xref: /netbsd-src/sys/ufs/ffs/ffs_inode.c (revision b78992537496bc71ee3d761f9fe0be0fc0a9a001)
1 /*	$NetBSD: ffs_inode.c,v 1.99 2008/08/30 08:25:53 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Wasabi Systems, Inc.
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 /*
33  * Copyright (c) 1982, 1986, 1989, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the University nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
61  */
62 
63 #include <sys/cdefs.h>
64 __KERNEL_RCSID(0, "$NetBSD: ffs_inode.c,v 1.99 2008/08/30 08:25:53 hannken Exp $");
65 
66 #if defined(_KERNEL_OPT)
67 #include "opt_ffs.h"
68 #include "opt_quota.h"
69 #endif
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/buf.h>
74 #include <sys/file.h>
75 #include <sys/fstrans.h>
76 #include <sys/kauth.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/mount.h>
80 #include <sys/proc.h>
81 #include <sys/resourcevar.h>
82 #include <sys/trace.h>
83 #include <sys/vnode.h>
84 #include <sys/wapbl.h>
85 
86 #include <ufs/ufs/quota.h>
87 #include <ufs/ufs/inode.h>
88 #include <ufs/ufs/ufsmount.h>
89 #include <ufs/ufs/ufs_extern.h>
90 #include <ufs/ufs/ufs_bswap.h>
91 #include <ufs/ufs/ufs_wapbl.h>
92 
93 #include <ufs/ffs/fs.h>
94 #include <ufs/ffs/ffs_extern.h>
95 
96 static int ffs_indirtrunc(struct inode *, daddr_t, daddr_t, daddr_t, int,
97 			  int64_t *);
98 
99 /*
100  * Update the access, modified, and inode change times as specified
101  * by the IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.
102  * The IN_MODIFIED flag is used to specify that the inode needs to be
103  * updated but that the times have already been set. The access
104  * and modified times are taken from the second and third parameters;
105  * the inode change time is always taken from the current time. If
106  * UPDATE_WAIT flag is set, or UPDATE_DIROP is set and we are not doing
107  * softupdates, then wait for the disk write of the inode to complete.
108  */
109 
110 int
111 ffs_update(struct vnode *vp, const struct timespec *acc,
112     const struct timespec *mod, int updflags)
113 {
114 	struct fs *fs;
115 	struct buf *bp;
116 	struct inode *ip;
117 	int error;
118 	void *cp;
119 	int waitfor, flags;
120 
121 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
122 		return (0);
123 	ip = VTOI(vp);
124 	FFS_ITIMES(ip, acc, mod, NULL);
125 	if (updflags & UPDATE_CLOSE)
126 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
127 	else
128 		flags = ip->i_flag & IN_MODIFIED;
129 	if (flags == 0)
130 		return (0);
131 	fs = ip->i_fs;
132 
133 	if ((flags & IN_MODIFIED) != 0 &&
134 	    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
135 		waitfor = updflags & UPDATE_WAIT;
136 		if ((updflags & UPDATE_DIROP) && !DOINGSOFTDEP(vp))
137 			waitfor |= UPDATE_WAIT;
138 	} else
139 		waitfor = 0;
140 
141 	/*
142 	 * Ensure that uid and gid are correct. This is a temporary
143 	 * fix until fsck has been changed to do the update.
144 	 */
145 	if (fs->fs_magic == FS_UFS1_MAGIC &&			/* XXX */
146 	    fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
147 		ip->i_ffs1_ouid = ip->i_uid;	/* XXX */
148 		ip->i_ffs1_ogid = ip->i_gid;	/* XXX */
149 	}							/* XXX */
150 	error = bread(ip->i_devvp,
151 		      fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
152 		      (int)fs->fs_bsize, NOCRED, B_MODIFY, &bp);
153 	if (error) {
154 		brelse(bp, 0);
155 		return (error);
156 	}
157 	ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED);
158 	if (DOINGSOFTDEP(vp)) {
159 		softdep_update_inodeblock(ip, bp, waitfor);
160 	} else if (ip->i_ffs_effnlink != ip->i_nlink)
161 		panic("ffs_update: bad link cnt");
162 	/* Keep unlinked inode list up to date */
163 	KDASSERT(DIP(ip, nlink) == ip->i_nlink);
164 	if (ip->i_mode) {
165 		if (ip->i_nlink > 0) {
166 			UFS_WAPBL_UNREGISTER_INODE(ip->i_ump->um_mountp,
167 			    ip->i_number, ip->i_mode);
168 		} else {
169 			UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp,
170 			    ip->i_number, ip->i_mode);
171 		}
172 	}
173 	if (fs->fs_magic == FS_UFS1_MAGIC) {
174 		cp = (char *)bp->b_data +
175 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
176 #ifdef FFS_EI
177 		if (UFS_FSNEEDSWAP(fs))
178 			ffs_dinode1_swap(ip->i_din.ffs1_din,
179 			    (struct ufs1_dinode *)cp);
180 		else
181 #endif
182 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
183 	} else {
184 		cp = (char *)bp->b_data +
185 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
186 #ifdef FFS_EI
187 		if (UFS_FSNEEDSWAP(fs))
188 			ffs_dinode2_swap(ip->i_din.ffs2_din,
189 			    (struct ufs2_dinode *)cp);
190 		else
191 #endif
192 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
193 	}
194 	if (waitfor) {
195 		return (bwrite(bp));
196 	} else {
197 		bdwrite(bp);
198 		return (0);
199 	}
200 }
201 
202 #define	SINGLE	0	/* index of single indirect block */
203 #define	DOUBLE	1	/* index of double indirect block */
204 #define	TRIPLE	2	/* index of triple indirect block */
205 /*
206  * Truncate the inode oip to at most length size, freeing the
207  * disk blocks.
208  */
209 int
210 ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
211 {
212 	daddr_t lastblock;
213 	struct inode *oip = VTOI(ovp);
214 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
215 	daddr_t blks[NDADDR + NIADDR];
216 	struct fs *fs;
217 	int offset, pgoffset, level;
218 	int64_t count, blocksreleased = 0;
219 	int i, aflag, nblocks;
220 	int error, allerror = 0;
221 	off_t osize;
222 	int sync;
223 	struct ufsmount *ump = oip->i_ump;
224 
225 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
226 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
227 		KASSERT(oip->i_size == 0);
228 		return 0;
229 	}
230 
231 	if (length < 0)
232 		return (EINVAL);
233 
234 	if (ovp->v_type == VLNK &&
235 	    (oip->i_size < ump->um_maxsymlinklen ||
236 	     (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
237 		KDASSERT(length == 0);
238 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
239 		oip->i_size = 0;
240 		DIP_ASSIGN(oip, size, 0);
241 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
242 		return (ffs_update(ovp, NULL, NULL, 0));
243 	}
244 	if (oip->i_size == length) {
245 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
246 		return (ffs_update(ovp, NULL, NULL, 0));
247 	}
248 	fs = oip->i_fs;
249 	if (length > ump->um_maxfilesize)
250 		return (EFBIG);
251 
252 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
253 		ffs_snapremove(ovp);
254 
255 	osize = oip->i_size;
256 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
257 
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 
264 	if (osize < length) {
265 		if (lblkno(fs, osize) < NDADDR &&
266 		    lblkno(fs, osize) != lblkno(fs, length) &&
267 		    blkroundup(fs, osize) != osize) {
268 			off_t eob;
269 
270 			eob = blkroundup(fs, osize);
271 			uvm_vnp_setwritesize(ovp, eob);
272 			error = ufs_balloc_range(ovp, osize, eob - osize,
273 			    cred, aflag);
274 			if (error)
275 				return error;
276 			if (ioflag & IO_SYNC) {
277 				mutex_enter(&ovp->v_interlock);
278 				VOP_PUTPAGES(ovp,
279 				    trunc_page(osize & fs->fs_bmask),
280 				    round_page(eob), PGO_CLEANIT | PGO_SYNCIO |
281 				    PGO_JOURNALLOCKED);
282 			}
283 		}
284 		uvm_vnp_setwritesize(ovp, length);
285 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
286 		if (error) {
287 			(void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred);
288 			return (error);
289 		}
290 		uvm_vnp_setsize(ovp, length);
291 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
292 		KASSERT(ovp->v_size == oip->i_size);
293 		return (ffs_update(ovp, NULL, NULL, 0));
294 	}
295 
296 	/*
297 	 * When truncating a regular file down to a non-block-aligned size,
298 	 * we must zero the part of last block which is past the new EOF.
299 	 * We must synchronously flush the zeroed pages to disk
300 	 * since the new pages will be invalidated as soon as we
301 	 * inform the VM system of the new, smaller size.
302 	 * We must do this before acquiring the GLOCK, since fetching
303 	 * the pages will acquire the GLOCK internally.
304 	 * So there is a window where another thread could see a whole
305 	 * zeroed page past EOF, but that's life.
306 	 */
307 
308 	offset = blkoff(fs, length);
309 	pgoffset = length & PAGE_MASK;
310 	if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
311 	    osize > length) {
312 		daddr_t lbn;
313 		voff_t eoz;
314 		int size;
315 
316 		if (offset != 0) {
317 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
318 			    aflag);
319 			if (error)
320 				return error;
321 		}
322 		lbn = lblkno(fs, length);
323 		size = blksize(fs, oip, lbn);
324 		eoz = MIN(MAX(lblktosize(fs, lbn) + size, round_page(pgoffset)),
325 		    osize);
326 		uvm_vnp_zerorange(ovp, length, eoz - length);
327 		if (round_page(eoz) > round_page(length)) {
328 			mutex_enter(&ovp->v_interlock);
329 			error = VOP_PUTPAGES(ovp, round_page(length),
330 			    round_page(eoz),
331 			    PGO_CLEANIT | PGO_DEACTIVATE | PGO_JOURNALLOCKED |
332 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
333 			if (error)
334 				return error;
335 		}
336 	}
337 
338 	genfs_node_wrlock(ovp);
339 
340 	if (DOINGSOFTDEP(ovp)) {
341 		if (length > 0) {
342 			/*
343 			 * If a file is only partially truncated, then
344 			 * we have to clean up the data structures
345 			 * describing the allocation past the truncation
346 			 * point. Finding and deallocating those structures
347 			 * is a lot of work. Since partial truncation occurs
348 			 * rarely, we solve the problem by syncing the file
349 			 * so that it will have no data structures left.
350 			 */
351 			if ((error = VOP_FSYNC(ovp, cred, FSYNC_WAIT,
352 			    0, 0)) != 0) {
353 				genfs_node_unlock(ovp);
354 				return (error);
355 			}
356 			mutex_enter(&ump->um_lock);
357 			if (oip->i_flag & IN_SPACECOUNTED)
358 				fs->fs_pendingblocks -= DIP(oip, blocks);
359 			mutex_exit(&ump->um_lock);
360 		} else {
361 			uvm_vnp_setsize(ovp, length);
362 #ifdef QUOTA
363  			(void) chkdq(oip, -DIP(oip, blocks), NOCRED, 0);
364 #endif
365 			softdep_setup_freeblocks(oip, length, 0);
366 			(void) vinvalbuf(ovp, 0, cred, curlwp, 0, 0);
367 			genfs_node_unlock(ovp);
368 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
369 			return (ffs_update(ovp, NULL, NULL, 0));
370 		}
371 	}
372 	oip->i_size = length;
373 	DIP_ASSIGN(oip, size, length);
374 	uvm_vnp_setsize(ovp, length);
375 	/*
376 	 * Calculate index into inode's block list of
377 	 * last direct and indirect blocks (if any)
378 	 * which we want to keep.  Lastblock is -1 when
379 	 * the file is truncated to 0.
380 	 */
381 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
382 	lastiblock[SINGLE] = lastblock - NDADDR;
383 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
384 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
385 	nblocks = btodb(fs->fs_bsize);
386 	/*
387 	 * Update file and block pointers on disk before we start freeing
388 	 * blocks.  If we crash before free'ing blocks below, the blocks
389 	 * will be returned to the free list.  lastiblock values are also
390 	 * normalized to -1 for calls to ffs_indirtrunc below.
391 	 */
392 	sync = 0;
393 	for (level = TRIPLE; level >= SINGLE; level--) {
394 		blks[NDADDR + level] = DIP(oip, ib[level]);
395 		if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
396 			sync = 1;
397 			DIP_ASSIGN(oip, ib[level], 0);
398 			lastiblock[level] = -1;
399 		}
400 	}
401 	for (i = 0; i < NDADDR; i++) {
402 		blks[i] = DIP(oip, db[i]);
403 		if (i > lastblock && blks[i] != 0) {
404 			sync = 1;
405 			DIP_ASSIGN(oip, db[i], 0);
406 		}
407 	}
408 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
409 	if (sync) {
410 		error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT);
411 		if (error && !allerror)
412 			allerror = error;
413 	}
414 
415 	/*
416 	 * Having written the new inode to disk, save its new configuration
417 	 * and put back the old block pointers long enough to process them.
418 	 * Note that we save the new block configuration so we can check it
419 	 * when we are done.
420 	 */
421 	for (i = 0; i < NDADDR; i++) {
422 		bn = DIP(oip, db[i]);
423 		DIP_ASSIGN(oip, db[i], blks[i]);
424 		blks[i] = bn;
425 	}
426 	for (i = 0; i < NIADDR; i++) {
427 		bn = DIP(oip, ib[i]);
428 		DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
429 		blks[NDADDR + i] = bn;
430 	}
431 
432 	oip->i_size = osize;
433 	DIP_ASSIGN(oip, size, osize);
434 	error = vtruncbuf(ovp, lastblock + 1, 0, 0);
435 	if (error && !allerror)
436 		allerror = error;
437 
438 	/*
439 	 * Indirect blocks first.
440 	 */
441 	indir_lbn[SINGLE] = -NDADDR;
442 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
443 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
444 	for (level = TRIPLE; level >= SINGLE; level--) {
445 		if (oip->i_ump->um_fstype == UFS1)
446 			bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
447 		else
448 			bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
449 		if (bn != 0) {
450 			error = ffs_indirtrunc(oip, indir_lbn[level],
451 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
452 			if (error)
453 				allerror = error;
454 			blocksreleased += count;
455 			if (lastiblock[level] < 0) {
456 				DIP_ASSIGN(oip, ib[level], 0);
457 				if (oip->i_ump->um_mountp->mnt_wapbl) {
458 					UFS_WAPBL_REGISTER_DEALLOCATION(
459 					    oip->i_ump->um_mountp,
460 					    fsbtodb(fs, bn), fs->fs_bsize);
461 				} else
462 					ffs_blkfree(fs, oip->i_devvp, bn,
463 					    fs->fs_bsize, oip->i_number);
464 				blocksreleased += nblocks;
465 			}
466 		}
467 		if (lastiblock[level] >= 0)
468 			goto done;
469 	}
470 
471 	/*
472 	 * All whole direct blocks or frags.
473 	 */
474 	for (i = NDADDR - 1; i > lastblock; i--) {
475 		long bsize;
476 
477 		if (oip->i_ump->um_fstype == UFS1)
478 			bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
479 		else
480 			bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
481 		if (bn == 0)
482 			continue;
483 		DIP_ASSIGN(oip, db[i], 0);
484 		bsize = blksize(fs, oip, i);
485 		if ((oip->i_ump->um_mountp->mnt_wapbl) &&
486 		    (ovp->v_type != VREG)) {
487 			UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp,
488 			    fsbtodb(fs, bn), bsize);
489 		} else
490 			ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
491 		blocksreleased += btodb(bsize);
492 	}
493 	if (lastblock < 0)
494 		goto done;
495 
496 	/*
497 	 * Finally, look for a change in size of the
498 	 * last direct block; release any frags.
499 	 */
500 	if (oip->i_ump->um_fstype == UFS1)
501 		bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
502 	else
503 		bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
504 	if (bn != 0) {
505 		long oldspace, newspace;
506 
507 		/*
508 		 * Calculate amount of space we're giving
509 		 * back as old block size minus new block size.
510 		 */
511 		oldspace = blksize(fs, oip, lastblock);
512 		oip->i_size = length;
513 		DIP_ASSIGN(oip, size, length);
514 		newspace = blksize(fs, oip, lastblock);
515 		if (newspace == 0)
516 			panic("itrunc: newspace");
517 		if (oldspace - newspace > 0) {
518 			/*
519 			 * Block number of space to be free'd is
520 			 * the old block # plus the number of frags
521 			 * required for the storage we're keeping.
522 			 */
523 			bn += numfrags(fs, newspace);
524 			if ((oip->i_ump->um_mountp->mnt_wapbl) &&
525 			    (ovp->v_type != VREG)) {
526 				UFS_WAPBL_REGISTER_DEALLOCATION(
527 				    oip->i_ump->um_mountp, fsbtodb(fs, bn),
528 				    oldspace - newspace);
529 			} else
530 				ffs_blkfree(fs, oip->i_devvp, bn,
531 				    oldspace - newspace, oip->i_number);
532 			blocksreleased += btodb(oldspace - newspace);
533 		}
534 	}
535 
536 done:
537 #ifdef DIAGNOSTIC
538 	for (level = SINGLE; level <= TRIPLE; level++)
539 		if (blks[NDADDR + level] != DIP(oip, ib[level]))
540 			panic("itrunc1");
541 	for (i = 0; i < NDADDR; i++)
542 		if (blks[i] != DIP(oip, db[i]))
543 			panic("itrunc2");
544 	if (length == 0 &&
545 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
546 		panic("itrunc3");
547 #endif /* DIAGNOSTIC */
548 	/*
549 	 * Put back the real size.
550 	 */
551 	oip->i_size = length;
552 	DIP_ASSIGN(oip, size, length);
553 	DIP_ADD(oip, blocks, -blocksreleased);
554 	genfs_node_unlock(ovp);
555 	oip->i_flag |= IN_CHANGE;
556 	UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0);
557 #ifdef QUOTA
558 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
559 #endif
560 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
561 	return (allerror);
562 }
563 
564 /*
565  * Release blocks associated with the inode ip and stored in the indirect
566  * block bn.  Blocks are free'd in LIFO order up to (but not including)
567  * lastbn.  If level is greater than SINGLE, the block is an indirect block
568  * and recursive calls to indirtrunc must be used to cleanse other indirect
569  * blocks.
570  *
571  * NB: triple indirect blocks are untested.
572  */
573 static int
574 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, daddr_t lastbn,
575     int level, int64_t *countp)
576 {
577 	int i;
578 	struct buf *bp;
579 	struct fs *fs = ip->i_fs;
580 	int32_t *bap1 = NULL;
581 	int64_t *bap2 = NULL;
582 	struct vnode *vp;
583 	daddr_t nb, nlbn, last;
584 	char *copy = NULL;
585 	int64_t blkcount, factor, blocksreleased = 0;
586 	int nblocks;
587 	int error = 0, allerror = 0;
588 #ifdef FFS_EI
589 	const int needswap = UFS_FSNEEDSWAP(fs);
590 #endif
591 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
592 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
593 #define BAP_ASSIGN(ip, i, value)					\
594 	do {								\
595 		if ((ip)->i_ump->um_fstype == UFS1)			\
596 			bap1[i] = (value);				\
597 		else							\
598 			bap2[i] = (value);				\
599 	} while(0)
600 
601 	/*
602 	 * Calculate index in current block of last
603 	 * block to be kept.  -1 indicates the entire
604 	 * block so we need not calculate the index.
605 	 */
606 	factor = 1;
607 	for (i = SINGLE; i < level; i++)
608 		factor *= NINDIR(fs);
609 	last = lastbn;
610 	if (lastbn > 0)
611 		last /= factor;
612 	nblocks = btodb(fs->fs_bsize);
613 	/*
614 	 * Get buffer of block pointers, zero those entries corresponding
615 	 * to blocks to be free'd, and update on disk copy first.  Since
616 	 * double(triple) indirect before single(double) indirect, calls
617 	 * to bmap on these blocks will fail.  However, we already have
618 	 * the on disk address, so we have to set the b_blkno field
619 	 * explicitly instead of letting bread do everything for us.
620 	 */
621 	vp = ITOV(ip);
622 	error = ffs_getblk(vp, lbn, FFS_NOBLK, fs->fs_bsize, false, &bp);
623 	if (error) {
624 		*countp = 0;
625 		return error;
626 	}
627 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
628 		/* Braces must be here in case trace evaluates to nothing. */
629 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
630 	} else {
631 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
632 		curlwp->l_ru.ru_inblock++;	/* pay for read */
633 		bp->b_flags |= B_READ;
634 		bp->b_flags &= ~B_COWDONE;	/* we change blkno below */
635 		if (bp->b_bcount > bp->b_bufsize)
636 			panic("ffs_indirtrunc: bad buffer size");
637 		bp->b_blkno = dbn;
638 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
639 		VOP_STRATEGY(vp, bp);
640 		error = biowait(bp);
641 		if (error == 0)
642 			error = fscow_run(bp, true);
643 	}
644 	if (error) {
645 		brelse(bp, 0);
646 		*countp = 0;
647 		return (error);
648 	}
649 
650 	if (ip->i_ump->um_fstype == UFS1)
651 		bap1 = (int32_t *)bp->b_data;
652 	else
653 		bap2 = (int64_t *)bp->b_data;
654 	if (lastbn >= 0) {
655 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
656 		memcpy((void *)copy, bp->b_data, (u_int)fs->fs_bsize);
657 		for (i = last + 1; i < NINDIR(fs); i++)
658 			BAP_ASSIGN(ip, i, 0);
659 		error = bwrite(bp);
660 		if (error)
661 			allerror = error;
662 		if (ip->i_ump->um_fstype == UFS1)
663 			bap1 = (int32_t *)copy;
664 		else
665 			bap2 = (int64_t *)copy;
666 	}
667 
668 	/*
669 	 * Recursively free totally unused blocks.
670 	 */
671 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
672 	    i--, nlbn += factor) {
673 		nb = RBAP(ip, i);
674 		if (nb == 0)
675 			continue;
676 		if (level > SINGLE) {
677 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
678 					       (daddr_t)-1, level - 1,
679 					       &blkcount);
680 			if (error)
681 				allerror = error;
682 			blocksreleased += blkcount;
683 		}
684 		if ((ip->i_ump->um_mountp->mnt_wapbl) &&
685 		    ((level > SINGLE) || (ITOV(ip)->v_type != VREG))) {
686 			UFS_WAPBL_REGISTER_DEALLOCATION(ip->i_ump->um_mountp,
687 			    fsbtodb(fs, nb), fs->fs_bsize);
688 		} else
689 			ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize,
690 			    ip->i_number);
691 		blocksreleased += nblocks;
692 	}
693 
694 	/*
695 	 * Recursively free last partial block.
696 	 */
697 	if (level > SINGLE && lastbn >= 0) {
698 		last = lastbn % factor;
699 		nb = RBAP(ip, i);
700 		if (nb != 0) {
701 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
702 					       last, level - 1, &blkcount);
703 			if (error)
704 				allerror = error;
705 			blocksreleased += blkcount;
706 		}
707 	}
708 
709 	if (copy != NULL) {
710 		FREE(copy, M_TEMP);
711 	} else {
712 		brelse(bp, BC_INVAL);
713 	}
714 
715 	*countp = blocksreleased;
716 	return (allerror);
717 }
718 
719 void
720 ffs_itimes(struct inode *ip, const struct timespec *acc,
721     const struct timespec *mod, const struct timespec *cre)
722 {
723 	struct timespec now;
724 
725 	if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY))) {
726 		return;
727 	}
728 
729 	vfs_timestamp(&now);
730 	if (ip->i_flag & IN_ACCESS) {
731 		if (acc == NULL)
732 			acc = &now;
733 		DIP_ASSIGN(ip, atime, acc->tv_sec);
734 		DIP_ASSIGN(ip, atimensec, acc->tv_nsec);
735 	}
736 	if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) {
737 		if ((ip->i_flags & SF_SNAPSHOT) == 0) {
738 			if (mod == NULL)
739 				mod = &now;
740 			DIP_ASSIGN(ip, mtime, mod->tv_sec);
741 			DIP_ASSIGN(ip, mtimensec, mod->tv_nsec);
742 		}
743 		ip->i_modrev++;
744 	}
745 	if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) {
746 		if (cre == NULL)
747 			cre = &now;
748 		DIP_ASSIGN(ip, ctime, cre->tv_sec);
749 		DIP_ASSIGN(ip, ctimensec, cre->tv_nsec);
750 	}
751 	if (ip->i_flag & (IN_ACCESS | IN_MODIFY))
752 		ip->i_flag |= IN_ACCESSED;
753 	if (ip->i_flag & (IN_UPDATE | IN_CHANGE))
754 		ip->i_flag |= IN_MODIFIED;
755 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY);
756 }
757