xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: lfs_inode.c,v 1.126 2011/11/23 19:42:10 bouyer 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 1.126 2011/11/23 19:42:10 bouyer 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 		/* still do a uvm_vnp_setsize() as writesize may be larger */
231 		uvm_vnp_setsize(ovp, length);
232 		return (0);
233 	}
234 
235 	if (ovp->v_type == VLNK &&
236 	    (oip->i_size < ump->um_maxsymlinklen ||
237 	     (ump->um_maxsymlinklen == 0 &&
238 	      oip->i_ffs1_blocks == 0))) {
239 #ifdef DIAGNOSTIC
240 		if (length != 0)
241 			panic("lfs_truncate: partial truncate of symlink");
242 #endif
243 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
244 		oip->i_size = oip->i_ffs1_size = 0;
245 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
246 		return (lfs_update(ovp, NULL, NULL, 0));
247 	}
248 	if (oip->i_size == length) {
249 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
250 		return (lfs_update(ovp, NULL, NULL, 0));
251 	}
252 	fs = oip->i_lfs;
253 	lfs_imtime(fs);
254 	osize = oip->i_size;
255 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
256 
257 	ASSERT_NO_SEGLOCK(fs);
258 	/*
259 	 * Lengthen the size of the file. We must ensure that the
260 	 * last byte of the file is allocated. Since the smallest
261 	 * value of osize is 0, length will be at least 1.
262 	 */
263 	if (osize < length) {
264 		if (length > ump->um_maxfilesize)
265 			return (EFBIG);
266 		aflags = B_CLRBUF;
267 		if (ioflag & IO_SYNC)
268 			aflags |= B_SYNC;
269 		if (usepc) {
270 			if (lblkno(fs, osize) < NDADDR &&
271 			    lblkno(fs, osize) != lblkno(fs, length) &&
272 			    blkroundup(fs, osize) != osize) {
273 				off_t eob;
274 
275 				eob = blkroundup(fs, osize);
276 				uvm_vnp_setwritesize(ovp, eob);
277 				error = ufs_balloc_range(ovp, osize,
278 				    eob - osize, cred, aflags);
279 				if (error) {
280 					(void) lfs_truncate(ovp, osize,
281 						    ioflag & IO_SYNC, cred);
282 					return error;
283 				}
284 				if (ioflag & IO_SYNC) {
285 					mutex_enter(ovp->v_interlock);
286 					VOP_PUTPAGES(ovp,
287 					    trunc_page(osize & fs->lfs_bmask),
288 					    round_page(eob),
289 					    PGO_CLEANIT | PGO_SYNCIO);
290 				}
291 			}
292 			uvm_vnp_setwritesize(ovp, length);
293 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
294 						 aflags);
295 			if (error) {
296 				(void) lfs_truncate(ovp, osize,
297 						    ioflag & IO_SYNC, cred);
298 				return error;
299 			}
300 			uvm_vnp_setsize(ovp, length);
301 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
302 			KASSERT(ovp->v_size == oip->i_size);
303 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
304 			return (lfs_update(ovp, NULL, NULL, 0));
305 		} else {
306 			error = lfs_reserve(fs, ovp, NULL,
307 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
308 			if (error)
309 				return (error);
310 			error = lfs_balloc(ovp, length - 1, 1, cred,
311 					   aflags, &bp);
312 			lfs_reserve(fs, ovp, NULL,
313 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
314 			if (error)
315 				return (error);
316 			oip->i_ffs1_size = oip->i_size = length;
317 			uvm_vnp_setsize(ovp, length);
318 			(void) VOP_BWRITE(bp->b_vp, bp);
319 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
320 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
321 			return (lfs_update(ovp, NULL, NULL, 0));
322 		}
323 	}
324 
325 	if ((error = lfs_reserve(fs, ovp, NULL,
326 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
327 		return (error);
328 
329 	/*
330 	 * Shorten the size of the file. If the file is not being
331 	 * truncated to a block boundary, the contents of the
332 	 * partial block following the end of the file must be
333 	 * zero'ed in case it ever becomes accessible again because
334 	 * of subsequent file growth. Directories however are not
335 	 * zero'ed as they should grow back initialized to empty.
336 	 */
337 	offset = blkoff(fs, length);
338 	lastseg = -1;
339 	bc = 0;
340 
341 	if (ovp != fs->lfs_ivnode)
342 		lfs_seglock(fs, SEGM_PROT);
343 	if (offset == 0) {
344 		oip->i_size = oip->i_ffs1_size = length;
345 	} else if (!usepc) {
346 		lbn = lblkno(fs, length);
347 		aflags = B_CLRBUF;
348 		if (ioflag & IO_SYNC)
349 			aflags |= B_SYNC;
350 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
351 		if (error) {
352 			lfs_reserve(fs, ovp, NULL,
353 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
354 			goto errout;
355 		}
356 		obufsize = bp->b_bufsize;
357 		odb = btofsb(fs, bp->b_bcount);
358 		oip->i_size = oip->i_ffs1_size = length;
359 		size = blksize(fs, oip, lbn);
360 		if (ovp->v_type != VDIR)
361 			memset((char *)bp->b_data + offset, 0,
362 			       (u_int)(size - offset));
363 		allocbuf(bp, size, 1);
364 		if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
365 			mutex_enter(&lfs_lock);
366 			locked_queue_bytes -= obufsize - bp->b_bufsize;
367 			mutex_exit(&lfs_lock);
368 		}
369 		if (bp->b_oflags & BO_DELWRI)
370 			fs->lfs_avail += odb - btofsb(fs, size);
371 		(void) VOP_BWRITE(bp->b_vp, bp);
372 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
373 		/*
374 		 * When truncating a regular file down to a non-block-aligned
375 		 * size, we must zero the part of last block which is past
376 		 * the new EOF.  We must synchronously flush the zeroed pages
377 		 * to disk since the new pages will be invalidated as soon
378 		 * as we inform the VM system of the new, smaller size.
379 		 * We must do this before acquiring the GLOCK, since fetching
380 		 * the pages will acquire the GLOCK internally.
381 		 * So there is a window where another thread could see a whole
382 		 * zeroed page past EOF, but that's life.
383 		 */
384 		daddr_t xlbn;
385 		voff_t eoz;
386 
387 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
388 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags);
389 		if (error) {
390 			lfs_reserve(fs, ovp, NULL,
391 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
392 			goto errout;
393 		}
394 		xlbn = lblkno(fs, length);
395 		size = blksize(fs, oip, xlbn);
396 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
397 		ubc_zerorange(&ovp->v_uobj, length, eoz - length,
398 		    UBC_UNMAP_FLAG(ovp));
399 		if (round_page(eoz) > round_page(length)) {
400 			mutex_enter(ovp->v_interlock);
401 			error = VOP_PUTPAGES(ovp, round_page(length),
402 			    round_page(eoz),
403 			    PGO_CLEANIT | PGO_DEACTIVATE |
404 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
405 			if (error) {
406 				lfs_reserve(fs, ovp, NULL,
407 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
408 				goto errout;
409 			}
410 		}
411 	}
412 
413 	genfs_node_wrlock(ovp);
414 
415 	oip->i_size = oip->i_ffs1_size = length;
416 	uvm_vnp_setsize(ovp, length);
417 
418 	/*
419 	 * Calculate index into inode's block list of
420 	 * last direct and indirect blocks (if any)
421 	 * which we want to keep.  Lastblock is -1 when
422 	 * the file is truncated to 0.
423 	 */
424 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
425 	if (length > QUAD_MAX - fs->lfs_bsize)
426 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
427 	else
428 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
429 	lastiblock[SINGLE] = lastblock - NDADDR;
430 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
431 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
432 	nblocks = btofsb(fs, fs->lfs_bsize);
433 	/*
434 	 * Record changed file and block pointers before we start
435 	 * freeing blocks.  lastiblock values are also normalized to -1
436 	 * for calls to lfs_indirtrunc below.
437 	 */
438 	memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
439 	for (level = TRIPLE; level >= SINGLE; level--)
440 		if (lastiblock[level] < 0) {
441 			newblks[NDADDR+level] = 0;
442 			lastiblock[level] = -1;
443 		}
444 	for (i = NDADDR - 1; i > lastblock; i--)
445 		newblks[i] = 0;
446 
447 	oip->i_size = oip->i_ffs1_size = osize;
448 	error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
449 	if (error && !allerror)
450 		allerror = error;
451 
452 	/*
453 	 * Indirect blocks first.
454 	 */
455 	indir_lbn[SINGLE] = -NDADDR;
456 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
457 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
458 	for (level = TRIPLE; level >= SINGLE; level--) {
459 		bn = oip->i_ffs1_ib[level];
460 		if (bn != 0) {
461 			error = lfs_indirtrunc(oip, indir_lbn[level],
462 					       bn, lastiblock[level],
463 					       level, &count, &rcount,
464 					       &lastseg, &bc);
465 			if (error)
466 				allerror = error;
467 			real_released += rcount;
468 			blocksreleased += count;
469 			if (lastiblock[level] < 0) {
470 				if (oip->i_ffs1_ib[level] > 0)
471 					real_released += nblocks;
472 				blocksreleased += nblocks;
473 				oip->i_ffs1_ib[level] = 0;
474 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
475 					    &lastseg, &bc);
476         			lfs_deregister_block(ovp, bn);
477 			}
478 		}
479 		if (lastiblock[level] >= 0)
480 			goto done;
481 	}
482 
483 	/*
484 	 * All whole direct blocks or frags.
485 	 */
486 	for (i = NDADDR - 1; i > lastblock; i--) {
487 		long bsize, obsize;
488 
489 		bn = oip->i_ffs1_db[i];
490 		if (bn == 0)
491 			continue;
492 		bsize = blksize(fs, oip, i);
493 		if (oip->i_ffs1_db[i] > 0) {
494 			/* Check for fragment size changes */
495 			obsize = oip->i_lfs_fragsize[i];
496 			real_released += btofsb(fs, obsize);
497 			oip->i_lfs_fragsize[i] = 0;
498 		} else
499 			obsize = 0;
500 		blocksreleased += btofsb(fs, bsize);
501 		oip->i_ffs1_db[i] = 0;
502 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
503         	lfs_deregister_block(ovp, bn);
504 	}
505 	if (lastblock < 0)
506 		goto done;
507 
508 	/*
509 	 * Finally, look for a change in size of the
510 	 * last direct block; release any frags.
511 	 */
512 	bn = oip->i_ffs1_db[lastblock];
513 	if (bn != 0) {
514 		long oldspace, newspace;
515 #if 0
516 		long olddspace;
517 #endif
518 
519 		/*
520 		 * Calculate amount of space we're giving
521 		 * back as old block size minus new block size.
522 		 */
523 		oldspace = blksize(fs, oip, lastblock);
524 #if 0
525 		olddspace = oip->i_lfs_fragsize[lastblock];
526 #endif
527 
528 		oip->i_size = oip->i_ffs1_size = length;
529 		newspace = blksize(fs, oip, lastblock);
530 		if (newspace == 0)
531 			panic("itrunc: newspace");
532 		if (oldspace - newspace > 0) {
533 			blocksreleased += btofsb(fs, oldspace - newspace);
534 		}
535 #if 0
536 		if (bn > 0 && olddspace - newspace > 0) {
537 			/* No segment accounting here, just vnode */
538 			real_released += btofsb(fs, olddspace - newspace);
539 		}
540 #endif
541 	}
542 
543 done:
544 	/* Finish segment accounting corrections */
545 	lfs_update_seguse(fs, oip, lastseg, bc);
546 #ifdef DIAGNOSTIC
547 	for (level = SINGLE; level <= TRIPLE; level++)
548 		if ((newblks[NDADDR + level] == 0) !=
549 		    ((oip->i_ffs1_ib[level]) == 0)) {
550 			panic("lfs itrunc1");
551 		}
552 	for (i = 0; i < NDADDR; i++)
553 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
554 			panic("lfs itrunc2");
555 		}
556 	if (length == 0 &&
557 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
558 		panic("lfs itrunc3");
559 #endif /* DIAGNOSTIC */
560 	/*
561 	 * Put back the real size.
562 	 */
563 	oip->i_size = oip->i_ffs1_size = length;
564 	oip->i_lfs_effnblks -= blocksreleased;
565 	oip->i_ffs1_blocks -= real_released;
566 	mutex_enter(&lfs_lock);
567 	fs->lfs_bfree += blocksreleased;
568 	mutex_exit(&lfs_lock);
569 #ifdef DIAGNOSTIC
570 	if (oip->i_size == 0 &&
571 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
572 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
573 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
574 		panic("lfs_truncate: persistent blocks");
575 	}
576 #endif
577 
578 	/*
579 	 * If we truncated to zero, take us off the paging queue.
580 	 */
581 	mutex_enter(&lfs_lock);
582 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
583 		oip->i_flags &= ~IN_PAGING;
584 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
585 	}
586 	mutex_exit(&lfs_lock);
587 
588 	oip->i_flag |= IN_CHANGE;
589 #ifdef QUOTA
590 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
591 #endif
592 	lfs_reserve(fs, ovp, NULL,
593 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
594 	genfs_node_unlock(ovp);
595   errout:
596 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
597 	if (ovp != fs->lfs_ivnode)
598 		lfs_segunlock(fs);
599 	return (allerror ? allerror : error);
600 }
601 
602 /* Update segment and avail usage information when removing a block. */
603 static int
604 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
605 	    size_t bsize, long *lastseg, size_t *num)
606 {
607 	long seg;
608 	int error = 0;
609 
610 	ASSERT_SEGLOCK(fs);
611 	bsize = fragroundup(fs, bsize);
612 	if (daddr > 0) {
613 		if (*lastseg != (seg = dtosn(fs, daddr))) {
614 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
615 			*num = bsize;
616 			*lastseg = seg;
617 		} else
618 			*num += bsize;
619 	}
620 
621 	return error;
622 }
623 
624 /* Finish the accounting updates for a segment. */
625 static int
626 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
627 {
628 	struct segdelta *sd;
629 	struct vnode *vp;
630 
631 	ASSERT_SEGLOCK(fs);
632 	if (lastseg < 0 || num == 0)
633 		return 0;
634 
635 	vp = ITOV(ip);
636 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
637 		if (sd->segnum == lastseg)
638 			break;
639 	if (sd == NULL) {
640 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
641 		sd->segnum = lastseg;
642 		sd->num = 0;
643 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
644 	}
645 	sd->num += num;
646 
647 	return 0;
648 }
649 
650 static void
651 lfs_finalize_seguse(struct lfs *fs, void *v)
652 {
653 	SEGUSE *sup;
654 	struct buf *bp;
655 	struct segdelta *sd;
656 	LIST_HEAD(, segdelta) *hd = v;
657 
658 	ASSERT_SEGLOCK(fs);
659 	while((sd = LIST_FIRST(hd)) != NULL) {
660 		LIST_REMOVE(sd, list);
661 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
662 		if (sd->num > sup->su_nbytes) {
663 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
664 				sd->segnum, (long)(sd->num - sup->su_nbytes));
665 			panic("lfs_finalize_seguse: negative bytes");
666 			sup->su_nbytes = sd->num;
667 		}
668 		sup->su_nbytes -= sd->num;
669 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
670 		free(sd, M_SEGMENT);
671 	}
672 }
673 
674 /* Finish the accounting updates for a segment. */
675 void
676 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
677 {
678 	ASSERT_SEGLOCK(fs);
679 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
680 }
681 
682 /* Finish the accounting updates for a segment. */
683 void
684 lfs_finalize_fs_seguse(struct lfs *fs)
685 {
686 	ASSERT_SEGLOCK(fs);
687 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
688 }
689 
690 /*
691  * Release blocks associated with the inode ip and stored in the indirect
692  * block bn.  Blocks are free'd in LIFO order up to (but not including)
693  * lastbn.  If level is greater than SINGLE, the block is an indirect block
694  * and recursive calls to indirtrunc must be used to cleanse other indirect
695  * blocks.
696  *
697  * NB: triple indirect blocks are untested.
698  */
699 static int
700 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
701 	       daddr_t lastbn, int level, long *countp,
702 	       long *rcountp, long *lastsegp, size_t *bcp)
703 {
704 	int i;
705 	struct buf *bp;
706 	struct lfs *fs = ip->i_lfs;
707 	int32_t *bap;	/* XXX ondisk32 */
708 	struct vnode *vp;
709 	daddr_t nb, nlbn, last;
710 	int32_t *copy = NULL;	/* XXX ondisk32 */
711 	long blkcount, rblkcount, factor;
712 	int nblocks, blocksreleased = 0, real_released = 0;
713 	int error = 0, allerror = 0;
714 
715 	ASSERT_SEGLOCK(fs);
716 	/*
717 	 * Calculate index in current block of last
718 	 * block to be kept.  -1 indicates the entire
719 	 * block so we need not calculate the index.
720 	 */
721 	factor = 1;
722 	for (i = SINGLE; i < level; i++)
723 		factor *= NINDIR(fs);
724 	last = lastbn;
725 	if (lastbn > 0)
726 		last /= factor;
727 	nblocks = btofsb(fs, fs->lfs_bsize);
728 	/*
729 	 * Get buffer of block pointers, zero those entries corresponding
730 	 * to blocks to be free'd, and update on disk copy first.  Since
731 	 * double(triple) indirect before single(double) indirect, calls
732 	 * to bmap on these blocks will fail.  However, we already have
733 	 * the on disk address, so we have to set the b_blkno field
734 	 * explicitly instead of letting bread do everything for us.
735 	 */
736 	vp = ITOV(ip);
737 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
738 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
739 		/* Braces must be here in case trace evaluates to nothing. */
740 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
741 	} else {
742 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
743 		curlwp->l_ru.ru_inblock++; /* pay for read */
744 		bp->b_flags |= B_READ;
745 		if (bp->b_bcount > bp->b_bufsize)
746 			panic("lfs_indirtrunc: bad buffer size");
747 		bp->b_blkno = fsbtodb(fs, dbn);
748 		VOP_STRATEGY(vp, bp);
749 		error = biowait(bp);
750 	}
751 	if (error) {
752 		brelse(bp, 0);
753 		*countp = *rcountp = 0;
754 		return (error);
755 	}
756 
757 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
758 	if (lastbn >= 0) {
759 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
760 		memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize);
761 		memset((void *)&bap[last + 1], 0,
762 		/* XXX ondisk32 */
763 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
764 		error = VOP_BWRITE(bp->b_vp, bp);
765 		if (error)
766 			allerror = error;
767 		bap = copy;
768 	}
769 
770 	/*
771 	 * Recursively free totally unused blocks.
772 	 */
773 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
774 	    i--, nlbn += factor) {
775 		nb = bap[i];
776 		if (nb == 0)
777 			continue;
778 		if (level > SINGLE) {
779 			error = lfs_indirtrunc(ip, nlbn, nb,
780 					       (daddr_t)-1, level - 1,
781 					       &blkcount, &rblkcount,
782 					       lastsegp, bcp);
783 			if (error)
784 				allerror = error;
785 			blocksreleased += blkcount;
786 			real_released += rblkcount;
787 		}
788 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
789 		if (bap[i] > 0)
790 			real_released += nblocks;
791 		blocksreleased += nblocks;
792 	}
793 
794 	/*
795 	 * Recursively free last partial block.
796 	 */
797 	if (level > SINGLE && lastbn >= 0) {
798 		last = lastbn % factor;
799 		nb = bap[i];
800 		if (nb != 0) {
801 			error = lfs_indirtrunc(ip, nlbn, nb,
802 					       last, level - 1, &blkcount,
803 					       &rblkcount, lastsegp, bcp);
804 			if (error)
805 				allerror = error;
806 			real_released += rblkcount;
807 			blocksreleased += blkcount;
808 		}
809 	}
810 
811 	if (copy != NULL) {
812 		lfs_free(fs, copy, LFS_NB_IBLOCK);
813 	} else {
814 		mutex_enter(&bufcache_lock);
815 		if (bp->b_oflags & BO_DELWRI) {
816 			LFS_UNLOCK_BUF(bp);
817 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
818 			wakeup(&fs->lfs_avail);
819 		}
820 		brelsel(bp, BC_INVAL);
821 		mutex_exit(&bufcache_lock);
822 	}
823 
824 	*countp = blocksreleased;
825 	*rcountp = real_released;
826 	return (allerror);
827 }
828 
829 /*
830  * Destroy any in core blocks past the truncation length.
831  * Inlined from vtruncbuf, so that lfs_avail could be updated.
832  * We take the seglock to prevent cleaning from occurring while we are
833  * invalidating blocks.
834  */
835 static int
836 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
837 {
838 	struct buf *bp, *nbp;
839 	int error;
840 	struct lfs *fs;
841 	voff_t off;
842 
843 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
844 	mutex_enter(vp->v_interlock);
845 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
846 	if (error)
847 		return error;
848 
849 	fs = VTOI(vp)->i_lfs;
850 
851 	ASSERT_SEGLOCK(fs);
852 
853 	mutex_enter(&bufcache_lock);
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 		error = bbusy(bp, catch, slptimeo, NULL);
860 		if (error == EPASSTHROUGH)
861 			goto restart;
862 		if (error != 0) {
863 			mutex_exit(&bufcache_lock);
864 			return (error);
865 		}
866 		mutex_enter(bp->b_objlock);
867 		if (bp->b_oflags & BO_DELWRI) {
868 			bp->b_oflags &= ~BO_DELWRI;
869 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
870 			wakeup(&fs->lfs_avail);
871 		}
872 		mutex_exit(bp->b_objlock);
873 		LFS_UNLOCK_BUF(bp);
874 		brelsel(bp, BC_INVAL | BC_VFLUSH);
875 	}
876 
877 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
878 		nbp = LIST_NEXT(bp, b_vnbufs);
879 		if (bp->b_lblkno < lbn)
880 			continue;
881 		error = bbusy(bp, catch, slptimeo, NULL);
882 		if (error == EPASSTHROUGH)
883 			goto restart;
884 		if (error != 0) {
885 			mutex_exit(&bufcache_lock);
886 			return (error);
887 		}
888 		mutex_enter(bp->b_objlock);
889 		if (bp->b_oflags & BO_DELWRI) {
890 			bp->b_oflags &= ~BO_DELWRI;
891 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
892 			wakeup(&fs->lfs_avail);
893 		}
894 		mutex_exit(bp->b_objlock);
895 		LFS_UNLOCK_BUF(bp);
896 		brelsel(bp, BC_INVAL | BC_VFLUSH);
897 	}
898 	mutex_exit(&bufcache_lock);
899 
900 	return (0);
901 }
902 
903