xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision cc16dea626cf2fc80cde667ac4798065108e596c)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_quota.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mount.h>
40 #include <sys/proc.h>
41 #include <sys/bio.h>
42 #include <sys/buf.h>
43 #include <sys/vnode.h>
44 #include <sys/malloc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/rwlock.h>
47 #include <sys/vmmeter.h>
48 #include <sys/stat.h>
49 
50 #include <vm/vm.h>
51 #include <vm/vm_extern.h>
52 #include <vm/vm_object.h>
53 
54 #include <ufs/ufs/extattr.h>
55 #include <ufs/ufs/quota.h>
56 #include <ufs/ufs/ufsmount.h>
57 #include <ufs/ufs/inode.h>
58 #include <ufs/ufs/ufs_extern.h>
59 
60 #include <ufs/ffs/fs.h>
61 #include <ufs/ffs/ffs_extern.h>
62 
63 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
64 	    ufs2_daddr_t, int, ufs2_daddr_t *);
65 
66 /*
67  * Update the access, modified, and inode change times as specified by the
68  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
69  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
70  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
71  * later if not now.  The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
72  * is currently being suspended (or is suspended) and vnode has been accessed.
73  * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
74  * reflect the presumably successful write, and if waitfor is set, then wait
75  * for the write to complete.
76  */
77 int
78 ffs_update(vp, waitfor)
79 	struct vnode *vp;
80 	int waitfor;
81 {
82 	struct fs *fs;
83 	struct buf *bp;
84 	struct inode *ip;
85 	int flags, error;
86 
87 	ASSERT_VOP_ELOCKED(vp, "ffs_update");
88 	ufs_itimes(vp);
89 	ip = VTOI(vp);
90 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
91 		return (0);
92 	ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
93 	fs = ip->i_fs;
94 	if (fs->fs_ronly && ip->i_ump->um_fsckpid == 0)
95 		return (0);
96 	/*
97 	 * If we are updating a snapshot and another process is currently
98 	 * writing the buffer containing the inode for this snapshot then
99 	 * a deadlock can occur when it tries to check the snapshot to see
100 	 * if that block needs to be copied. Thus when updating a snapshot
101 	 * we check to see if the buffer is already locked, and if it is
102 	 * we drop the snapshot lock until the buffer has been written
103 	 * and is available to us. We have to grab a reference to the
104 	 * snapshot vnode to prevent it from being removed while we are
105 	 * waiting for the buffer.
106 	 */
107 	flags = 0;
108 	if (IS_SNAPSHOT(ip))
109 		flags = GB_LOCK_NOWAIT;
110 loop:
111 	error = breadn_flags(ip->i_devvp,
112 	     fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
113 	     (int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
114 	if (error != 0) {
115 		if (error != EBUSY) {
116 			brelse(bp);
117 			return (error);
118 		}
119 		KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
120 		/*
121 		 * Wait for our inode block to become available.
122 		 *
123 		 * Hold a reference to the vnode to protect against
124 		 * ffs_snapgone(). Since we hold a reference, it can only
125 		 * get reclaimed (VI_DOOMED flag) in a forcible downgrade
126 		 * or unmount. For an unmount, the entire filesystem will be
127 		 * gone, so we cannot attempt to touch anything associated
128 		 * with it while the vnode is unlocked; all we can do is
129 		 * pause briefly and try again. If when we relock the vnode
130 		 * we discover that it has been reclaimed, updating it is no
131 		 * longer necessary and we can just return an error.
132 		 */
133 		vref(vp);
134 		VOP_UNLOCK(vp, 0);
135 		pause("ffsupd", 1);
136 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
137 		vrele(vp);
138 		if ((vp->v_iflag & VI_DOOMED) != 0)
139 			return (ENOENT);
140 		goto loop;
141 	}
142 	if (DOINGSOFTDEP(vp))
143 		softdep_update_inodeblock(ip, bp, waitfor);
144 	else if (ip->i_effnlink != ip->i_nlink)
145 		panic("ffs_update: bad link cnt");
146 	if (ip->i_ump->um_fstype == UFS1)
147 		*((struct ufs1_dinode *)bp->b_data +
148 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
149 	else
150 		*((struct ufs2_dinode *)bp->b_data +
151 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
152 	if (waitfor && !DOINGASYNC(vp))
153 		error = bwrite(bp);
154 	else if (vm_page_count_severe() || buf_dirty_count_severe()) {
155 		bawrite(bp);
156 		error = 0;
157 	} else {
158 		if (bp->b_bufsize == fs->fs_bsize)
159 			bp->b_flags |= B_CLUSTEROK;
160 		bdwrite(bp);
161 		error = 0;
162 	}
163 	return (error);
164 }
165 
166 #define	SINGLE	0	/* index of single indirect block */
167 #define	DOUBLE	1	/* index of double indirect block */
168 #define	TRIPLE	2	/* index of triple indirect block */
169 /*
170  * Truncate the inode ip to at most length size, freeing the
171  * disk blocks.
172  */
173 int
174 ffs_truncate(vp, length, flags, cred)
175 	struct vnode *vp;
176 	off_t length;
177 	int flags;
178 	struct ucred *cred;
179 {
180 	struct inode *ip;
181 	ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
182 	ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
183 	ufs2_daddr_t count, blocksreleased = 0, datablocks;
184 	struct bufobj *bo;
185 	struct fs *fs;
186 	struct buf *bp;
187 	struct ufsmount *ump;
188 	int softdeptrunc, journaltrunc;
189 	int needextclean, extblocks;
190 	int offset, size, level, nblocks;
191 	int i, error, allerror;
192 	off_t osize;
193 
194 	ip = VTOI(vp);
195 	fs = ip->i_fs;
196 	ump = ip->i_ump;
197 	bo = &vp->v_bufobj;
198 
199 	ASSERT_VOP_LOCKED(vp, "ffs_truncate");
200 
201 	if (length < 0)
202 		return (EINVAL);
203 	if (length > fs->fs_maxfilesize)
204 		return (EFBIG);
205 #ifdef QUOTA
206 	error = getinoquota(ip);
207 	if (error)
208 		return (error);
209 #endif
210 	/*
211 	 * Historically clients did not have to specify which data
212 	 * they were truncating. So, if not specified, we assume
213 	 * traditional behavior, e.g., just the normal data.
214 	 */
215 	if ((flags & (IO_EXT | IO_NORMAL)) == 0)
216 		flags |= IO_NORMAL;
217 	if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
218 		flags |= IO_SYNC;
219 	/*
220 	 * If we are truncating the extended-attributes, and cannot
221 	 * do it with soft updates, then do it slowly here. If we are
222 	 * truncating both the extended attributes and the file contents
223 	 * (e.g., the file is being unlinked), then pick it off with
224 	 * soft updates below.
225 	 */
226 	allerror = 0;
227 	needextclean = 0;
228 	softdeptrunc = 0;
229 	journaltrunc = DOINGSUJ(vp);
230 	if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
231 		softdeptrunc = !softdep_slowdown(vp);
232 	extblocks = 0;
233 	datablocks = DIP(ip, i_blocks);
234 	if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
235 		extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
236 		datablocks -= extblocks;
237 	}
238 	if ((flags & IO_EXT) && extblocks > 0) {
239 		if (length != 0)
240 			panic("ffs_truncate: partial trunc of extdata");
241 		if (softdeptrunc || journaltrunc) {
242 			if ((flags & IO_NORMAL) == 0)
243 				goto extclean;
244 			needextclean = 1;
245 		} else {
246 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
247 				return (error);
248 #ifdef QUOTA
249 			(void) chkdq(ip, -extblocks, NOCRED, 0);
250 #endif
251 			vinvalbuf(vp, V_ALT, 0, 0);
252 			vn_pages_remove(vp,
253 			    OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
254 			osize = ip->i_din2->di_extsize;
255 			ip->i_din2->di_blocks -= extblocks;
256 			ip->i_din2->di_extsize = 0;
257 			for (i = 0; i < NXADDR; i++) {
258 				oldblks[i] = ip->i_din2->di_extb[i];
259 				ip->i_din2->di_extb[i] = 0;
260 			}
261 			ip->i_flag |= IN_CHANGE;
262 			if ((error = ffs_update(vp, !DOINGASYNC(vp))))
263 				return (error);
264 			for (i = 0; i < NXADDR; i++) {
265 				if (oldblks[i] == 0)
266 					continue;
267 				ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
268 				    sblksize(fs, osize, i), ip->i_number,
269 				    vp->v_type, NULL);
270 			}
271 		}
272 	}
273 	if ((flags & IO_NORMAL) == 0)
274 		return (0);
275 	if (vp->v_type == VLNK &&
276 	    (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
277 	     datablocks == 0)) {
278 #ifdef INVARIANTS
279 		if (length != 0)
280 			panic("ffs_truncate: partial truncate of symlink");
281 #endif
282 		bzero(SHORTLINK(ip), (u_int)ip->i_size);
283 		ip->i_size = 0;
284 		DIP_SET(ip, i_size, 0);
285 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
286 		if (needextclean)
287 			goto extclean;
288 		return (ffs_update(vp, !DOINGASYNC(vp)));
289 	}
290 	if (ip->i_size == length) {
291 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
292 		if (needextclean)
293 			goto extclean;
294 		return (ffs_update(vp, 0));
295 	}
296 	if (fs->fs_ronly)
297 		panic("ffs_truncate: read-only filesystem");
298 	if (IS_SNAPSHOT(ip))
299 		ffs_snapremove(vp);
300 	vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
301 	osize = ip->i_size;
302 	/*
303 	 * Lengthen the size of the file. We must ensure that the
304 	 * last byte of the file is allocated. Since the smallest
305 	 * value of osize is 0, length will be at least 1.
306 	 */
307 	if (osize < length) {
308 		vnode_pager_setsize(vp, length);
309 		flags |= BA_CLRBUF;
310 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
311 		if (error) {
312 			vnode_pager_setsize(vp, osize);
313 			return (error);
314 		}
315 		ip->i_size = length;
316 		DIP_SET(ip, i_size, length);
317 		if (bp->b_bufsize == fs->fs_bsize)
318 			bp->b_flags |= B_CLUSTEROK;
319 		if (flags & IO_SYNC)
320 			bwrite(bp);
321 		else if (DOINGASYNC(vp))
322 			bdwrite(bp);
323 		else
324 			bawrite(bp);
325 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
326 		return (ffs_update(vp, !DOINGASYNC(vp)));
327 	}
328 	if (DOINGSOFTDEP(vp)) {
329 		if (softdeptrunc == 0 && journaltrunc == 0) {
330 			/*
331 			 * If a file is only partially truncated, then
332 			 * we have to clean up the data structures
333 			 * describing the allocation past the truncation
334 			 * point. Finding and deallocating those structures
335 			 * is a lot of work. Since partial truncation occurs
336 			 * rarely, we solve the problem by syncing the file
337 			 * so that it will have no data structures left.
338 			 */
339 			if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
340 				return (error);
341 		} else {
342 			flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
343 			if (journaltrunc)
344 				softdep_journal_freeblocks(ip, cred, length,
345 				    flags);
346 			else
347 				softdep_setup_freeblocks(ip, length, flags);
348 			ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
349 			if (journaltrunc == 0) {
350 				ip->i_flag |= IN_CHANGE | IN_UPDATE;
351 				error = ffs_update(vp, 0);
352 			}
353 			return (error);
354 		}
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 	if (offset == 0) {
366 		ip->i_size = length;
367 		DIP_SET(ip, i_size, length);
368 	} else {
369 		lbn = lblkno(fs, length);
370 		flags |= BA_CLRBUF;
371 		error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
372 		if (error)
373 			return (error);
374 		/*
375 		 * When we are doing soft updates and the UFS_BALLOC
376 		 * above fills in a direct block hole with a full sized
377 		 * block that will be truncated down to a fragment below,
378 		 * we must flush out the block dependency with an FSYNC
379 		 * so that we do not get a soft updates inconsistency
380 		 * when we create the fragment below.
381 		 */
382 		if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
383 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
384 		    (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
385 			return (error);
386 		ip->i_size = length;
387 		DIP_SET(ip, i_size, length);
388 		size = blksize(fs, ip, lbn);
389 		if (vp->v_type != VDIR)
390 			bzero((char *)bp->b_data + offset,
391 			    (u_int)(size - offset));
392 		/* Kirk's code has reallocbuf(bp, size, 1) here */
393 		allocbuf(bp, size);
394 		if (bp->b_bufsize == fs->fs_bsize)
395 			bp->b_flags |= B_CLUSTEROK;
396 		if (flags & IO_SYNC)
397 			bwrite(bp);
398 		else if (DOINGASYNC(vp))
399 			bdwrite(bp);
400 		else
401 			bawrite(bp);
402 	}
403 	/*
404 	 * Calculate index into inode's block list of
405 	 * last direct and indirect blocks (if any)
406 	 * which we want to keep.  Lastblock is -1 when
407 	 * the file is truncated to 0.
408 	 */
409 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
410 	lastiblock[SINGLE] = lastblock - NDADDR;
411 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
412 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
413 	nblocks = btodb(fs->fs_bsize);
414 	/*
415 	 * Update file and block pointers on disk before we start freeing
416 	 * blocks.  If we crash before free'ing blocks below, the blocks
417 	 * will be returned to the free list.  lastiblock values are also
418 	 * normalized to -1 for calls to ffs_indirtrunc below.
419 	 */
420 	for (level = TRIPLE; level >= SINGLE; level--) {
421 		oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
422 		if (lastiblock[level] < 0) {
423 			DIP_SET(ip, i_ib[level], 0);
424 			lastiblock[level] = -1;
425 		}
426 	}
427 	for (i = 0; i < NDADDR; i++) {
428 		oldblks[i] = DIP(ip, i_db[i]);
429 		if (i > lastblock)
430 			DIP_SET(ip, i_db[i], 0);
431 	}
432 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
433 	allerror = ffs_update(vp, !DOINGASYNC(vp));
434 
435 	/*
436 	 * Having written the new inode to disk, save its new configuration
437 	 * and put back the old block pointers long enough to process them.
438 	 * Note that we save the new block configuration so we can check it
439 	 * when we are done.
440 	 */
441 	for (i = 0; i < NDADDR; i++) {
442 		newblks[i] = DIP(ip, i_db[i]);
443 		DIP_SET(ip, i_db[i], oldblks[i]);
444 	}
445 	for (i = 0; i < NIADDR; i++) {
446 		newblks[NDADDR + i] = DIP(ip, i_ib[i]);
447 		DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
448 	}
449 	ip->i_size = osize;
450 	DIP_SET(ip, i_size, osize);
451 
452 	error = vtruncbuf(vp, cred, length, fs->fs_bsize);
453 	if (error && (allerror == 0))
454 		allerror = error;
455 
456 	/*
457 	 * Indirect blocks first.
458 	 */
459 	indir_lbn[SINGLE] = -NDADDR;
460 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
461 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
462 	for (level = TRIPLE; level >= SINGLE; level--) {
463 		bn = DIP(ip, i_ib[level]);
464 		if (bn != 0) {
465 			error = ffs_indirtrunc(ip, indir_lbn[level],
466 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
467 			if (error)
468 				allerror = error;
469 			blocksreleased += count;
470 			if (lastiblock[level] < 0) {
471 				DIP_SET(ip, i_ib[level], 0);
472 				ffs_blkfree(ump, fs, ip->i_devvp, bn,
473 				    fs->fs_bsize, ip->i_number,
474 				    vp->v_type, NULL);
475 				blocksreleased += nblocks;
476 			}
477 		}
478 		if (lastiblock[level] >= 0)
479 			goto done;
480 	}
481 
482 	/*
483 	 * All whole direct blocks or frags.
484 	 */
485 	for (i = NDADDR - 1; i > lastblock; i--) {
486 		long bsize;
487 
488 		bn = DIP(ip, i_db[i]);
489 		if (bn == 0)
490 			continue;
491 		DIP_SET(ip, i_db[i], 0);
492 		bsize = blksize(fs, ip, i);
493 		ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number,
494 		    vp->v_type, NULL);
495 		blocksreleased += btodb(bsize);
496 	}
497 	if (lastblock < 0)
498 		goto done;
499 
500 	/*
501 	 * Finally, look for a change in size of the
502 	 * last direct block; release any frags.
503 	 */
504 	bn = DIP(ip, i_db[lastblock]);
505 	if (bn != 0) {
506 		long oldspace, newspace;
507 
508 		/*
509 		 * Calculate amount of space we're giving
510 		 * back as old block size minus new block size.
511 		 */
512 		oldspace = blksize(fs, ip, lastblock);
513 		ip->i_size = length;
514 		DIP_SET(ip, i_size, length);
515 		newspace = blksize(fs, ip, lastblock);
516 		if (newspace == 0)
517 			panic("ffs_truncate: newspace");
518 		if (oldspace - newspace > 0) {
519 			/*
520 			 * Block number of space to be free'd is
521 			 * the old block # plus the number of frags
522 			 * required for the storage we're keeping.
523 			 */
524 			bn += numfrags(fs, newspace);
525 			ffs_blkfree(ump, fs, ip->i_devvp, bn,
526 			   oldspace - newspace, ip->i_number, vp->v_type, NULL);
527 			blocksreleased += btodb(oldspace - newspace);
528 		}
529 	}
530 done:
531 #ifdef INVARIANTS
532 	for (level = SINGLE; level <= TRIPLE; level++)
533 		if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
534 			panic("ffs_truncate1");
535 	for (i = 0; i < NDADDR; i++)
536 		if (newblks[i] != DIP(ip, i_db[i]))
537 			panic("ffs_truncate2");
538 	BO_LOCK(bo);
539 	if (length == 0 &&
540 	    (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
541 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
542 		panic("ffs_truncate3");
543 	BO_UNLOCK(bo);
544 #endif /* INVARIANTS */
545 	/*
546 	 * Put back the real size.
547 	 */
548 	ip->i_size = length;
549 	DIP_SET(ip, i_size, length);
550 	if (DIP(ip, i_blocks) >= blocksreleased)
551 		DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
552 	else	/* sanity */
553 		DIP_SET(ip, i_blocks, 0);
554 	ip->i_flag |= IN_CHANGE;
555 #ifdef QUOTA
556 	(void) chkdq(ip, -blocksreleased, NOCRED, 0);
557 #endif
558 	return (allerror);
559 
560 extclean:
561 	if (journaltrunc)
562 		softdep_journal_freeblocks(ip, cred, length, IO_EXT);
563 	else
564 		softdep_setup_freeblocks(ip, length, IO_EXT);
565 	return (ffs_update(vp, !DOINGASYNC(vp)));
566 }
567 
568 /*
569  * Release blocks associated with the inode ip and stored in the indirect
570  * block bn.  Blocks are free'd in LIFO order up to (but not including)
571  * lastbn.  If level is greater than SINGLE, the block is an indirect block
572  * and recursive calls to indirtrunc must be used to cleanse other indirect
573  * blocks.
574  */
575 static int
576 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
577 	struct inode *ip;
578 	ufs2_daddr_t lbn, lastbn;
579 	ufs2_daddr_t dbn;
580 	int level;
581 	ufs2_daddr_t *countp;
582 {
583 	struct buf *bp;
584 	struct fs *fs = ip->i_fs;
585 	struct vnode *vp;
586 	caddr_t copy = NULL;
587 	int i, nblocks, error = 0, allerror = 0;
588 	ufs2_daddr_t nb, nlbn, last;
589 	ufs2_daddr_t blkcount, factor, blocksreleased = 0;
590 	ufs1_daddr_t *bap1 = NULL;
591 	ufs2_daddr_t *bap2 = NULL;
592 #	define BAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? bap1[i] : bap2[i])
593 
594 	/*
595 	 * Calculate index in current block of last
596 	 * block to be kept.  -1 indicates the entire
597 	 * block so we need not calculate the index.
598 	 */
599 	factor = lbn_offset(fs, level);
600 	last = lastbn;
601 	if (lastbn > 0)
602 		last /= factor;
603 	nblocks = btodb(fs->fs_bsize);
604 	/*
605 	 * Get buffer of block pointers, zero those entries corresponding
606 	 * to blocks to be free'd, and update on disk copy first.  Since
607 	 * double(triple) indirect before single(double) indirect, calls
608 	 * to bmap on these blocks will fail.  However, we already have
609 	 * the on disk address, so we have to set the b_blkno field
610 	 * explicitly instead of letting bread do everything for us.
611 	 */
612 	vp = ITOV(ip);
613 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
614 	if ((bp->b_flags & B_CACHE) == 0) {
615 		curthread->td_ru.ru_inblock++;	/* pay for read */
616 		bp->b_iocmd = BIO_READ;
617 		bp->b_flags &= ~B_INVAL;
618 		bp->b_ioflags &= ~BIO_ERROR;
619 		if (bp->b_bcount > bp->b_bufsize)
620 			panic("ffs_indirtrunc: bad buffer size");
621 		bp->b_blkno = dbn;
622 		vfs_busy_pages(bp, 0);
623 		bp->b_iooffset = dbtob(bp->b_blkno);
624 		bstrategy(bp);
625 		error = bufwait(bp);
626 	}
627 	if (error) {
628 		brelse(bp);
629 		*countp = 0;
630 		return (error);
631 	}
632 
633 	if (ip->i_ump->um_fstype == UFS1)
634 		bap1 = (ufs1_daddr_t *)bp->b_data;
635 	else
636 		bap2 = (ufs2_daddr_t *)bp->b_data;
637 	if (lastbn != -1) {
638 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
639 		bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
640 		for (i = last + 1; i < NINDIR(fs); i++)
641 			if (ip->i_ump->um_fstype == UFS1)
642 				bap1[i] = 0;
643 			else
644 				bap2[i] = 0;
645 		if (DOINGASYNC(vp)) {
646 			bdwrite(bp);
647 		} else {
648 			error = bwrite(bp);
649 			if (error)
650 				allerror = error;
651 		}
652 		if (ip->i_ump->um_fstype == UFS1)
653 			bap1 = (ufs1_daddr_t *)copy;
654 		else
655 			bap2 = (ufs2_daddr_t *)copy;
656 	}
657 
658 	/*
659 	 * Recursively free totally unused blocks.
660 	 */
661 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
662 	    i--, nlbn += factor) {
663 		nb = BAP(ip, i);
664 		if (nb == 0)
665 			continue;
666 		if (level > SINGLE) {
667 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
668 			    (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
669 				allerror = error;
670 			blocksreleased += blkcount;
671 		}
672 		ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
673 		    ip->i_number, vp->v_type, NULL);
674 		blocksreleased += nblocks;
675 	}
676 
677 	/*
678 	 * Recursively free last partial block.
679 	 */
680 	if (level > SINGLE && lastbn >= 0) {
681 		last = lastbn % factor;
682 		nb = BAP(ip, i);
683 		if (nb != 0) {
684 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
685 			    last, level - 1, &blkcount);
686 			if (error)
687 				allerror = error;
688 			blocksreleased += blkcount;
689 		}
690 	}
691 	if (copy != NULL) {
692 		free(copy, M_TEMP);
693 	} else {
694 		bp->b_flags |= B_INVAL | B_NOCACHE;
695 		brelse(bp);
696 	}
697 
698 	*countp = blocksreleased;
699 	return (allerror);
700 }
701 
702 int
703 ffs_rdonly(struct inode *ip)
704 {
705 
706 	return (ip->i_ump->um_fs->fs_ronly != 0);
707 }
708 
709