xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: lfs_inode.c,v 1.122 2010/02/16 23:20:30 mlelstv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
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  * Copyright (c) 1986, 1989, 1991, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
60  */
61 
62 #include <sys/cdefs.h>
63 __KERNEL_RCSID(0, "$NetBSD: lfs_inode.c,v 1.122 2010/02/16 23:20:30 mlelstv Exp $");
64 
65 #if defined(_KERNEL_OPT)
66 #include "opt_quota.h"
67 #endif
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/mount.h>
72 #include <sys/malloc.h>
73 #include <sys/proc.h>
74 #include <sys/file.h>
75 #include <sys/buf.h>
76 #include <sys/vnode.h>
77 #include <sys/kernel.h>
78 #include <sys/trace.h>
79 #include <sys/resourcevar.h>
80 #include <sys/kauth.h>
81 
82 #include <ufs/ufs/quota.h>
83 #include <ufs/ufs/inode.h>
84 #include <ufs/ufs/ufsmount.h>
85 #include <ufs/ufs/ufs_extern.h>
86 
87 #include <ufs/lfs/lfs.h>
88 #include <ufs/lfs/lfs_extern.h>
89 
90 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t);
91 static int lfs_indirtrunc (struct inode *, daddr_t, daddr_t,
92 			   daddr_t, int, long *, long *, long *, size_t *);
93 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *);
94 static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int);
95 
96 /* Search a block for a specific dinode. */
97 struct ufs1_dinode *
98 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
99 {
100 	struct ufs1_dinode *dip = (struct ufs1_dinode *)bp->b_data;
101 	struct ufs1_dinode *ldip, *fin;
102 
103 	ASSERT_NO_SEGLOCK(fs);
104 	/*
105 	 * Read the inode block backwards, since later versions of the
106 	 * inode will supercede earlier ones.  Though it is unlikely, it is
107 	 * possible that the same inode will appear in the same inode block.
108 	 */
109 	fin = dip + INOPB(fs);
110 	for (ldip = fin - 1; ldip >= dip; --ldip)
111 		if (ldip->di_inumber == ino)
112 			return (ldip);
113 
114 	printf("searched %d entries\n", (int)(fin - dip));
115 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
116 	       dtosn(fs, fs->lfs_offset));
117 	printf("block is 0x%llx (seg %lld)\n",
118 	       (unsigned long long)dbtofsb(fs, bp->b_blkno),
119 	       (long long)dtosn(fs, dbtofsb(fs, bp->b_blkno)));
120 
121 	return NULL;
122 }
123 
124 int
125 lfs_update(struct vnode *vp, const struct timespec *acc,
126     const struct timespec *mod, int updflags)
127 {
128 	struct inode *ip;
129 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
130 	int flags;
131 
132 	ASSERT_NO_SEGLOCK(fs);
133 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
134 		return (0);
135 	ip = VTOI(vp);
136 
137 	/*
138 	 * If we are called from vinvalbuf, and the file's blocks have
139 	 * already been scheduled for writing, but the writes have not
140 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
141 	 * will cause a panic.	So, we must wait until any pending write
142 	 * for our inode completes, if we are called with UPDATE_WAIT set.
143 	 */
144 	mutex_enter(&vp->v_interlock);
145 	while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
146 	    WRITEINPROG(vp)) {
147 		DLOG((DLOG_SEG, "lfs_update: sleeping on ino %d"
148 		      " (in progress)\n", ip->i_number));
149 		cv_wait(&vp->v_cv, &vp->v_interlock);
150 	}
151 	mutex_exit(&vp->v_interlock);
152 	LFS_ITIMES(ip, acc, mod, NULL);
153 	if (updflags & UPDATE_CLOSE)
154 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING);
155 	else
156 		flags = ip->i_flag & (IN_MODIFIED | IN_CLEANING);
157 	if (flags == 0)
158 		return (0);
159 
160 	/* If sync, push back the vnode and any dirty blocks it may have. */
161 	if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) {
162 		/* Avoid flushing VU_DIROP. */
163 		mutex_enter(&lfs_lock);
164 		++fs->lfs_diropwait;
165 		while (vp->v_uflag & VU_DIROP) {
166 			DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %d"
167 			      " (dirops)\n", ip->i_number));
168 			DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags"
169 			      " 0x%x\n",
170 			      vp->v_iflag | vp->v_vflag | vp->v_uflag,
171 			      ip->i_flag));
172 			if (fs->lfs_dirops == 0)
173 				lfs_flush_fs(fs, SEGM_SYNC);
174 			else
175 				mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
176 					0, &lfs_lock);
177 			/* XXX KS - by falling out here, are we writing the vn
178 			twice? */
179 		}
180 		--fs->lfs_diropwait;
181 		mutex_exit(&lfs_lock);
182 		return lfs_vflush(vp);
183 	}
184 	return 0;
185 }
186 
187 #define	SINGLE	0	/* index of single indirect block */
188 #define	DOUBLE	1	/* index of double indirect block */
189 #define	TRIPLE	2	/* index of triple indirect block */
190 /*
191  * Truncate the inode oip to at most length size, freeing the
192  * disk blocks.
193  */
194 /* VOP_BWRITE 1 + NIADDR + lfs_balloc == 2 + 2*NIADDR times */
195 
196 int
197 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
198 {
199 	daddr_t lastblock;
200 	struct inode *oip = VTOI(ovp);
201 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
202 	/* XXX ondisk32 */
203 	int32_t newblks[NDADDR + NIADDR];
204 	struct lfs *fs;
205 	struct buf *bp;
206 	int offset, size, level;
207 	long count, rcount, blocksreleased = 0, real_released = 0;
208 	int i, nblocks;
209 	int aflags, error, allerror = 0;
210 	off_t osize;
211 	long lastseg;
212 	size_t bc;
213 	int obufsize, odb;
214 	int usepc;
215 	struct ufsmount *ump = oip->i_ump;
216 
217 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
218 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
219 		KASSERT(oip->i_size == 0);
220 		return 0;
221 	}
222 
223 	if (length < 0)
224 		return (EINVAL);
225 
226 	/*
227 	 * Just return and not update modification times.
228 	 */
229 	if (oip->i_size == length) {
230 		/* still do a uvm_vnp_setsize() as writesize may be larger */
231 		uvm_vnp_setsize(ovp, length);
232 		return (0);
233 	}
234 
235 	if (ovp->v_type == VLNK &&
236 	    (oip->i_size < ump->um_maxsymlinklen ||
237 	     (ump->um_maxsymlinklen == 0 &&
238 	      oip->i_ffs1_blocks == 0))) {
239 #ifdef DIAGNOSTIC
240 		if (length != 0)
241 			panic("lfs_truncate: partial truncate of symlink");
242 #endif
243 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
244 		oip->i_size = oip->i_ffs1_size = 0;
245 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
246 		return (lfs_update(ovp, NULL, NULL, 0));
247 	}
248 	if (oip->i_size == length) {
249 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
250 		return (lfs_update(ovp, NULL, NULL, 0));
251 	}
252 	fs = oip->i_lfs;
253 	lfs_imtime(fs);
254 	osize = oip->i_size;
255 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
256 
257 	ASSERT_NO_SEGLOCK(fs);
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 	if (osize < length) {
264 		if (length > ump->um_maxfilesize)
265 			return (EFBIG);
266 		aflags = B_CLRBUF;
267 		if (ioflag & IO_SYNC)
268 			aflags |= B_SYNC;
269 		if (usepc) {
270 			if (lblkno(fs, osize) < NDADDR &&
271 			    lblkno(fs, osize) != lblkno(fs, length) &&
272 			    blkroundup(fs, osize) != osize) {
273 				off_t eob;
274 
275 				eob = blkroundup(fs, osize);
276 				uvm_vnp_setwritesize(ovp, eob);
277 				error = ufs_balloc_range(ovp, osize,
278 				    eob - osize, cred, aflags);
279 				if (error)
280 					return error;
281 				if (ioflag & IO_SYNC) {
282 					mutex_enter(&ovp->v_interlock);
283 					VOP_PUTPAGES(ovp,
284 					    trunc_page(osize & fs->lfs_bmask),
285 					    round_page(eob),
286 					    PGO_CLEANIT | PGO_SYNCIO);
287 				}
288 			}
289 			uvm_vnp_setwritesize(ovp, length);
290 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
291 						 aflags);
292 			if (error) {
293 				(void) lfs_truncate(ovp, osize,
294 						    ioflag & IO_SYNC, cred);
295 				return error;
296 			}
297 			uvm_vnp_setsize(ovp, length);
298 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
299 			KASSERT(ovp->v_size == oip->i_size);
300 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
301 			return (lfs_update(ovp, NULL, NULL, 0));
302 		} else {
303 			error = lfs_reserve(fs, ovp, NULL,
304 			    btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
305 			if (error)
306 				return (error);
307 			error = lfs_balloc(ovp, length - 1, 1, cred,
308 					   aflags, &bp);
309 			lfs_reserve(fs, ovp, NULL,
310 			    -btofsb(fs, (NIADDR + 2) << fs->lfs_bshift));
311 			if (error)
312 				return (error);
313 			oip->i_ffs1_size = oip->i_size = length;
314 			uvm_vnp_setsize(ovp, length);
315 			(void) VOP_BWRITE(bp);
316 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
317 			oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
318 			return (lfs_update(ovp, NULL, NULL, 0));
319 		}
320 	}
321 
322 	if ((error = lfs_reserve(fs, ovp, NULL,
323 	    btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift))) != 0)
324 		return (error);
325 
326 	/*
327 	 * Shorten the size of the file. If the file is not being
328 	 * truncated to a block boundary, the contents of the
329 	 * partial block following the end of the file must be
330 	 * zero'ed in case it ever becomes accessible again because
331 	 * of subsequent file growth. Directories however are not
332 	 * zero'ed as they should grow back initialized to empty.
333 	 */
334 	offset = blkoff(fs, length);
335 	lastseg = -1;
336 	bc = 0;
337 
338 	if (ovp != fs->lfs_ivnode)
339 		lfs_seglock(fs, SEGM_PROT);
340 	if (offset == 0) {
341 		oip->i_size = oip->i_ffs1_size = length;
342 	} else if (!usepc) {
343 		lbn = lblkno(fs, length);
344 		aflags = B_CLRBUF;
345 		if (ioflag & IO_SYNC)
346 			aflags |= B_SYNC;
347 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
348 		if (error) {
349 			lfs_reserve(fs, ovp, NULL,
350 			    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
351 			goto errout;
352 		}
353 		obufsize = bp->b_bufsize;
354 		odb = btofsb(fs, bp->b_bcount);
355 		oip->i_size = oip->i_ffs1_size = length;
356 		size = blksize(fs, oip, lbn);
357 		if (ovp->v_type != VDIR)
358 			memset((char *)bp->b_data + offset, 0,
359 			       (u_int)(size - offset));
360 		allocbuf(bp, size, 1);
361 		if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
362 			mutex_enter(&lfs_lock);
363 			locked_queue_bytes -= obufsize - bp->b_bufsize;
364 			mutex_exit(&lfs_lock);
365 		}
366 		if (bp->b_oflags & BO_DELWRI)
367 			fs->lfs_avail += odb - btofsb(fs, size);
368 		(void) VOP_BWRITE(bp);
369 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
370 		/*
371 		 * When truncating a regular file down to a non-block-aligned
372 		 * size, we must zero the part of last block which is past
373 		 * the new EOF.  We must synchronously flush the zeroed pages
374 		 * to disk since the new pages will be invalidated as soon
375 		 * as we inform the VM system of the new, smaller size.
376 		 * We must do this before acquiring the GLOCK, since fetching
377 		 * the pages will acquire the GLOCK internally.
378 		 * So there is a window where another thread could see a whole
379 		 * zeroed page past EOF, but that's life.
380 		 */
381 		daddr_t xlbn;
382 		voff_t eoz;
383 
384 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
385 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflags);
386 		if (error) {
387 			lfs_reserve(fs, ovp, NULL,
388 				    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
389 			goto errout;
390 		}
391 		xlbn = lblkno(fs, length);
392 		size = blksize(fs, oip, xlbn);
393 		eoz = MIN(lblktosize(fs, xlbn) + size, osize);
394 		uvm_vnp_zerorange(ovp, length, eoz - length);
395 		if (round_page(eoz) > round_page(length)) {
396 			mutex_enter(&ovp->v_interlock);
397 			error = VOP_PUTPAGES(ovp, round_page(length),
398 			    round_page(eoz),
399 			    PGO_CLEANIT | PGO_DEACTIVATE |
400 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
401 			if (error) {
402 				lfs_reserve(fs, ovp, NULL,
403 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
404 				goto errout;
405 			}
406 		}
407 	}
408 
409 	genfs_node_wrlock(ovp);
410 
411 	oip->i_size = oip->i_ffs1_size = length;
412 	uvm_vnp_setsize(ovp, length);
413 
414 	/*
415 	 * Calculate index into inode's block list of
416 	 * last direct and indirect blocks (if any)
417 	 * which we want to keep.  Lastblock is -1 when
418 	 * the file is truncated to 0.
419 	 */
420 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
421 	if (length > QUAD_MAX - fs->lfs_bsize)
422 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
423 	else
424 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
425 	lastiblock[SINGLE] = lastblock - NDADDR;
426 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
427 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
428 	nblocks = btofsb(fs, fs->lfs_bsize);
429 	/*
430 	 * Record changed file and block pointers before we start
431 	 * freeing blocks.  lastiblock values are also normalized to -1
432 	 * for calls to lfs_indirtrunc below.
433 	 */
434 	memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
435 	for (level = TRIPLE; level >= SINGLE; level--)
436 		if (lastiblock[level] < 0) {
437 			newblks[NDADDR+level] = 0;
438 			lastiblock[level] = -1;
439 		}
440 	for (i = NDADDR - 1; i > lastblock; i--)
441 		newblks[i] = 0;
442 
443 	oip->i_size = oip->i_ffs1_size = osize;
444 	error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
445 	if (error && !allerror)
446 		allerror = error;
447 
448 	/*
449 	 * Indirect blocks first.
450 	 */
451 	indir_lbn[SINGLE] = -NDADDR;
452 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
453 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
454 	for (level = TRIPLE; level >= SINGLE; level--) {
455 		bn = oip->i_ffs1_ib[level];
456 		if (bn != 0) {
457 			error = lfs_indirtrunc(oip, indir_lbn[level],
458 					       bn, lastiblock[level],
459 					       level, &count, &rcount,
460 					       &lastseg, &bc);
461 			if (error)
462 				allerror = error;
463 			real_released += rcount;
464 			blocksreleased += count;
465 			if (lastiblock[level] < 0) {
466 				if (oip->i_ffs1_ib[level] > 0)
467 					real_released += nblocks;
468 				blocksreleased += nblocks;
469 				oip->i_ffs1_ib[level] = 0;
470 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
471 					    &lastseg, &bc);
472         			lfs_deregister_block(ovp, bn);
473 			}
474 		}
475 		if (lastiblock[level] >= 0)
476 			goto done;
477 	}
478 
479 	/*
480 	 * All whole direct blocks or frags.
481 	 */
482 	for (i = NDADDR - 1; i > lastblock; i--) {
483 		long bsize, obsize;
484 
485 		bn = oip->i_ffs1_db[i];
486 		if (bn == 0)
487 			continue;
488 		bsize = blksize(fs, oip, i);
489 		if (oip->i_ffs1_db[i] > 0) {
490 			/* Check for fragment size changes */
491 			obsize = oip->i_lfs_fragsize[i];
492 			real_released += btofsb(fs, obsize);
493 			oip->i_lfs_fragsize[i] = 0;
494 		} else
495 			obsize = 0;
496 		blocksreleased += btofsb(fs, bsize);
497 		oip->i_ffs1_db[i] = 0;
498 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
499         	lfs_deregister_block(ovp, bn);
500 	}
501 	if (lastblock < 0)
502 		goto done;
503 
504 	/*
505 	 * Finally, look for a change in size of the
506 	 * last direct block; release any frags.
507 	 */
508 	bn = oip->i_ffs1_db[lastblock];
509 	if (bn != 0) {
510 		long oldspace, newspace;
511 #if 0
512 		long olddspace;
513 #endif
514 
515 		/*
516 		 * Calculate amount of space we're giving
517 		 * back as old block size minus new block size.
518 		 */
519 		oldspace = blksize(fs, oip, lastblock);
520 #if 0
521 		olddspace = oip->i_lfs_fragsize[lastblock];
522 #endif
523 
524 		oip->i_size = oip->i_ffs1_size = length;
525 		newspace = blksize(fs, oip, lastblock);
526 		if (newspace == 0)
527 			panic("itrunc: newspace");
528 		if (oldspace - newspace > 0) {
529 			blocksreleased += btofsb(fs, oldspace - newspace);
530 		}
531 #if 0
532 		if (bn > 0 && olddspace - newspace > 0) {
533 			/* No segment accounting here, just vnode */
534 			real_released += btofsb(fs, olddspace - newspace);
535 		}
536 #endif
537 	}
538 
539 done:
540 	/* Finish segment accounting corrections */
541 	lfs_update_seguse(fs, oip, lastseg, bc);
542 #ifdef DIAGNOSTIC
543 	for (level = SINGLE; level <= TRIPLE; level++)
544 		if ((newblks[NDADDR + level] == 0) !=
545 		    ((oip->i_ffs1_ib[level]) == 0)) {
546 			panic("lfs itrunc1");
547 		}
548 	for (i = 0; i < NDADDR; i++)
549 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
550 			panic("lfs itrunc2");
551 		}
552 	if (length == 0 &&
553 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
554 		panic("lfs itrunc3");
555 #endif /* DIAGNOSTIC */
556 	/*
557 	 * Put back the real size.
558 	 */
559 	oip->i_size = oip->i_ffs1_size = length;
560 	oip->i_lfs_effnblks -= blocksreleased;
561 	oip->i_ffs1_blocks -= real_released;
562 	mutex_enter(&lfs_lock);
563 	fs->lfs_bfree += blocksreleased;
564 	mutex_exit(&lfs_lock);
565 #ifdef DIAGNOSTIC
566 	if (oip->i_size == 0 &&
567 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
568 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
569 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
570 		panic("lfs_truncate: persistent blocks");
571 	}
572 #endif
573 
574 	/*
575 	 * If we truncated to zero, take us off the paging queue.
576 	 */
577 	mutex_enter(&lfs_lock);
578 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
579 		oip->i_flags &= ~IN_PAGING;
580 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
581 	}
582 	mutex_exit(&lfs_lock);
583 
584 	oip->i_flag |= IN_CHANGE;
585 #ifdef QUOTA
586 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
587 #endif
588 	lfs_reserve(fs, ovp, NULL,
589 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
590 	genfs_node_unlock(ovp);
591   errout:
592 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
593 	if (ovp != fs->lfs_ivnode)
594 		lfs_segunlock(fs);
595 	return (allerror ? allerror : error);
596 }
597 
598 /* Update segment and avail usage information when removing a block. */
599 static int
600 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
601 	    size_t bsize, long *lastseg, size_t *num)
602 {
603 	long seg;
604 	int error = 0;
605 
606 	ASSERT_SEGLOCK(fs);
607 	bsize = fragroundup(fs, bsize);
608 	if (daddr > 0) {
609 		if (*lastseg != (seg = dtosn(fs, daddr))) {
610 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
611 			*num = bsize;
612 			*lastseg = seg;
613 		} else
614 			*num += bsize;
615 	}
616 
617 	return error;
618 }
619 
620 /* Finish the accounting updates for a segment. */
621 static int
622 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
623 {
624 	struct segdelta *sd;
625 	struct vnode *vp;
626 
627 	ASSERT_SEGLOCK(fs);
628 	if (lastseg < 0 || num == 0)
629 		return 0;
630 
631 	vp = ITOV(ip);
632 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
633 		if (sd->segnum == lastseg)
634 			break;
635 	if (sd == NULL) {
636 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
637 		sd->segnum = lastseg;
638 		sd->num = 0;
639 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
640 	}
641 	sd->num += num;
642 
643 	return 0;
644 }
645 
646 static void
647 lfs_finalize_seguse(struct lfs *fs, void *v)
648 {
649 	SEGUSE *sup;
650 	struct buf *bp;
651 	struct segdelta *sd;
652 	LIST_HEAD(, segdelta) *hd = v;
653 
654 	ASSERT_SEGLOCK(fs);
655 	while((sd = LIST_FIRST(hd)) != NULL) {
656 		LIST_REMOVE(sd, list);
657 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
658 		if (sd->num > sup->su_nbytes) {
659 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
660 				sd->segnum, (long)(sd->num - sup->su_nbytes));
661 			panic("lfs_finalize_seguse: negative bytes");
662 			sup->su_nbytes = sd->num;
663 		}
664 		sup->su_nbytes -= sd->num;
665 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
666 		free(sd, M_SEGMENT);
667 	}
668 }
669 
670 /* Finish the accounting updates for a segment. */
671 void
672 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
673 {
674 	ASSERT_SEGLOCK(fs);
675 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
676 }
677 
678 /* Finish the accounting updates for a segment. */
679 void
680 lfs_finalize_fs_seguse(struct lfs *fs)
681 {
682 	ASSERT_SEGLOCK(fs);
683 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
684 }
685 
686 /*
687  * Release blocks associated with the inode ip and stored in the indirect
688  * block bn.  Blocks are free'd in LIFO order up to (but not including)
689  * lastbn.  If level is greater than SINGLE, the block is an indirect block
690  * and recursive calls to indirtrunc must be used to cleanse other indirect
691  * blocks.
692  *
693  * NB: triple indirect blocks are untested.
694  */
695 static int
696 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
697 	       daddr_t lastbn, int level, long *countp,
698 	       long *rcountp, long *lastsegp, size_t *bcp)
699 {
700 	int i;
701 	struct buf *bp;
702 	struct lfs *fs = ip->i_lfs;
703 	int32_t *bap;	/* XXX ondisk32 */
704 	struct vnode *vp;
705 	daddr_t nb, nlbn, last;
706 	int32_t *copy = NULL;	/* XXX ondisk32 */
707 	long blkcount, rblkcount, factor;
708 	int nblocks, blocksreleased = 0, real_released = 0;
709 	int error = 0, allerror = 0;
710 
711 	ASSERT_SEGLOCK(fs);
712 	/*
713 	 * Calculate index in current block of last
714 	 * block to be kept.  -1 indicates the entire
715 	 * block so we need not calculate the index.
716 	 */
717 	factor = 1;
718 	for (i = SINGLE; i < level; i++)
719 		factor *= NINDIR(fs);
720 	last = lastbn;
721 	if (lastbn > 0)
722 		last /= factor;
723 	nblocks = btofsb(fs, fs->lfs_bsize);
724 	/*
725 	 * Get buffer of block pointers, zero those entries corresponding
726 	 * to blocks to be free'd, and update on disk copy first.  Since
727 	 * double(triple) indirect before single(double) indirect, calls
728 	 * to bmap on these blocks will fail.  However, we already have
729 	 * the on disk address, so we have to set the b_blkno field
730 	 * explicitly instead of letting bread do everything for us.
731 	 */
732 	vp = ITOV(ip);
733 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
734 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
735 		/* Braces must be here in case trace evaluates to nothing. */
736 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
737 	} else {
738 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
739 		curlwp->l_ru.ru_inblock++; /* pay for read */
740 		bp->b_flags |= B_READ;
741 		if (bp->b_bcount > bp->b_bufsize)
742 			panic("lfs_indirtrunc: bad buffer size");
743 		bp->b_blkno = fsbtodb(fs, dbn);
744 		VOP_STRATEGY(vp, bp);
745 		error = biowait(bp);
746 	}
747 	if (error) {
748 		brelse(bp, 0);
749 		*countp = *rcountp = 0;
750 		return (error);
751 	}
752 
753 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
754 	if (lastbn >= 0) {
755 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
756 		memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize);
757 		memset((void *)&bap[last + 1], 0,
758 		/* XXX ondisk32 */
759 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
760 		error = VOP_BWRITE(bp);
761 		if (error)
762 			allerror = error;
763 		bap = copy;
764 	}
765 
766 	/*
767 	 * Recursively free totally unused blocks.
768 	 */
769 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
770 	    i--, nlbn += factor) {
771 		nb = bap[i];
772 		if (nb == 0)
773 			continue;
774 		if (level > SINGLE) {
775 			error = lfs_indirtrunc(ip, nlbn, nb,
776 					       (daddr_t)-1, level - 1,
777 					       &blkcount, &rblkcount,
778 					       lastsegp, bcp);
779 			if (error)
780 				allerror = error;
781 			blocksreleased += blkcount;
782 			real_released += rblkcount;
783 		}
784 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
785 		if (bap[i] > 0)
786 			real_released += nblocks;
787 		blocksreleased += nblocks;
788 	}
789 
790 	/*
791 	 * Recursively free last partial block.
792 	 */
793 	if (level > SINGLE && lastbn >= 0) {
794 		last = lastbn % factor;
795 		nb = bap[i];
796 		if (nb != 0) {
797 			error = lfs_indirtrunc(ip, nlbn, nb,
798 					       last, level - 1, &blkcount,
799 					       &rblkcount, lastsegp, bcp);
800 			if (error)
801 				allerror = error;
802 			real_released += rblkcount;
803 			blocksreleased += blkcount;
804 		}
805 	}
806 
807 	if (copy != NULL) {
808 		lfs_free(fs, copy, LFS_NB_IBLOCK);
809 	} else {
810 		mutex_enter(&bufcache_lock);
811 		if (bp->b_oflags & BO_DELWRI) {
812 			LFS_UNLOCK_BUF(bp);
813 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
814 			wakeup(&fs->lfs_avail);
815 		}
816 		brelsel(bp, BC_INVAL);
817 		mutex_exit(&bufcache_lock);
818 	}
819 
820 	*countp = blocksreleased;
821 	*rcountp = real_released;
822 	return (allerror);
823 }
824 
825 /*
826  * Destroy any in core blocks past the truncation length.
827  * Inlined from vtruncbuf, so that lfs_avail could be updated.
828  * We take the seglock to prevent cleaning from occurring while we are
829  * invalidating blocks.
830  */
831 static int
832 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
833 {
834 	struct buf *bp, *nbp;
835 	int error;
836 	struct lfs *fs;
837 	voff_t off;
838 
839 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
840 	mutex_enter(&vp->v_interlock);
841 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
842 	if (error)
843 		return error;
844 
845 	fs = VTOI(vp)->i_lfs;
846 
847 	ASSERT_SEGLOCK(fs);
848 
849 	mutex_enter(&bufcache_lock);
850 restart:
851 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
852 		nbp = LIST_NEXT(bp, b_vnbufs);
853 		if (bp->b_lblkno < lbn)
854 			continue;
855 		error = bbusy(bp, catch, slptimeo, NULL);
856 		if (error == EPASSTHROUGH)
857 			goto restart;
858 		if (error != 0) {
859 			mutex_exit(&bufcache_lock);
860 			return (error);
861 		}
862 		mutex_enter(bp->b_objlock);
863 		if (bp->b_oflags & BO_DELWRI) {
864 			bp->b_oflags &= ~BO_DELWRI;
865 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
866 			wakeup(&fs->lfs_avail);
867 		}
868 		mutex_exit(bp->b_objlock);
869 		LFS_UNLOCK_BUF(bp);
870 		brelsel(bp, BC_INVAL | BC_VFLUSH);
871 	}
872 
873 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
874 		nbp = LIST_NEXT(bp, b_vnbufs);
875 		if (bp->b_lblkno < lbn)
876 			continue;
877 		error = bbusy(bp, catch, slptimeo, NULL);
878 		if (error == EPASSTHROUGH)
879 			goto restart;
880 		if (error != 0) {
881 			mutex_exit(&bufcache_lock);
882 			return (error);
883 		}
884 		mutex_enter(bp->b_objlock);
885 		if (bp->b_oflags & BO_DELWRI) {
886 			bp->b_oflags &= ~BO_DELWRI;
887 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
888 			wakeup(&fs->lfs_avail);
889 		}
890 		mutex_exit(bp->b_objlock);
891 		LFS_UNLOCK_BUF(bp);
892 		brelsel(bp, BC_INVAL | BC_VFLUSH);
893 	}
894 	mutex_exit(&bufcache_lock);
895 
896 	return (0);
897 }
898 
899