lfs_inode.c revision 1.41 1 1.41 perseant /* $NetBSD: lfs_inode.c,v 1.41 2000/07/04 22:30:37 perseant Exp $ */
2 1.2 cgd
3 1.18 perseant /*-
4 1.18 perseant * Copyright (c) 1999 The NetBSD Foundation, Inc.
5 1.18 perseant * All rights reserved.
6 1.18 perseant *
7 1.18 perseant * This code is derived from software contributed to The NetBSD Foundation
8 1.18 perseant * by Konrad E. Schroder <perseant (at) hhhh.org>.
9 1.18 perseant *
10 1.18 perseant * Redistribution and use in source and binary forms, with or without
11 1.18 perseant * modification, are permitted provided that the following conditions
12 1.18 perseant * are met:
13 1.18 perseant * 1. Redistributions of source code must retain the above copyright
14 1.18 perseant * notice, this list of conditions and the following disclaimer.
15 1.18 perseant * 2. Redistributions in binary form must reproduce the above copyright
16 1.18 perseant * notice, this list of conditions and the following disclaimer in the
17 1.18 perseant * documentation and/or other materials provided with the distribution.
18 1.18 perseant * 3. All advertising materials mentioning features or use of this software
19 1.18 perseant * must display the following acknowledgement:
20 1.18 perseant * This product includes software developed by the NetBSD
21 1.18 perseant * Foundation, Inc. and its contributors.
22 1.18 perseant * 4. Neither the name of The NetBSD Foundation nor the names of its
23 1.18 perseant * contributors may be used to endorse or promote products derived
24 1.18 perseant * from this software without specific prior written permission.
25 1.18 perseant *
26 1.18 perseant * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 1.18 perseant * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 1.18 perseant * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 1.18 perseant * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 1.18 perseant * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 1.18 perseant * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 1.18 perseant * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 1.18 perseant * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 1.18 perseant * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 1.18 perseant * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 1.18 perseant * POSSIBILITY OF SUCH DAMAGE.
37 1.18 perseant */
38 1.1 mycroft /*
39 1.1 mycroft * Copyright (c) 1986, 1989, 1991, 1993
40 1.1 mycroft * The Regents of the University of California. All rights reserved.
41 1.1 mycroft *
42 1.1 mycroft * Redistribution and use in source and binary forms, with or without
43 1.1 mycroft * modification, are permitted provided that the following conditions
44 1.1 mycroft * are met:
45 1.1 mycroft * 1. Redistributions of source code must retain the above copyright
46 1.1 mycroft * notice, this list of conditions and the following disclaimer.
47 1.1 mycroft * 2. Redistributions in binary form must reproduce the above copyright
48 1.1 mycroft * notice, this list of conditions and the following disclaimer in the
49 1.1 mycroft * documentation and/or other materials provided with the distribution.
50 1.1 mycroft * 3. All advertising materials mentioning features or use of this software
51 1.1 mycroft * must display the following acknowledgement:
52 1.1 mycroft * This product includes software developed by the University of
53 1.1 mycroft * California, Berkeley and its contributors.
54 1.1 mycroft * 4. Neither the name of the University nor the names of its contributors
55 1.1 mycroft * may be used to endorse or promote products derived from this software
56 1.1 mycroft * without specific prior written permission.
57 1.1 mycroft *
58 1.1 mycroft * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59 1.1 mycroft * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60 1.1 mycroft * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61 1.1 mycroft * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62 1.1 mycroft * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63 1.1 mycroft * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64 1.1 mycroft * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65 1.1 mycroft * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66 1.1 mycroft * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67 1.1 mycroft * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68 1.1 mycroft * SUCH DAMAGE.
69 1.1 mycroft *
70 1.12 fvdl * @(#)lfs_inode.c 8.9 (Berkeley) 5/8/95
71 1.1 mycroft */
72 1.13 scottr
73 1.14 scottr #if defined(_KERNEL) && !defined(_LKM)
74 1.13 scottr #include "opt_quota.h"
75 1.14 scottr #endif
76 1.1 mycroft
77 1.1 mycroft #include <sys/param.h>
78 1.1 mycroft #include <sys/systm.h>
79 1.1 mycroft #include <sys/mount.h>
80 1.1 mycroft #include <sys/proc.h>
81 1.1 mycroft #include <sys/file.h>
82 1.1 mycroft #include <sys/buf.h>
83 1.1 mycroft #include <sys/vnode.h>
84 1.1 mycroft #include <sys/kernel.h>
85 1.1 mycroft #include <sys/malloc.h>
86 1.38 perseant #include <sys/trace.h>
87 1.38 perseant #include <sys/resourcevar.h>
88 1.1 mycroft
89 1.1 mycroft #include <ufs/ufs/quota.h>
90 1.1 mycroft #include <ufs/ufs/inode.h>
91 1.1 mycroft #include <ufs/ufs/ufsmount.h>
92 1.1 mycroft #include <ufs/ufs/ufs_extern.h>
93 1.1 mycroft
94 1.1 mycroft #include <ufs/lfs/lfs.h>
95 1.1 mycroft #include <ufs/lfs/lfs_extern.h>
96 1.1 mycroft
97 1.35 perseant extern int locked_queue_count;
98 1.35 perseant extern long locked_queue_bytes;
99 1.21 perseant
100 1.38 perseant static int lfs_update_seguse(struct lfs *, long, size_t);
101 1.38 perseant static int lfs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t,
102 1.40 perseant ufs_daddr_t, int, long *, long *, long *, size_t *);
103 1.38 perseant static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *);
104 1.38 perseant static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
105 1.38 perseant
106 1.1 mycroft /* Search a block for a specific dinode. */
107 1.1 mycroft struct dinode *
108 1.30 perseant lfs_ifind(fs, ino, bp)
109 1.1 mycroft struct lfs *fs;
110 1.1 mycroft ino_t ino;
111 1.30 perseant struct buf *bp;
112 1.1 mycroft {
113 1.32 augustss int cnt;
114 1.32 augustss struct dinode *dip = (struct dinode *)bp->b_data;
115 1.32 augustss struct dinode *ldip;
116 1.18 perseant
117 1.1 mycroft for (cnt = INOPB(fs), ldip = dip + (cnt - 1); cnt--; --ldip)
118 1.1 mycroft if (ldip->di_inumber == ino)
119 1.1 mycroft return (ldip);
120 1.18 perseant
121 1.40 perseant printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
122 1.40 perseant datosn(fs,fs->lfs_offset));
123 1.40 perseant printf("block is 0x%x (seg %d)\n", bp->b_blkno,
124 1.40 perseant datosn(fs,bp->b_blkno));
125 1.1 mycroft panic("lfs_ifind: dinode %u not found", ino);
126 1.1 mycroft /* NOTREACHED */
127 1.1 mycroft }
128 1.1 mycroft
129 1.1 mycroft int
130 1.4 christos lfs_update(v)
131 1.4 christos void *v;
132 1.4 christos {
133 1.1 mycroft struct vop_update_args /* {
134 1.18 perseant struct vnode *a_vp;
135 1.18 perseant struct timespec *a_access;
136 1.18 perseant struct timespec *a_modify;
137 1.36 perseant int a_flags;
138 1.18 perseant } */ *ap = v;
139 1.1 mycroft struct inode *ip;
140 1.12 fvdl struct vnode *vp = ap->a_vp;
141 1.18 perseant int mod, oflag;
142 1.6 mycroft struct timespec ts;
143 1.25 perseant struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
144 1.18 perseant
145 1.12 fvdl if (vp->v_mount->mnt_flag & MNT_RDONLY)
146 1.1 mycroft return (0);
147 1.12 fvdl ip = VTOI(vp);
148 1.18 perseant
149 1.18 perseant /*
150 1.18 perseant * If we are called from vinvalbuf, and the file's blocks have
151 1.18 perseant * already been scheduled for writing, but the writes have not
152 1.18 perseant * yet completed, lfs_vflush will not be called, and vinvalbuf
153 1.18 perseant * will cause a panic. So, we must wait until any pending write
154 1.36 perseant * for our inode completes, if we are called with UPDATE_WAIT set.
155 1.18 perseant */
156 1.36 perseant while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
157 1.36 perseant WRITEINPROG(vp)) {
158 1.23 perseant #ifdef DEBUG_LFS
159 1.38 perseant printf("lfs_update: sleeping on inode %d (in-progress)\n",
160 1.38 perseant ip->i_number);
161 1.23 perseant #endif
162 1.18 perseant tsleep(vp, (PRIBIO+1), "lfs_update", 0);
163 1.18 perseant }
164 1.37 perseant mod = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
165 1.18 perseant oflag = ip->i_flag;
166 1.6 mycroft TIMEVAL_TO_TIMESPEC(&time, &ts);
167 1.18 perseant LFS_ITIMES(ip,
168 1.18 perseant ap->a_access ? ap->a_access : &ts,
169 1.18 perseant ap->a_modify ? ap->a_modify : &ts, &ts);
170 1.37 perseant if (!mod && (ip->i_flag & (IN_MODIFIED | IN_ACCESSED)))
171 1.6 mycroft ip->i_lfs->lfs_uinodes++;
172 1.37 perseant if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) {
173 1.1 mycroft return (0);
174 1.18 perseant }
175 1.18 perseant
176 1.1 mycroft /* If sync, push back the vnode and any dirty blocks it may have. */
177 1.36 perseant if((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP))==UPDATE_WAIT) {
178 1.25 perseant /* Avoid flushing VDIROP. */
179 1.25 perseant ++fs->lfs_diropwait;
180 1.25 perseant while(vp->v_flag & VDIROP) {
181 1.25 perseant #ifdef DEBUG_LFS
182 1.38 perseant printf("lfs_update: sleeping on inode %d (dirops)\n",
183 1.38 perseant ip->i_number);
184 1.38 perseant printf("lfs_update: vflags 0x%lx, iflags 0x%x\n",
185 1.38 perseant vp->v_flag, ip->i_flag);
186 1.25 perseant #endif
187 1.38 perseant if(fs->lfs_dirops == 0)
188 1.38 perseant lfs_flush_fs(fs, SEGM_SYNC);
189 1.25 perseant else
190 1.38 perseant tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
191 1.38 perseant 0);
192 1.25 perseant /* XXX KS - by falling out here, are we writing the vn
193 1.25 perseant twice? */
194 1.25 perseant }
195 1.25 perseant --fs->lfs_diropwait;
196 1.25 perseant return lfs_vflush(vp);
197 1.25 perseant }
198 1.25 perseant return 0;
199 1.1 mycroft }
200 1.1 mycroft
201 1.38 perseant #define SINGLE 0 /* index of single indirect block */
202 1.38 perseant #define DOUBLE 1 /* index of double indirect block */
203 1.38 perseant #define TRIPLE 2 /* index of triple indirect block */
204 1.1 mycroft /*
205 1.38 perseant * Truncate the inode oip to at most length size, freeing the
206 1.38 perseant * disk blocks.
207 1.1 mycroft */
208 1.1 mycroft int
209 1.4 christos lfs_truncate(v)
210 1.4 christos void *v;
211 1.4 christos {
212 1.1 mycroft struct vop_truncate_args /* {
213 1.1 mycroft struct vnode *a_vp;
214 1.1 mycroft off_t a_length;
215 1.1 mycroft int a_flags;
216 1.1 mycroft struct ucred *a_cred;
217 1.1 mycroft struct proc *a_p;
218 1.4 christos } */ *ap = v;
219 1.38 perseant struct vnode *ovp = ap->a_vp;
220 1.38 perseant ufs_daddr_t lastblock;
221 1.38 perseant struct inode *oip;
222 1.38 perseant ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
223 1.38 perseant ufs_daddr_t newblks[NDADDR + NIADDR];
224 1.1 mycroft off_t length = ap->a_length;
225 1.1 mycroft struct lfs *fs;
226 1.38 perseant struct buf *bp;
227 1.38 perseant int offset, size, level;
228 1.40 perseant long count, rcount, nblocks, blocksreleased = 0, real_released = 0;
229 1.38 perseant int i;
230 1.38 perseant int aflags, error, allerror = 0;
231 1.38 perseant off_t osize;
232 1.38 perseant long lastseg;
233 1.38 perseant size_t bc;
234 1.38 perseant
235 1.33 perseant if (length < 0)
236 1.33 perseant return (EINVAL);
237 1.38 perseant oip = VTOI(ovp);
238 1.33 perseant
239 1.38 perseant /*
240 1.38 perseant * Just return and not update modification times.
241 1.38 perseant */
242 1.38 perseant if (oip->i_ffs_size == length)
243 1.38 perseant return (0);
244 1.1 mycroft
245 1.38 perseant if (ovp->v_type == VLNK &&
246 1.38 perseant (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen ||
247 1.38 perseant (ovp->v_mount->mnt_maxsymlinklen == 0 &&
248 1.38 perseant oip->i_din.ffs_din.di_blocks == 0))) {
249 1.1 mycroft #ifdef DIAGNOSTIC
250 1.1 mycroft if (length != 0)
251 1.1 mycroft panic("lfs_truncate: partial truncate of symlink");
252 1.1 mycroft #endif
253 1.38 perseant memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size);
254 1.38 perseant oip->i_ffs_size = 0;
255 1.38 perseant oip->i_flag |= IN_CHANGE | IN_UPDATE;
256 1.38 perseant return (VOP_UPDATE(ovp, NULL, NULL, 0));
257 1.38 perseant }
258 1.38 perseant if (oip->i_ffs_size == length) {
259 1.38 perseant oip->i_flag |= IN_CHANGE | IN_UPDATE;
260 1.38 perseant return (VOP_UPDATE(ovp, NULL, NULL, 0));
261 1.1 mycroft }
262 1.38 perseant #ifdef QUOTA
263 1.38 perseant if ((error = getinoquota(oip)) != 0)
264 1.38 perseant return (error);
265 1.38 perseant #endif
266 1.38 perseant fs = oip->i_lfs;
267 1.18 perseant lfs_imtime(fs);
268 1.38 perseant osize = oip->i_ffs_size;
269 1.38 perseant ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
270 1.38 perseant
271 1.38 perseant /*
272 1.38 perseant * Lengthen the size of the file. We must ensure that the
273 1.38 perseant * last byte of the file is allocated. Since the smallest
274 1.38 perseant * value of osize is 0, length will be at least 1.
275 1.38 perseant */
276 1.38 perseant if (osize < length) {
277 1.33 perseant if (length > fs->lfs_maxfilesize)
278 1.33 perseant return (EFBIG);
279 1.38 perseant aflags = B_CLRBUF;
280 1.38 perseant if (ap->a_flags & IO_SYNC)
281 1.38 perseant aflags |= B_SYNC;
282 1.38 perseant error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
283 1.33 perseant if (error)
284 1.38 perseant return (error);
285 1.38 perseant oip->i_ffs_size = length;
286 1.38 perseant uvm_vnp_setsize(ovp, length);
287 1.38 perseant (void) uvm_vnp_uncache(ovp);
288 1.33 perseant VOP_BWRITE(bp);
289 1.38 perseant oip->i_flag |= IN_CHANGE | IN_UPDATE;
290 1.38 perseant return (VOP_UPDATE(ovp, NULL, NULL, 0));
291 1.1 mycroft }
292 1.1 mycroft
293 1.18 perseant /*
294 1.35 perseant * Make sure no writes to this inode can happen while we're
295 1.35 perseant * truncating. Otherwise, blocks which are accounted for on the
296 1.35 perseant * inode *and* which have been created for cleaning can coexist,
297 1.35 perseant * and cause an overcounting.
298 1.20 perseant *
299 1.35 perseant * (We don't need to *hold* the seglock, though, because we already
300 1.35 perseant * hold the inode lock; draining the seglock is sufficient.)
301 1.18 perseant */
302 1.38 perseant if (ovp != fs->lfs_unlockvp) {
303 1.35 perseant while(fs->lfs_seglock) {
304 1.35 perseant tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0);
305 1.35 perseant }
306 1.18 perseant }
307 1.18 perseant
308 1.1 mycroft /*
309 1.38 perseant * Shorten the size of the file. If the file is not being
310 1.38 perseant * truncated to a block boundary, the contents of the
311 1.38 perseant * partial block following the end of the file must be
312 1.38 perseant * zero'ed in case it ever becomes accessible again because
313 1.38 perseant * of subsequent file growth. Directories however are not
314 1.38 perseant * zero'ed as they should grow back initialized to empty.
315 1.38 perseant */
316 1.38 perseant offset = blkoff(fs, length);
317 1.38 perseant lastseg = -1;
318 1.38 perseant bc = 0;
319 1.38 perseant if (offset == 0) {
320 1.38 perseant oip->i_ffs_size = length;
321 1.38 perseant } else {
322 1.38 perseant lbn = lblkno(fs, length);
323 1.38 perseant aflags = B_CLRBUF;
324 1.38 perseant if (ap->a_flags & IO_SYNC)
325 1.38 perseant aflags |= B_SYNC;
326 1.38 perseant error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
327 1.38 perseant if (error)
328 1.38 perseant return (error);
329 1.38 perseant oip->i_ffs_size = length;
330 1.38 perseant size = blksize(fs, oip, lbn);
331 1.38 perseant (void) uvm_vnp_uncache(ovp);
332 1.38 perseant if (ovp->v_type != VDIR)
333 1.38 perseant memset((char *)bp->b_data + offset, 0,
334 1.38 perseant (u_int)(size - offset));
335 1.38 perseant allocbuf(bp, size);
336 1.38 perseant VOP_BWRITE(bp);
337 1.38 perseant }
338 1.38 perseant uvm_vnp_setsize(ovp, length);
339 1.38 perseant /*
340 1.38 perseant * Calculate index into inode's block list of
341 1.38 perseant * last direct and indirect blocks (if any)
342 1.38 perseant * which we want to keep. Lastblock is -1 when
343 1.38 perseant * the file is truncated to 0.
344 1.38 perseant */
345 1.38 perseant lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
346 1.38 perseant lastiblock[SINGLE] = lastblock - NDADDR;
347 1.38 perseant lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
348 1.38 perseant lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
349 1.38 perseant nblocks = btodb(fs->lfs_bsize);
350 1.38 perseant /*
351 1.38 perseant * Record changed file and block pointers before we start
352 1.38 perseant * freeing blocks. lastiblock values are also normalized to -1
353 1.38 perseant * for calls to lfs_indirtrunc below.
354 1.38 perseant */
355 1.38 perseant memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks);
356 1.38 perseant for (level = TRIPLE; level >= SINGLE; level--)
357 1.38 perseant if (lastiblock[level] < 0) {
358 1.38 perseant newblks[NDADDR+level] = 0;
359 1.38 perseant lastiblock[level] = -1;
360 1.38 perseant }
361 1.38 perseant for (i = NDADDR - 1; i > lastblock; i--)
362 1.38 perseant newblks[i] = 0;
363 1.38 perseant
364 1.38 perseant oip->i_ffs_size = osize;
365 1.38 perseant error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
366 1.38 perseant if (error && !allerror)
367 1.38 perseant allerror = error;
368 1.1 mycroft
369 1.1 mycroft /*
370 1.38 perseant * Indirect blocks first.
371 1.1 mycroft */
372 1.38 perseant indir_lbn[SINGLE] = -NDADDR;
373 1.38 perseant indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
374 1.38 perseant indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
375 1.38 perseant for (level = TRIPLE; level >= SINGLE; level--) {
376 1.38 perseant bn = oip->i_ffs_ib[level];
377 1.38 perseant if (bn != 0) {
378 1.38 perseant error = lfs_indirtrunc(oip, indir_lbn[level],
379 1.40 perseant bn, lastiblock[level],
380 1.40 perseant level, &count, &rcount,
381 1.40 perseant &lastseg, &bc);
382 1.38 perseant if (error)
383 1.38 perseant allerror = error;
384 1.40 perseant real_released += rcount;
385 1.38 perseant blocksreleased += count;
386 1.38 perseant if (lastiblock[level] < 0) {
387 1.40 perseant if (oip->i_ffs_ib[level] > 0)
388 1.40 perseant real_released += nblocks;
389 1.40 perseant blocksreleased += nblocks;
390 1.38 perseant oip->i_ffs_ib[level] = 0;
391 1.38 perseant lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc);
392 1.38 perseant }
393 1.38 perseant }
394 1.38 perseant if (lastiblock[level] >= 0)
395 1.38 perseant goto done;
396 1.38 perseant }
397 1.38 perseant
398 1.38 perseant /*
399 1.38 perseant * All whole direct blocks or frags.
400 1.38 perseant */
401 1.38 perseant for (i = NDADDR - 1; i > lastblock; i--) {
402 1.38 perseant long bsize;
403 1.12 fvdl
404 1.38 perseant bn = oip->i_ffs_db[i];
405 1.38 perseant if (bn == 0)
406 1.38 perseant continue;
407 1.40 perseant bsize = blksize(fs, oip, i);
408 1.40 perseant if (oip->i_ffs_db[i] > 0)
409 1.40 perseant real_released += btodb(bsize);
410 1.40 perseant blocksreleased += btodb(bsize);
411 1.38 perseant oip->i_ffs_db[i] = 0;
412 1.38 perseant lfs_blkfree(fs, bn, bsize, &lastseg, &bc);
413 1.1 mycroft }
414 1.38 perseant if (lastblock < 0)
415 1.38 perseant goto done;
416 1.38 perseant
417 1.1 mycroft /*
418 1.38 perseant * Finally, look for a change in size of the
419 1.38 perseant * last direct block; release any frags.
420 1.1 mycroft */
421 1.38 perseant bn = oip->i_ffs_db[lastblock];
422 1.38 perseant if (bn != 0) {
423 1.38 perseant long oldspace, newspace;
424 1.1 mycroft
425 1.38 perseant /*
426 1.38 perseant * Calculate amount of space we're giving
427 1.38 perseant * back as old block size minus new block size.
428 1.38 perseant */
429 1.38 perseant oldspace = blksize(fs, oip, lastblock);
430 1.38 perseant oip->i_ffs_size = length;
431 1.38 perseant newspace = blksize(fs, oip, lastblock);
432 1.38 perseant if (newspace == 0)
433 1.38 perseant panic("itrunc: newspace");
434 1.38 perseant if (oldspace - newspace > 0) {
435 1.38 perseant lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc);
436 1.40 perseant real_released += btodb(oldspace - newspace);
437 1.38 perseant blocksreleased += btodb(oldspace - newspace);
438 1.1 mycroft }
439 1.1 mycroft }
440 1.38 perseant
441 1.38 perseant done:
442 1.38 perseant /* Finish segment accounting corrections */
443 1.38 perseant lfs_update_seguse(fs, lastseg, bc);
444 1.38 perseant #ifdef DIAGNOSTIC
445 1.38 perseant for (level = SINGLE; level <= TRIPLE; level++)
446 1.38 perseant if (newblks[NDADDR + level] != oip->i_ffs_ib[level])
447 1.38 perseant panic("lfs itrunc1");
448 1.38 perseant for (i = 0; i < NDADDR; i++)
449 1.38 perseant if (newblks[i] != oip->i_ffs_db[i])
450 1.38 perseant panic("lfs itrunc2");
451 1.38 perseant if (length == 0 &&
452 1.38 perseant (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
453 1.38 perseant panic("lfs itrunc3");
454 1.38 perseant #endif /* DIAGNOSTIC */
455 1.38 perseant /*
456 1.38 perseant * Put back the real size.
457 1.38 perseant */
458 1.38 perseant oip->i_ffs_size = length;
459 1.40 perseant oip->i_lfs_effnblks -= blocksreleased;
460 1.40 perseant oip->i_ffs_blocks -= real_released;
461 1.38 perseant fs->lfs_bfree += blocksreleased;
462 1.38 perseant #ifdef DIAGNOSTIC
463 1.38 perseant if (oip->i_ffs_size == 0 && oip->i_ffs_blocks > 0) {
464 1.38 perseant printf("lfs_tuncate: truncate to 0 but %d blocks on inode\n",
465 1.38 perseant oip->i_ffs_blocks);
466 1.38 perseant panic("lfs_truncate: persistent blocks\n");
467 1.1 mycroft }
468 1.38 perseant #endif
469 1.38 perseant if (oip->i_ffs_blocks < 0) {
470 1.1 mycroft #ifdef DIAGNOSTIC
471 1.38 perseant panic("lfs_truncate: negative block count\n");
472 1.38 perseant #endif
473 1.38 perseant oip->i_ffs_blocks = 0;
474 1.1 mycroft }
475 1.38 perseant oip->i_flag |= IN_CHANGE;
476 1.38 perseant #ifdef QUOTA
477 1.38 perseant (void) chkdq(oip, -blocksreleased, NOCRED, 0);
478 1.1 mycroft #endif
479 1.38 perseant return (allerror);
480 1.38 perseant }
481 1.28 fvdl
482 1.38 perseant /* Update segment usage information when removing a block. */
483 1.38 perseant static int
484 1.38 perseant lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg,
485 1.38 perseant size_t *num)
486 1.38 perseant {
487 1.38 perseant long seg;
488 1.38 perseant int error = 0;
489 1.24 perseant
490 1.38 perseant bsize = fragroundup(fs, bsize);
491 1.38 perseant if (daddr > 0) {
492 1.38 perseant if (*lastseg != (seg = datosn(fs, daddr))) {
493 1.38 perseant error = lfs_update_seguse(fs, *lastseg, *num);
494 1.38 perseant *num = bsize;
495 1.38 perseant *lastseg = seg;
496 1.38 perseant } else
497 1.38 perseant *num += bsize;
498 1.18 perseant }
499 1.38 perseant return error;
500 1.38 perseant }
501 1.38 perseant
502 1.38 perseant /* Finish the accounting updates for a segment. */
503 1.38 perseant static int
504 1.38 perseant lfs_update_seguse(struct lfs *fs, long lastseg, size_t num)
505 1.38 perseant {
506 1.38 perseant SEGUSE *sup;
507 1.38 perseant struct buf *bp;
508 1.38 perseant
509 1.38 perseant if (lastseg < 0 || num == 0)
510 1.38 perseant return 0;
511 1.35 perseant
512 1.38 perseant
513 1.38 perseant LFS_SEGENTRY(sup, fs, lastseg, bp);
514 1.38 perseant if (num > sup->su_nbytes) {
515 1.38 perseant printf("lfs_truncate: segment %ld short by %ld\n",
516 1.38 perseant lastseg, (long)num - sup->su_nbytes);
517 1.38 perseant panic("lfs_truncate: negative bytes");
518 1.38 perseant sup->su_nbytes = num;
519 1.12 fvdl }
520 1.38 perseant sup->su_nbytes -= num;
521 1.38 perseant return (VOP_BWRITE(bp));
522 1.38 perseant }
523 1.38 perseant
524 1.38 perseant /*
525 1.38 perseant * Release blocks associated with the inode ip and stored in the indirect
526 1.38 perseant * block bn. Blocks are free'd in LIFO order up to (but not including)
527 1.38 perseant * lastbn. If level is greater than SINGLE, the block is an indirect block
528 1.38 perseant * and recursive calls to indirtrunc must be used to cleanse other indirect
529 1.38 perseant * blocks.
530 1.38 perseant *
531 1.38 perseant * NB: triple indirect blocks are untested.
532 1.38 perseant */
533 1.38 perseant static int
534 1.38 perseant lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn,
535 1.38 perseant ufs_daddr_t lastbn, int level, long *countp,
536 1.40 perseant long *rcountp, long *lastsegp, size_t *bcp)
537 1.38 perseant {
538 1.38 perseant int i;
539 1.38 perseant struct buf *bp;
540 1.38 perseant struct lfs *fs = ip->i_lfs;
541 1.38 perseant ufs_daddr_t *bap;
542 1.38 perseant struct vnode *vp;
543 1.38 perseant ufs_daddr_t *copy = NULL, nb, nlbn, last;
544 1.40 perseant long blkcount, rblkcount, factor;
545 1.40 perseant int nblocks, blocksreleased = 0, real_released = 0;
546 1.38 perseant int error = 0, allerror = 0;
547 1.38 perseant
548 1.38 perseant /*
549 1.38 perseant * Calculate index in current block of last
550 1.38 perseant * block to be kept. -1 indicates the entire
551 1.38 perseant * block so we need not calculate the index.
552 1.38 perseant */
553 1.38 perseant factor = 1;
554 1.38 perseant for (i = SINGLE; i < level; i++)
555 1.38 perseant factor *= NINDIR(fs);
556 1.38 perseant last = lastbn;
557 1.38 perseant if (lastbn > 0)
558 1.38 perseant last /= factor;
559 1.38 perseant nblocks = btodb(fs->lfs_bsize);
560 1.38 perseant /*
561 1.38 perseant * Get buffer of block pointers, zero those entries corresponding
562 1.38 perseant * to blocks to be free'd, and update on disk copy first. Since
563 1.38 perseant * double(triple) indirect before single(double) indirect, calls
564 1.38 perseant * to bmap on these blocks will fail. However, we already have
565 1.38 perseant * the on disk address, so we have to set the b_blkno field
566 1.38 perseant * explicitly instead of letting bread do everything for us.
567 1.38 perseant */
568 1.38 perseant vp = ITOV(ip);
569 1.38 perseant bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
570 1.38 perseant if (bp->b_flags & (B_DONE | B_DELWRI)) {
571 1.38 perseant /* Braces must be here in case trace evaluates to nothing. */
572 1.38 perseant trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
573 1.38 perseant } else {
574 1.38 perseant trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
575 1.38 perseant curproc->p_stats->p_ru.ru_inblock++; /* pay for read */
576 1.38 perseant bp->b_flags |= B_READ;
577 1.38 perseant if (bp->b_bcount > bp->b_bufsize)
578 1.38 perseant panic("lfs_indirtrunc: bad buffer size");
579 1.38 perseant bp->b_blkno = dbn;
580 1.38 perseant VOP_STRATEGY(bp);
581 1.38 perseant error = biowait(bp);
582 1.38 perseant }
583 1.38 perseant if (error) {
584 1.38 perseant brelse(bp);
585 1.40 perseant *countp = *rcountp = 0;
586 1.38 perseant return (error);
587 1.38 perseant }
588 1.38 perseant
589 1.38 perseant bap = (ufs_daddr_t *)bp->b_data;
590 1.38 perseant if (lastbn >= 0) {
591 1.38 perseant MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK);
592 1.38 perseant memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
593 1.38 perseant memset((caddr_t)&bap[last + 1], 0,
594 1.38 perseant (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
595 1.38 perseant error = VOP_BWRITE(bp);
596 1.38 perseant if (error)
597 1.38 perseant allerror = error;
598 1.38 perseant bap = copy;
599 1.35 perseant }
600 1.35 perseant
601 1.38 perseant /*
602 1.38 perseant * Recursively free totally unused blocks.
603 1.38 perseant */
604 1.38 perseant for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
605 1.38 perseant i--, nlbn += factor) {
606 1.38 perseant nb = bap[i];
607 1.38 perseant if (nb == 0)
608 1.38 perseant continue;
609 1.38 perseant if (level > SINGLE) {
610 1.38 perseant error = lfs_indirtrunc(ip, nlbn, nb,
611 1.38 perseant (ufs_daddr_t)-1, level - 1,
612 1.40 perseant &blkcount, &rblkcount,
613 1.40 perseant lastsegp, bcp);
614 1.38 perseant if (error)
615 1.38 perseant allerror = error;
616 1.38 perseant blocksreleased += blkcount;
617 1.40 perseant real_released += rblkcount;
618 1.38 perseant }
619 1.38 perseant lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp);
620 1.40 perseant if (bap[i] > 0)
621 1.40 perseant real_released += nblocks;
622 1.38 perseant blocksreleased += nblocks;
623 1.35 perseant }
624 1.35 perseant
625 1.38 perseant /*
626 1.38 perseant * Recursively free last partial block.
627 1.38 perseant */
628 1.38 perseant if (level > SINGLE && lastbn >= 0) {
629 1.38 perseant last = lastbn % factor;
630 1.38 perseant nb = bap[i];
631 1.38 perseant if (nb != 0) {
632 1.38 perseant error = lfs_indirtrunc(ip, nlbn, nb,
633 1.38 perseant last, level - 1, &blkcount,
634 1.40 perseant &rblkcount, lastsegp, bcp);
635 1.38 perseant if (error)
636 1.38 perseant allerror = error;
637 1.40 perseant real_released += rblkcount;
638 1.38 perseant blocksreleased += blkcount;
639 1.38 perseant }
640 1.38 perseant }
641 1.35 perseant
642 1.38 perseant if (copy != NULL) {
643 1.38 perseant FREE(copy, M_TEMP);
644 1.38 perseant } else {
645 1.38 perseant bp->b_flags |= B_INVAL;
646 1.38 perseant brelse(bp);
647 1.38 perseant }
648 1.18 perseant
649 1.38 perseant *countp = blocksreleased;
650 1.40 perseant *rcountp = real_released;
651 1.38 perseant return (allerror);
652 1.21 perseant }
653 1.21 perseant
654 1.21 perseant /*
655 1.38 perseant * Destroy any in core blocks past the truncation length.
656 1.38 perseant * Inlined from vtruncbuf, so that lfs_avail could be updated.
657 1.21 perseant */
658 1.38 perseant static int
659 1.38 perseant lfs_vtruncbuf(vp, lbn, slpflag, slptimeo)
660 1.32 augustss struct vnode *vp;
661 1.38 perseant daddr_t lbn;
662 1.38 perseant int slpflag, slptimeo;
663 1.21 perseant {
664 1.38 perseant struct buf *bp, *nbp;
665 1.38 perseant int s, error;
666 1.38 perseant struct lfs *fs;
667 1.38 perseant
668 1.38 perseant fs = VTOI(vp)->i_lfs;
669 1.38 perseant s = splbio();
670 1.21 perseant
671 1.38 perseant restart:
672 1.38 perseant for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
673 1.38 perseant nbp = LIST_NEXT(bp, b_vnbufs);
674 1.38 perseant if (bp->b_lblkno < lbn)
675 1.38 perseant continue;
676 1.38 perseant if (bp->b_flags & B_BUSY) {
677 1.38 perseant bp->b_flags |= B_WANTED;
678 1.38 perseant error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
679 1.38 perseant "lfs_vtruncbuf", slptimeo);
680 1.38 perseant if (error) {
681 1.22 perseant splx(s);
682 1.38 perseant return (error);
683 1.22 perseant }
684 1.38 perseant goto restart;
685 1.38 perseant }
686 1.38 perseant bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
687 1.38 perseant if (bp->b_flags & B_DELWRI)
688 1.41 perseant fs->lfs_avail += btodb(bp->b_bcount);
689 1.38 perseant if (bp->b_flags & B_LOCKED) {
690 1.38 perseant bp->b_flags &= ~B_LOCKED;
691 1.38 perseant --locked_queue_count;
692 1.38 perseant locked_queue_bytes -= bp->b_bcount;
693 1.38 perseant }
694 1.38 perseant brelse(bp);
695 1.38 perseant }
696 1.38 perseant
697 1.38 perseant for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
698 1.38 perseant nbp = LIST_NEXT(bp, b_vnbufs);
699 1.38 perseant if (bp->b_lblkno < lbn)
700 1.38 perseant continue;
701 1.38 perseant if (bp->b_flags & B_BUSY) {
702 1.38 perseant bp->b_flags |= B_WANTED;
703 1.38 perseant error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
704 1.38 perseant "lfs_vtruncbuf", slptimeo);
705 1.38 perseant if (error) {
706 1.35 perseant splx(s);
707 1.38 perseant return (error);
708 1.21 perseant }
709 1.38 perseant goto restart;
710 1.38 perseant }
711 1.38 perseant bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
712 1.38 perseant if (bp->b_flags & B_DELWRI)
713 1.41 perseant fs->lfs_avail += btodb(bp->b_bcount);
714 1.38 perseant if (bp->b_flags & B_LOCKED) {
715 1.38 perseant bp->b_flags &= ~B_LOCKED;
716 1.38 perseant --locked_queue_count;
717 1.38 perseant locked_queue_bytes -= bp->b_bcount;
718 1.21 perseant }
719 1.38 perseant brelse(bp);
720 1.21 perseant }
721 1.38 perseant
722 1.38 perseant splx(s);
723 1.38 perseant
724 1.21 perseant return (0);
725 1.1 mycroft }
726 1.38 perseant
727