
298 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
299 						 aflags);
300 			if (error) {
301 				(void) lfs_truncate(ovp, osize,
302 						    ioflag & IO_SYNC, cred, l);
303 				return error;
304 			}
305 			uvm_vnp_setsize(ovp, length);
306 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
307 			KASSERT(ovp->v_size == oip->i_size);
308 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
309 			return (lfs_update(ovp, NULL, NULL, 0));
310 		} else {
311 			error = lfs_reserve(fs, ovp, NULL,
312 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
313 			if (error)
314 				return (error);
315 			error = lfs_balloc(ovp, length - 1, 1, cred,
316 					   aflags, &bp);
317 			lfs_reserve(fs, ovp, NULL,
318 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
319 			if (error)
320 				return (error);
321 			oip->i_ffs1_size = oip->i_size = length;
322 			uvm_vnp_setsize(ovp, length);
323 			(void) VOP_BWRITE(bp);
324 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
325 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
326 			return (lfs_update(ovp, NULL, NULL, 0));
327 		}
328 	}
329 
330 	if ((error = lfs_reserve(fs, ovp, NULL,
331 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
332 		return (error);
333 
334 	/*
335 	 * Shorten the size of the file. If the file is not being
336 	 * truncated to a block boundary, the contents of the
337 	 * partial block following the end of the file must be
338 	 * zero'ed in case it ever becomes accessible again because
339 	 * of subsequent file growth. Directories however are not
340 	 * zero'ed as they should grow back initialized to empty.
341 	 */
342 	offset = blkoff(fs, length);
343 	lastseg = -1;
344 	bc = 0;
345 
346 	if (ovp != fs->lfs_ivnode)
347 		lfs_seglock(fs, SEGM_PROT);
348 	if (offset == 0) {
349 		oip->i_size = oip->i_ffs1_size = length;
350 	} else if (!usepc) {
351 		lbn = lblkno(fs, length);
352 		aflags = B_CLRBUF;
353 		if (ioflag & IO_SYNC)
354 			aflags |= B_SYNC;
355 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
356 		if (error) {
357 			lfs_reserve(fs, ovp, NULL,
358 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
359 			goto errout;
360 		}
361 		obufsize = bp->b_bufsize;
362 		odb = btofsb(fs, bp->b_bcount);
363 		oip->i_size = oip->i_ffs1_size = length;
364 		size = blksize(fs, oip, lbn);
365 		if (ovp->v_type != VDIR)
366 			memset((char *)bp->b_data + offset, 0,
367 			       (u_int)(size - offset));
368 		allocbuf(bp, size, 1);
369 		if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) {
370 			simple_lock(&lfs_subsys_lock);
371 			locked_queue_bytes -= obufsize - bp->b_bufsize;
372 			simple_unlock(&lfs_subsys_lock);
373 		}
374 		if (bp->b_flags & B_DELWRI)
375 			fs->lfs_avail += odb - btofsb(fs, size);
376 		(void) VOP_BWRITE(bp);
377 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
378 		/*
379 		 * When truncating a regular file down to a non-block-aligned
380 		 * size, we must zero the part of last block which is past
381 		 * the new EOF.  We must synchronously flush the zeroed pages
382 		 * to disk since the new pages will be invalidated as soon
383 		 * as we inform the VM system of the new, smaller size.
384 		 * We must do this before acquiring the GLOCK, since fetching
385 		 * the pages will acquire the GLOCK internally.
386 		 * So there is a window where another thread could see a whole
387 		 * zeroed page past EOF, but that's life.
388 		 */
389 		daddr_t xlbn;
390 		voff_t eoz;
391 
392 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
393 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags);
394 		if (error) {
395 			lfs_reserve(fs, ovp, NULL,
396 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
397 			goto errout;
398 		}
399 		xlbn = lblkno(fs, length);
400 		size = blksize(fs, oip, xlbn);
401 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
402 		uvm_vnp_zerorange(ovp, length, eoz - length);
403 		if (round_page(eoz) > round_page(length)) {
404 			simple_lock(&ovp->v_interlock);
405 			error = VOP_PUTPAGES(ovp, round_page(length),
406 			    round_page(eoz),
407 			    PGO_CLEANIT | PGO_DEACTIVATE |
408 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
409 			if (error) {
410 				lfs_reserve(fs, ovp, NULL,
411 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
412 				goto errout;
413 			}
414 		}
415 	}
416 
417 	genfs_node_wrlock(ovp);
418 
419 	oip->i_size = oip->i_ffs1_size = length;
420 	uvm_vnp_setsize(ovp, length);
421 	/*
422 	 * Calculate index into inode's block list of
423 	 * last direct and indirect blocks (if any)
424 	 * which we want to keep.  Lastblock is -1 when
425 	 * the file is truncated to 0.
426 	 */
427 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
428 	if (length > QUAD_MAX - fs->lfs_bsize)
429 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
430 	else
431 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
432 	lastiblock[SINGLE] = lastblock - NDADDR;
433 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
434 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
435 	nblocks = btofsb(fs, fs->lfs_bsize);
436 	/*
437 	 * Record changed file and block pointers before we start
438 	 * freeing blocks.  lastiblock values are also normalized to -1
439 	 * for calls to lfs_indirtrunc below.
440 	 */
441 	memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
442 	for (level = TRIPLE; level >= SINGLE; level--)
443 		if (lastiblock[level] < 0) {
444 			newblks[NDADDR+level] = 0;
445 			lastiblock[level] = -1;
446 		}
447 	for (i = NDADDR - 1; i > lastblock; i--)
448 		newblks[i] = 0;
449 
450 	oip->i_size = oip->i_ffs1_size = osize;
451 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
452 	if (error && !allerror)
453 		allerror = error;
454 
455 	/*
456 	 * Indirect blocks first.
457 	 */
458 	indir_lbn[SINGLE] = -NDADDR;
459 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
460 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
461 	for (level = TRIPLE; level >= SINGLE; level--) {
462 		bn = oip->i_ffs1_ib[level];
463 		if (bn != 0) {
464 			error = lfs_indirtrunc(oip, indir_lbn[level],
465 					       bn, lastiblock[level],
466 					       level, &count, &rcount,
467 					       &lastseg, &bc, l);
468 			if (error)
469 				allerror = error;
470 			real_released += rcount;
471 			blocksreleased += count;
472 			if (lastiblock[level] < 0) {
473 				if (oip->i_ffs1_ib[level] > 0)
474 					real_released += nblocks;
475 				blocksreleased += nblocks;
476 				oip->i_ffs1_ib[level] = 0;
477 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
478 					    &lastseg, &bc);
479         			lfs_deregister_block(ovp, bn);
480 			}
481 		}
482 		if (lastiblock[level] >= 0)
483 			goto done;
484 	}
485 
486 	/*
487 	 * All whole direct blocks or frags.
488 	 */
489 	for (i = NDADDR - 1; i > lastblock; i--) {
490 		long bsize, obsize;
491 
492 		bn = oip->i_ffs1_db[i];
493 		if (bn == 0)
494 			continue;
495 		bsize = blksize(fs, oip, i);
496 		if (oip->i_ffs1_db[i] > 0) {
497 			/* Check for fragment size changes */
498 			obsize = oip->i_lfs_fragsize[i];
499 			real_released += btofsb(fs, obsize);
500 			oip->i_lfs_fragsize[i] = 0;
501 		} else
502 			obsize = 0;
503 		blocksreleased += btofsb(fs, bsize);
504 		oip->i_ffs1_db[i] = 0;
505 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
506         	lfs_deregister_block(ovp, bn);
507 	}
508 	if (lastblock < 0)
509 		goto done;
510 
511 	/*
512 	 * Finally, look for a change in size of the
513 	 * last direct block; release any frags.
514 	 */
515 	bn = oip->i_ffs1_db[lastblock];
516 	if (bn != 0) {
517 		long oldspace, newspace;
518 #if 0
519 		long olddspace;
520 #endif
521 
522 		/*
523 		 * Calculate amount of space we're giving
524 		 * back as old block size minus new block size.
525 		 */
526 		oldspace = blksize(fs, oip, lastblock);
527 #if 0
528 		olddspace = oip->i_lfs_fragsize[lastblock];
529 #endif
530 
531 		oip->i_size = oip->i_ffs1_size = length;
532 		newspace = blksize(fs, oip, lastblock);
533 		if (newspace == 0)
534 			panic("itrunc: newspace");
535 		if (oldspace - newspace > 0) {
536 			blocksreleased += btofsb(fs, oldspace - newspace);
537 		}
538 #if 0
539 		if (bn > 0 && olddspace - newspace > 0) {
540 			/* No segment accounting here, just vnode */
541 			real_released += btofsb(fs, olddspace - newspace);
542 		}
543 #endif
544 	}
545 
546 done:
547 	/* Finish segment accounting corrections */
548 	lfs_update_seguse(fs, oip, lastseg, bc);
549 #ifdef DIAGNOSTIC
550 	for (level = SINGLE; level <= TRIPLE; level++)
551 		if ((newblks[NDADDR + level] == 0) !=
552 		    ((oip->i_ffs1_ib[level]) == 0)) {
553 			panic("lfs itrunc1");
554 		}
555 	for (i = 0; i < NDADDR; i++)
556 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
557 			panic("lfs itrunc2");
558 		}
559 	if (length == 0 &&
560 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
561 		panic("lfs itrunc3");
562 #endif /* DIAGNOSTIC */
563 	/*
564 	 * Put back the real size.
565 	 */
566 	oip->i_size = oip->i_ffs1_size = length;
567 	oip->i_lfs_effnblks -= blocksreleased;
568 	oip->i_ffs1_blocks -= real_released;
569 	simple_lock(&fs->lfs_interlock);
570 	fs->lfs_bfree += blocksreleased;
571 	simple_unlock(&fs->lfs_interlock);
572 #ifdef DIAGNOSTIC
573 	if (oip->i_size == 0 &&
574 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
575 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
576 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
577 		panic("lfs_truncate: persistent blocks");
578 	}
579 #endif
580 
581 	/*
582 	 * If we truncated to zero, take us off the paging queue.
583 	 */
584 	simple_lock(&fs->lfs_interlock);
585 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
586 		oip->i_flags &= ~IN_PAGING;
587 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
588 	}
589 	simple_unlock(&fs->lfs_interlock);
590 
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 	genfs_node_unlock(ovp);
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, struct inode *ip, daddr_t daddr,
608 	    size_t bsize, long *lastseg, 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, ip, *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, struct inode *ip, long lastseg, size_t num)
630 {
631 	struct segdelta *sd;
632 	struct vnode *vp;
633 
634 	ASSERT_SEGLOCK(fs);
635 	if (lastseg < 0 || num == 0)
636 		return 0;
637 
638 	vp = ITOV(ip);
639 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
640 		if (sd->segnum == lastseg)
641 			break;
642 	if (sd == NULL) {
643 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
644 		sd->segnum = lastseg;
645 		sd->num = 0;
646 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
647 	}
648 	sd->num += num;
649 
650 	return 0;
651 }
652 
653 static void
654 lfs_finalize_seguse(struct lfs *fs, void *v)
655 {
656 	SEGUSE *sup;
657 	struct buf *bp;
658 	struct segdelta *sd;
659 	LIST_HEAD(, segdelta) *hd = v;
660 
661 	ASSERT_SEGLOCK(fs);
662 	while((sd = LIST_FIRST(hd)) != NULL) {
663 		LIST_REMOVE(sd, list);
664 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
665 		if (sd->num > sup->su_nbytes) {
666 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
667 				sd->segnum, (long)(sd->num - sup->su_nbytes));
668 			panic("lfs_finalize_seguse: negative bytes");
669 			sup->su_nbytes = sd->num;
670 		}
671 		sup->su_nbytes -= sd->num;
672 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
673 		free(sd, M_SEGMENT);
674 	}
675 }
676 
677 /* Finish the accounting updates for a segment. */
678 void
679 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
680 {
681 	ASSERT_SEGLOCK(fs);
682 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
683 }
684 
685 /* Finish the accounting updates for a segment. */
686 void
687 lfs_finalize_fs_seguse(struct lfs *fs)
688 {
689 	ASSERT_SEGLOCK(fs);
690 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
691 }
692 
693 /*
694  * Release blocks associated with the inode ip and stored in the indirect
695  * block bn.  Blocks are free'd in LIFO order up to (but not including)
696  * lastbn.  If level is greater than SINGLE, the block is an indirect block
697  * and recursive calls to indirtrunc must be used to cleanse other indirect
698  * blocks.
699  *
700  * NB: triple indirect blocks are untested.
701  */
702 static int
703 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
704 	       daddr_t lastbn, int level, long *countp,
705 	       long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l)
706 {
707 	int i;
708 	struct buf *bp;
709 	struct lfs *fs = ip->i_lfs;
710 	int32_t *bap;	/* XXX ondisk32 */
711 	struct vnode *vp;
712 	daddr_t nb, nlbn, last;
713 	int32_t *copy = NULL;	/* XXX ondisk32 */
714 	long blkcount, rblkcount, factor;
715 	int nblocks, blocksreleased = 0, real_released = 0;
716 	int error = 0, allerror = 0;
717 
718 	ASSERT_SEGLOCK(fs);
719 	/*
720 	 * Calculate index in current block of last
721 	 * block to be kept.  -1 indicates the entire
722 	 * block so we need not calculate the index.
723 	 */
724 	factor = 1;
725 	for (i = SINGLE; i < level; i++)
726 		factor *= NINDIR(fs);
727 	last = lastbn;
728 	if (lastbn > 0)
729 		last /= factor;
730 	nblocks = btofsb(fs, fs->lfs_bsize);
731 	/*
732 	 * Get buffer of block pointers, zero those entries corresponding
733 	 * to blocks to be free'd, and update on disk copy first.  Since
734 	 * double(triple) indirect before single(double) indirect, calls
735 	 * to bmap on these blocks will fail.  However, we already have
736 	 * the on disk address, so we have to set the b_blkno field
737 	 * explicitly instead of letting bread do everything for us.
738 	 */
739 	vp = ITOV(ip);
740 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
741 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
742 		/* Braces must be here in case trace evaluates to nothing. */
743 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
744 	} else {
745 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
746 		l->l_proc->p_stats->p_ru.ru_inblock++;	/* pay for read */
747 		bp->b_flags |= B_READ;
748 		if (bp->b_bcount > bp->b_bufsize)
749 			panic("lfs_indirtrunc: bad buffer size");
750 		bp->b_blkno = fsbtodb(fs, dbn);
751 		VOP_STRATEGY(vp, bp);
752 		error = biowait(bp);
753 	}
754 	if (error) {
755 		brelse(bp, 0);
756 		*countp = *rcountp = 0;
757 		return (error);
758 	}
759 
760 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
761 	if (lastbn >= 0) {
762 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
763 		memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize);
764 		memset((void *)&bap[last + 1], 0,
765 		/* XXX ondisk32 */
766 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
767 		error = VOP_BWRITE(bp);
768 		if (error)
769 			allerror = error;
770 		bap = copy;
771 	}
772 
773 	/*
774 	 * Recursively free totally unused blocks.
775 	 */
776 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
777 	    i--, nlbn += factor) {
778 		nb = bap[i];
779 		if (nb == 0)
780 			continue;
781 		if (level > SINGLE) {
782 			error = lfs_indirtrunc(ip, nlbn, nb,
783 					       (daddr_t)-1, level - 1,
784 					       &blkcount, &rblkcount,
785 					       lastsegp, bcp, l);
786 			if (error)
787 				allerror = error;
788 			blocksreleased += blkcount;
789 			real_released += rblkcount;
790 		}
791 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
792 		if (bap[i] > 0)
793 			real_released += nblocks;
794 		blocksreleased += nblocks;
795 	}
796 
797 	/*
798 	 * Recursively free last partial block.
799 	 */
800 	if (level > SINGLE && lastbn >= 0) {
801 		last = lastbn % factor;
802 		nb = bap[i];
803 		if (nb != 0) {
804 			error = lfs_indirtrunc(ip, nlbn, nb,
805 					       last, level - 1, &blkcount,
806 					       &rblkcount, lastsegp, bcp, l);
807 			if (error)
808 				allerror = error;
809 			real_released += rblkcount;
810 			blocksreleased += blkcount;
811 		}
812 	}
813 
814 	if (copy != NULL) {
815 		lfs_free(fs, copy, LFS_NB_IBLOCK);
816 	} else {
817 		if (bp->b_flags & B_DELWRI) {
818 			LFS_UNLOCK_BUF(bp);
819 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
820 			wakeup(&fs->lfs_avail);
821 		}
822 		brelse(bp, BC_INVAL);
823 	}
824 
825 	*countp = blocksreleased;
826 	*rcountp = real_released;
827 	return (allerror);
828 }
829 
830 /*
831  * Destroy any in core blocks past the truncation length.
832  * Inlined from vtruncbuf, so that lfs_avail could be updated.
833  * We take the seglock to prevent cleaning from occurring while we are
834  * invalidating blocks.
835  */
836 static int
837 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo)
838 {
839 	struct buf *bp, *nbp;
840 	int s, error;
841 	struct lfs *fs;
842 	voff_t off;
843 
844 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
845 	simple_lock(&vp->v_interlock);
846 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
847 	if (error)
848 		return error;
849 
850 	fs = VTOI(vp)->i_lfs;
851 	s = splbio();
852 
853 	ASSERT_SEGLOCK(fs);
854 restart:
855 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
856 		nbp = LIST_NEXT(bp, b_vnbufs);
857 		if (bp->b_lblkno < lbn)
858 			continue;
859 		simple_lock(&bp->b_interlock);
860 		if (bp->b_flags & B_BUSY) {
861 			bp->b_flags |= B_WANTED;
862 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
863 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
864 			if (error) {
865 				splx(s);
866 				return (error);
867 			}
868 			goto restart;
869 		}
870 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
871 		if (bp->b_flags & B_DELWRI) {
872 			bp->b_flags &= ~B_DELWRI;
873 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
874 			wakeup(&fs->lfs_avail);
875 		}
876 		LFS_UNLOCK_BUF(bp);
877 		simple_unlock(&bp->b_interlock);
878 		brelse(bp, 0);
879 	}
880 
881 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
882 		nbp = LIST_NEXT(bp, b_vnbufs);
883 		if (bp->b_lblkno < lbn)
884 			continue;
885 		simple_lock(&bp->b_interlock);
886 		if (bp->b_flags & B_BUSY) {
887 			bp->b_flags |= B_WANTED;
888 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
889 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
890 			if (error) {
891 				splx(s);
892 				return (error);
893 			}
894 			goto restart;
895 		}
896 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
897 		if (bp->b_flags & B_DELWRI) {
898 			bp->b_flags &= ~B_DELWRI;
899 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
900 			wakeup(&fs->lfs_avail);
901 		}
902 		LFS_UNLOCK_BUF(bp);
903 		simple_unlock(&bp->b_interlock);
904 		brelse(bp, 0);
905 	}
906 
907 	splx(s);
908 
909 	return (0);
910 }
911 
912