xref: /netbsd/sys/ufs/ffs/ffs_inode.c (revision 6550d01e)
1 /*	$NetBSD: ffs_inode.c,v 1.104 2010/02/07 17:12:40 bouyer 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 1.104 2010/02/07 17:12:40 bouyer 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 then wait for the
107  * 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) != 0)
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 	/* Keep unlinked inode list up to date */
159 	KDASSERT(DIP(ip, nlink) == ip->i_nlink);
160 	if (ip->i_mode) {
161 		if (ip->i_nlink > 0) {
162 			UFS_WAPBL_UNREGISTER_INODE(ip->i_ump->um_mountp,
163 			    ip->i_number, ip->i_mode);
164 		} else {
165 			UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp,
166 			    ip->i_number, ip->i_mode);
167 		}
168 	}
169 	if (fs->fs_magic == FS_UFS1_MAGIC) {
170 		cp = (char *)bp->b_data +
171 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
172 #ifdef FFS_EI
173 		if (UFS_FSNEEDSWAP(fs))
174 			ffs_dinode1_swap(ip->i_din.ffs1_din,
175 			    (struct ufs1_dinode *)cp);
176 		else
177 #endif
178 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
179 	} else {
180 		cp = (char *)bp->b_data +
181 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
182 #ifdef FFS_EI
183 		if (UFS_FSNEEDSWAP(fs))
184 			ffs_dinode2_swap(ip->i_din.ffs2_din,
185 			    (struct ufs2_dinode *)cp);
186 		else
187 #endif
188 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
189 	}
190 	if (waitfor) {
191 		return (bwrite(bp));
192 	} else {
193 		bdwrite(bp);
194 		return (0);
195 	}
196 }
197 
198 #define	SINGLE	0	/* index of single indirect block */
199 #define	DOUBLE	1	/* index of double indirect block */
200 #define	TRIPLE	2	/* index of triple indirect block */
201 /*
202  * Truncate the inode oip to at most length size, freeing the
203  * disk blocks.
204  */
205 int
206 ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
207 {
208 	daddr_t lastblock;
209 	struct inode *oip = VTOI(ovp);
210 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
211 	daddr_t blks[NDADDR + NIADDR];
212 	struct fs *fs;
213 	int offset, pgoffset, level;
214 	int64_t count, blocksreleased = 0;
215 	int i, aflag, nblocks;
216 	int error, allerror = 0;
217 	off_t osize;
218 	int sync;
219 	struct ufsmount *ump = oip->i_ump;
220 
221 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
222 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
223 		KASSERT(oip->i_size == 0);
224 		return 0;
225 	}
226 
227 	if (length < 0)
228 		return (EINVAL);
229 
230 	if (ovp->v_type == VLNK &&
231 	    (oip->i_size < ump->um_maxsymlinklen ||
232 	     (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
233 		KDASSERT(length == 0);
234 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
235 		oip->i_size = 0;
236 		DIP_ASSIGN(oip, size, 0);
237 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
238 		return (ffs_update(ovp, NULL, NULL, 0));
239 	}
240 	if (oip->i_size == length) {
241 		/* still do a uvm_vnp_setsize() as writesize may be larger */
242 		uvm_vnp_setsize(ovp, length);
243 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
244 		return (ffs_update(ovp, NULL, NULL, 0));
245 	}
246 	fs = oip->i_fs;
247 	if (length > ump->um_maxfilesize)
248 		return (EFBIG);
249 
250 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
251 		ffs_snapremove(ovp);
252 
253 	osize = oip->i_size;
254 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
255 
256 	/*
257 	 * Lengthen the size of the file. We must ensure that the
258 	 * last byte of the file is allocated. Since the smallest
259 	 * value of osize is 0, length will be at least 1.
260 	 */
261 
262 	if (osize < length) {
263 		if (lblkno(fs, osize) < NDADDR &&
264 		    lblkno(fs, osize) != lblkno(fs, length) &&
265 		    blkroundup(fs, osize) != osize) {
266 			off_t eob;
267 
268 			eob = blkroundup(fs, osize);
269 			uvm_vnp_setwritesize(ovp, eob);
270 			error = ufs_balloc_range(ovp, osize, eob - osize,
271 			    cred, aflag);
272 			if (error)
273 				return error;
274 			if (ioflag & IO_SYNC) {
275 				mutex_enter(&ovp->v_interlock);
276 				VOP_PUTPAGES(ovp,
277 				    trunc_page(osize & fs->fs_bmask),
278 				    round_page(eob), PGO_CLEANIT | PGO_SYNCIO |
279 				    PGO_JOURNALLOCKED);
280 			}
281 		}
282 		uvm_vnp_setwritesize(ovp, length);
283 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
284 		if (error) {
285 			(void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred);
286 			return (error);
287 		}
288 		uvm_vnp_setsize(ovp, length);
289 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
290 		KASSERT(ovp->v_size == oip->i_size);
291 		return (ffs_update(ovp, NULL, NULL, 0));
292 	}
293 
294 	/*
295 	 * When truncating a regular file down to a non-block-aligned size,
296 	 * we must zero the part of last block which is past the new EOF.
297 	 * We must synchronously flush the zeroed pages to disk
298 	 * since the new pages will be invalidated as soon as we
299 	 * inform the VM system of the new, smaller size.
300 	 * We must do this before acquiring the GLOCK, since fetching
301 	 * the pages will acquire the GLOCK internally.
302 	 * So there is a window where another thread could see a whole
303 	 * zeroed page past EOF, but that's life.
304 	 */
305 
306 	offset = blkoff(fs, length);
307 	pgoffset = length & PAGE_MASK;
308 	if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
309 	    osize > length) {
310 		daddr_t lbn;
311 		voff_t eoz;
312 		int size;
313 
314 		if (offset != 0) {
315 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
316 			    aflag);
317 			if (error)
318 				return error;
319 		}
320 		lbn = lblkno(fs, length);
321 		size = blksize(fs, oip, lbn);
322 		eoz = MIN(MAX(lblktosize(fs, lbn) + size, round_page(pgoffset)),
323 		    osize);
324 		uvm_vnp_zerorange(ovp, length, eoz - length);
325 		if (round_page(eoz) > round_page(length)) {
326 			mutex_enter(&ovp->v_interlock);
327 			error = VOP_PUTPAGES(ovp, round_page(length),
328 			    round_page(eoz),
329 			    PGO_CLEANIT | PGO_DEACTIVATE | PGO_JOURNALLOCKED |
330 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
331 			if (error)
332 				return error;
333 		}
334 	}
335 
336 	genfs_node_wrlock(ovp);
337 	oip->i_size = length;
338 	DIP_ASSIGN(oip, size, length);
339 	uvm_vnp_setsize(ovp, length);
340 	/*
341 	 * Calculate index into inode's block list of
342 	 * last direct and indirect blocks (if any)
343 	 * which we want to keep.  Lastblock is -1 when
344 	 * the file is truncated to 0.
345 	 */
346 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
347 	lastiblock[SINGLE] = lastblock - NDADDR;
348 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
349 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
350 	nblocks = btodb(fs->fs_bsize);
351 	/*
352 	 * Update file and block pointers on disk before we start freeing
353 	 * blocks.  If we crash before free'ing blocks below, the blocks
354 	 * will be returned to the free list.  lastiblock values are also
355 	 * normalized to -1 for calls to ffs_indirtrunc below.
356 	 */
357 	sync = 0;
358 	for (level = TRIPLE; level >= SINGLE; level--) {
359 		blks[NDADDR + level] = DIP(oip, ib[level]);
360 		if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
361 			sync = 1;
362 			DIP_ASSIGN(oip, ib[level], 0);
363 			lastiblock[level] = -1;
364 		}
365 	}
366 	for (i = 0; i < NDADDR; i++) {
367 		blks[i] = DIP(oip, db[i]);
368 		if (i > lastblock && blks[i] != 0) {
369 			sync = 1;
370 			DIP_ASSIGN(oip, db[i], 0);
371 		}
372 	}
373 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
374 	if (sync) {
375 		error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT);
376 		if (error && !allerror)
377 			allerror = error;
378 	}
379 
380 	/*
381 	 * Having written the new inode to disk, save its new configuration
382 	 * and put back the old block pointers long enough to process them.
383 	 * Note that we save the new block configuration so we can check it
384 	 * when we are done.
385 	 */
386 	for (i = 0; i < NDADDR; i++) {
387 		bn = DIP(oip, db[i]);
388 		DIP_ASSIGN(oip, db[i], blks[i]);
389 		blks[i] = bn;
390 	}
391 	for (i = 0; i < NIADDR; i++) {
392 		bn = DIP(oip, ib[i]);
393 		DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
394 		blks[NDADDR + i] = bn;
395 	}
396 
397 	oip->i_size = osize;
398 	DIP_ASSIGN(oip, size, osize);
399 	error = vtruncbuf(ovp, lastblock + 1, 0, 0);
400 	if (error && !allerror)
401 		allerror = error;
402 
403 	/*
404 	 * Indirect blocks first.
405 	 */
406 	indir_lbn[SINGLE] = -NDADDR;
407 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
408 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
409 	for (level = TRIPLE; level >= SINGLE; level--) {
410 		if (oip->i_ump->um_fstype == UFS1)
411 			bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
412 		else
413 			bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
414 		if (bn != 0) {
415 			error = ffs_indirtrunc(oip, indir_lbn[level],
416 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
417 			if (error)
418 				allerror = error;
419 			blocksreleased += count;
420 			if (lastiblock[level] < 0) {
421 				DIP_ASSIGN(oip, ib[level], 0);
422 				if (oip->i_ump->um_mountp->mnt_wapbl) {
423 					UFS_WAPBL_REGISTER_DEALLOCATION(
424 					    oip->i_ump->um_mountp,
425 					    fsbtodb(fs, bn), fs->fs_bsize);
426 				} else
427 					ffs_blkfree(fs, oip->i_devvp, bn,
428 					    fs->fs_bsize, oip->i_number);
429 				blocksreleased += nblocks;
430 			}
431 		}
432 		if (lastiblock[level] >= 0)
433 			goto done;
434 	}
435 
436 	/*
437 	 * All whole direct blocks or frags.
438 	 */
439 	for (i = NDADDR - 1; i > lastblock; i--) {
440 		long bsize;
441 
442 		if (oip->i_ump->um_fstype == UFS1)
443 			bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
444 		else
445 			bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
446 		if (bn == 0)
447 			continue;
448 		DIP_ASSIGN(oip, db[i], 0);
449 		bsize = blksize(fs, oip, i);
450 		if ((oip->i_ump->um_mountp->mnt_wapbl) &&
451 		    (ovp->v_type != VREG)) {
452 			UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp,
453 			    fsbtodb(fs, bn), bsize);
454 		} else
455 			ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
456 		blocksreleased += btodb(bsize);
457 	}
458 	if (lastblock < 0)
459 		goto done;
460 
461 	/*
462 	 * Finally, look for a change in size of the
463 	 * last direct block; release any frags.
464 	 */
465 	if (oip->i_ump->um_fstype == UFS1)
466 		bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
467 	else
468 		bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
469 	if (bn != 0) {
470 		long oldspace, newspace;
471 
472 		/*
473 		 * Calculate amount of space we're giving
474 		 * back as old block size minus new block size.
475 		 */
476 		oldspace = blksize(fs, oip, lastblock);
477 		oip->i_size = length;
478 		DIP_ASSIGN(oip, size, length);
479 		newspace = blksize(fs, oip, lastblock);
480 		if (newspace == 0)
481 			panic("itrunc: newspace");
482 		if (oldspace - newspace > 0) {
483 			/*
484 			 * Block number of space to be free'd is
485 			 * the old block # plus the number of frags
486 			 * required for the storage we're keeping.
487 			 */
488 			bn += numfrags(fs, newspace);
489 			if ((oip->i_ump->um_mountp->mnt_wapbl) &&
490 			    (ovp->v_type != VREG)) {
491 				UFS_WAPBL_REGISTER_DEALLOCATION(
492 				    oip->i_ump->um_mountp, fsbtodb(fs, bn),
493 				    oldspace - newspace);
494 			} else
495 				ffs_blkfree(fs, oip->i_devvp, bn,
496 				    oldspace - newspace, oip->i_number);
497 			blocksreleased += btodb(oldspace - newspace);
498 		}
499 	}
500 
501 done:
502 #ifdef DIAGNOSTIC
503 	for (level = SINGLE; level <= TRIPLE; level++)
504 		if (blks[NDADDR + level] != DIP(oip, ib[level]))
505 			panic("itrunc1");
506 	for (i = 0; i < NDADDR; i++)
507 		if (blks[i] != DIP(oip, db[i]))
508 			panic("itrunc2");
509 	if (length == 0 &&
510 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
511 		panic("itrunc3");
512 #endif /* DIAGNOSTIC */
513 	/*
514 	 * Put back the real size.
515 	 */
516 	oip->i_size = length;
517 	DIP_ASSIGN(oip, size, length);
518 	DIP_ADD(oip, blocks, -blocksreleased);
519 	genfs_node_unlock(ovp);
520 	oip->i_flag |= IN_CHANGE;
521 	UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0);
522 #ifdef QUOTA
523 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
524 #endif
525 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
526 	return (allerror);
527 }
528 
529 /*
530  * Release blocks associated with the inode ip and stored in the indirect
531  * block bn.  Blocks are free'd in LIFO order up to (but not including)
532  * lastbn.  If level is greater than SINGLE, the block is an indirect block
533  * and recursive calls to indirtrunc must be used to cleanse other indirect
534  * blocks.
535  *
536  * NB: triple indirect blocks are untested.
537  */
538 static int
539 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, daddr_t lastbn,
540     int level, int64_t *countp)
541 {
542 	int i;
543 	struct buf *bp;
544 	struct fs *fs = ip->i_fs;
545 	int32_t *bap1 = NULL;
546 	int64_t *bap2 = NULL;
547 	struct vnode *vp;
548 	daddr_t nb, nlbn, last;
549 	char *copy = NULL;
550 	int64_t blkcount, factor, blocksreleased = 0;
551 	int nblocks;
552 	int error = 0, allerror = 0;
553 #ifdef FFS_EI
554 	const int needswap = UFS_FSNEEDSWAP(fs);
555 #endif
556 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
557 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
558 #define BAP_ASSIGN(ip, i, value)					\
559 	do {								\
560 		if ((ip)->i_ump->um_fstype == UFS1)			\
561 			bap1[i] = (value);				\
562 		else							\
563 			bap2[i] = (value);				\
564 	} while(0)
565 
566 	/*
567 	 * Calculate index in current block of last
568 	 * block to be kept.  -1 indicates the entire
569 	 * block so we need not calculate the index.
570 	 */
571 	factor = 1;
572 	for (i = SINGLE; i < level; i++)
573 		factor *= NINDIR(fs);
574 	last = lastbn;
575 	if (lastbn > 0)
576 		last /= factor;
577 	nblocks = btodb(fs->fs_bsize);
578 	/*
579 	 * Get buffer of block pointers, zero those entries corresponding
580 	 * to blocks to be free'd, and update on disk copy first.  Since
581 	 * double(triple) indirect before single(double) indirect, calls
582 	 * to bmap on these blocks will fail.  However, we already have
583 	 * the on disk address, so we have to set the b_blkno field
584 	 * explicitly instead of letting bread do everything for us.
585 	 */
586 	vp = ITOV(ip);
587 	error = ffs_getblk(vp, lbn, FFS_NOBLK, fs->fs_bsize, false, &bp);
588 	if (error) {
589 		*countp = 0;
590 		return error;
591 	}
592 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
593 		/* Braces must be here in case trace evaluates to nothing. */
594 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
595 	} else {
596 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
597 		curlwp->l_ru.ru_inblock++;	/* pay for read */
598 		bp->b_flags |= B_READ;
599 		bp->b_flags &= ~B_COWDONE;	/* we change blkno below */
600 		if (bp->b_bcount > bp->b_bufsize)
601 			panic("ffs_indirtrunc: bad buffer size");
602 		bp->b_blkno = dbn;
603 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
604 		VOP_STRATEGY(vp, bp);
605 		error = biowait(bp);
606 		if (error == 0)
607 			error = fscow_run(bp, true);
608 	}
609 	if (error) {
610 		brelse(bp, 0);
611 		*countp = 0;
612 		return (error);
613 	}
614 
615 	if (ip->i_ump->um_fstype == UFS1)
616 		bap1 = (int32_t *)bp->b_data;
617 	else
618 		bap2 = (int64_t *)bp->b_data;
619 	if (lastbn >= 0) {
620 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
621 		memcpy((void *)copy, bp->b_data, (u_int)fs->fs_bsize);
622 		for (i = last + 1; i < NINDIR(fs); i++)
623 			BAP_ASSIGN(ip, i, 0);
624 		error = bwrite(bp);
625 		if (error)
626 			allerror = error;
627 		if (ip->i_ump->um_fstype == UFS1)
628 			bap1 = (int32_t *)copy;
629 		else
630 			bap2 = (int64_t *)copy;
631 	}
632 
633 	/*
634 	 * Recursively free totally unused blocks.
635 	 */
636 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
637 	    i--, nlbn += factor) {
638 		nb = RBAP(ip, i);
639 		if (nb == 0)
640 			continue;
641 		if (level > SINGLE) {
642 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
643 					       (daddr_t)-1, level - 1,
644 					       &blkcount);
645 			if (error)
646 				allerror = error;
647 			blocksreleased += blkcount;
648 		}
649 		if ((ip->i_ump->um_mountp->mnt_wapbl) &&
650 		    ((level > SINGLE) || (ITOV(ip)->v_type != VREG))) {
651 			UFS_WAPBL_REGISTER_DEALLOCATION(ip->i_ump->um_mountp,
652 			    fsbtodb(fs, nb), fs->fs_bsize);
653 		} else
654 			ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize,
655 			    ip->i_number);
656 		blocksreleased += nblocks;
657 	}
658 
659 	/*
660 	 * Recursively free last partial block.
661 	 */
662 	if (level > SINGLE && lastbn >= 0) {
663 		last = lastbn % factor;
664 		nb = RBAP(ip, i);
665 		if (nb != 0) {
666 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
667 					       last, level - 1, &blkcount);
668 			if (error)
669 				allerror = error;
670 			blocksreleased += blkcount;
671 		}
672 	}
673 
674 	if (copy != NULL) {
675 		free(copy, M_TEMP);
676 	} else {
677 		brelse(bp, BC_INVAL);
678 	}
679 
680 	*countp = blocksreleased;
681 	return (allerror);
682 }
683 
684 void
685 ffs_itimes(struct inode *ip, const struct timespec *acc,
686     const struct timespec *mod, const struct timespec *cre)
687 {
688 	struct timespec now;
689 
690 	if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY))) {
691 		return;
692 	}
693 
694 	vfs_timestamp(&now);
695 	if (ip->i_flag & IN_ACCESS) {
696 		if (acc == NULL)
697 			acc = &now;
698 		DIP_ASSIGN(ip, atime, acc->tv_sec);
699 		DIP_ASSIGN(ip, atimensec, acc->tv_nsec);
700 	}
701 	if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) {
702 		if ((ip->i_flags & SF_SNAPSHOT) == 0) {
703 			if (mod == NULL)
704 				mod = &now;
705 			DIP_ASSIGN(ip, mtime, mod->tv_sec);
706 			DIP_ASSIGN(ip, mtimensec, mod->tv_nsec);
707 		}
708 		ip->i_modrev++;
709 	}
710 	if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) {
711 		if (cre == NULL)
712 			cre = &now;
713 		DIP_ASSIGN(ip, ctime, cre->tv_sec);
714 		DIP_ASSIGN(ip, ctimensec, cre->tv_nsec);
715 	}
716 	if (ip->i_flag & (IN_ACCESS | IN_MODIFY))
717 		ip->i_flag |= IN_ACCESSED;
718 	if (ip->i_flag & (IN_UPDATE | IN_CHANGE))
719 		ip->i_flag |= IN_MODIFIED;
720 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY);
721 }
722