xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: lfs_inode.c,v 1.120 2008/04/28 20:24:11 martin 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*
32  * Copyright (c) 1986, 1989, 1991, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
60  */
61 
62 #include <sys/cdefs.h>
63 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.120 2008/04/28 20:24:11 martin Exp $");
64 
65 #if defined(_KERNEL_OPT)
66 #include "opt_quota.h"
67 #endif
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/mount.h>
72 #include <sys/malloc.h>
73 #include <sys/proc.h>
74 #include <sys/file.h>
75 #include <sys/buf.h>
76 #include <sys/vnode.h>
77 #include <sys/kernel.h>
78 #include <sys/trace.h>
79 #include <sys/resourcevar.h>
80 #include <sys/kauth.h>
81 
82 #include <ufs/ufs/quota.h>
83 #include <ufs/ufs/inode.h>
84 #include <ufs/ufs/ufsmount.h>
85 #include <ufs/ufs/ufs_extern.h>
86 
87 #include <ufs/lfs/lfs.h>
88 #include <ufs/lfs/lfs_extern.h>
89 
90 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t);
91 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t,
92 			   daddr_t, int, long *, long *, long *, size_t *);
93 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *);
94 static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int);
95 
96 /* Search a block for a specific dinode. */
97 struct ufs1_dinode *
98 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
99 {
100 	struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data;
101 	struct ufs1_dinode *ldip, *fin;
102 
103 	ASSERT_NO_SEGLOCK(fs);
104 	/*
105 	 * Read the inode block backwards, since later versions of the
106 	 * inode will supercede earlier ones.  Though it is unlikely, it is
107 	 * possible that the same inode will appear in the same inode block.
108 	 */
109 	fin = dip + INOPB(fs);
110 	for (ldip = fin - 1; ldip >= dip; --ldip)
111 		if (ldip->di_inumber == ino)
112 			return (ldip);
113 
114 	printf("searched %d entries\n", (int)(fin - dip));
115 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
116 	       dtosn(fs, fs->lfs_offset));
117 	printf("block is 0x%llx (seg %lld)\n",
118 	       (unsigned long long)dbtofsb(fs, bp->b_blkno),
119 	       (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno)));
120 
121 	return NULL;
122 }
123 
124 int
125 lfs_update(struct vnode *vp, const struct timespec *acc,
126     const struct timespec *mod, int updflags)
127 {
128 	struct inode *ip;
129 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
130 	int flags;
131 
132 	ASSERT_NO_SEGLOCK(fs);
133 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
134 		return (0);
135 	ip = VTOI(vp);
136 
137 	/*
138 	 * If we are called from vinvalbuf, and the file's blocks have
139 	 * already been scheduled for writing, but the writes have not
140 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
141 	 * will cause a panic.	So, we must wait until any pending write
142 	 * for our inode completes, if we are called with UPDATE_WAIT set.
143 	 */
144 	mutex_enter(&vp->v_interlock);
145 	while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
146 	    WRITEINPROG(vp)) {
147 		DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d"
148 		      " (in progress)\n", ip->i_number));
149 		cv_wait(&vp->v_cv, &vp->v_interlock);
150 	}
151 	mutex_exit(&vp->v_interlock);
152 	LFS_ITIMES(ip, acc, mod, NULL);
153 	if (updflags & UPDATE_CLOSE)
154 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING);
155 	else
156 		flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING);
157 	if (flags == 0)
158 		return (0);
159 
160 	/* If sync, push back the vnode and any dirty blocks it may have. */
161 	if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) {
162 		/* Avoid flushing VU_DIROP. */
163 		mutex_enter(&lfs_lock);
164 		++fs->lfs_diropwait;
165 		while (vp->v_uflag & VU_DIROP) {
166 			DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d"
167 			      " (dirops)\n", ip->i_number));
168 			DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags"
169 			      " 0x%x\n",
170 			      vp->v_iflag | vp->v_vflag | vp->v_uflag,
171 			      ip->i_flag));
172 			if (fs->lfs_dirops == 0)
173 				lfs_flush_fs(fs, SEGM_SYNC);
174 			else
175 				mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
176 					0, &lfs_lock);
177 			/* XXX KS - by falling out here, are we writing the vn
178 			twice? */
179 		}
180 		--fs->lfs_diropwait;
181 		mutex_exit(&lfs_lock);
182 		return lfs_vflush(vp);
183 	}
184 	return 0;
185 }
186 
187 #define	SINGLE	0	/* index of single indirect block */
188 #define	DOUBLE	1	/* index of double indirect block */
189 #define	TRIPLE	2	/* index of triple indirect block */
190 /*
191  * Truncate the inode oip to at most length size, freeing the
192  * disk blocks.
193  */
194 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */
195 
196 int
197 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
198 {
199 	daddr_t lastblock;
200 	struct inode *oip = VTOI(ovp);
201 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
202 	/* XXX ondisk32 */
203 	int32_t newblks[NDADDR + NIADDR];
204 	struct lfs *fs;
205 	struct buf *bp;
206 	int offset, size, level;
207 	long count, rcount, blocksreleased = 0, real_released = 0;
208 	int i, nblocks;
209 	int aflags, error, allerror = 0;
210 	off_t osize;
211 	long lastseg;
212 	size_t bc;
213 	int obufsize, odb;
214 	int usepc;
215 	struct ufsmount *ump = oip->i_ump;
216 
217 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
218 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
219 		KASSERT(oip->i_size == 0);
220 		return 0;
221 	}
222 
223 	if (length < 0)
224 		return (EINVAL);
225 
226 	/*
227 	 * Just return and not update modification times.
228 	 */
229 	if (oip->i_size == length)
230 		return (0);
231 
232 	if (ovp->v_type == VLNK &&
233 	    (oip->i_size < ump->um_maxsymlinklen ||
234 	     (ump->um_maxsymlinklen == 0 &&
235 	      oip->i_ffs1_blocks == 0))) {
236 #ifdef DIAGNOSTIC
237 		if (length != 0)
238 			panic("lfs_truncate: partial truncate of symlink");
239 #endif
240 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
241 		oip->i_size = oip->i_ffs1_size = 0;
242 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
243 		return (lfs_update(ovp, NULL, NULL, 0));
244 	}
245 	if (oip->i_size == length) {
246 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
247 		return (lfs_update(ovp, NULL, NULL, 0));
248 	}
249 	fs = oip->i_lfs;
250 	lfs_imtime(fs);
251 	osize = oip->i_size;
252 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
253 
254 	ASSERT_NO_SEGLOCK(fs);
255 	/*
256 	 * Lengthen the size of the file. We must ensure that the
257 	 * last byte of the file is allocated. Since the smallest
258 	 * value of osize is 0, length will be at least 1.
259 	 */
260 	if (osize < length) {
261 		if (length > ump->um_maxfilesize)
262 			return (EFBIG);
263 		aflags = B_CLRBUF;
264 		if (ioflag & IO_SYNC)
265 			aflags |= B_SYNC;
266 		if (usepc) {
267 			if (lblkno(fs, osize) < NDADDR &&
268 			    lblkno(fs, osize) != lblkno(fs, length) &&
269 			    blkroundup(fs, osize) != osize) {
270 				off_t eob;
271 
272 				eob = blkroundup(fs, osize);
273 				uvm_vnp_setwritesize(ovp, eob);
274 				error = ufs_balloc_range(ovp, osize,
275 				    eob - osize, cred, aflags);
276 				if (error)
277 					return error;
278 				if (ioflag & IO_SYNC) {
279 					mutex_enter(&ovp->v_interlock);
280 					VOP_PUTPAGES(ovp,
281 					    trunc_page(osize & fs->lfs_bmask),
282 					    round_page(eob),
283 					    PGO_CLEANIT | PGO_SYNCIO);
284 				}
285 			}
286 			uvm_vnp_setwritesize(ovp, length);
287 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
288 						 aflags);
289 			if (error) {
290 				(void) lfs_truncate(ovp, osize,
291 						    ioflag & IO_SYNC, cred);
292 				return error;
293 			}
294 			uvm_vnp_setsize(ovp, length);
295 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
296 			KASSERT(ovp->v_size == oip->i_size);
297 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
298 			return (lfs_update(ovp, NULL, NULL, 0));
299 		} else {
300 			error = lfs_reserve(fs, ovp, NULL,
301 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
302 			if (error)
303 				return (error);
304 			error = lfs_balloc(ovp, length - 1, 1, cred,
305 					   aflags, &bp);
306 			lfs_reserve(fs, ovp, NULL,
307 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
308 			if (error)
309 				return (error);
310 			oip->i_ffs1_size = oip->i_size = length;
311 			uvm_vnp_setsize(ovp, length);
312 			(void) VOP_BWRITE(bp);
313 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
314 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
315 			return (lfs_update(ovp, NULL, NULL, 0));
316 		}
317 	}
318 
319 	if ((error = lfs_reserve(fs, ovp, NULL,
320 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
321 		return (error);
322 
323 	/*
324 	 * Shorten the size of the file. If the file is not being
325 	 * truncated to a block boundary, the contents of the
326 	 * partial block following the end of the file must be
327 	 * zero'ed in case it ever becomes accessible again because
328 	 * of subsequent file growth. Directories however are not
329 	 * zero'ed as they should grow back initialized to empty.
330 	 */
331 	offset = blkoff(fs, length);
332 	lastseg = -1;
333 	bc = 0;
334 
335 	if (ovp != fs->lfs_ivnode)
336 		lfs_seglock(fs, SEGM_PROT);
337 	if (offset == 0) {
338 		oip->i_size = oip->i_ffs1_size = length;
339 	} else if (!usepc) {
340 		lbn = lblkno(fs, length);
341 		aflags = B_CLRBUF;
342 		if (ioflag & IO_SYNC)
343 			aflags |= B_SYNC;
344 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
345 		if (error) {
346 			lfs_reserve(fs, ovp, NULL,
347 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
348 			goto errout;
349 		}
350 		obufsize = bp->b_bufsize;
351 		odb = btofsb(fs, bp->b_bcount);
352 		oip->i_size = oip->i_ffs1_size = length;
353 		size = blksize(fs, oip, lbn);
354 		if (ovp->v_type != VDIR)
355 			memset((char *)bp->b_data + offset, 0,
356 			       (u_int)(size - offset));
357 		allocbuf(bp, size, 1);
358 		if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
359 			mutex_enter(&lfs_lock);
360 			locked_queue_bytes -= obufsize - bp->b_bufsize;
361 			mutex_exit(&lfs_lock);
362 		}
363 		if (bp->b_oflags & BO_DELWRI)
364 			fs->lfs_avail += odb - btofsb(fs, size);
365 		(void) VOP_BWRITE(bp);
366 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
367 		/*
368 		 * When truncating a regular file down to a non-block-aligned
369 		 * size, we must zero the part of last block which is past
370 		 * the new EOF.  We must synchronously flush the zeroed pages
371 		 * to disk since the new pages will be invalidated as soon
372 		 * as we inform the VM system of the new, smaller size.
373 		 * We must do this before acquiring the GLOCK, since fetching
374 		 * the pages will acquire the GLOCK internally.
375 		 * So there is a window where another thread could see a whole
376 		 * zeroed page past EOF, but that's life.
377 		 */
378 		daddr_t xlbn;
379 		voff_t eoz;
380 
381 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
382 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags);
383 		if (error) {
384 			lfs_reserve(fs, ovp, NULL,
385 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
386 			goto errout;
387 		}
388 		xlbn = lblkno(fs, length);
389 		size = blksize(fs, oip, xlbn);
390 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
391 		uvm_vnp_zerorange(ovp, length, eoz - length);
392 		if (round_page(eoz) > round_page(length)) {
393 			mutex_enter(&ovp->v_interlock);
394 			error = VOP_PUTPAGES(ovp, round_page(length),
395 			    round_page(eoz),
396 			    PGO_CLEANIT | PGO_DEACTIVATE |
397 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
398 			if (error) {
399 				lfs_reserve(fs, ovp, NULL,
400 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
401 				goto errout;
402 			}
403 		}
404 	}
405 
406 	genfs_node_wrlock(ovp);
407 
408 	oip->i_size = oip->i_ffs1_size = length;
409 	uvm_vnp_setsize(ovp, length);
410 	/*
411 	 * Calculate index into inode's block list of
412 	 * last direct and indirect blocks (if any)
413 	 * which we want to keep.  Lastblock is -1 when
414 	 * the file is truncated to 0.
415 	 */
416 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
417 	if (length > QUAD_MAX - fs->lfs_bsize)
418 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
419 	else
420 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
421 	lastiblock[SINGLE] = lastblock - NDADDR;
422 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
423 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
424 	nblocks = btofsb(fs, fs->lfs_bsize);
425 	/*
426 	 * Record changed file and block pointers before we start
427 	 * freeing blocks.  lastiblock values are also normalized to -1
428 	 * for calls to lfs_indirtrunc below.
429 	 */
430 	memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
431 	for (level = TRIPLE; level >= SINGLE; level--)
432 		if (lastiblock[level] < 0) {
433 			newblks[NDADDR+level] = 0;
434 			lastiblock[level] = -1;
435 		}
436 	for (i = NDADDR - 1; i > lastblock; i--)
437 		newblks[i] = 0;
438 
439 	oip->i_size = oip->i_ffs1_size = osize;
440 	error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
441 	if (error && !allerror)
442 		allerror = error;
443 
444 	/*
445 	 * Indirect blocks first.
446 	 */
447 	indir_lbn[SINGLE] = -NDADDR;
448 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
449 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
450 	for (level = TRIPLE; level >= SINGLE; level--) {
451 		bn = oip->i_ffs1_ib[level];
452 		if (bn != 0) {
453 			error = lfs_indirtrunc(oip, indir_lbn[level],
454 					       bn, lastiblock[level],
455 					       level, &count, &rcount,
456 					       &lastseg, &bc);
457 			if (error)
458 				allerror = error;
459 			real_released += rcount;
460 			blocksreleased += count;
461 			if (lastiblock[level] < 0) {
462 				if (oip->i_ffs1_ib[level] > 0)
463 					real_released += nblocks;
464 				blocksreleased += nblocks;
465 				oip->i_ffs1_ib[level] = 0;
466 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
467 					    &lastseg, &bc);
468         			lfs_deregister_block(ovp, bn);
469 			}
470 		}
471 		if (lastiblock[level] >= 0)
472 			goto done;
473 	}
474 
475 	/*
476 	 * All whole direct blocks or frags.
477 	 */
478 	for (i = NDADDR - 1; i > lastblock; i--) {
479 		long bsize, obsize;
480 
481 		bn = oip->i_ffs1_db[i];
482 		if (bn == 0)
483 			continue;
484 		bsize = blksize(fs, oip, i);
485 		if (oip->i_ffs1_db[i] > 0) {
486 			/* Check for fragment size changes */
487 			obsize = oip->i_lfs_fragsize[i];
488 			real_released += btofsb(fs, obsize);
489 			oip->i_lfs_fragsize[i] = 0;
490 		} else
491 			obsize = 0;
492 		blocksreleased += btofsb(fs, bsize);
493 		oip->i_ffs1_db[i] = 0;
494 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
495         	lfs_deregister_block(ovp, bn);
496 	}
497 	if (lastblock < 0)
498 		goto done;
499 
500 	/*
501 	 * Finally, look for a change in size of the
502 	 * last direct block; release any frags.
503 	 */
504 	bn = oip->i_ffs1_db[lastblock];
505 	if (bn != 0) {
506 		long oldspace, newspace;
507 #if 0
508 		long olddspace;
509 #endif
510 
511 		/*
512 		 * Calculate amount of space we're giving
513 		 * back as old block size minus new block size.
514 		 */
515 		oldspace = blksize(fs, oip, lastblock);
516 #if 0
517 		olddspace = oip->i_lfs_fragsize[lastblock];
518 #endif
519 
520 		oip->i_size = oip->i_ffs1_size = length;
521 		newspace = blksize(fs, oip, lastblock);
522 		if (newspace == 0)
523 			panic("itrunc: newspace");
524 		if (oldspace - newspace > 0) {
525 			blocksreleased += btofsb(fs, oldspace - newspace);
526 		}
527 #if 0
528 		if (bn > 0 && olddspace - newspace > 0) {
529 			/* No segment accounting here, just vnode */
530 			real_released += btofsb(fs, olddspace - newspace);
531 		}
532 #endif
533 	}
534 
535 done:
536 	/* Finish segment accounting corrections */
537 	lfs_update_seguse(fs, oip, lastseg, bc);
538 #ifdef DIAGNOSTIC
539 	for (level = SINGLE; level <= TRIPLE; level++)
540 		if ((newblks[NDADDR + level] == 0) !=
541 		    ((oip->i_ffs1_ib[level]) == 0)) {
542 			panic("lfs itrunc1");
543 		}
544 	for (i = 0; i < NDADDR; i++)
545 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
546 			panic("lfs itrunc2");
547 		}
548 	if (length == 0 &&
549 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
550 		panic("lfs itrunc3");
551 #endif /* DIAGNOSTIC */
552 	/*
553 	 * Put back the real size.
554 	 */
555 	oip->i_size = oip->i_ffs1_size = length;
556 	oip->i_lfs_effnblks -= blocksreleased;
557 	oip->i_ffs1_blocks -= real_released;
558 	mutex_enter(&lfs_lock);
559 	fs->lfs_bfree += blocksreleased;
560 	mutex_exit(&lfs_lock);
561 #ifdef DIAGNOSTIC
562 	if (oip->i_size == 0 &&
563 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
564 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
565 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
566 		panic("lfs_truncate: persistent blocks");
567 	}
568 #endif
569 
570 	/*
571 	 * If we truncated to zero, take us off the paging queue.
572 	 */
573 	mutex_enter(&lfs_lock);
574 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
575 		oip->i_flags &= ~IN_PAGING;
576 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
577 	}
578 	mutex_exit(&lfs_lock);
579 
580 	oip->i_flag |= IN_CHANGE;
581 #ifdef QUOTA
582 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
583 #endif
584 	lfs_reserve(fs, ovp, NULL,
585 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
586 	genfs_node_unlock(ovp);
587   errout:
588 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
589 	if (ovp != fs->lfs_ivnode)
590 		lfs_segunlock(fs);
591 	return (allerror ? allerror : error);
592 }
593 
594 /* Update segment and avail usage information when removing a block. */
595 static int
596 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
597 	    size_t bsize, long *lastseg, size_t *num)
598 {
599 	long seg;
600 	int error = 0;
601 
602 	ASSERT_SEGLOCK(fs);
603 	bsize = fragroundup(fs, bsize);
604 	if (daddr > 0) {
605 		if (*lastseg != (seg = dtosn(fs, daddr))) {
606 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
607 			*num = bsize;
608 			*lastseg = seg;
609 		} else
610 			*num += bsize;
611 	}
612 
613 	return error;
614 }
615 
616 /* Finish the accounting updates for a segment. */
617 static int
618 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
619 {
620 	struct segdelta *sd;
621 	struct vnode *vp;
622 
623 	ASSERT_SEGLOCK(fs);
624 	if (lastseg < 0 || num == 0)
625 		return 0;
626 
627 	vp = ITOV(ip);
628 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
629 		if (sd->segnum == lastseg)
630 			break;
631 	if (sd == NULL) {
632 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
633 		sd->segnum = lastseg;
634 		sd->num = 0;
635 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
636 	}
637 	sd->num += num;
638 
639 	return 0;
640 }
641 
642 static void
643 lfs_finalize_seguse(struct lfs *fs, void *v)
644 {
645 	SEGUSE *sup;
646 	struct buf *bp;
647 	struct segdelta *sd;
648 	LIST_HEAD(, segdelta) *hd = v;
649 
650 	ASSERT_SEGLOCK(fs);
651 	while((sd = LIST_FIRST(hd)) != NULL) {
652 		LIST_REMOVE(sd, list);
653 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
654 		if (sd->num > sup->su_nbytes) {
655 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
656 				sd->segnum, (long)(sd->num - sup->su_nbytes));
657 			panic("lfs_finalize_seguse: negative bytes");
658 			sup->su_nbytes = sd->num;
659 		}
660 		sup->su_nbytes -= sd->num;
661 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
662 		free(sd, M_SEGMENT);
663 	}
664 }
665 
666 /* Finish the accounting updates for a segment. */
667 void
668 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
669 {
670 	ASSERT_SEGLOCK(fs);
671 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
672 }
673 
674 /* Finish the accounting updates for a segment. */
675 void
676 lfs_finalize_fs_seguse(struct lfs *fs)
677 {
678 	ASSERT_SEGLOCK(fs);
679 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
680 }
681 
682 /*
683  * Release blocks associated with the inode ip and stored in the indirect
684  * block bn.  Blocks are free'd in LIFO order up to (but not including)
685  * lastbn.  If level is greater than SINGLE, the block is an indirect block
686  * and recursive calls to indirtrunc must be used to cleanse other indirect
687  * blocks.
688  *
689  * NB: triple indirect blocks are untested.
690  */
691 static int
692 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
693 	       daddr_t lastbn, int level, long *countp,
694 	       long *rcountp, long *lastsegp, size_t *bcp)
695 {
696 	int i;
697 	struct buf *bp;
698 	struct lfs *fs = ip->i_lfs;
699 	int32_t *bap;	/* XXX ondisk32 */
700 	struct vnode *vp;
701 	daddr_t nb, nlbn, last;
702 	int32_t *copy = NULL;	/* XXX ondisk32 */
703 	long blkcount, rblkcount, factor;
704 	int nblocks, blocksreleased = 0, real_released = 0;
705 	int error = 0, allerror = 0;
706 
707 	ASSERT_SEGLOCK(fs);
708 	/*
709 	 * Calculate index in current block of last
710 	 * block to be kept.  -1 indicates the entire
711 	 * block so we need not calculate the index.
712 	 */
713 	factor = 1;
714 	for (i = SINGLE; i < level; i++)
715 		factor *= NINDIR(fs);
716 	last = lastbn;
717 	if (lastbn > 0)
718 		last /= factor;
719 	nblocks = btofsb(fs, fs->lfs_bsize);
720 	/*
721 	 * Get buffer of block pointers, zero those entries corresponding
722 	 * to blocks to be free'd, and update on disk copy first.  Since
723 	 * double(triple) indirect before single(double) indirect, calls
724 	 * to bmap on these blocks will fail.  However, we already have
725 	 * the on disk address, so we have to set the b_blkno field
726 	 * explicitly instead of letting bread do everything for us.
727 	 */
728 	vp = ITOV(ip);
729 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
730 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
731 		/* Braces must be here in case trace evaluates to nothing. */
732 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
733 	} else {
734 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
735 		curlwp->l_ru.ru_inblock++; /* pay for read */
736 		bp->b_flags |= B_READ;
737 		if (bp->b_bcount > bp->b_bufsize)
738 			panic("lfs_indirtrunc: bad buffer size");
739 		bp->b_blkno = fsbtodb(fs, dbn);
740 		VOP_STRATEGY(vp, bp);
741 		error = biowait(bp);
742 	}
743 	if (error) {
744 		brelse(bp, 0);
745 		*countp = *rcountp = 0;
746 		return (error);
747 	}
748 
749 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
750 	if (lastbn >= 0) {
751 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
752 		memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize);
753 		memset((void *)&bap[last + 1], 0,
754 		/* XXX ondisk32 */
755 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
756 		error = VOP_BWRITE(bp);
757 		if (error)
758 			allerror = error;
759 		bap = copy;
760 	}
761 
762 	/*
763 	 * Recursively free totally unused blocks.
764 	 */
765 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
766 	    i--, nlbn += factor) {
767 		nb = bap[i];
768 		if (nb == 0)
769 			continue;
770 		if (level > SINGLE) {
771 			error = lfs_indirtrunc(ip, nlbn, nb,
772 					       (daddr_t)-1, level - 1,
773 					       &blkcount, &rblkcount,
774 					       lastsegp, bcp);
775 			if (error)
776 				allerror = error;
777 			blocksreleased += blkcount;
778 			real_released += rblkcount;
779 		}
780 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
781 		if (bap[i] > 0)
782 			real_released += nblocks;
783 		blocksreleased += nblocks;
784 	}
785 
786 	/*
787 	 * Recursively free last partial block.
788 	 */
789 	if (level > SINGLE && lastbn >= 0) {
790 		last = lastbn % factor;
791 		nb = bap[i];
792 		if (nb != 0) {
793 			error = lfs_indirtrunc(ip, nlbn, nb,
794 					       last, level - 1, &blkcount,
795 					       &rblkcount, lastsegp, bcp);
796 			if (error)
797 				allerror = error;
798 			real_released += rblkcount;
799 			blocksreleased += blkcount;
800 		}
801 	}
802 
803 	if (copy != NULL) {
804 		lfs_free(fs, copy, LFS_NB_IBLOCK);
805 	} else {
806 		mutex_enter(&bufcache_lock);
807 		if (bp->b_oflags & BO_DELWRI) {
808 			LFS_UNLOCK_BUF(bp);
809 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
810 			wakeup(&fs->lfs_avail);
811 		}
812 		brelsel(bp, BC_INVAL);
813 		mutex_exit(&bufcache_lock);
814 	}
815 
816 	*countp = blocksreleased;
817 	*rcountp = real_released;
818 	return (allerror);
819 }
820 
821 /*
822  * Destroy any in core blocks past the truncation length.
823  * Inlined from vtruncbuf, so that lfs_avail could be updated.
824  * We take the seglock to prevent cleaning from occurring while we are
825  * invalidating blocks.
826  */
827 static int
828 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
829 {
830 	struct buf *bp, *nbp;
831 	int error;
832 	struct lfs *fs;
833 	voff_t off;
834 
835 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
836 	mutex_enter(&vp->v_interlock);
837 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
838 	if (error)
839 		return error;
840 
841 	fs = VTOI(vp)->i_lfs;
842 
843 	ASSERT_SEGLOCK(fs);
844 
845 	mutex_enter(&bufcache_lock);
846 restart:
847 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
848 		nbp = LIST_NEXT(bp, b_vnbufs);
849 		if (bp->b_lblkno < lbn)
850 			continue;
851 		error = bbusy(bp, catch, slptimeo, NULL);
852 		if (error == EPASSTHROUGH)
853 			goto restart;
854 		if (error != 0) {
855 			mutex_exit(&bufcache_lock);
856 			return (error);
857 		}
858 		mutex_enter(bp->b_objlock);
859 		if (bp->b_oflags & BO_DELWRI) {
860 			bp->b_oflags &= ~BO_DELWRI;
861 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
862 			wakeup(&fs->lfs_avail);
863 		}
864 		mutex_exit(bp->b_objlock);
865 		LFS_UNLOCK_BUF(bp);
866 		brelsel(bp, BC_INVAL | BC_VFLUSH);
867 	}
868 
869 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
870 		nbp = LIST_NEXT(bp, b_vnbufs);
871 		if (bp->b_lblkno < lbn)
872 			continue;
873 		error = bbusy(bp, catch, slptimeo, NULL);
874 		if (error == EPASSTHROUGH)
875 			goto restart;
876 		if (error != 0) {
877 			mutex_exit(&bufcache_lock);
878 			return (error);
879 		}
880 		mutex_enter(bp->b_objlock);
881 		if (bp->b_oflags & BO_DELWRI) {
882 			bp->b_oflags &= ~BO_DELWRI;
883 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
884 			wakeup(&fs->lfs_avail);
885 		}
886 		mutex_exit(bp->b_objlock);
887 		LFS_UNLOCK_BUF(bp);
888 		brelsel(bp, BC_INVAL | BC_VFLUSH);
889 	}
890 	mutex_exit(&bufcache_lock);
891 
892 	return (0);
893 }
894 
895