xref: /netbsd-src/sys/ufs/lfs/lfs_inode.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: lfs_inode.c,v 1.157 2017/06/10 05:29:36 maya 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 1.157 2017/06/10 05:29:36 maya 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/lfs/ulfs_quotacommon.h>
83 #include <ufs/lfs/ulfs_inode.h>
84 #include <ufs/lfs/ulfsmount.h>
85 #include <ufs/lfs/ulfs_extern.h>
86 
87 #include <ufs/lfs/lfs.h>
88 #include <ufs/lfs/lfs_accessors.h>
89 #include <ufs/lfs/lfs_extern.h>
90 #include <ufs/lfs/lfs_kernel.h>
91 
92 static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t);
93 static int lfs_indirtrunc(struct inode *, daddr_t, daddr_t,
94 			  daddr_t, int, daddr_t *, daddr_t *,
95 			  long *, size_t *);
96 static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *);
97 static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int);
98 
99 /* Search a block for a specific dinode. */
100 union lfs_dinode *
101 lfs_ifind(struct lfs *fs, ino_t ino, struct buf *bp)
102 {
103 	union lfs_dinode *ldip;
104 	unsigned num, i;
105 
106 	ASSERT_NO_SEGLOCK(fs);
107 	/*
108 	 * Read the inode block backwards, since later versions of the
109 	 * inode will supercede earlier ones.  Though it is unlikely, it is
110 	 * possible that the same inode will appear in the same inode block.
111 	 */
112 	num = LFS_INOPB(fs);
113 	for (i = num; i-- > 0; ) {
114 		ldip = DINO_IN_BLOCK(fs, bp->b_data, i);
115 		if (lfs_dino_getinumber(fs, ldip) == ino)
116 			return (ldip);
117 	}
118 
119 	printf("searched %u entries for %ju\n", num, (uintmax_t)ino);
120 	printf("offset is 0x%jx (seg %d)\n", (uintmax_t)lfs_sb_getoffset(fs),
121 	       lfs_dtosn(fs, lfs_sb_getoffset(fs)));
122 	printf("block is 0x%jx (seg %d)\n",
123 	       (uintmax_t)LFS_DBTOFSB(fs, bp->b_blkno),
124 	       lfs_dtosn(fs, LFS_DBTOFSB(fs, bp->b_blkno)));
125 
126 	return NULL;
127 }
128 
129 int
130 lfs_update(struct vnode *vp, const struct timespec *acc,
131     const struct timespec *mod, int updflags)
132 {
133 	struct inode *ip;
134 	struct lfs *fs = VFSTOULFS(vp->v_mount)->um_lfs;
135 	int flags;
136 
137 	ASSERT_NO_SEGLOCK(fs);
138 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
139 		return (0);
140 	ip = VTOI(vp);
141 
142 	/*
143 	 * If we are called from vinvalbuf, and the file's blocks have
144 	 * already been scheduled for writing, but the writes have not
145 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
146 	 * will cause a panic.	So, we must wait until any pending write
147 	 * for our inode completes, if we are called with UPDATE_WAIT set.
148 	 */
149 	mutex_enter(vp->v_interlock);
150 	while ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
151 	    WRITEINPROG(vp)) {
152 		DLOG((DLOG_SEG, "lfs_update: sleeping on ino %llu"
153 		      " (in progress)\n", (unsigned long long) ip->i_number));
154 		cv_wait(&vp->v_cv, vp->v_interlock);
155 	}
156 	mutex_exit(vp->v_interlock);
157 	LFS_ITIMES(ip, acc, mod, NULL);
158 	if (updflags & UPDATE_CLOSE)
159 		flags = ip->i_state & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING);
160 	else
161 		flags = ip->i_state & (IN_MODIFIED | IN_CLEANING);
162 	if (flags == 0)
163 		return (0);
164 
165 	/* If sync, push back the vnode and any dirty blocks it may have. */
166 	if ((updflags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT) {
167 		/* Avoid flushing VU_DIROP. */
168 		mutex_enter(&lfs_lock);
169 		++fs->lfs_diropwait;
170 		while (vp->v_uflag & VU_DIROP) {
171 			DLOG((DLOG_DIROP, "lfs_update: sleeping on inode %llu "
172 			      "(dirops)\n", (unsigned long long) ip->i_number));
173 			DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, i_state"
174 			      " 0x%x\n",
175 			      vp->v_iflag | vp->v_vflag | vp->v_uflag,
176 			      ip->i_state));
177 			if (fs->lfs_dirops == 0)
178 				lfs_flush_fs(fs, SEGM_SYNC);
179 			else
180 				mtsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
181 					0, &lfs_lock);
182 			/* XXX KS - by falling out here, are we writing the vn
183 			twice? */
184 		}
185 		--fs->lfs_diropwait;
186 		mutex_exit(&lfs_lock);
187 		return lfs_vflush(vp);
188 	}
189 	return 0;
190 }
191 
192 #define	SINGLE	0	/* index of single indirect block */
193 #define	DOUBLE	1	/* index of double indirect block */
194 #define	TRIPLE	2	/* index of triple indirect block */
195 /*
196  * Truncate the inode oip to at most length size, freeing the
197  * disk blocks.
198  */
199 /* VOP_BWRITE 1 + ULFS_NIADDR + lfs_balloc == 2 + 2*ULFS_NIADDR times */
200 
201 int
202 lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
203 {
204 	daddr_t lastblock;
205 	struct inode *oip = VTOI(ovp);
206 	daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR];
207 	/* note: newblks is set but only actually used if DIAGNOSTIC */
208 	daddr_t newblks[ULFS_NDADDR + ULFS_NIADDR] __diagused;
209 	struct lfs *fs;
210 	struct buf *bp;
211 	int offset, size, level;
212 	daddr_t count, rcount;
213 	daddr_t blocksreleased = 0, real_released = 0;
214 	int i, nblocks;
215 	int aflags, error, allerror = 0;
216 	off_t osize;
217 	long lastseg;
218 	size_t bc;
219 	int obufsize, odb;
220 	int usepc;
221 
222 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
223 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
224 		KASSERT(oip->i_size == 0);
225 		return 0;
226 	}
227 
228 	if (length < 0)
229 		return (EINVAL);
230 
231 	fs = oip->i_lfs;
232 
233 	if (ovp->v_type == VLNK &&
234 	    (oip->i_size < fs->um_maxsymlinklen ||
235 	     (fs->um_maxsymlinklen == 0 &&
236 	      lfs_dino_getblocks(fs, oip->i_din) == 0))) {
237 		KASSERTMSG((length == 0),
238 		    "partial truncate of symlink: %jd", (intmax_t)length);
239 		memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
240 		oip->i_size = 0;
241 		lfs_dino_setsize(fs, oip->i_din, 0);
242 		oip->i_state |= IN_CHANGE | IN_UPDATE;
243 		return (lfs_update(ovp, NULL, NULL, 0));
244 	}
245 	if (oip->i_size == length) {
246 		/* still do a uvm_vnp_setsize() as writesize may be larger */
247 		uvm_vnp_setsize(ovp, length);
248 		oip->i_state |= IN_CHANGE | IN_UPDATE;
249 		return (lfs_update(ovp, NULL, NULL, 0));
250 	}
251 	lfs_imtime(fs);
252 	osize = oip->i_size;
253 	usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
254 
255 	ASSERT_NO_SEGLOCK(fs);
256 	/*
257 	 * Lengthen the size of the file. We must ensure that the
258 	 * last byte of the file is allocated. Since the smallest
259 	 * value of osize is 0, length will be at least 1.
260 	 */
261 	if (osize < length) {
262 		if (length > fs->um_maxfilesize)
263 			return (EFBIG);
264 		aflags = B_CLRBUF;
265 		if (ioflag & IO_SYNC)
266 			aflags |= B_SYNC;
267 		if (usepc) {
268 			if (lfs_lblkno(fs, osize) < ULFS_NDADDR &&
269 			    lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) &&
270 			    lfs_blkroundup(fs, osize) != osize) {
271 				off_t eob;
272 
273 				eob = lfs_blkroundup(fs, osize);
274 				uvm_vnp_setwritesize(ovp, eob);
275 				error = ulfs_balloc_range(ovp, osize,
276 				    eob - osize, cred, aflags);
277 				if (error) {
278 					(void) lfs_truncate(ovp, osize,
279 						    ioflag & IO_SYNC, cred);
280 					return error;
281 				}
282 				if (ioflag & IO_SYNC) {
283 					mutex_enter(ovp->v_interlock);
284 					VOP_PUTPAGES(ovp,
285 					    trunc_page(osize & lfs_sb_getbmask(fs)),
286 					    round_page(eob),
287 					    PGO_CLEANIT | PGO_SYNCIO);
288 				}
289 			}
290 			uvm_vnp_setwritesize(ovp, length);
291 			error = ulfs_balloc_range(ovp, length - 1, 1, cred,
292 						 aflags);
293 			if (error) {
294 				(void) lfs_truncate(ovp, osize,
295 						    ioflag & IO_SYNC, cred);
296 				return error;
297 			}
298 			uvm_vnp_setsize(ovp, length);
299 			oip->i_state |= IN_CHANGE | IN_UPDATE;
300 			KASSERT(ovp->v_size == oip->i_size);
301 			oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
302 			return (lfs_update(ovp, NULL, NULL, 0));
303 		} else {
304 			error = lfs_reserve(fs, ovp, NULL,
305 			    lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
306 			if (error)
307 				return (error);
308 			error = lfs_balloc(ovp, length - 1, 1, cred,
309 					   aflags, &bp);
310 			lfs_reserve(fs, ovp, NULL,
311 			    -lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
312 			if (error)
313 				return (error);
314 			oip->i_size = length;
315 			lfs_dino_setsize(fs, oip->i_din, oip->i_size);
316 			uvm_vnp_setsize(ovp, length);
317 			(void) VOP_BWRITE(bp->b_vp, bp);
318 			oip->i_state |= IN_CHANGE | IN_UPDATE;
319 			oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
320 			return (lfs_update(ovp, NULL, NULL, 0));
321 		}
322 	}
323 
324 	if ((error = lfs_reserve(fs, ovp, NULL,
325 	    lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0)
326 		return (error);
327 
328 	/*
329 	 * Shorten the size of the file. If the file is not being
330 	 * truncated to a block boundary, the contents of the
331 	 * partial block following the end of the file must be
332 	 * zero'ed in case it ever becomes accessible again because
333 	 * of subsequent file growth. Directories however are not
334 	 * zero'ed as they should grow back initialized to empty.
335 	 */
336 	offset = lfs_blkoff(fs, length);
337 	lastseg = -1;
338 	bc = 0;
339 
340 	if (ovp != fs->lfs_ivnode)
341 		lfs_seglock(fs, SEGM_PROT);
342 	if (offset == 0) {
343 		oip->i_size = length;
344 		lfs_dino_setsize(fs, oip->i_din, oip->i_size);
345 	} else if (!usepc) {
346 		lbn = lfs_lblkno(fs, length);
347 		aflags = B_CLRBUF;
348 		if (ioflag & IO_SYNC)
349 			aflags |= B_SYNC;
350 		error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
351 		if (error) {
352 			lfs_reserve(fs, ovp, NULL,
353 			    -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
354 			goto errout;
355 		}
356 		obufsize = bp->b_bufsize;
357 		odb = lfs_btofsb(fs, bp->b_bcount);
358 		oip->i_size = length;
359 		lfs_dino_setsize(fs, oip->i_din, oip->i_size);
360 		size = lfs_blksize(fs, oip, lbn);
361 		if (ovp->v_type != VDIR)
362 			memset((char *)bp->b_data + offset, 0,
363 			       (u_int)(size - offset));
364 		allocbuf(bp, size, 1);
365 		if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
366 			mutex_enter(&lfs_lock);
367 			locked_queue_bytes -= obufsize - bp->b_bufsize;
368 			mutex_exit(&lfs_lock);
369 		}
370 		if (bp->b_oflags & BO_DELWRI) {
371 			lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size));
372 			/* XXX shouldn't this wake up on lfs_availsleep? */
373 		}
374 		(void) VOP_BWRITE(bp->b_vp, bp);
375 	} else { /* vp->v_type == VREG && length < osize && offset != 0 */
376 		/*
377 		 * When truncating a regular file down to a non-block-aligned
378 		 * size, we must zero the part of last block which is past
379 		 * the new EOF.  We must synchronously flush the zeroed pages
380 		 * to disk since the new pages will be invalidated as soon
381 		 * as we inform the VM system of the new, smaller size.
382 		 * We must do this before acquiring the GLOCK, since fetching
383 		 * the pages will acquire the GLOCK internally.
384 		 * So there is a window where another thread could see a whole
385 		 * zeroed page past EOF, but that's life.
386 		 */
387 		daddr_t xlbn;
388 		voff_t eoz;
389 
390 		aflags = ioflag & IO_SYNC ? B_SYNC : 0;
391 		error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags);
392 		if (error) {
393 			lfs_reserve(fs, ovp, NULL,
394 				    -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
395 			goto errout;
396 		}
397 		xlbn = lfs_lblkno(fs, length);
398 		size = lfs_blksize(fs, oip, xlbn);
399 		eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize);
400 		ubc_zerorange(&ovp->v_uobj, length, eoz - length,
401 		    UBC_UNMAP_FLAG(ovp));
402 		if (round_page(eoz) > round_page(length)) {
403 			mutex_enter(ovp->v_interlock);
404 			error = VOP_PUTPAGES(ovp, round_page(length),
405 			    round_page(eoz),
406 			    PGO_CLEANIT | PGO_DEACTIVATE |
407 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
408 			if (error) {
409 				lfs_reserve(fs, ovp, NULL,
410 					    -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
411 				goto errout;
412 			}
413 		}
414 	}
415 
416 	genfs_node_wrlock(ovp);
417 
418 	oip->i_size = length;
419 	lfs_dino_setsize(fs, oip->i_din, oip->i_size);
420 	uvm_vnp_setsize(ovp, length);
421 
422 	/*
423 	 * Calculate index into inode's block list of
424 	 * last direct and indirect blocks (if any)
425 	 * which we want to keep.  Lastblock is -1 when
426 	 * the file is truncated to 0.
427 	 */
428 	/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
429 	if (length > QUAD_MAX - lfs_sb_getbsize(fs))
430 		lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs));
431 	else
432 		lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1;
433 	lastiblock[SINGLE] = lastblock - ULFS_NDADDR;
434 	lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs);
435 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs);
436 	nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
437 	/*
438 	 * Record changed file and block pointers before we start
439 	 * freeing blocks.  lastiblock values are also normalized to -1
440 	 * for calls to lfs_indirtrunc below.
441 	 */
442 	for (i=0; i<ULFS_NDADDR; i++) {
443 		newblks[i] = lfs_dino_getdb(fs, oip->i_din, i);
444 	}
445 	for (i=0; i<ULFS_NIADDR; i++) {
446 		newblks[ULFS_NDADDR + i] = lfs_dino_getib(fs, oip->i_din, i);
447 	}
448 	for (level = TRIPLE; level >= SINGLE; level--)
449 		if (lastiblock[level] < 0) {
450 			newblks[ULFS_NDADDR+level] = 0;
451 			lastiblock[level] = -1;
452 		}
453 	for (i = ULFS_NDADDR - 1; i > lastblock; i--)
454 		newblks[i] = 0;
455 
456 	oip->i_size = osize;
457 	lfs_dino_setsize(fs, oip->i_din, oip->i_size);
458 	error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
459 	if (error && !allerror)
460 		allerror = error;
461 
462 	/*
463 	 * Indirect blocks first.
464 	 */
465 	indir_lbn[SINGLE] = -ULFS_NDADDR;
466 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1;
467 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1;
468 	for (level = TRIPLE; level >= SINGLE; level--) {
469 		bn = lfs_dino_getib(fs, oip->i_din, level);
470 		if (bn != 0) {
471 			error = lfs_indirtrunc(oip, indir_lbn[level],
472 					       bn, lastiblock[level],
473 					       level, &count, &rcount,
474 					       &lastseg, &bc);
475 			if (error)
476 				allerror = error;
477 			real_released += rcount;
478 			blocksreleased += count;
479 			if (lastiblock[level] < 0) {
480 				if (lfs_dino_getib(fs, oip->i_din, level) > 0)
481 					real_released += nblocks;
482 				blocksreleased += nblocks;
483 				lfs_dino_setib(fs, oip->i_din, level, 0);
484 				lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs),
485 					    &lastseg, &bc);
486         			lfs_deregister_block(ovp, bn);
487 			}
488 		}
489 		if (lastiblock[level] >= 0)
490 			goto done;
491 	}
492 
493 	/*
494 	 * All whole direct blocks or frags.
495 	 */
496 	for (i = ULFS_NDADDR - 1; i > lastblock; i--) {
497 		long bsize, obsize;
498 
499 		bn = lfs_dino_getdb(fs, oip->i_din, i);
500 		if (bn == 0)
501 			continue;
502 		bsize = lfs_blksize(fs, oip, i);
503 		if (lfs_dino_getdb(fs, oip->i_din, i) > 0) {
504 			/* Check for fragment size changes */
505 			obsize = oip->i_lfs_fragsize[i];
506 			real_released += lfs_btofsb(fs, obsize);
507 			oip->i_lfs_fragsize[i] = 0;
508 		} else
509 			obsize = 0;
510 		blocksreleased += lfs_btofsb(fs, bsize);
511 		lfs_dino_setdb(fs, oip->i_din, i, 0);
512 		lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
513         	lfs_deregister_block(ovp, bn);
514 	}
515 	if (lastblock < 0)
516 		goto done;
517 
518 	/*
519 	 * Finally, look for a change in size of the
520 	 * last direct block; release any frags.
521 	 */
522 	bn = lfs_dino_getdb(fs, oip->i_din, lastblock);
523 	if (bn != 0) {
524 		long oldspace, newspace;
525 #if 0
526 		long olddspace;
527 #endif
528 
529 		/*
530 		 * Calculate amount of space we're giving
531 		 * back as old block size minus new block size.
532 		 */
533 		oldspace = lfs_blksize(fs, oip, lastblock);
534 #if 0
535 		olddspace = oip->i_lfs_fragsize[lastblock];
536 #endif
537 
538 		oip->i_size = length;
539 		lfs_dino_setsize(fs, oip->i_din, oip->i_size);
540 		newspace = lfs_blksize(fs, oip, lastblock);
541 		if (newspace == 0)
542 			panic("itrunc: newspace");
543 		if (oldspace - newspace > 0) {
544 			blocksreleased += lfs_btofsb(fs, oldspace - newspace);
545 		}
546 #if 0
547 		if (bn > 0 && olddspace - newspace > 0) {
548 			/* No segment accounting here, just vnode */
549 			real_released += lfs_btofsb(fs, olddspace - newspace);
550 		}
551 #endif
552 	}
553 
554 done:
555 	/* Finish segment accounting corrections */
556 	lfs_update_seguse(fs, oip, lastseg, bc);
557 	for (level = SINGLE; level <= TRIPLE; level++)
558 		KASSERTMSG(((newblks[ULFS_NDADDR + level] == 0) ==
559 			(lfs_dino_getib(fs, oip->i_din, level) == 0)),
560 		    "lfs itrunc1");
561 	for (i = 0; i < ULFS_NDADDR; i++)
562 		KASSERTMSG(((newblks[i] == 0) ==
563 			(lfs_dino_getdb(fs, oip->i_din, i) == 0)),
564 		    "lfs itrunc2");
565 	KASSERTMSG((length != 0 || LIST_EMPTY(&ovp->v_cleanblkhd)),
566 	    "lfs itrunc3a");
567 	KASSERTMSG((length != 0 || LIST_EMPTY(&ovp->v_dirtyblkhd)),
568 	    "lfs itrunc3b");
569 
570 	/*
571 	 * Put back the real size.
572 	 */
573 	oip->i_size = length;
574 	lfs_dino_setsize(fs, oip->i_din, oip->i_size);
575 	oip->i_lfs_effnblks -= blocksreleased;
576 
577 	mutex_enter(&lfs_lock);
578 	lfs_dino_setblocks(fs, oip->i_din,
579 	    lfs_dino_getblocks(fs, oip->i_din) - real_released);
580 	lfs_sb_addbfree(fs, blocksreleased);
581 
582 	KASSERTMSG((oip->i_size != 0 ||
583 		lfs_dino_getblocks(fs, oip->i_din) == 0),
584 	    "ino %llu truncate to 0 but %jd blks/%jd effblks",
585 	    (unsigned long long) oip->i_number,
586 	    lfs_dino_getblocks(fs, oip->i_din), oip->i_lfs_effnblks);
587 	KASSERTMSG((oip->i_size != 0 || oip->i_lfs_effnblks == 0),
588 	    "ino %llu truncate to 0 but %jd blks/%jd effblks",
589 	    (unsigned long long) oip->i_number,
590 	    lfs_dino_getblocks(fs, oip->i_din), oip->i_lfs_effnblks);
591 
592 	/*
593 	 * If we truncated to zero, take us off the paging queue.
594 	 */
595 	if (oip->i_size == 0 && oip->i_state & IN_PAGING) {
596 		oip->i_state &= ~IN_PAGING;
597 		TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
598 	}
599 	mutex_exit(&lfs_lock);
600 
601 	oip->i_state |= IN_CHANGE;
602 #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
603 	(void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0);
604 #endif
605 	lfs_reserve(fs, ovp, NULL,
606 	    -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
607 	genfs_node_unlock(ovp);
608   errout:
609 	oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
610 	if (ovp != fs->lfs_ivnode)
611 		lfs_segunlock(fs);
612 	return (allerror ? allerror : error);
613 }
614 
615 /* Update segment and avail usage information when removing a block. */
616 static int
617 lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
618 	    size_t bsize, long *lastseg, size_t *num)
619 {
620 	long seg;
621 	int error = 0;
622 
623 	ASSERT_SEGLOCK(fs);
624 	bsize = lfs_fragroundup(fs, bsize);
625 	if (daddr > 0) {
626 		if (*lastseg != (seg = lfs_dtosn(fs, daddr))) {
627 			error = lfs_update_seguse(fs, ip, *lastseg, *num);
628 			*num = bsize;
629 			*lastseg = seg;
630 		} else
631 			*num += bsize;
632 	}
633 
634 	return error;
635 }
636 
637 /* Finish the accounting updates for a segment. */
638 static int
639 lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
640 {
641 	struct segdelta *sd;
642 
643 	ASSERT_SEGLOCK(fs);
644 	if (lastseg < 0 || num == 0)
645 		return 0;
646 
647 	LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
648 		if (sd->segnum == lastseg)
649 			break;
650 	if (sd == NULL) {
651 		sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
652 		sd->segnum = lastseg;
653 		sd->num = 0;
654 		LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
655 	}
656 	sd->num += num;
657 
658 	return 0;
659 }
660 
661 static void
662 lfs_finalize_seguse(struct lfs *fs, void *v)
663 {
664 	SEGUSE *sup;
665 	struct buf *bp;
666 	struct segdelta *sd;
667 	LIST_HEAD(, segdelta) *hd = v;
668 
669 	ASSERT_SEGLOCK(fs);
670 	while((sd = LIST_FIRST(hd)) != NULL) {
671 		LIST_REMOVE(sd, list);
672 		LFS_SEGENTRY(sup, fs, sd->segnum, bp);
673 		if (sd->num > sup->su_nbytes) {
674 			printf("lfs_finalize_seguse: segment %ld short by %ld\n",
675 				sd->segnum, (long)(sd->num - sup->su_nbytes));
676 			panic("lfs_finalize_seguse: negative bytes");
677 			sup->su_nbytes = sd->num;
678 		}
679 		sup->su_nbytes -= sd->num;
680 		LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
681 		free(sd, M_SEGMENT);
682 	}
683 }
684 
685 /* Finish the accounting updates for a segment. */
686 void
687 lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
688 {
689 	ASSERT_SEGLOCK(fs);
690 	lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
691 }
692 
693 /* Finish the accounting updates for a segment. */
694 void
695 lfs_finalize_fs_seguse(struct lfs *fs)
696 {
697 	ASSERT_SEGLOCK(fs);
698 	lfs_finalize_seguse(fs, &fs->lfs_segdhd);
699 }
700 
701 /*
702  * Release blocks associated with the inode ip and stored in the indirect
703  * block bn.  Blocks are free'd in LIFO order up to (but not including)
704  * lastbn.  If level is greater than SINGLE, the block is an indirect block
705  * and recursive calls to indirtrunc must be used to cleanse other indirect
706  * blocks.
707  *
708  * NB: triple indirect blocks are untested.
709  */
710 static int
711 lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
712 	       daddr_t lastbn, int level, daddr_t *countp,
713 	       daddr_t *rcountp, long *lastsegp, size_t *bcp)
714 {
715 	int i;
716 	struct buf *bp;
717 	struct lfs *fs = ip->i_lfs;
718 	void *bap;
719 	bool bap_needs_free;
720 	struct vnode *vp;
721 	daddr_t nb, nlbn, last;
722 	daddr_t blkcount, rblkcount, factor;
723 	int nblocks;
724 	daddr_t blocksreleased = 0, real_released = 0;
725 	int error = 0, allerror = 0;
726 
727 	ASSERT_SEGLOCK(fs);
728 	/*
729 	 * Calculate index in current block of last
730 	 * block to be kept.  -1 indicates the entire
731 	 * block so we need not calculate the index.
732 	 */
733 	factor = 1;
734 	for (i = SINGLE; i < level; i++)
735 		factor *= LFS_NINDIR(fs);
736 	last = lastbn;
737 	if (lastbn > 0)
738 		last /= factor;
739 	nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
740 	/*
741 	 * Get buffer of block pointers, zero those entries corresponding
742 	 * to blocks to be free'd, and update on disk copy first.  Since
743 	 * double(triple) indirect before single(double) indirect, calls
744 	 * to bmap on these blocks will fail.  However, we already have
745 	 * the on disk address, so we have to set the b_blkno field
746 	 * explicitly instead of letting bread do everything for us.
747 	 */
748 	vp = ITOV(ip);
749 	bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0);
750 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
751 		/* Braces must be here in case trace evaluates to nothing. */
752 		trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn);
753 	} else {
754 		trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn);
755 		curlwp->l_ru.ru_inblock++; /* pay for read */
756 		bp->b_flags |= B_READ;
757 		if (bp->b_bcount > bp->b_bufsize)
758 			panic("lfs_indirtrunc: bad buffer size");
759 		bp->b_blkno = LFS_FSBTODB(fs, dbn);
760 		VOP_STRATEGY(vp, bp);
761 		error = biowait(bp);
762 	}
763 	if (error) {
764 		brelse(bp, 0);
765 		*countp = *rcountp = 0;
766 		return (error);
767 	}
768 
769 	if (lastbn >= 0) {
770 		/*
771 		 * We still need this block, so copy the data for
772 		 * subsequent processing; then in the original block,
773 		 * zero out the dying block pointers and send it off.
774 		 */
775 		bap = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK);
776 		memcpy(bap, bp->b_data, lfs_sb_getbsize(fs));
777 		bap_needs_free = true;
778 
779 		for (i = last + 1; i < LFS_NINDIR(fs); i++) {
780 			lfs_iblock_set(fs, bp->b_data, i, 0);
781 		}
782 		error = VOP_BWRITE(bp->b_vp, bp);
783 		if (error)
784 			allerror = error;
785 	} else {
786 		bap = bp->b_data;
787 		bap_needs_free = false;
788 	}
789 
790 	/*
791 	 * Recursively free totally unused blocks.
792 	 */
793 	for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
794 	    i--, nlbn += factor) {
795 		nb = lfs_iblock_get(fs, bap, i);
796 		if (nb == 0)
797 			continue;
798 		if (level > SINGLE) {
799 			error = lfs_indirtrunc(ip, nlbn, nb,
800 					       (daddr_t)-1, level - 1,
801 					       &blkcount, &rblkcount,
802 					       lastsegp, bcp);
803 			if (error)
804 				allerror = error;
805 			blocksreleased += blkcount;
806 			real_released += rblkcount;
807 		}
808 		lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp);
809 		if (lfs_iblock_get(fs, bap, i) > 0)
810 			real_released += nblocks;
811 		blocksreleased += nblocks;
812 	}
813 
814 	/*
815 	 * Recursively free last partial block.
816 	 */
817 	if (level > SINGLE && lastbn >= 0) {
818 		last = lastbn % factor;
819 		nb = lfs_iblock_get(fs, bap, i);
820 		if (nb != 0) {
821 			error = lfs_indirtrunc(ip, nlbn, nb,
822 					       last, level - 1, &blkcount,
823 					       &rblkcount, lastsegp, bcp);
824 			if (error)
825 				allerror = error;
826 			real_released += rblkcount;
827 			blocksreleased += blkcount;
828 		}
829 	}
830 
831 	if (bap_needs_free) {
832 		lfs_free(fs, bap, LFS_NB_IBLOCK);
833 	} else {
834 		mutex_enter(&bufcache_lock);
835 		if (bp->b_oflags & BO_DELWRI) {
836 			LFS_UNLOCK_BUF(bp);
837 			lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
838 			wakeup(&fs->lfs_availsleep);
839 		}
840 		brelsel(bp, BC_INVAL);
841 		mutex_exit(&bufcache_lock);
842 	}
843 
844 	*countp = blocksreleased;
845 	*rcountp = real_released;
846 	return (allerror);
847 }
848 
849 /*
850  * Destroy any in core blocks past the truncation length.
851  * Inlined from vtruncbuf, so that lfs_avail could be updated.
852  * We take the seglock to prevent cleaning from occurring while we are
853  * invalidating blocks.
854  */
855 static int
856 lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
857 {
858 	struct buf *bp, *nbp;
859 	int error = 0;
860 	struct lfs *fs;
861 	voff_t off;
862 
863 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
864 	mutex_enter(vp->v_interlock);
865 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
866 	if (error)
867 		return error;
868 
869 	fs = VTOI(vp)->i_lfs;
870 
871 	ASSERT_SEGLOCK(fs);
872 
873 	mutex_enter(&bufcache_lock);
874 restart:
875 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
876 		nbp = LIST_NEXT(bp, b_vnbufs);
877 		if (bp->b_lblkno < lbn)
878 			continue;
879 		error = bbusy(bp, catch, slptimeo, NULL);
880 		if (error == EPASSTHROUGH)
881 			goto restart;
882 		if (error)
883 			goto exit;
884 
885 		mutex_enter(bp->b_objlock);
886 		if (bp->b_oflags & BO_DELWRI) {
887 			bp->b_oflags &= ~BO_DELWRI;
888 			lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
889 			wakeup(&fs->lfs_availsleep);
890 		}
891 		mutex_exit(bp->b_objlock);
892 		LFS_UNLOCK_BUF(bp);
893 		brelsel(bp, BC_INVAL | BC_VFLUSH);
894 	}
895 
896 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
897 		nbp = LIST_NEXT(bp, b_vnbufs);
898 		if (bp->b_lblkno < lbn)
899 			continue;
900 		error = bbusy(bp, catch, slptimeo, NULL);
901 		if (error == EPASSTHROUGH)
902 			goto restart;
903 		if (error)
904 			goto exit;
905 
906 		mutex_enter(bp->b_objlock);
907 		if (bp->b_oflags & BO_DELWRI) {
908 			bp->b_oflags &= ~BO_DELWRI;
909 			lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
910 			wakeup(&fs->lfs_availsleep);
911 		}
912 		mutex_exit(bp->b_objlock);
913 		LFS_UNLOCK_BUF(bp);
914 		brelsel(bp, BC_INVAL | BC_VFLUSH);
915 	}
916 exit:
917 	mutex_exit(&bufcache_lock);
918 
919 	return error;
920 }
921 
922