1/* $NetBSD: lfs_inode.c,v 1.147 2015/09/01 06:13:09 dholland 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 1.147 2015/09/01 06:13:09 dholland 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
92static int lfs_update_seguse(struct lfs *, struct inode *ip, long, size_t);
93static int lfs_indirtrunc(struct inode *, daddr_t, daddr_t,
94 daddr_t, int, daddr_t *, daddr_t *,
95 long *, size_t *);
96static int lfs_blkfree (struct lfs *, struct inode *, daddr_t, size_t, long *, size_t *);
97static int lfs_vtruncbuf(struct vnode *, daddr_t, bool, int);
98
99/* Search a block for a specific dinode. */
100union lfs_dinode *
101lfs_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
129int
130lfs_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 %d"
153 " (in progress)\n", 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_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING);
160 else
161 flags = ip->i_flag & (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 %d"
172 " (dirops)\n", ip->i_number));
173 DLOG((DLOG_DIROP, "lfs_update: vflags 0x%x, iflags"
174 " 0x%x\n",
175 vp->v_iflag | vp->v_vflag | vp->v_uflag,
176 ip->i_flag));
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
201int
202lfs_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 /*
232 * Just return and not update modification times.
233 */
234 if (oip->i_size == length) {
235 /* still do a uvm_vnp_setsize() as writesize may be larger */
236 uvm_vnp_setsize(ovp, length);
237 return (0);
238 }
239
240 fs = oip->i_lfs;
241
242 if (ovp->v_type == VLNK &&
243 (oip->i_size < fs->um_maxsymlinklen ||
244 (fs->um_maxsymlinklen == 0 &&
245 lfs_dino_getblocks(fs, oip->i_din) == 0))) {
246#ifdef DIAGNOSTIC
247 if (length != 0)
248 panic("lfs_truncate: partial truncate of symlink");
249#endif
250 memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
251 oip->i_size = 0;
252 lfs_dino_setsize(fs, oip->i_din, 0);
253 oip->i_flag |= IN_CHANGE | IN_UPDATE;
254 return (lfs_update(ovp, NULL, NULL, 0));
255 }
256 if (oip->i_size == length) {
257 oip->i_flag |= IN_CHANGE | IN_UPDATE;
258 return (lfs_update(ovp, NULL, NULL, 0));
259 }
260 lfs_imtime(fs);
261 osize = oip->i_size;
262 usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
263
264 ASSERT_NO_SEGLOCK(fs);
265 /*
266 * Lengthen the size of the file. We must ensure that the
267 * last byte of the file is allocated. Since the smallest
268 * value of osize is 0, length will be at least 1.
269 */
270 if (osize < length) {
271 if (length > fs->um_maxfilesize)
272 return (EFBIG);
273 aflags = B_CLRBUF;
274 if (ioflag & IO_SYNC)
275 aflags |= B_SYNC;
276 if (usepc) {
277 if (lfs_lblkno(fs, osize) < ULFS_NDADDR &&
278 lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) &&
279 lfs_blkroundup(fs, osize) != osize) {
280 off_t eob;
281
282 eob = lfs_blkroundup(fs, osize);
283 uvm_vnp_setwritesize(ovp, eob);
284 error = ulfs_balloc_range(ovp, osize,
285 eob - osize, cred, aflags);
286 if (error) {
287 (void) lfs_truncate(ovp, osize,
288 ioflag & IO_SYNC, cred);
289 return error;
290 }
291 if (ioflag & IO_SYNC) {
292 mutex_enter(ovp->v_interlock);
293 VOP_PUTPAGES(ovp,
294 trunc_page(osize & lfs_sb_getbmask(fs)),
295 round_page(eob),
296 PGO_CLEANIT | PGO_SYNCIO);
297 }
298 }
299 uvm_vnp_setwritesize(ovp, length);
300 error = ulfs_balloc_range(ovp, length - 1, 1, cred,
301 aflags);
302 if (error) {
303 (void) lfs_truncate(ovp, osize,
304 ioflag & IO_SYNC, cred);
305 return error;
306 }
307 uvm_vnp_setsize(ovp, length);
308 oip->i_flag |= IN_CHANGE | IN_UPDATE;
309 KASSERT(ovp->v_size == oip->i_size);
310 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
311 return (lfs_update(ovp, NULL, NULL, 0));
312 } else {
313 error = lfs_reserve(fs, ovp, NULL,
314 lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
315 if (error)
316 return (error);
317 error = lfs_balloc(ovp, length - 1, 1, cred,
318 aflags, &bp);
319 lfs_reserve(fs, ovp, NULL,
320 -lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
321 if (error)
322 return (error);
323 oip->i_size = length;
324 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
325 uvm_vnp_setsize(ovp, length);
326 (void) VOP_BWRITE(bp->b_vp, bp);
327 oip->i_flag |= IN_CHANGE | IN_UPDATE;
328 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
329 return (lfs_update(ovp, NULL, NULL, 0));
330 }
331 }
332
333 if ((error = lfs_reserve(fs, ovp, NULL,
334 lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0)
335 return (error);
336
337 /*
338 * Shorten the size of the file. If the file is not being
339 * truncated to a block boundary, the contents of the
340 * partial block following the end of the file must be
341 * zero'ed in case it ever becomes accessible again because
342 * of subsequent file growth. Directories however are not
343 * zero'ed as they should grow back initialized to empty.
344 */
345 offset = lfs_blkoff(fs, length);
346 lastseg = -1;
347 bc = 0;
348
349 if (ovp != fs->lfs_ivnode)
350 lfs_seglock(fs, SEGM_PROT);
351 if (offset == 0) {
352 oip->i_size = length;
353 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
354 } else if (!usepc) {
355 lbn = lfs_lblkno(fs, length);
356 aflags = B_CLRBUF;
357 if (ioflag & IO_SYNC)
358 aflags |= B_SYNC;
359 error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
360 if (error) {
361 lfs_reserve(fs, ovp, NULL,
362 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
363 goto errout;
364 }
365 obufsize = bp->b_bufsize;
366 odb = lfs_btofsb(fs, bp->b_bcount);
367 oip->i_size = length;
368 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
369 size = lfs_blksize(fs, oip, lbn);
370 if (ovp->v_type != VDIR)
371 memset((char *)bp->b_data + offset, 0,
372 (u_int)(size - offset));
373 allocbuf(bp, size, 1);
374 if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
375 mutex_enter(&lfs_lock);
376 locked_queue_bytes -= obufsize - bp->b_bufsize;
377 mutex_exit(&lfs_lock);
378 }
379 if (bp->b_oflags & BO_DELWRI) {
380 lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size));
381 /* XXX shouldn't this wake up on lfs_availsleep? */
382 }
383 (void) VOP_BWRITE(bp->b_vp, bp);
384 } else { /* vp->v_type == VREG && length < osize && offset != 0 */
385 /*
386 * When truncating a regular file down to a non-block-aligned
387 * size, we must zero the part of last block which is past
388 * the new EOF. We must synchronously flush the zeroed pages
389 * to disk since the new pages will be invalidated as soon
390 * as we inform the VM system of the new, smaller size.
391 * We must do this before acquiring the GLOCK, since fetching
392 * the pages will acquire the GLOCK internally.
393 * So there is a window where another thread could see a whole
394 * zeroed page past EOF, but that's life.
395 */
396 daddr_t xlbn;
397 voff_t eoz;
398
399 aflags = ioflag & IO_SYNC ? B_SYNC : 0;
400 error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags);
401 if (error) {
402 lfs_reserve(fs, ovp, NULL,
403 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
404 goto errout;
405 }
406 xlbn = lfs_lblkno(fs, length);
407 size = lfs_blksize(fs, oip, xlbn);
408 eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize);
409 ubc_zerorange(&ovp->v_uobj, length, eoz - length,
410 UBC_UNMAP_FLAG(ovp));
411 if (round_page(eoz) > round_page(length)) {
412 mutex_enter(ovp->v_interlock);
413 error = VOP_PUTPAGES(ovp, round_page(length),
414 round_page(eoz),
415 PGO_CLEANIT | PGO_DEACTIVATE |
416 ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
417 if (error) {
418 lfs_reserve(fs, ovp, NULL,
419 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
420 goto errout;
421 }
422 }
423 }
424
425 genfs_node_wrlock(ovp);
426
427 oip->i_size = length;
428 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
429 uvm_vnp_setsize(ovp, length);
430
431 /*
432 * Calculate index into inode's block list of
433 * last direct and indirect blocks (if any)
434 * which we want to keep. Lastblock is -1 when
435 * the file is truncated to 0.
436 */
437 /* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
438 if (length > QUAD_MAX - lfs_sb_getbsize(fs))
439 lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs));
440 else
441 lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1;
442 lastiblock[SINGLE] = lastblock - ULFS_NDADDR;
443 lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs);
444 lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs);
445 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
446 /*
447 * Record changed file and block pointers before we start
448 * freeing blocks. lastiblock values are also normalized to -1
449 * for calls to lfs_indirtrunc below.
450 */
451 for (i=0; i<ULFS_NDADDR; i++) {
452 newblks[i] = lfs_dino_getdb(fs, oip->i_din, i);
453 }
454 for (i=0; i<ULFS_NIADDR; i++) {
455 newblks[ULFS_NDADDR + i] = lfs_dino_getib(fs, oip->i_din, i);
456 }
457 for (level = TRIPLE; level >= SINGLE; level--)
458 if (lastiblock[level] < 0) {
459 newblks[ULFS_NDADDR+level] = 0;
460 lastiblock[level] = -1;
461 }
462 for (i = ULFS_NDADDR - 1; i > lastblock; i--)
463 newblks[i] = 0;
464
465 oip->i_size = osize;
466 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
467 error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
468 if (error && !allerror)
469 allerror = error;
470
471 /*
472 * Indirect blocks first.
473 */
474 indir_lbn[SINGLE] = -ULFS_NDADDR;
475 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1;
476 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1;
477 for (level = TRIPLE; level >= SINGLE; level--) {
478 bn = lfs_dino_getib(fs, oip->i_din, level);
479 if (bn != 0) {
480 error = lfs_indirtrunc(oip, indir_lbn[level],
481 bn, lastiblock[level],
482 level, &count, &rcount,
483 &lastseg, &bc);
484 if (error)
485 allerror = error;
486 real_released += rcount;
487 blocksreleased += count;
488 if (lastiblock[level] < 0) {
489 if (lfs_dino_getib(fs, oip->i_din, level) > 0)
490 real_released += nblocks;
491 blocksreleased += nblocks;
492 lfs_dino_setib(fs, oip->i_din, level, 0);
493 lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs),
494 &lastseg, &bc);
495 lfs_deregister_block(ovp, bn);
496 }
497 }
498 if (lastiblock[level] >= 0)
499 goto done;
500 }
501
502 /*
503 * All whole direct blocks or frags.
504 */
505 for (i = ULFS_NDADDR - 1; i > lastblock; i--) {
506 long bsize, obsize;
507
508 bn = lfs_dino_getdb(fs, oip->i_din, i);
509 if (bn == 0)
510 continue;
511 bsize = lfs_blksize(fs, oip, i);
512 if (lfs_dino_getdb(fs, oip->i_din, i) > 0) {
513 /* Check for fragment size changes */
514 obsize = oip->i_lfs_fragsize[i];
515 real_released += lfs_btofsb(fs, obsize);
516 oip->i_lfs_fragsize[i] = 0;
517 } else
518 obsize = 0;
519 blocksreleased += lfs_btofsb(fs, bsize);
520 lfs_dino_setdb(fs, oip->i_din, i, 0);
521 lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
522 lfs_deregister_block(ovp, bn);
523 }
524 if (lastblock < 0)
525 goto done;
526
527 /*
528 * Finally, look for a change in size of the
529 * last direct block; release any frags.
530 */
531 bn = lfs_dino_getdb(fs, oip->i_din, lastblock);
532 if (bn != 0) {
533 long oldspace, newspace;
534#if 0
535 long olddspace;
536#endif
537
538 /*
539 * Calculate amount of space we're giving
540 * back as old block size minus new block size.
541 */
542 oldspace = lfs_blksize(fs, oip, lastblock);
543#if 0
544 olddspace = oip->i_lfs_fragsize[lastblock];
545#endif
546
547 oip->i_size = length;
548 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
549 newspace = lfs_blksize(fs, oip, lastblock);
550 if (newspace == 0)
551 panic("itrunc: newspace");
552 if (oldspace - newspace > 0) {
553 blocksreleased += lfs_btofsb(fs, oldspace - newspace);
554 }
555#if 0
556 if (bn > 0 && olddspace - newspace > 0) {
557 /* No segment accounting here, just vnode */
558 real_released += lfs_btofsb(fs, olddspace - newspace);
559 }
560#endif
561 }
562
563done:
564 /* Finish segment accounting corrections */
565 lfs_update_seguse(fs, oip, lastseg, bc);
566#ifdef DIAGNOSTIC
567 for (level = SINGLE; level <= TRIPLE; level++)
568 if ((newblks[ULFS_NDADDR + level] == 0) !=
569 (lfs_dino_getib(fs, oip->i_din, level) == 0)) {
570 panic("lfs itrunc1");
571 }
572 for (i = 0; i < ULFS_NDADDR; i++)
573 if ((newblks[i] == 0) !=
574 (lfs_dino_getdb(fs, oip->i_din, i) == 0)) {
575 panic("lfs itrunc2");
576 }
577 if (length == 0 &&
578 (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
579 panic("lfs itrunc3");
580#endif /* DIAGNOSTIC */
581 /*
582 * Put back the real size.
583 */
584 oip->i_size = length;
585 lfs_dino_setsize(fs, oip->i_din, oip->i_size);
586 oip->i_lfs_effnblks -= blocksreleased;
587 lfs_dino_setblocks(fs, oip->i_din,
588 lfs_dino_getblocks(fs, oip->i_din) - real_released);
589 mutex_enter(&lfs_lock);
590 lfs_sb_addbfree(fs, blocksreleased);
591 mutex_exit(&lfs_lock);
592#ifdef DIAGNOSTIC
593 if (oip->i_size == 0 &&
594 (lfs_dino_getblocks(fs, oip->i_din) != 0 || oip->i_lfs_effnblks != 0)) {
595 printf("lfs_truncate: truncate to 0 but %jd blks/%jd effblks\n",
596 (intmax_t)lfs_dino_getblocks(fs, oip->i_din),
597 (intmax_t)oip->i_lfs_effnblks);
598 panic("lfs_truncate: persistent blocks");
599 }
600#endif
601
602 /*
603 * If we truncated to zero, take us off the paging queue.
604 */
605 mutex_enter(&lfs_lock);
606 if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
607 oip->i_flags &= ~IN_PAGING;
608 TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
609 }
610 mutex_exit(&lfs_lock);
611
612 oip->i_flag |= IN_CHANGE;
613#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
614 (void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0);
615#endif
616 lfs_reserve(fs, ovp, NULL,
617 -lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
618 genfs_node_unlock(ovp);
619 errout:
620 oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
621 if (ovp != fs->lfs_ivnode)
622 lfs_segunlock(fs);
623 return (allerror ? allerror : error);
624}
625
626/* Update segment and avail usage information when removing a block. */
627static int
628lfs_blkfree(struct lfs *fs, struct inode *ip, daddr_t daddr,
629 size_t bsize, long *lastseg, size_t *num)
630{
631 long seg;
632 int error = 0;
633
634 ASSERT_SEGLOCK(fs);
635 bsize = lfs_fragroundup(fs, bsize);
636 if (daddr > 0) {
637 if (*lastseg != (seg = lfs_dtosn(fs, daddr))) {
638 error = lfs_update_seguse(fs, ip, *lastseg, *num);
639 *num = bsize;
640 *lastseg = seg;
641 } else
642 *num += bsize;
643 }
644
645 return error;
646}
647
648/* Finish the accounting updates for a segment. */
649static int
650lfs_update_seguse(struct lfs *fs, struct inode *ip, long lastseg, size_t num)
651{
652 struct segdelta *sd;
653
654 ASSERT_SEGLOCK(fs);
655 if (lastseg < 0 || num == 0)
656 return 0;
657
658 LIST_FOREACH(sd, &ip->i_lfs_segdhd, list)
659 if (sd->segnum == lastseg)
660 break;
661 if (sd == NULL) {
662 sd = malloc(sizeof(*sd), M_SEGMENT, M_WAITOK);
663 sd->segnum = lastseg;
664 sd->num = 0;
665 LIST_INSERT_HEAD(&ip->i_lfs_segdhd, sd, list);
666 }
667 sd->num += num;
668
669 return 0;
670}
671
672static void
673lfs_finalize_seguse(struct lfs *fs, void *v)
674{
675 SEGUSE *sup;
676 struct buf *bp;
677 struct segdelta *sd;
678 LIST_HEAD(, segdelta) *hd = v;
679
680 ASSERT_SEGLOCK(fs);
681 while((sd = LIST_FIRST(hd)) != NULL) {
682 LIST_REMOVE(sd, list);
683 LFS_SEGENTRY(sup, fs, sd->segnum, bp);
684 if (sd->num > sup->su_nbytes) {
685 printf("lfs_finalize_seguse: segment %ld short by %ld\n",
686 sd->segnum, (long)(sd->num - sup->su_nbytes));
687 panic("lfs_finalize_seguse: negative bytes");
688 sup->su_nbytes = sd->num;
689 }
690 sup->su_nbytes -= sd->num;
691 LFS_WRITESEGENTRY(sup, fs, sd->segnum, bp);
692 free(sd, M_SEGMENT);
693 }
694}
695
696/* Finish the accounting updates for a segment. */
697void
698lfs_finalize_ino_seguse(struct lfs *fs, struct inode *ip)
699{
700 ASSERT_SEGLOCK(fs);
701 lfs_finalize_seguse(fs, &ip->i_lfs_segdhd);
702}
703
704/* Finish the accounting updates for a segment. */
705void
706lfs_finalize_fs_seguse(struct lfs *fs)
707{
708 ASSERT_SEGLOCK(fs);
709 lfs_finalize_seguse(fs, &fs->lfs_segdhd);
710}
711
712/*
713 * Release blocks associated with the inode ip and stored in the indirect
714 * block bn. Blocks are free'd in LIFO order up to (but not including)
715 * lastbn. If level is greater than SINGLE, the block is an indirect block
716 * and recursive calls to indirtrunc must be used to cleanse other indirect
717 * blocks.
718 *
719 * NB: triple indirect blocks are untested.
720 */
721static int
722lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
723 daddr_t lastbn, int level, daddr_t *countp,
724 daddr_t *rcountp, long *lastsegp, size_t *bcp)
725{
726 int i;
727 struct buf *bp;
728 struct lfs *fs = ip->i_lfs;
729 void *bap;
730 bool bap_needs_free;
731 struct vnode *vp;
732 daddr_t nb, nlbn, last;
733 daddr_t blkcount, rblkcount, factor;
734 int nblocks;
735 daddr_t blocksreleased = 0, real_released = 0;
736 int error = 0, allerror = 0;
737
738 ASSERT_SEGLOCK(fs);
739 /*
740 * Calculate index in current block of last
741 * block to be kept. -1 indicates the entire
742 * block so we need not calculate the index.
743 */
744 factor = 1;
745 for (i = SINGLE; i < level; i++)
746 factor *= LFS_NINDIR(fs);
747 last = lastbn;
748 if (lastbn > 0)
749 last /= factor;
750 nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
751 /*
752 * Get buffer of block pointers, zero those entries corresponding
753 * to blocks to be free'd, and update on disk copy first. Since
754 * double(triple) indirect before single(double) indirect, calls
755 * to bmap on these blocks will fail. However, we already have
756 * the on disk address, so we have to set the b_blkno field
757 * explicitly instead of letting bread do everything for us.
758 */
759 vp = ITOV(ip);
760 bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0);
761 if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
762 /* Braces must be here in case trace evaluates to nothing. */
763 trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn);
764 } else {
765 trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn);
766 curlwp->l_ru.ru_inblock++; /* pay for read */
767 bp->b_flags |= B_READ;
768 if (bp->b_bcount > bp->b_bufsize)
769 panic("lfs_indirtrunc: bad buffer size");
770 bp->b_blkno = LFS_FSBTODB(fs, dbn);
771 VOP_STRATEGY(vp, bp);
772 error = biowait(bp);
773 }
774 if (error) {
775 brelse(bp, 0);
776 *countp = *rcountp = 0;
777 return (error);
778 }
779
780 if (lastbn >= 0) {
781 /*
782 * We still need this block, so copy the data for
783 * subsequent processing; then in the original block,
784 * zero out the dying block pointers and send it off.
785 */
786 bap = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK);
787 memcpy(bap, bp->b_data, lfs_sb_getbsize(fs));
788 bap_needs_free = true;
789
790 for (i = last + 1; i < LFS_NINDIR(fs); i++) {
791 lfs_iblock_set(fs, bp->b_data, i, 0);
792 }
793 error = VOP_BWRITE(bp->b_vp, bp);
794 if (error)
795 allerror = error;
796 } else {
797 bap = bp->b_data;
798 bap_needs_free = false;
799 }
800
801 /*
802 * Recursively free totally unused blocks.
803 */
804 for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
805 i--, nlbn += factor) {
806 nb = lfs_iblock_get(fs, bap, i);
807 if (nb == 0)
808 continue;
809 if (level > SINGLE) {
810 error = lfs_indirtrunc(ip, nlbn, nb,
811 (daddr_t)-1, level - 1,
812 &blkcount, &rblkcount,
813 lastsegp, bcp);
814 if (error)
815 allerror = error;
816 blocksreleased += blkcount;
817 real_released += rblkcount;
818 }
819 lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp);
820 if (lfs_iblock_get(fs, bap, i) > 0)
821 real_released += nblocks;
822 blocksreleased += nblocks;
823 }
824
825 /*
826 * Recursively free last partial block.
827 */
828 if (level > SINGLE && lastbn >= 0) {
829 last = lastbn % factor;
830 nb = lfs_iblock_get(fs, bap, i);
831 if (nb != 0) {
832 error = lfs_indirtrunc(ip, nlbn, nb,
833 last, level - 1, &blkcount,
834 &rblkcount, lastsegp, bcp);
835 if (error)
836 allerror = error;
837 real_released += rblkcount;
838 blocksreleased += blkcount;
839 }
840 }
841
842 if (bap_needs_free) {
843 lfs_free(fs, bap, LFS_NB_IBLOCK);
844 } else {
845 mutex_enter(&bufcache_lock);
846 if (bp->b_oflags & BO_DELWRI) {
847 LFS_UNLOCK_BUF(bp);
848 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
849 wakeup(&fs->lfs_availsleep);
850 }
851 brelsel(bp, BC_INVAL);
852 mutex_exit(&bufcache_lock);
853 }
854
855 *countp = blocksreleased;
856 *rcountp = real_released;
857 return (allerror);
858}
859
860/*
861 * Destroy any in core blocks past the truncation length.
862 * Inlined from vtruncbuf, so that lfs_avail could be updated.
863 * We take the seglock to prevent cleaning from occurring while we are
864 * invalidating blocks.
865 */
866static int
867lfs_vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo)
868{
869 struct buf *bp, *nbp;
870 int error;
871 struct lfs *fs;
872 voff_t off;
873
874 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
875 mutex_enter(vp->v_interlock);
876 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
877 if (error)
878 return error;
879
880 fs = VTOI(vp)->i_lfs;
881
882 ASSERT_SEGLOCK(fs);
883
884 mutex_enter(&bufcache_lock);
885restart:
886 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
887 nbp = LIST_NEXT(bp, b_vnbufs);
888 if (bp->b_lblkno < lbn)
889 continue;
890 error = bbusy(bp, catch, slptimeo, NULL);
891 if (error == EPASSTHROUGH)
892 goto restart;
893 if (error != 0) {
894 mutex_exit(&bufcache_lock);
895 return (error);
896 }
897 mutex_enter(bp->b_objlock);
898 if (bp->b_oflags & BO_DELWRI) {
899 bp->b_oflags &= ~BO_DELWRI;
900 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
901 wakeup(&fs->lfs_availsleep);
902 }
903 mutex_exit(bp->b_objlock);
904 LFS_UNLOCK_BUF(bp);
905 brelsel(bp, BC_INVAL | BC_VFLUSH);
906 }
907
908 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
909 nbp = LIST_NEXT(bp, b_vnbufs);
910 if (bp->b_lblkno < lbn)
911 continue;
912 error = bbusy(bp, catch, slptimeo, NULL);
913 if (error == EPASSTHROUGH)
914 goto restart;
915 if (error != 0) {
916 mutex_exit(&bufcache_lock);
917 return (error);
918 }
919 mutex_enter(bp->b_objlock);
920 if (bp->b_oflags & BO_DELWRI) {
921 bp->b_oflags &= ~BO_DELWRI;
922 lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
923 wakeup(&fs->lfs_availsleep);
924 }
925 mutex_exit(bp->b_objlock);
926 LFS_UNLOCK_BUF(bp);
927 brelsel(bp, BC_INVAL | BC_VFLUSH);
928 }
929 mutex_exit(&bufcache_lock);
930
931 return (0);
932}
933
934