xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: lfs_inode.c,v 1.75 2003/04/27 06:46:38 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 /*
39  * Copyright (c) 1986, 1989, 1991, 1993
40  *	The Regents of the University of California.  All rights reserved.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. All advertising materials mentioning features or use of this software
51  *    must display the following acknowledgement:
52  *	This product includes software developed by the University of
53  *	California, Berkeley and its contributors.
54  * 4. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
71  */
72 
73 #include <sys/cdefs.h>
74 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.75 2003/04/27 06:46:38 yamt Exp $");
75 
76 #if defined(_KERNEL_OPT)
77 #include "opt_quota.h"
78 #endif
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/mount.h>
83 #include <sys/proc.h>
84 #include <sys/file.h>
85 #include <sys/buf.h>
86 #include <sys/vnode.h>
87 #include <sys/kernel.h>
88 #include <sys/malloc.h>
89 #include <sys/trace.h>
90 #include <sys/resourcevar.h>
91 
92 #include <ufs/ufs/quota.h>
93 #include <ufs/ufs/inode.h>
94 #include <ufs/ufs/ufsmount.h>
95 #include <ufs/ufs/ufs_extern.h>
96 
97 #include <ufs/lfs/lfs.h>
98 #include <ufs/lfs/lfs_extern.h>
99 
100 extern int locked_queue_count;
101 extern long locked_queue_bytes;
102 
103 static int lfs_update_seguse(struct lfs *, long, size_t);
104 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t,
105 			   daddr_t, int, long *, long *, long *, size_t *,
106 			   struct proc *);
107 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *);
108 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
109 
110 /* Search a block for a specific dinode. */
111 struct ufs1_dinode *
112 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
113 {
114 	struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data;
115 	struct ufs1_dinode *ldip, *fin;
116 
117 #ifdef LFS_IFILE_FRAG_ADDRESSING
118 	if (fs->lfs_version == 1)
119 		fin = dip + INOPB(fs);
120 	else
121 		fin = dip + INOPF(fs);
122 #else
123 	fin = dip + INOPB(fs);
124 #endif
125 
126 	/*
127 	 * Read the inode block backwards, since later versions of the
128 	 * inode will supercede earlier ones.  Though it is unlikely, it is
129 	 * possible that the same inode will appear in the same inode block.
130 	 */
131 	for (ldip = fin - 1; ldip >= dip; --ldip)
132 		if (ldip->di_inumber == ino)
133 			return (ldip);
134 
135 	printf("searched %d entries\n", (int)(fin - dip));
136 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
137 	       dtosn(fs, fs->lfs_offset));
138 	printf("block is 0x%llx (seg %lld)\n",
139 	       (unsigned long long)dbtofsb(fs, bp->b_blkno),
140 	       (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno)));
141 
142 	return NULL;
143 }
144 
145 int
146 lfs_update(void *v)
147 {
148 	struct vop_update_args /* {
149 				  struct vnode *a_vp;
150 				  struct timespec *a_access;
151 				  struct timespec *a_modify;
152 				  int a_flags;
153 				  } */ *ap = v;
154 	struct inode *ip;
155 	struct vnode *vp = ap->a_vp;
156 	struct timespec ts;
157 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
158 	int s;
159 
160 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
161 		return (0);
162 	ip = VTOI(vp);
163 
164 	/*
165 	 * If we are called from vinvalbuf, and the file's blocks have
166 	 * already been scheduled for writing, but the writes have not
167 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
168 	 * will cause a panic.	So, we must wait until any pending write
169 	 * for our inode completes, if we are called with UPDATE_WAIT set.
170 	 */
171 	s = splbio();
172 	while ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
173 	    WRITEINPROG(vp)) {
174 #ifdef DEBUG_LFS
175 		printf("lfs_update: sleeping on inode %d (in-progress)\n",
176 		       ip->i_number);
177 #endif
178 		tsleep(vp, (PRIBIO+1), "lfs_update", 0);
179 	}
180 	splx(s);
181 	TIMEVAL_TO_TIMESPEC(&time, &ts);
182 	LFS_ITIMES(ip,
183 		   ap->a_access ? ap->a_access : &ts,
184 		   ap->a_modify ? ap->a_modify : &ts, &ts);
185 	if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) {
186 		return (0);
187 	}
188 
189 	/* If sync, push back the vnode and any dirty blocks it may have. */
190 	if ((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) {
191 		/* Avoid flushing VDIROP. */
192 		++fs->lfs_diropwait;
193 		while (vp->v_flag & VDIROP) {
194 #ifdef DEBUG_LFS
195 			printf("lfs_update: sleeping on inode %d (dirops)\n",
196 			       ip->i_number);
197 			printf("lfs_update: vflags 0x%x, iflags 0x%x\n",
198 				vp->v_flag, ip->i_flag);
199 #endif
200 			if (fs->lfs_dirops == 0)
201 				lfs_flush_fs(fs, SEGM_SYNC);
202 			else
203 				tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
204 				       0);
205 			/* XXX KS - by falling out here, are we writing the vn
206 			twice? */
207 		}
208 		--fs->lfs_diropwait;
209 		return lfs_vflush(vp);
210 	}
211 	return 0;
212 }
213 
214 #define	SINGLE	0	/* index of single indirect block */
215 #define	DOUBLE	1	/* index of double indirect block */
216 #define	TRIPLE	2	/* index of triple indirect block */
217 /*
218  * Truncate the inode oip to at most length size, freeing the
219  * disk blocks.
220  */
221 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */
222 
223 int
224 lfs_truncate(void *v)
225 {
226 	struct vop_truncate_args /* {
227 		struct vnode *a_vp;
228 		off_t a_length;
229 		int a_flags;
230 		struct ucred *a_cred;
231 		struct proc *a_p;
232 	} */ *ap = v;
233 	struct vnode *ovp = ap->a_vp;
234 	struct genfs_node *gp = VTOG(ovp);
235 	daddr_t lastblock;
236 	struct inode *oip;
237 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
238 	/* XXX ondisk32 */
239 	int32_t newblks[NDADDR + NIADDR];
240 	off_t length = ap->a_length;
241 	struct lfs *fs;
242 	struct buf *bp;
243 	int offset, size, level;
244 	long count, rcount, nblocks, blocksreleased = 0, real_released = 0;
245 	int i;
246 	int aflags, error, allerror = 0;
247 	off_t osize;
248 	voff_t eoz;
249 	long lastseg;
250 	size_t bc;
251 	int obufsize, odb;
252 	int usepc;
253 
254 	if (length < 0)
255 		return (EINVAL);
256 	oip = VTOI(ovp);
257 
258 	/*
259 	 * Just return and not update modification times.
260 	 */
261 	if (oip->i_size == length)
262 		return (0);
263 
264 	if (ovp->v_type == VLNK &&
265 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen ||
266 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
267 	      oip->i_ffs1_blocks == 0))) {
268 #ifdef DIAGNOSTIC
269 		if (length != 0)
270 			panic("lfs_truncate: partial truncate of symlink");
271 #endif
272 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
273 		oip->i_size = oip->i_ffs1_size = 0;
274 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
275 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
276 	}
277 	if (oip->i_size == length) {
278 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
279 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
280 	}
281 #ifdef QUOTA
282 	if ((error = getinoquota(oip)) != 0)
283 		return (error);
284 #endif
285 	fs = oip->i_lfs;
286 	lfs_imtime(fs);
287 	osize = oip->i_size;
288 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
289 
290 	/*
291 	 * Lengthen the size of the file. We must ensure that the
292 	 * last byte of the file is allocated. Since the smallest
293 	 * value of osize is 0, length will be at least 1.
294 	 */
295 	if (osize < length) {
296 		if (length > fs->lfs_maxfilesize)
297 			return (EFBIG);
298 		aflags = B_CLRBUF;
299 		if (ap->a_flags & IO_SYNC)
300 			aflags |= B_SYNC;
301 		if (usepc) {
302 			if (lblkno(fs, osize) < NDADDR &&
303 			    lblkno(fs, osize) != lblkno(fs, length) &&
304 			    blkroundup(fs, osize) != osize) {
305 				error = ufs_balloc_range(ovp, osize,
306 							 blkroundup(fs, osize) -
307 							 osize, ap->a_cred,
308 							 aflags);
309 				if (error) {
310 					return error;
311 				}
312 				if (ap->a_flags & IO_SYNC) {
313 					ovp->v_size = blkroundup(fs, osize);
314 					simple_lock(&ovp->v_interlock);
315 					VOP_PUTPAGES(ovp,
316 					    trunc_page(osize & fs->lfs_bmask),
317 					    round_page(ovp->v_size),
318 					    PGO_CLEANIT | PGO_SYNCIO);
319 				}
320 			}
321 			error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
322 						 aflags);
323 			if (error) {
324 				(void) VOP_TRUNCATE(ovp, osize,
325 						    ap->a_flags & IO_SYNC,
326 				    		    ap->a_cred, ap->a_p);
327 				return error;
328 			}
329 			uvm_vnp_setsize(ovp, length);
330 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
331 			KASSERT(ovp->v_size == oip->i_size);
332 			return (VOP_UPDATE(ovp, NULL, NULL, 0));
333 		} else {
334 			error = lfs_reserve(fs, ovp, NULL,
335 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
336 			if (error)
337 				return (error);
338 			error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred,
339 					   aflags, &bp);
340 			lfs_reserve(fs, ovp, NULL,
341 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
342 			if (error)
343 				return (error);
344 			oip->i_ffs1_size = oip->i_size = length;
345 			uvm_vnp_setsize(ovp, length);
346 			(void) VOP_BWRITE(bp);
347 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
348 			return (VOP_UPDATE(ovp, NULL, NULL, 0));
349 		}
350 	}
351 
352 	if ((error = lfs_reserve(fs, ovp, NULL,
353 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
354 		return (error);
355 
356 	/*
357 	 * Shorten the size of the file. If the file is not being
358 	 * truncated to a block boundary, the contents of the
359 	 * partial block following the end of the file must be
360 	 * zero'ed in case it ever becomes accessible again because
361 	 * of subsequent file growth. Directories however are not
362 	 * zero'ed as they should grow back initialized to empty.
363 	 */
364 	offset = blkoff(fs, length);
365 	lastseg = -1;
366 	bc = 0;
367 
368 	lfs_seglock(fs, SEGM_PROT);
369 	if (offset == 0) {
370 		oip->i_size = oip->i_ffs1_size = length;
371 	} else if (!usepc) {
372 		lbn = lblkno(fs, length);
373 		aflags = B_CLRBUF;
374 		if (ap->a_flags & IO_SYNC)
375 			aflags |= B_SYNC;
376 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
377 		if (error) {
378 			lfs_reserve(fs, ovp, NULL,
379 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
380 			goto errout;
381 		}
382 		obufsize = bp->b_bufsize;
383 		odb = btofsb(fs, bp->b_bcount);
384 		oip->i_size = oip->i_ffs1_size = length;
385 		size = blksize(fs, oip, lbn);
386 		if (ovp->v_type != VDIR)
387 			memset((char *)bp->b_data + offset, 0,
388 			       (u_int)(size - offset));
389 		allocbuf(bp, size);
390 		if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED)
391 			locked_queue_bytes -= obufsize - bp->b_bufsize;
392 		if (bp->b_flags & B_DELWRI)
393 			fs->lfs_avail += odb - btofsb(fs, size);
394 		(void) VOP_BWRITE(bp);
395 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
396 		/*
397 		 * When truncating a regular file down to a non-block-aligned
398 		 * size, we must zero the part of last block which is past
399 		 * the new EOF.  We must synchronously flush the zeroed pages
400 		 * to disk since the new pages will be invalidated as soon
401 		 * as we inform the VM system of the new, smaller size.
402 		 * We must do this before acquiring the GLOCK, since fetching
403 		 * the pages will acquire the GLOCK internally.
404 		 * So there is a window where another thread could see a whole
405 		 * zeroed page past EOF, but that's life.
406 		 */
407 		aflags = ap->a_flags & IO_SYNC ? B_SYNC : 0;
408 		error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
409 					 aflags);
410 		if (error) {
411 			goto errout;
412 		}
413 		eoz = blkroundup(fs, length);
414 		uvm_vnp_zerorange(ovp, length, eoz - length);
415 		simple_lock(&ovp->v_interlock);
416 		error = VOP_PUTPAGES(ovp, trunc_page(length), round_page(eoz),
417 		    PGO_CLEANIT | PGO_DEACTIVATE | (aflags ? PGO_SYNCIO : 0));
418 		if (error) {
419 			goto errout;
420 		}
421 	}
422 
423 	lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
424 
425 	oip->i_size = oip->i_ffs1_size = length;
426 	uvm_vnp_setsize(ovp, length);
427 	/*
428 	 * Calculate index into inode's block list of
429 	 * last direct and indirect blocks (if any)
430 	 * which we want to keep.  Lastblock is -1 when
431 	 * the file is truncated to 0.
432 	 */
433 	lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
434 	lastiblock[SINGLE] = lastblock - NDADDR;
435 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
436 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
437 	nblocks = btofsb(fs, fs->lfs_bsize);
438 	/*
439 	 * Record changed file and block pointers before we start
440 	 * freeing blocks.  lastiblock values are also normalized to -1
441 	 * for calls to lfs_indirtrunc below.
442 	 */
443 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs1_db[0], sizeof newblks);
444 	for (level = TRIPLE; level >= SINGLE; level--)
445 		if (lastiblock[level] < 0) {
446 			newblks[NDADDR+level] = 0;
447 			lastiblock[level] = -1;
448 		}
449 	for (i = NDADDR - 1; i > lastblock; i--)
450 		newblks[i] = 0;
451 
452 	oip->i_size = oip->i_ffs1_size = osize;
453 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
454 	if (error && !allerror)
455 		allerror = error;
456 
457 	/*
458 	 * Indirect blocks first.
459 	 */
460 	indir_lbn[SINGLE] = -NDADDR;
461 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
462 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
463 	for (level = TRIPLE; level >= SINGLE; level--) {
464 		bn = oip->i_ffs1_ib[level];
465 		if (bn != 0) {
466 			error = lfs_indirtrunc(oip, indir_lbn[level],
467 					       bn, lastiblock[level],
468 					       level, &count, &rcount,
469 					       &lastseg, &bc, ap->a_p);
470 			if (error)
471 				allerror = error;
472 			real_released += rcount;
473 			blocksreleased += count;
474 			if (lastiblock[level] < 0) {
475 				if (oip->i_ffs1_ib[level] > 0)
476 					real_released += nblocks;
477 				blocksreleased += nblocks;
478 				oip->i_ffs1_ib[level] = 0;
479 				lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc);
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, bn, obsize, &lastseg, &bc);
506 	}
507 	if (lastblock < 0)
508 		goto done;
509 
510 	/*
511 	 * Finally, look for a change in size of the
512 	 * last direct block; release any frags.
513 	 */
514 	bn = oip->i_ffs1_db[lastblock];
515 	if (bn != 0) {
516 		long oldspace, newspace;
517 #if 0
518 		long olddspace;
519 #endif
520 
521 		/*
522 		 * Calculate amount of space we're giving
523 		 * back as old block size minus new block size.
524 		 */
525 		oldspace = blksize(fs, oip, lastblock);
526 #if 0
527 		olddspace = oip->i_lfs_fragsize[lastblock];
528 #endif
529 
530 		oip->i_size = oip->i_ffs1_size = length;
531 		newspace = blksize(fs, oip, lastblock);
532 		if (newspace == 0)
533 			panic("itrunc: newspace");
534 		if (oldspace - newspace > 0) {
535 			blocksreleased += btofsb(fs, oldspace - newspace);
536 		}
537 #if 0
538 		if (bn > 0 && olddspace - newspace > 0) {
539 			/* No segment accounting here, just vnode */
540 			real_released += btofsb(fs, olddspace - newspace);
541 		}
542 #endif
543 	}
544 
545 done:
546 	/* Finish segment accounting corrections */
547 	lfs_update_seguse(fs, lastseg, bc);
548 #ifdef DIAGNOSTIC
549 	for (level = SINGLE; level <= TRIPLE; level++)
550 		if ((newblks[NDADDR + level] == 0) !=
551 		    (oip->i_ffs1_ib[level]) == 0) {
552 			panic("lfs itrunc1");
553 		}
554 	for (i = 0; i < NDADDR; i++)
555 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
556 			panic("lfs itrunc2");
557 		}
558 	if (length == 0 &&
559 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
560 		panic("lfs itrunc3");
561 #endif /* DIAGNOSTIC */
562 	/*
563 	 * Put back the real size.
564 	 */
565 	oip->i_size = oip->i_ffs1_size = length;
566 	oip->i_lfs_effnblks -= blocksreleased;
567 	oip->i_ffs1_blocks -= real_released;
568 	fs->lfs_bfree += blocksreleased;
569 #ifdef DIAGNOSTIC
570 	if (oip->i_size == 0 && oip->i_ffs1_blocks != 0) {
571 		printf("lfs_truncate: truncate to 0 but %d blocks on inode\n",
572 		       oip->i_ffs1_blocks);
573 		panic("lfs_truncate: persistent blocks");
574 	}
575 #endif
576 	oip->i_flag |= IN_CHANGE;
577 #ifdef QUOTA
578 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
579 #endif
580 	lfs_reserve(fs, ovp, NULL,
581 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
582 	lockmgr(&gp->g_glock, LK_RELEASE, NULL);
583   errout:
584 	lfs_segunlock(fs);
585 	return (allerror ? allerror : error);
586 }
587 
588 /* Update segment usage information when removing a block. */
589 static int
590 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg,
591 	    size_t *num)
592 {
593 	long seg;
594 	int error = 0;
595 
596 	bsize = fragroundup(fs, bsize);
597 	if (daddr > 0) {
598 		if (*lastseg != (seg = dtosn(fs, daddr))) {
599 			error = lfs_update_seguse(fs, *lastseg, *num);
600 			*num = bsize;
601 			*lastseg = seg;
602 		} else
603 			*num += bsize;
604 	}
605 	return error;
606 }
607 
608 /* Finish the accounting updates for a segment. */
609 static int
610 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num)
611 {
612 	SEGUSE *sup;
613 	struct buf *bp;
614 
615 	if (lastseg < 0 || num == 0)
616 		return 0;
617 
618 	LFS_SEGENTRY(sup, fs, lastseg, bp);
619 	if (num > sup->su_nbytes) {
620 		printf("lfs_truncate: segment %ld short by %ld\n",
621 		       lastseg, (long)num - sup->su_nbytes);
622 		panic("lfs_truncate: negative bytes");
623 		sup->su_nbytes = num;
624 	}
625 	sup->su_nbytes -= num;
626 	LFS_WRITESEGENTRY(sup, fs, lastseg, bp);
627 
628 	return 0;
629 }
630 
631 /*
632  * Release blocks associated with the inode ip and stored in the indirect
633  * block bn.  Blocks are free'd in LIFO order up to (but not including)
634  * lastbn.  If level is greater than SINGLE, the block is an indirect block
635  * and recursive calls to indirtrunc must be used to cleanse other indirect
636  * blocks.
637  *
638  * NB: triple indirect blocks are untested.
639  */
640 static int
641 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
642 	       daddr_t lastbn, int level, long *countp,
643 	       long *rcountp, long *lastsegp, size_t *bcp, struct proc *p)
644 {
645 	int i;
646 	struct buf *bp;
647 	struct lfs *fs = ip->i_lfs;
648 	int32_t *bap;	/* XXX ondisk32 */
649 	struct vnode *vp;
650 	daddr_t nb, nlbn, last;
651 	int32_t *copy = NULL;	/* XXX ondisk32 */
652 	long blkcount, rblkcount, factor;
653 	int nblocks, blocksreleased = 0, real_released = 0;
654 	int error = 0, allerror = 0;
655 
656 	/*
657 	 * Calculate index in current block of last
658 	 * block to be kept.  -1 indicates the entire
659 	 * block so we need not calculate the index.
660 	 */
661 	factor = 1;
662 	for (i = SINGLE; i < level; i++)
663 		factor *= NINDIR(fs);
664 	last = lastbn;
665 	if (lastbn > 0)
666 		last /= factor;
667 	nblocks = btofsb(fs, fs->lfs_bsize);
668 	/*
669 	 * Get buffer of block pointers, zero those entries corresponding
670 	 * to blocks to be free'd, and update on disk copy first.  Since
671 	 * double(triple) indirect before single(double) indirect, calls
672 	 * to bmap on these blocks will fail.  However, we already have
673 	 * the on disk address, so we have to set the b_blkno field
674 	 * explicitly instead of letting bread do everything for us.
675 	 */
676 	vp = ITOV(ip);
677 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
678 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
679 		/* Braces must be here in case trace evaluates to nothing. */
680 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
681 	} else {
682 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
683 		p->p_stats->p_ru.ru_inblock++;	/* pay for read */
684 		bp->b_flags |= B_READ;
685 		if (bp->b_bcount > bp->b_bufsize)
686 			panic("lfs_indirtrunc: bad buffer size");
687 		bp->b_blkno = fsbtodb(fs, dbn);
688 		VOP_STRATEGY(bp);
689 		error = biowait(bp);
690 	}
691 	if (error) {
692 		brelse(bp);
693 		*countp = *rcountp = 0;
694 		return (error);
695 	}
696 
697 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
698 	if (lastbn >= 0) {
699 		MALLOC(copy, int32_t *, fs->lfs_bsize, M_TEMP, M_WAITOK);
700 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
701 		memset((caddr_t)&bap[last + 1], 0,
702 		/* XXX ondisk32 */
703 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
704 		error = VOP_BWRITE(bp);
705 		if (error)
706 			allerror = error;
707 		bap = copy;
708 	}
709 
710 	/*
711 	 * Recursively free totally unused blocks.
712 	 */
713 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
714 	    i--, nlbn += factor) {
715 		nb = bap[i];
716 		if (nb == 0)
717 			continue;
718 		if (level > SINGLE) {
719 			error = lfs_indirtrunc(ip, nlbn, nb,
720 					       (daddr_t)-1, level - 1,
721 					       &blkcount, &rblkcount,
722 					       lastsegp, bcp, p);
723 			if (error)
724 				allerror = error;
725 			blocksreleased += blkcount;
726 			real_released += rblkcount;
727 		}
728 		lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp);
729 		if (bap[i] > 0)
730 			real_released += nblocks;
731 		blocksreleased += nblocks;
732 	}
733 
734 	/*
735 	 * Recursively free last partial block.
736 	 */
737 	if (level > SINGLE && lastbn >= 0) {
738 		last = lastbn % factor;
739 		nb = bap[i];
740 		if (nb != 0) {
741 			error = lfs_indirtrunc(ip, nlbn, nb,
742 					       last, level - 1, &blkcount,
743 					       &rblkcount, lastsegp, bcp, p);
744 			if (error)
745 				allerror = error;
746 			real_released += rblkcount;
747 			blocksreleased += blkcount;
748 		}
749 	}
750 
751 	if (copy != NULL) {
752 		FREE(copy, M_TEMP);
753 	} else {
754 		if (bp->b_flags & B_DELWRI) {
755 			LFS_UNLOCK_BUF(bp);
756 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
757 			wakeup(&fs->lfs_avail);
758 		}
759 		bp->b_flags |= B_INVAL;
760 		brelse(bp);
761 	}
762 
763 	*countp = blocksreleased;
764 	*rcountp = real_released;
765 	return (allerror);
766 }
767 
768 /*
769  * Destroy any in core blocks past the truncation length.
770  * Inlined from vtruncbuf, so that lfs_avail could be updated.
771  * We take the seglock to prevent cleaning from occurring while we are
772  * invalidating blocks.
773  */
774 static int
775 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, int slpflag, int slptimeo)
776 {
777 	struct buf *bp, *nbp;
778 	int s, error;
779 	struct lfs *fs;
780 	voff_t off;
781 
782 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
783 	simple_lock(&vp->v_interlock);
784 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
785 	if (error) {
786 		return error;
787 	}
788 
789 	fs = VTOI(vp)->i_lfs;
790 	s = splbio();
791 
792 restart:
793 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
794 		nbp = LIST_NEXT(bp, b_vnbufs);
795 		if (bp->b_lblkno < lbn)
796 			continue;
797 		simple_lock(&bp->b_interlock);
798 		if (bp->b_flags & B_BUSY) {
799 			bp->b_flags |= B_WANTED;
800 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
801 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
802 			if (error) {
803 				splx(s);
804 				return (error);
805 			}
806 			goto restart;
807 		}
808 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
809 		if (bp->b_flags & B_DELWRI) {
810 			bp->b_flags &= ~B_DELWRI;
811 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
812 			wakeup(&fs->lfs_avail);
813 		}
814 		LFS_UNLOCK_BUF(bp);
815 		simple_unlock(&bp->b_interlock);
816 		brelse(bp);
817 	}
818 
819 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
820 		nbp = LIST_NEXT(bp, b_vnbufs);
821 		if (bp->b_lblkno < lbn)
822 			continue;
823 		simple_lock(&bp->b_interlock);
824 		if (bp->b_flags & B_BUSY) {
825 			bp->b_flags |= B_WANTED;
826 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
827 			    "lfs_vtruncbuf", slptimeo, &bp->b_interlock);
828 			if (error) {
829 				splx(s);
830 				return (error);
831 			}
832 			goto restart;
833 		}
834 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
835 		if (bp->b_flags & B_DELWRI) {
836 			bp->b_flags &= ~B_DELWRI;
837 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
838 			wakeup(&fs->lfs_avail);
839 		}
840 		LFS_UNLOCK_BUF(bp);
841 		simple_unlock(&bp->b_interlock);
842 		brelse(bp);
843 	}
844 
845 	splx(s);
846 
847 	return (0);
848 }
849 
850