xref: /netbsd-src/sys/ufs/ffs/ffs_inode.c (revision e61202360d5611414dd6f6115934a96aa1f50b1a)
1 /*	$NetBSD: ffs_inode.c,v 1.110 2012/07/09 11:20:22 matt 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 1.110 2012/07/09 11:20:22 matt 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/kmem.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 	KDASSERTMSG(DIP(ip, nlink) == ip->i_nlink,
160 	    "DIP(ip, nlink) [%d] == ip->i_nlink [%d]",
161 	    DIP(ip, nlink), ip->i_nlink);
162 	if (ip->i_mode) {
163 		if (ip->i_nlink > 0) {
164 			UFS_WAPBL_UNREGISTER_INODE(ip->i_ump->um_mountp,
165 			    ip->i_number, ip->i_mode);
166 		} else {
167 			UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp,
168 			    ip->i_number, ip->i_mode);
169 		}
170 	}
171 	if (fs->fs_magic == FS_UFS1_MAGIC) {
172 		cp = (char *)bp->b_data +
173 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
174 #ifdef FFS_EI
175 		if (UFS_FSNEEDSWAP(fs))
176 			ffs_dinode1_swap(ip->i_din.ffs1_din,
177 			    (struct ufs1_dinode *)cp);
178 		else
179 #endif
180 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
181 	} else {
182 		cp = (char *)bp->b_data +
183 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
184 #ifdef FFS_EI
185 		if (UFS_FSNEEDSWAP(fs))
186 			ffs_dinode2_swap(ip->i_din.ffs2_din,
187 			    (struct ufs2_dinode *)cp);
188 		else
189 #endif
190 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
191 	}
192 	if (waitfor) {
193 		return (bwrite(bp));
194 	} else {
195 		bdwrite(bp);
196 		return (0);
197 	}
198 }
199 
200 #define	SINGLE	0	/* index of single indirect block */
201 #define	DOUBLE	1	/* index of double indirect block */
202 #define	TRIPLE	2	/* index of triple indirect block */
203 /*
204  * Truncate the inode oip to at most length size, freeing the
205  * disk blocks.
206  */
207 int
208 ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
209 {
210 	daddr_t lastblock;
211 	struct inode *oip = VTOI(ovp);
212 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
213 	daddr_t blks[NDADDR + NIADDR];
214 	struct fs *fs;
215 	int offset, pgoffset, level;
216 	int64_t count, blocksreleased = 0;
217 	int i, aflag, nblocks;
218 	int error, allerror = 0;
219 	off_t osize;
220 	int sync;
221 	struct ufsmount *ump = oip->i_ump;
222 
223 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
224 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
225 		KASSERT(oip->i_size == 0);
226 		return 0;
227 	}
228 
229 	if (length < 0)
230 		return (EINVAL);
231 
232 	if (ovp->v_type == VLNK &&
233 	    (oip->i_size < ump->um_maxsymlinklen ||
234 	     (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
235 		KDASSERT(length == 0);
236 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
237 		oip->i_size = 0;
238 		DIP_ASSIGN(oip, size, 0);
239 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
240 		return (ffs_update(ovp, NULL, NULL, 0));
241 	}
242 	if (oip->i_size == length) {
243 		/* still do a uvm_vnp_setsize() as writesize may be larger */
244 		uvm_vnp_setsize(ovp, 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 				(void) ffs_truncate(ovp, osize,
276 				    ioflag & IO_SYNC, cred);
277 				return error;
278 			}
279 			if (ioflag & IO_SYNC) {
280 				mutex_enter(ovp->v_interlock);
281 				VOP_PUTPAGES(ovp,
282 				    trunc_page(osize & fs->fs_bmask),
283 				    round_page(eob), PGO_CLEANIT | PGO_SYNCIO |
284 				    PGO_JOURNALLOCKED);
285 			}
286 		}
287 		uvm_vnp_setwritesize(ovp, length);
288 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
289 		if (error) {
290 			(void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred);
291 			return (error);
292 		}
293 		uvm_vnp_setsize(ovp, length);
294 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
295 		KASSERT(ovp->v_size == oip->i_size);
296 		return (ffs_update(ovp, NULL, NULL, 0));
297 	}
298 
299 	/*
300 	 * When truncating a regular file down to a non-block-aligned size,
301 	 * we must zero the part of last block which is past the new EOF.
302 	 * We must synchronously flush the zeroed pages to disk
303 	 * since the new pages will be invalidated as soon as we
304 	 * inform the VM system of the new, smaller size.
305 	 * We must do this before acquiring the GLOCK, since fetching
306 	 * the pages will acquire the GLOCK internally.
307 	 * So there is a window where another thread could see a whole
308 	 * zeroed page past EOF, but that's life.
309 	 */
310 
311 	offset = blkoff(fs, length);
312 	pgoffset = length & PAGE_MASK;
313 	if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
314 	    osize > length) {
315 		daddr_t lbn;
316 		voff_t eoz;
317 		int size;
318 
319 		if (offset != 0) {
320 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
321 			    aflag);
322 			if (error)
323 				return error;
324 		}
325 		lbn = lblkno(fs, length);
326 		size = blksize(fs, oip, lbn);
327 		eoz = MIN(MAX(lblktosize(fs, lbn) + size, round_page(pgoffset)),
328 		    osize);
329 		ubc_zerorange(&ovp->v_uobj, length, eoz - length,
330 		    UBC_UNMAP_FLAG(ovp));
331 		if (round_page(eoz) > round_page(length)) {
332 			mutex_enter(ovp->v_interlock);
333 			error = VOP_PUTPAGES(ovp, round_page(length),
334 			    round_page(eoz),
335 			    PGO_CLEANIT | PGO_DEACTIVATE | PGO_JOURNALLOCKED |
336 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
337 			if (error)
338 				return error;
339 		}
340 	}
341 
342 	genfs_node_wrlock(ovp);
343 	oip->i_size = length;
344 	DIP_ASSIGN(oip, size, length);
345 	uvm_vnp_setsize(ovp, length);
346 	/*
347 	 * Calculate index into inode's block list of
348 	 * last direct and indirect blocks (if any)
349 	 * which we want to keep.  Lastblock is -1 when
350 	 * the file is truncated to 0.
351 	 */
352 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
353 	lastiblock[SINGLE] = lastblock - NDADDR;
354 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
355 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
356 	nblocks = btodb(fs->fs_bsize);
357 	/*
358 	 * Update file and block pointers on disk before we start freeing
359 	 * blocks.  If we crash before free'ing blocks below, the blocks
360 	 * will be returned to the free list.  lastiblock values are also
361 	 * normalized to -1 for calls to ffs_indirtrunc below.
362 	 */
363 	sync = 0;
364 	for (level = TRIPLE; level >= SINGLE; level--) {
365 		blks[NDADDR + level] = DIP(oip, ib[level]);
366 		if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
367 			sync = 1;
368 			DIP_ASSIGN(oip, ib[level], 0);
369 			lastiblock[level] = -1;
370 		}
371 	}
372 	for (i = 0; i < NDADDR; i++) {
373 		blks[i] = DIP(oip, db[i]);
374 		if (i > lastblock && blks[i] != 0) {
375 			sync = 1;
376 			DIP_ASSIGN(oip, db[i], 0);
377 		}
378 	}
379 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
380 	if (sync) {
381 		error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT);
382 		if (error && !allerror)
383 			allerror = error;
384 	}
385 
386 	/*
387 	 * Having written the new inode to disk, save its new configuration
388 	 * and put back the old block pointers long enough to process them.
389 	 * Note that we save the new block configuration so we can check it
390 	 * when we are done.
391 	 */
392 	for (i = 0; i < NDADDR; i++) {
393 		bn = DIP(oip, db[i]);
394 		DIP_ASSIGN(oip, db[i], blks[i]);
395 		blks[i] = bn;
396 	}
397 	for (i = 0; i < NIADDR; i++) {
398 		bn = DIP(oip, ib[i]);
399 		DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
400 		blks[NDADDR + i] = bn;
401 	}
402 
403 	oip->i_size = osize;
404 	DIP_ASSIGN(oip, size, osize);
405 	error = vtruncbuf(ovp, lastblock + 1, 0, 0);
406 	if (error && !allerror)
407 		allerror = error;
408 
409 	/*
410 	 * Indirect blocks first.
411 	 */
412 	indir_lbn[SINGLE] = -NDADDR;
413 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
414 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
415 	for (level = TRIPLE; level >= SINGLE; level--) {
416 		if (oip->i_ump->um_fstype == UFS1)
417 			bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
418 		else
419 			bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
420 		if (bn != 0) {
421 			error = ffs_indirtrunc(oip, indir_lbn[level],
422 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
423 			if (error)
424 				allerror = error;
425 			blocksreleased += count;
426 			if (lastiblock[level] < 0) {
427 				DIP_ASSIGN(oip, ib[level], 0);
428 				if (oip->i_ump->um_mountp->mnt_wapbl) {
429 					UFS_WAPBL_REGISTER_DEALLOCATION(
430 					    oip->i_ump->um_mountp,
431 					    fsbtodb(fs, bn), fs->fs_bsize);
432 				} else
433 					ffs_blkfree(fs, oip->i_devvp, bn,
434 					    fs->fs_bsize, oip->i_number);
435 				blocksreleased += nblocks;
436 			}
437 		}
438 		if (lastiblock[level] >= 0)
439 			goto done;
440 	}
441 
442 	/*
443 	 * All whole direct blocks or frags.
444 	 */
445 	for (i = NDADDR - 1; i > lastblock; i--) {
446 		long bsize;
447 
448 		if (oip->i_ump->um_fstype == UFS1)
449 			bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
450 		else
451 			bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
452 		if (bn == 0)
453 			continue;
454 		DIP_ASSIGN(oip, db[i], 0);
455 		bsize = blksize(fs, oip, i);
456 		if ((oip->i_ump->um_mountp->mnt_wapbl) &&
457 		    (ovp->v_type != VREG)) {
458 			UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp,
459 			    fsbtodb(fs, bn), bsize);
460 		} else
461 			ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
462 		blocksreleased += btodb(bsize);
463 	}
464 	if (lastblock < 0)
465 		goto done;
466 
467 	/*
468 	 * Finally, look for a change in size of the
469 	 * last direct block; release any frags.
470 	 */
471 	if (oip->i_ump->um_fstype == UFS1)
472 		bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
473 	else
474 		bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
475 	if (bn != 0) {
476 		long oldspace, newspace;
477 
478 		/*
479 		 * Calculate amount of space we're giving
480 		 * back as old block size minus new block size.
481 		 */
482 		oldspace = blksize(fs, oip, lastblock);
483 		oip->i_size = length;
484 		DIP_ASSIGN(oip, size, length);
485 		newspace = blksize(fs, oip, lastblock);
486 		if (newspace == 0)
487 			panic("itrunc: newspace");
488 		if (oldspace - newspace > 0) {
489 			/*
490 			 * Block number of space to be free'd is
491 			 * the old block # plus the number of frags
492 			 * required for the storage we're keeping.
493 			 */
494 			bn += numfrags(fs, newspace);
495 			if ((oip->i_ump->um_mountp->mnt_wapbl) &&
496 			    (ovp->v_type != VREG)) {
497 				UFS_WAPBL_REGISTER_DEALLOCATION(
498 				    oip->i_ump->um_mountp, fsbtodb(fs, bn),
499 				    oldspace - newspace);
500 			} else
501 				ffs_blkfree(fs, oip->i_devvp, bn,
502 				    oldspace - newspace, oip->i_number);
503 			blocksreleased += btodb(oldspace - newspace);
504 		}
505 	}
506 
507 done:
508 #ifdef DIAGNOSTIC
509 	for (level = SINGLE; level <= TRIPLE; level++)
510 		if (blks[NDADDR + level] != DIP(oip, ib[level]))
511 			panic("itrunc1");
512 	for (i = 0; i < NDADDR; i++)
513 		if (blks[i] != DIP(oip, db[i]))
514 			panic("itrunc2");
515 	if (length == 0 &&
516 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
517 		panic("itrunc3");
518 #endif /* DIAGNOSTIC */
519 	/*
520 	 * Put back the real size.
521 	 */
522 	oip->i_size = length;
523 	DIP_ASSIGN(oip, size, length);
524 	DIP_ADD(oip, blocks, -blocksreleased);
525 	genfs_node_unlock(ovp);
526 	oip->i_flag |= IN_CHANGE;
527 	UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0);
528 #if defined(QUOTA) || defined(QUOTA2)
529 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
530 #endif
531 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
532 	return (allerror);
533 }
534 
535 /*
536  * Release blocks associated with the inode ip and stored in the indirect
537  * block bn.  Blocks are free'd in LIFO order up to (but not including)
538  * lastbn.  If level is greater than SINGLE, the block is an indirect block
539  * and recursive calls to indirtrunc must be used to cleanse other indirect
540  * blocks.
541  *
542  * NB: triple indirect blocks are untested.
543  */
544 static int
545 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, daddr_t lastbn,
546     int level, int64_t *countp)
547 {
548 	int i;
549 	struct buf *bp;
550 	struct fs *fs = ip->i_fs;
551 	int32_t *bap1 = NULL;
552 	int64_t *bap2 = NULL;
553 	struct vnode *vp;
554 	daddr_t nb, nlbn, last;
555 	char *copy = NULL;
556 	int64_t blkcount, factor, blocksreleased = 0;
557 	int nblocks;
558 	int error = 0, allerror = 0;
559 #ifdef FFS_EI
560 	const int needswap = UFS_FSNEEDSWAP(fs);
561 #endif
562 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
563 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
564 #define BAP_ASSIGN(ip, i, value)					\
565 	do {								\
566 		if ((ip)->i_ump->um_fstype == UFS1)			\
567 			bap1[i] = (value);				\
568 		else							\
569 			bap2[i] = (value);				\
570 	} while(0)
571 
572 	/*
573 	 * Calculate index in current block of last
574 	 * block to be kept.  -1 indicates the entire
575 	 * block so we need not calculate the index.
576 	 */
577 	factor = 1;
578 	for (i = SINGLE; i < level; i++)
579 		factor *= NINDIR(fs);
580 	last = lastbn;
581 	if (lastbn > 0)
582 		last /= factor;
583 	nblocks = btodb(fs->fs_bsize);
584 	/*
585 	 * Get buffer of block pointers, zero those entries corresponding
586 	 * to blocks to be free'd, and update on disk copy first.  Since
587 	 * double(triple) indirect before single(double) indirect, calls
588 	 * to bmap on these blocks will fail.  However, we already have
589 	 * the on disk address, so we have to set the b_blkno field
590 	 * explicitly instead of letting bread do everything for us.
591 	 */
592 	vp = ITOV(ip);
593 	error = ffs_getblk(vp, lbn, FFS_NOBLK, fs->fs_bsize, false, &bp);
594 	if (error) {
595 		*countp = 0;
596 		return error;
597 	}
598 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
599 		/* Braces must be here in case trace evaluates to nothing. */
600 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
601 	} else {
602 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
603 		curlwp->l_ru.ru_inblock++;	/* pay for read */
604 		bp->b_flags |= B_READ;
605 		bp->b_flags &= ~B_COWDONE;	/* we change blkno below */
606 		if (bp->b_bcount > bp->b_bufsize)
607 			panic("ffs_indirtrunc: bad buffer size");
608 		bp->b_blkno = dbn;
609 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
610 		VOP_STRATEGY(vp, bp);
611 		error = biowait(bp);
612 		if (error == 0)
613 			error = fscow_run(bp, true);
614 	}
615 	if (error) {
616 		brelse(bp, 0);
617 		*countp = 0;
618 		return (error);
619 	}
620 
621 	if (ip->i_ump->um_fstype == UFS1)
622 		bap1 = (int32_t *)bp->b_data;
623 	else
624 		bap2 = (int64_t *)bp->b_data;
625 	if (lastbn >= 0) {
626 		copy = kmem_alloc(fs->fs_bsize, KM_SLEEP);
627 		memcpy((void *)copy, bp->b_data, (u_int)fs->fs_bsize);
628 		for (i = last + 1; i < NINDIR(fs); i++)
629 			BAP_ASSIGN(ip, i, 0);
630 		error = bwrite(bp);
631 		if (error)
632 			allerror = error;
633 		if (ip->i_ump->um_fstype == UFS1)
634 			bap1 = (int32_t *)copy;
635 		else
636 			bap2 = (int64_t *)copy;
637 	}
638 
639 	/*
640 	 * Recursively free totally unused blocks.
641 	 */
642 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
643 	    i--, nlbn += factor) {
644 		nb = RBAP(ip, i);
645 		if (nb == 0)
646 			continue;
647 		if (level > SINGLE) {
648 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
649 					       (daddr_t)-1, level - 1,
650 					       &blkcount);
651 			if (error)
652 				allerror = error;
653 			blocksreleased += blkcount;
654 		}
655 		if ((ip->i_ump->um_mountp->mnt_wapbl) &&
656 		    ((level > SINGLE) || (ITOV(ip)->v_type != VREG))) {
657 			UFS_WAPBL_REGISTER_DEALLOCATION(ip->i_ump->um_mountp,
658 			    fsbtodb(fs, nb), fs->fs_bsize);
659 		} else
660 			ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize,
661 			    ip->i_number);
662 		blocksreleased += nblocks;
663 	}
664 
665 	/*
666 	 * Recursively free last partial block.
667 	 */
668 	if (level > SINGLE && lastbn >= 0) {
669 		last = lastbn % factor;
670 		nb = RBAP(ip, i);
671 		if (nb != 0) {
672 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
673 					       last, level - 1, &blkcount);
674 			if (error)
675 				allerror = error;
676 			blocksreleased += blkcount;
677 		}
678 	}
679 
680 	if (copy != NULL) {
681 		kmem_free(copy, fs->fs_bsize);
682 	} else {
683 		brelse(bp, BC_INVAL);
684 	}
685 
686 	*countp = blocksreleased;
687 	return (allerror);
688 }
689 
690 void
691 ffs_itimes(struct inode *ip, const struct timespec *acc,
692     const struct timespec *mod, const struct timespec *cre)
693 {
694 	struct timespec now;
695 
696 	if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY))) {
697 		return;
698 	}
699 
700 	vfs_timestamp(&now);
701 	if (ip->i_flag & IN_ACCESS) {
702 		if (acc == NULL)
703 			acc = &now;
704 		DIP_ASSIGN(ip, atime, acc->tv_sec);
705 		DIP_ASSIGN(ip, atimensec, acc->tv_nsec);
706 	}
707 	if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) {
708 		if ((ip->i_flags & SF_SNAPSHOT) == 0) {
709 			if (mod == NULL)
710 				mod = &now;
711 			DIP_ASSIGN(ip, mtime, mod->tv_sec);
712 			DIP_ASSIGN(ip, mtimensec, mod->tv_nsec);
713 		}
714 		ip->i_modrev++;
715 	}
716 	if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) {
717 		if (cre == NULL)
718 			cre = &now;
719 		DIP_ASSIGN(ip, ctime, cre->tv_sec);
720 		DIP_ASSIGN(ip, ctimensec, cre->tv_nsec);
721 	}
722 	if (ip->i_flag & (IN_ACCESS | IN_MODIFY))
723 		ip->i_flag |= IN_ACCESSED;
724 	if (ip->i_flag & (IN_UPDATE | IN_CHANGE))
725 		ip->i_flag |= IN_MODIFIED;
726 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY);
727 }
728