xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision fad4c9f71477ae11cea2ee75ec82151ac770a534)
1 /*	$NetBSD: lfs_inode.c,v 1.105 2006/05/14 21:32:45 elad 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 1.105 2006/05/14 21:32:45 elad 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 	struct genfs_node *gp = VTOG(ovp);
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 #ifdef QUOTA
261 	if ((error = getinoquota(oip)) != 0)
262 		return (error);
263 #endif
264 	fs = oip->i_lfs;
265 	lfs_imtime(fs);
266 	osize = oip->i_size;
267 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
268 
269 	ASSERT_NO_SEGLOCK(fs);
270 	/*
271 	 * Lengthen the size of the file. We must ensure that the
272 	 * last byte of the file is allocated. Since the smallest
273 	 * value of osize is 0, length will be at least 1.
274 	 */
275 	if (osize < length) {
276 		if (length > ump->um_maxfilesize)
277 			return (EFBIG);
278 		aflags = B_CLRBUF;
279 		if (ioflag & IO_SYNC)
280 			aflags |= B_SYNC;
281 		if (usepc) {
282 			if (lblkno(fs, osize) < NDADDR &&
283 			    lblkno(fs, osize) != lblkno(fs, length) &&
284 			    blkroundup(fs, osize) != osize) {
285 				off_t eob;
286 
287 				eob = blkroundup(fs, osize);
288 				error = ufs_balloc_range(ovp, osize,
289 				    eob - osize, cred, aflags);
290 				if (error)
291 					return error;
292 				if (ioflag & IO_SYNC) {
293 					ovp->v_size = eob;
294 					simple_lock(&ovp->v_interlock);
295 					VOP_PUTPAGES(ovp,
296 					    trunc_page(osize & fs->lfs_bmask),
297 					    round_page(eob),
298 					    PGO_CLEANIT | PGO_SYNCIO);
299 				}
300 			}
301 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
302 						 aflags);
303 			if (error) {
304 				(void) lfs_truncate(ovp, osize,
305 						    ioflag & IO_SYNC, cred, l);
306 				return error;
307 			}
308 			uvm_vnp_setsize(ovp, length);
309 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
310 			KASSERT(ovp->v_size == oip->i_size);
311 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
312 			return (lfs_update(ovp, NULL, NULL, 0));
313 		} else {
314 			error = lfs_reserve(fs, ovp, NULL,
315 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
316 			if (error)
317 				return (error);
318 			error = lfs_balloc(ovp, length - 1, 1, cred,
319 					   aflags, &bp);
320 			lfs_reserve(fs, ovp, NULL,
321 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
322 			if (error)
323 				return (error);
324 			oip->i_ffs1_size = oip->i_size = length;
325 			uvm_vnp_setsize(ovp, length);
326 			(void) VOP_BWRITE(bp);
327 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
328 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
329 			return (lfs_update(ovp, NULL, NULL, 0));
330 		}
331 	}
332 
333 	if ((error = lfs_reserve(fs, ovp, NULL,
334 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
335 		return (error);
336 
337 	/*
338 	 * Shorten the size of the file. If the file is not being
339 	 * truncated to a block boundary, the contents of the
340 	 * partial block following the end of the file must be
341 	 * zero'ed in case it ever becomes accessible again because
342 	 * of subsequent file growth. Directories however are not
343 	 * zero'ed as they should grow back initialized to empty.
344 	 */
345 	offset = blkoff(fs, length);
346 	lastseg = -1;
347 	bc = 0;
348 
349 	if (ovp != fs->lfs_ivnode)
350 		lfs_seglock(fs, SEGM_PROT);
351 	if (offset == 0) {
352 		oip->i_size = oip->i_ffs1_size = length;
353 	} else if (!usepc) {
354 		lbn = lblkno(fs, length);
355 		aflags = B_CLRBUF;
356 		if (ioflag & IO_SYNC)
357 			aflags |= B_SYNC;
358 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
359 		if (error) {
360 			lfs_reserve(fs, ovp, NULL,
361 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
362 			goto errout;
363 		}
364 		obufsize = bp->b_bufsize;
365 		odb = btofsb(fs, bp->b_bcount);
366 		oip->i_size = oip->i_ffs1_size = length;
367 		size = blksize(fs, oip, lbn);
368 		if (ovp->v_type != VDIR)
369 			memset((char *)bp->b_data + offset, 0,
370 			       (u_int)(size - offset));
371 		allocbuf(bp, size, 1);
372 		if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) {
373 			simple_lock(&lfs_subsys_lock);
374 			locked_queue_bytes -= obufsize - bp->b_bufsize;
375 			simple_unlock(&lfs_subsys_lock);
376 		}
377 		if (bp->b_flags & B_DELWRI)
378 			fs->lfs_avail += odb - btofsb(fs, size);
379 		(void) VOP_BWRITE(bp);
380 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
381 		/*
382 		 * When truncating a regular file down to a non-block-aligned
383 		 * size, we must zero the part of last block which is past
384 		 * the new EOF.  We must synchronously flush the zeroed pages
385 		 * to disk since the new pages will be invalidated as soon
386 		 * as we inform the VM system of the new, smaller size.
387 		 * We must do this before acquiring the GLOCK, since fetching
388 		 * the pages will acquire the GLOCK internally.
389 		 * So there is a window where another thread could see a whole
390 		 * zeroed page past EOF, but that's life.
391 		 */
392 		daddr_t xlbn;
393 		voff_t eoz;
394 
395 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
396 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags);
397 		if (error) {
398 			lfs_reserve(fs, ovp, NULL,
399 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
400 			goto errout;
401 		}
402 		xlbn = lblkno(fs, length);
403 		size = blksize(fs, oip, xlbn);
404 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
405 		uvm_vnp_zerorange(ovp, length, eoz - length);
406 		if (round_page(eoz) > round_page(length)) {
407 			simple_lock(&ovp->v_interlock);
408 			error = VOP_PUTPAGES(ovp, round_page(length),
409 			    round_page(eoz),
410 			    PGO_CLEANIT | PGO_DEACTIVATE |
411 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
412 			if (error) {
413 				lfs_reserve(fs, ovp, NULL,
414 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
415 				goto errout;
416 			}
417 		}
418 	}
419 
420 	lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
421 
422 	oip->i_size = oip->i_ffs1_size = length;
423 	uvm_vnp_setsize(ovp, length);
424 	/*
425 	 * Calculate index into inode's block list of
426 	 * last direct and indirect blocks (if any)
427 	 * which we want to keep.  Lastblock is -1 when
428 	 * the file is truncated to 0.
429 	 */
430 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
431 	if (length > QUAD_MAX - fs->lfs_bsize)
432 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
433 	else
434 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
435 	lastiblock[SINGLE] = lastblock - NDADDR;
436 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
437 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
438 	nblocks = btofsb(fs, fs->lfs_bsize);
439 	/*
440 	 * Record changed file and block pointers before we start
441 	 * freeing blocks.  lastiblock values are also normalized to -1
442 	 * for calls to lfs_indirtrunc below.
443 	 */
444 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks);
445 	for (level = TRIPLE; level >= SINGLE; level--)
446 		if (lastiblock[level] < 0) {
447 			newblks[NDADDR+level] = 0;
448 			lastiblock[level] = -1;
449 		}
450 	for (i = NDADDR - 1; i > lastblock; i--)
451 		newblks[i] = 0;
452 
453 	oip->i_size = oip->i_ffs1_size = osize;
454 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
455 	if (error && !allerror)
456 		allerror = error;
457 
458 	/*
459 	 * Indirect blocks first.
460 	 */
461 	indir_lbn[SINGLE] = -NDADDR;
462 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
463 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
464 	for (level = TRIPLE; level >= SINGLE; level--) {
465 		bn = oip->i_ffs1_ib[level];
466 		if (bn != 0) {
467 			error = lfs_indirtrunc(oip, indir_lbn[level],
468 					       bn, lastiblock[level],
469 					       level, &count, &rcount,
470 					       &lastseg, &bc, l);
471 			if (error)
472 				allerror = error;
473 			real_released += rcount;
474 			blocksreleased += count;
475 			if (lastiblock[level] < 0) {
476 				if (oip->i_ffs1_ib[level] > 0)
477 					real_released += nblocks;
478 				blocksreleased += nblocks;
479 				oip->i_ffs1_ib[level] = 0;
480 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
481 					    &lastseg, &bc);
482         			lfs_deregister_block(ovp, bn);
483 			}
484 		}
485 		if (lastiblock[level] >= 0)
486 			goto done;
487 	}
488 
489 	/*
490 	 * All whole direct blocks or frags.
491 	 */
492 	for (i = NDADDR - 1; i > lastblock; i--) {
493 		long bsize, obsize;
494 
495 		bn = oip->i_ffs1_db[i];
496 		if (bn == 0)
497 			continue;
498 		bsize = blksize(fs, oip, i);
499 		if (oip->i_ffs1_db[i] > 0) {
500 			/* Check for fragment size changes */
501 			obsize = oip->i_lfs_fragsize[i];
502 			real_released += btofsb(fs, obsize);
503 			oip->i_lfs_fragsize[i] = 0;
504 		} else
505 			obsize = 0;
506 		blocksreleased += btofsb(fs, bsize);
507 		oip->i_ffs1_db[i] = 0;
508 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
509         	lfs_deregister_block(ovp, bn);
510 	}
511 	if (lastblock < 0)
512 		goto done;
513 
514 	/*
515 	 * Finally, look for a change in size of the
516 	 * last direct block; release any frags.
517 	 */
518 	bn = oip->i_ffs1_db[lastblock];
519 	if (bn != 0) {
520 		long oldspace, newspace;
521 #if 0
522 		long olddspace;
523 #endif
524 
525 		/*
526 		 * Calculate amount of space we're giving
527 		 * back as old block size minus new block size.
528 		 */
529 		oldspace = blksize(fs, oip, lastblock);
530 #if 0
531 		olddspace = oip->i_lfs_fragsize[lastblock];
532 #endif
533 
534 		oip->i_size = oip->i_ffs1_size = length;
535 		newspace = blksize(fs, oip, lastblock);
536 		if (newspace == 0)
537 			panic("itrunc: newspace");
538 		if (oldspace - newspace > 0) {
539 			blocksreleased += btofsb(fs, oldspace - newspace);
540 		}
541 #if 0
542 		if (bn > 0 && olddspace - newspace > 0) {
543 			/* No segment accounting here, just vnode */
544 			real_released += btofsb(fs, olddspace - newspace);
545 		}
546 #endif
547 	}
548 
549 done:
550 	/* Finish segment accounting corrections */
551 	lfs_update_seguse(fs, oip, lastseg, bc);
552 #ifdef DIAGNOSTIC
553 	for (level = SINGLE; level <= TRIPLE; level++)
554 		if ((newblks[NDADDR + level] == 0) !=
555 		    ((oip->i_ffs1_ib[level]) == 0)) {
556 			panic("lfs itrunc1");
557 		}
558 	for (i = 0; i < NDADDR; i++)
559 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
560 			panic("lfs itrunc2");
561 		}
562 	if (length == 0 &&
563 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
564 		panic("lfs itrunc3");
565 #endif /* DIAGNOSTIC */
566 	/*
567 	 * Put back the real size.
568 	 */
569 	oip->i_size = oip->i_ffs1_size = length;
570 	oip->i_lfs_effnblks -= blocksreleased;
571 	oip->i_ffs1_blocks -= real_released;
572 	simple_lock(&fs->lfs_interlock);
573 	fs->lfs_bfree += blocksreleased;
574 	simple_unlock(&fs->lfs_interlock);
575 #ifdef DIAGNOSTIC
576 	if (oip->i_size == 0 &&
577 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
578 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
579 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
580 		panic("lfs_truncate: persistent blocks");
581 	}
582 #endif
583 
584 	/*
585 	 * If we truncated to zero, take us off the paging queue.
586 	 */
587 	simple_lock(&fs->lfs_interlock);
588 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
589 		oip->i_flags &= ~IN_PAGING;
590 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
591 	}
592 	simple_unlock(&fs->lfs_interlock);
593 
594 	oip->i_flag |= IN_CHANGE;
595 #ifdef QUOTA
596 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
597 #endif
598 	lfs_reserve(fs, ovp, NULL,
599 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
600 	lockmgr(&gp->g_glock, LK_RELEASE, NULL);
601   errout:
602 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
603 	if (ovp != fs->lfs_ivnode)
604 		lfs_segunlock(fs);
605 	return (allerror ? allerror : error);
606 }
607 
608 /* Update segment and avail usage information when removing a block. */
609 static int
610 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
611 	    size_t bsize, long *lastseg, size_t *num)
612 {
613 	long seg;
614 	int error = 0;
615 
616 	ASSERT_SEGLOCK(fs);
617 	bsize = fragroundup(fs, bsize);
618 	if (daddr > 0) {
619 		if (*lastseg != (seg = dtosn(fs, daddr))) {
620 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
621 			*num = bsize;
622 			*lastseg = seg;
623 		} else
624 			*num += bsize;
625 	}
626 
627 	return error;
628 }
629 
630 /* Finish the accounting updates for a segment. */
631 static int
632 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
633 {
634 	struct segdelta *sd;
635 	struct vnode *vp;
636 
637 	ASSERT_SEGLOCK(fs);
638 	if (lastseg < 0 || num == 0)
639 		return 0;
640 
641 	vp = ITOV(ip);
642 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
643 		if (sd->segnum == lastseg)
644 			break;
645 	if (sd == NULL) {
646 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
647 		sd->segnum = lastseg;
648 		sd->num = 0;
649 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
650 	}
651 	sd->num += num;
652 
653 	return 0;
654 }
655 
656 static void
657 lfs_finalize_seguse(struct lfs *fs, void *v)
658 {
659 	SEGUSE *sup;
660 	struct buf *bp;
661 	struct segdelta *sd;
662 	LIST_HEAD(, segdelta) *hd = v;
663 
664 	ASSERT_SEGLOCK(fs);
665 	while((sd = LIST_FIRST(hd)) != NULL) {
666 		LIST_REMOVE(sd, list);
667 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
668 		if (sd->num > sup->su_nbytes) {
669 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
670 				sd->segnum, (long)(sd->num - sup->su_nbytes));
671 			panic("lfs_finalize_seguse: negative bytes");
672 			sup->su_nbytes = sd->num;
673 		}
674 		sup->su_nbytes -= sd->num;
675 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
676 		free(sd, M_SEGMENT);
677 	}
678 }
679 
680 /* Finish the accounting updates for a segment. */
681 void
682 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
683 {
684 	ASSERT_SEGLOCK(fs);
685 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
686 }
687 
688 /* Finish the accounting updates for a segment. */
689 void
690 lfs_finalize_fs_seguse(struct lfs *fs)
691 {
692 	ASSERT_SEGLOCK(fs);
693 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
694 }
695 
696 /*
697  * Release blocks associated with the inode ip and stored in the indirect
698  * block bn.  Blocks are free'd in LIFO order up to (but not including)
699  * lastbn.  If level is greater than SINGLE, the block is an indirect block
700  * and recursive calls to indirtrunc must be used to cleanse other indirect
701  * blocks.
702  *
703  * NB: triple indirect blocks are untested.
704  */
705 static int
706 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
707 	       daddr_t lastbn, int level, long *countp,
708 	       long *rcountp, long *lastsegp, size_t *bcp, struct lwp *l)
709 {
710 	int i;
711 	struct buf *bp;
712 	struct lfs *fs = ip->i_lfs;
713 	int32_t *bap;	/* XXX ondisk32 */
714 	struct vnode *vp;
715 	daddr_t nb, nlbn, last;
716 	int32_t *copy = NULL;	/* XXX ondisk32 */
717 	long blkcount, rblkcount, factor;
718 	int nblocks, blocksreleased = 0, real_released = 0;
719 	int error = 0, allerror = 0;
720 
721 	ASSERT_SEGLOCK(fs);
722 	/*
723 	 * Calculate index in current block of last
724 	 * block to be kept.  -1 indicates the entire
725 	 * block so we need not calculate the index.
726 	 */
727 	factor = 1;
728 	for (i = SINGLE; i < level; i++)
729 		factor *= NINDIR(fs);
730 	last = lastbn;
731 	if (lastbn > 0)
732 		last /= factor;
733 	nblocks = btofsb(fs, fs->lfs_bsize);
734 	/*
735 	 * Get buffer of block pointers, zero those entries corresponding
736 	 * to blocks to be free'd, and update on disk copy first.  Since
737 	 * double(triple) indirect before single(double) indirect, calls
738 	 * to bmap on these blocks will fail.  However, we already have
739 	 * the on disk address, so we have to set the b_blkno field
740 	 * explicitly instead of letting bread do everything for us.
741 	 */
742 	vp = ITOV(ip);
743 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
744 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
745 		/* Braces must be here in case trace evaluates to nothing. */
746 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
747 	} else {
748 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
749 		l->l_proc->p_stats->p_ru.ru_inblock++;	/* pay for read */
750 		bp->b_flags |= B_READ;
751 		if (bp->b_bcount > bp->b_bufsize)
752 			panic("lfs_indirtrunc: bad buffer size");
753 		bp->b_blkno = fsbtodb(fs, dbn);
754 		VOP_STRATEGY(vp, bp);
755 		error = biowait(bp);
756 	}
757 	if (error) {
758 		brelse(bp);
759 		*countp = *rcountp = 0;
760 		return (error);
761 	}
762 
763 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
764 	if (lastbn >= 0) {
765 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
766 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
767 		memset((caddr_t)&bap[last + 1], 0,
768 		/* XXX ondisk32 */
769 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
770 		error = VOP_BWRITE(bp);
771 		if (error)
772 			allerror = error;
773 		bap = copy;
774 	}
775 
776 	/*
777 	 * Recursively free totally unused blocks.
778 	 */
779 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
780 	    i--, nlbn += factor) {
781 		nb = bap[i];
782 		if (nb == 0)
783 			continue;
784 		if (level > SINGLE) {
785 			error = lfs_indirtrunc(ip, nlbn, nb,
786 					       (daddr_t)-1, level - 1,
787 					       &blkcount, &rblkcount,
788 					       lastsegp, bcp, l);
789 			if (error)
790 				allerror = error;
791 			blocksreleased += blkcount;
792 			real_released += rblkcount;
793 		}
794 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
795 		if (bap[i] > 0)
796 			real_released += nblocks;
797 		blocksreleased += nblocks;
798 	}
799 
800 	/*
801 	 * Recursively free last partial block.
802 	 */
803 	if (level > SINGLE && lastbn >= 0) {
804 		last = lastbn % factor;
805 		nb = bap[i];
806 		if (nb != 0) {
807 			error = lfs_indirtrunc(ip, nlbn, nb,
808 					       last, level - 1, &blkcount,
809 					       &rblkcount, lastsegp, bcp, l);
810 			if (error)
811 				allerror = error;
812 			real_released += rblkcount;
813 			blocksreleased += blkcount;
814 		}
815 	}
816 
817 	if (copy != NULL) {
818 		lfs_free(fs, copy, LFS_NB_IBLOCK);
819 	} else {
820 		if (bp->b_flags & B_DELWRI) {
821 			LFS_UNLOCK_BUF(bp);
822 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
823 			wakeup(&fs->lfs_avail);
824 		}
825 		bp->b_flags |= B_INVAL;
826 		brelse(bp);
827 	}
828 
829 	*countp = blocksreleased;
830 	*rcountp = real_released;
831 	return (allerror);
832 }
833 
834 /*
835  * Destroy any in core blocks past the truncation length.
836  * Inlined from vtruncbuf, so that lfs_avail could be updated.
837  * We take the seglock to prevent cleaning from occurring while we are
838  * invalidating blocks.
839  */
840 static int
841 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo)
842 {
843 	struct buf *bp, *nbp;
844 	int s, error;
845 	struct lfs *fs;
846 	voff_t off;
847 
848 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
849 	simple_lock(&vp->v_interlock);
850 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
851 	if (error)
852 		return error;
853 
854 	fs = VTOI(vp)->i_lfs;
855 	s = splbio();
856 
857 	ASSERT_SEGLOCK(fs);
858 restart:
859 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
860 		nbp = LIST_NEXT(bp, b_vnbufs);
861 		if (bp->b_lblkno < lbn)
862 			continue;
863 		simple_lock(&bp->b_interlock);
864 		if (bp->b_flags & B_BUSY) {
865 			bp->b_flags |= B_WANTED;
866 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
867 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
868 			if (error) {
869 				splx(s);
870 				return (error);
871 			}
872 			goto restart;
873 		}
874 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
875 		if (bp->b_flags & B_DELWRI) {
876 			bp->b_flags &= ~B_DELWRI;
877 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
878 			wakeup(&fs->lfs_avail);
879 		}
880 		LFS_UNLOCK_BUF(bp);
881 		simple_unlock(&bp->b_interlock);
882 		brelse(bp);
883 	}
884 
885 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
886 		nbp = LIST_NEXT(bp, b_vnbufs);
887 		if (bp->b_lblkno < lbn)
888 			continue;
889 		simple_lock(&bp->b_interlock);
890 		if (bp->b_flags & B_BUSY) {
891 			bp->b_flags |= B_WANTED;
892 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
893 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
894 			if (error) {
895 				splx(s);
896 				return (error);
897 			}
898 			goto restart;
899 		}
900 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
901 		if (bp->b_flags & B_DELWRI) {
902 			bp->b_flags &= ~B_DELWRI;
903 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
904 			wakeup(&fs->lfs_avail);
905 		}
906 		LFS_UNLOCK_BUF(bp);
907 		simple_unlock(&bp->b_interlock);
908 		brelse(bp);
909 	}
910 
911 	splx(s);
912 
913 	return (0);
914 }
915 
916