xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision 1b9578b8c2c1f848eeb16dabbfd7d1f0d9fdefbd)
1 /*	$NetBSD: lfs_inode.c,v 1.124 2011/06/16 09:21:03 hannken 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 1.124 2011/06/16 09:21:03 hannken 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 		ubc_zerorange(&ovp->v_uobj, length, eoz - length,
395 		    UBC_UNMAP_FLAG(ovp));
396 		if (round_page(eoz) > round_page(length)) {
397 			mutex_enter(ovp->v_interlock);
398 			error = VOP_PUTPAGES(ovp, round_page(length),
399 			    round_page(eoz),
400 			    PGO_CLEANIT | PGO_DEACTIVATE |
401 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
402 			if (error) {
403 				lfs_reserve(fs, ovp, NULL,
404 					    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
405 				goto errout;
406 			}
407 		}
408 	}
409 
410 	genfs_node_wrlock(ovp);
411 
412 	oip->i_size = oip->i_ffs1_size = length;
413 	uvm_vnp_setsize(ovp, length);
414 
415 	/*
416 	 * Calculate index into inode's block list of
417 	 * last direct and indirect blocks (if any)
418 	 * which we want to keep.  Lastblock is -1 when
419 	 * the file is truncated to 0.
420 	 */
421 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
422 	if (length > QUAD_MAX - fs->lfs_bsize)
423 		lastblock = lblkno(fs, QUAD_MAX - fs->lfs_bsize);
424 	else
425 		lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
426 	lastiblock[SINGLE] = lastblock - NDADDR;
427 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
428 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
429 	nblocks = btofsb(fs, fs->lfs_bsize);
430 	/*
431 	 * Record changed file and block pointers before we start
432 	 * freeing blocks.  lastiblock values are also normalized to -1
433 	 * for calls to lfs_indirtrunc below.
434 	 */
435 	memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
436 	for (level = TRIPLE; level >= SINGLE; level--)
437 		if (lastiblock[level] < 0) {
438 			newblks[NDADDR+level] = 0;
439 			lastiblock[level] = -1;
440 		}
441 	for (i = NDADDR - 1; i > lastblock; i--)
442 		newblks[i] = 0;
443 
444 	oip->i_size = oip->i_ffs1_size = osize;
445 	error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
446 	if (error && !allerror)
447 		allerror = error;
448 
449 	/*
450 	 * Indirect blocks first.
451 	 */
452 	indir_lbn[SINGLE] = -NDADDR;
453 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
454 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
455 	for (level = TRIPLE; level >= SINGLE; level--) {
456 		bn = oip->i_ffs1_ib[level];
457 		if (bn != 0) {
458 			error = lfs_indirtrunc(oip, indir_lbn[level],
459 					       bn, lastiblock[level],
460 					       level, &count, &rcount,
461 					       &lastseg, &bc);
462 			if (error)
463 				allerror = error;
464 			real_released += rcount;
465 			blocksreleased += count;
466 			if (lastiblock[level] < 0) {
467 				if (oip->i_ffs1_ib[level] > 0)
468 					real_released += nblocks;
469 				blocksreleased += nblocks;
470 				oip->i_ffs1_ib[level] = 0;
471 				lfs_blkfree(fs, oip, bn, fs->lfs_bsize,
472 					    &lastseg, &bc);
473         			lfs_deregister_block(ovp, bn);
474 			}
475 		}
476 		if (lastiblock[level] >= 0)
477 			goto done;
478 	}
479 
480 	/*
481 	 * All whole direct blocks or frags.
482 	 */
483 	for (i = NDADDR - 1; i > lastblock; i--) {
484 		long bsize, obsize;
485 
486 		bn = oip->i_ffs1_db[i];
487 		if (bn == 0)
488 			continue;
489 		bsize = blksize(fs, oip, i);
490 		if (oip->i_ffs1_db[i] > 0) {
491 			/* Check for fragment size changes */
492 			obsize = oip->i_lfs_fragsize[i];
493 			real_released += btofsb(fs, obsize);
494 			oip->i_lfs_fragsize[i] = 0;
495 		} else
496 			obsize = 0;
497 		blocksreleased += btofsb(fs, bsize);
498 		oip->i_ffs1_db[i] = 0;
499 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
500         	lfs_deregister_block(ovp, bn);
501 	}
502 	if (lastblock < 0)
503 		goto done;
504 
505 	/*
506 	 * Finally, look for a change in size of the
507 	 * last direct block; release any frags.
508 	 */
509 	bn = oip->i_ffs1_db[lastblock];
510 	if (bn != 0) {
511 		long oldspace, newspace;
512 #if 0
513 		long olddspace;
514 #endif
515 
516 		/*
517 		 * Calculate amount of space we're giving
518 		 * back as old block size minus new block size.
519 		 */
520 		oldspace = blksize(fs, oip, lastblock);
521 #if 0
522 		olddspace = oip->i_lfs_fragsize[lastblock];
523 #endif
524 
525 		oip->i_size = oip->i_ffs1_size = length;
526 		newspace = blksize(fs, oip, lastblock);
527 		if (newspace == 0)
528 			panic("itrunc: newspace");
529 		if (oldspace - newspace > 0) {
530 			blocksreleased += btofsb(fs, oldspace - newspace);
531 		}
532 #if 0
533 		if (bn > 0 && olddspace - newspace > 0) {
534 			/* No segment accounting here, just vnode */
535 			real_released += btofsb(fs, olddspace - newspace);
536 		}
537 #endif
538 	}
539 
540 done:
541 	/* Finish segment accounting corrections */
542 	lfs_update_seguse(fs, oip, lastseg, bc);
543 #ifdef DIAGNOSTIC
544 	for (level = SINGLE; level <= TRIPLE; level++)
545 		if ((newblks[NDADDR + level] == 0) !=
546 		    ((oip->i_ffs1_ib[level]) == 0)) {
547 			panic("lfs itrunc1");
548 		}
549 	for (i = 0; i < NDADDR; i++)
550 		if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
551 			panic("lfs itrunc2");
552 		}
553 	if (length == 0 &&
554 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
555 		panic("lfs itrunc3");
556 #endif /* DIAGNOSTIC */
557 	/*
558 	 * Put back the real size.
559 	 */
560 	oip->i_size = oip->i_ffs1_size = length;
561 	oip->i_lfs_effnblks -= blocksreleased;
562 	oip->i_ffs1_blocks -= real_released;
563 	mutex_enter(&lfs_lock);
564 	fs->lfs_bfree += blocksreleased;
565 	mutex_exit(&lfs_lock);
566 #ifdef DIAGNOSTIC
567 	if (oip->i_size == 0 &&
568 	    (oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
569 		printf("lfs_truncate: truncate to 0 but %d blks/%d effblks\n",
570 		       oip->i_ffs1_blocks, oip->i_lfs_effnblks);
571 		panic("lfs_truncate: persistent blocks");
572 	}
573 #endif
574 
575 	/*
576 	 * If we truncated to zero, take us off the paging queue.
577 	 */
578 	mutex_enter(&lfs_lock);
579 	if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
580 		oip->i_flags &= ~IN_PAGING;
581 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
582 	}
583 	mutex_exit(&lfs_lock);
584 
585 	oip->i_flag |= IN_CHANGE;
586 #ifdef QUOTA
587 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
588 #endif
589 	lfs_reserve(fs, ovp, NULL,
590 	    -btofsb(fs, (2 * NIADDR + 3) << fs->lfs_bshift));
591 	genfs_node_unlock(ovp);
592   errout:
593 	oip->i_lfs_hiblk = lblkno(fs, oip->i_size + fs->lfs_bsize - 1) - 1;
594 	if (ovp != fs->lfs_ivnode)
595 		lfs_segunlock(fs);
596 	return (allerror ? allerror : error);
597 }
598 
599 /* Update segment and avail usage information when removing a block. */
600 static int
601 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
602 	    size_t bsize, long *lastseg, size_t *num)
603 {
604 	long seg;
605 	int error = 0;
606 
607 	ASSERT_SEGLOCK(fs);
608 	bsize = fragroundup(fs, bsize);
609 	if (daddr > 0) {
610 		if (*lastseg != (seg = dtosn(fs, daddr))) {
611 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
612 			*num = bsize;
613 			*lastseg = seg;
614 		} else
615 			*num += bsize;
616 	}
617 
618 	return error;
619 }
620 
621 /* Finish the accounting updates for a segment. */
622 static int
623 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
624 {
625 	struct segdelta *sd;
626 	struct vnode *vp;
627 
628 	ASSERT_SEGLOCK(fs);
629 	if (lastseg < 0 || num == 0)
630 		return 0;
631 
632 	vp = ITOV(ip);
633 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
634 		if (sd->segnum == lastseg)
635 			break;
636 	if (sd == NULL) {
637 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
638 		sd->segnum = lastseg;
639 		sd->num = 0;
640 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
641 	}
642 	sd->num += num;
643 
644 	return 0;
645 }
646 
647 static void
648 lfs_finalize_seguse(struct lfs *fs, void *v)
649 {
650 	SEGUSE *sup;
651 	struct buf *bp;
652 	struct segdelta *sd;
653 	LIST_HEAD(, segdelta) *hd = v;
654 
655 	ASSERT_SEGLOCK(fs);
656 	while((sd = LIST_FIRST(hd)) != NULL) {
657 		LIST_REMOVE(sd, list);
658 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
659 		if (sd->num > sup->su_nbytes) {
660 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
661 				sd->segnum, (long)(sd->num - sup->su_nbytes));
662 			panic("lfs_finalize_seguse: negative bytes");
663 			sup->su_nbytes = sd->num;
664 		}
665 		sup->su_nbytes -= sd->num;
666 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
667 		free(sd, M_SEGMENT);
668 	}
669 }
670 
671 /* Finish the accounting updates for a segment. */
672 void
673 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
674 {
675 	ASSERT_SEGLOCK(fs);
676 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
677 }
678 
679 /* Finish the accounting updates for a segment. */
680 void
681 lfs_finalize_fs_seguse(struct lfs *fs)
682 {
683 	ASSERT_SEGLOCK(fs);
684 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
685 }
686 
687 /*
688  * Release blocks associated with the inode ip and stored in the indirect
689  * block bn.  Blocks are free'd in LIFO order up to (but not including)
690  * lastbn.  If level is greater than SINGLE, the block is an indirect block
691  * and recursive calls to indirtrunc must be used to cleanse other indirect
692  * blocks.
693  *
694  * NB: triple indirect blocks are untested.
695  */
696 static int
697 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
698 	       daddr_t lastbn, int level, long *countp,
699 	       long *rcountp, long *lastsegp, size_t *bcp)
700 {
701 	int i;
702 	struct buf *bp;
703 	struct lfs *fs = ip->i_lfs;
704 	int32_t *bap;	/* XXX ondisk32 */
705 	struct vnode *vp;
706 	daddr_t nb, nlbn, last;
707 	int32_t *copy = NULL;	/* XXX ondisk32 */
708 	long blkcount, rblkcount, factor;
709 	int nblocks, blocksreleased = 0, real_released = 0;
710 	int error = 0, allerror = 0;
711 
712 	ASSERT_SEGLOCK(fs);
713 	/*
714 	 * Calculate index in current block of last
715 	 * block to be kept.  -1 indicates the entire
716 	 * block so we need not calculate the index.
717 	 */
718 	factor = 1;
719 	for (i = SINGLE; i < level; i++)
720 		factor *= NINDIR(fs);
721 	last = lastbn;
722 	if (lastbn > 0)
723 		last /= factor;
724 	nblocks = btofsb(fs, fs->lfs_bsize);
725 	/*
726 	 * Get buffer of block pointers, zero those entries corresponding
727 	 * to blocks to be free'd, and update on disk copy first.  Since
728 	 * double(triple) indirect before single(double) indirect, calls
729 	 * to bmap on these blocks will fail.  However, we already have
730 	 * the on disk address, so we have to set the b_blkno field
731 	 * explicitly instead of letting bread do everything for us.
732 	 */
733 	vp = ITOV(ip);
734 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
735 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
736 		/* Braces must be here in case trace evaluates to nothing. */
737 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
738 	} else {
739 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
740 		curlwp->l_ru.ru_inblock++; /* pay for read */
741 		bp->b_flags |= B_READ;
742 		if (bp->b_bcount > bp->b_bufsize)
743 			panic("lfs_indirtrunc: bad buffer size");
744 		bp->b_blkno = fsbtodb(fs, dbn);
745 		VOP_STRATEGY(vp, bp);
746 		error = biowait(bp);
747 	}
748 	if (error) {
749 		brelse(bp, 0);
750 		*countp = *rcountp = 0;
751 		return (error);
752 	}
753 
754 	bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
755 	if (lastbn >= 0) {
756 		copy = (int32_t *)lfs_malloc(fs, fs->lfs_bsize, LFS_NB_IBLOCK);
757 		memcpy((void *)copy, (void *)bap, (u_int)fs->lfs_bsize);
758 		memset((void *)&bap[last + 1], 0,
759 		/* XXX ondisk32 */
760 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (int32_t));
761 		error = VOP_BWRITE(bp);
762 		if (error)
763 			allerror = error;
764 		bap = copy;
765 	}
766 
767 	/*
768 	 * Recursively free totally unused blocks.
769 	 */
770 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
771 	    i--, nlbn += factor) {
772 		nb = bap[i];
773 		if (nb == 0)
774 			continue;
775 		if (level > SINGLE) {
776 			error = lfs_indirtrunc(ip, nlbn, nb,
777 					       (daddr_t)-1, level - 1,
778 					       &blkcount, &rblkcount,
779 					       lastsegp, bcp);
780 			if (error)
781 				allerror = error;
782 			blocksreleased += blkcount;
783 			real_released += rblkcount;
784 		}
785 		lfs_blkfree(fs, ip, nb, fs->lfs_bsize, lastsegp, bcp);
786 		if (bap[i] > 0)
787 			real_released += nblocks;
788 		blocksreleased += nblocks;
789 	}
790 
791 	/*
792 	 * Recursively free last partial block.
793 	 */
794 	if (level > SINGLE && lastbn >= 0) {
795 		last = lastbn % factor;
796 		nb = bap[i];
797 		if (nb != 0) {
798 			error = lfs_indirtrunc(ip, nlbn, nb,
799 					       last, level - 1, &blkcount,
800 					       &rblkcount, lastsegp, bcp);
801 			if (error)
802 				allerror = error;
803 			real_released += rblkcount;
804 			blocksreleased += blkcount;
805 		}
806 	}
807 
808 	if (copy != NULL) {
809 		lfs_free(fs, copy, LFS_NB_IBLOCK);
810 	} else {
811 		mutex_enter(&bufcache_lock);
812 		if (bp->b_oflags & BO_DELWRI) {
813 			LFS_UNLOCK_BUF(bp);
814 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
815 			wakeup(&fs->lfs_avail);
816 		}
817 		brelsel(bp, BC_INVAL);
818 		mutex_exit(&bufcache_lock);
819 	}
820 
821 	*countp = blocksreleased;
822 	*rcountp = real_released;
823 	return (allerror);
824 }
825 
826 /*
827  * Destroy any in core blocks past the truncation length.
828  * Inlined from vtruncbuf, so that lfs_avail could be updated.
829  * We take the seglock to prevent cleaning from occurring while we are
830  * invalidating blocks.
831  */
832 static int
833 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
834 {
835 	struct buf *bp, *nbp;
836 	int error;
837 	struct lfs *fs;
838 	voff_t off;
839 
840 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
841 	mutex_enter(vp->v_interlock);
842 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
843 	if (error)
844 		return error;
845 
846 	fs = VTOI(vp)->i_lfs;
847 
848 	ASSERT_SEGLOCK(fs);
849 
850 	mutex_enter(&bufcache_lock);
851 restart:
852 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
853 		nbp = LIST_NEXT(bp, b_vnbufs);
854 		if (bp->b_lblkno < lbn)
855 			continue;
856 		error = bbusy(bp, catch, slptimeo, NULL);
857 		if (error == EPASSTHROUGH)
858 			goto restart;
859 		if (error != 0) {
860 			mutex_exit(&bufcache_lock);
861 			return (error);
862 		}
863 		mutex_enter(bp->b_objlock);
864 		if (bp->b_oflags & BO_DELWRI) {
865 			bp->b_oflags &= ~BO_DELWRI;
866 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
867 			wakeup(&fs->lfs_avail);
868 		}
869 		mutex_exit(bp->b_objlock);
870 		LFS_UNLOCK_BUF(bp);
871 		brelsel(bp, BC_INVAL | BC_VFLUSH);
872 	}
873 
874 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
875 		nbp = LIST_NEXT(bp, b_vnbufs);
876 		if (bp->b_lblkno < lbn)
877 			continue;
878 		error = bbusy(bp, catch, slptimeo, NULL);
879 		if (error == EPASSTHROUGH)
880 			goto restart;
881 		if (error != 0) {
882 			mutex_exit(&bufcache_lock);
883 			return (error);
884 		}
885 		mutex_enter(bp->b_objlock);
886 		if (bp->b_oflags & BO_DELWRI) {
887 			bp->b_oflags &= ~BO_DELWRI;
888 			fs->lfs_avail += btofsb(fs, bp->b_bcount);
889 			wakeup(&fs->lfs_avail);
890 		}
891 		mutex_exit(bp->b_objlock);
892 		LFS_UNLOCK_BUF(bp);
893 		brelsel(bp, BC_INVAL | BC_VFLUSH);
894 	}
895 	mutex_exit(&bufcache_lock);
896 
897 	return (0);
898 }
899 
900