ffs_inode.c revision 1.71 1 1.71 mycroft /* $NetBSD: ffs_inode.c,v 1.71 2004/08/15 19:01:16 mycroft Exp $ */
2 1.5 cgd
3 1.1 mycroft /*
4 1.1 mycroft * Copyright (c) 1982, 1986, 1989, 1993
5 1.1 mycroft * The Regents of the University of California. All rights reserved.
6 1.1 mycroft *
7 1.1 mycroft * Redistribution and use in source and binary forms, with or without
8 1.1 mycroft * modification, are permitted provided that the following conditions
9 1.1 mycroft * are met:
10 1.1 mycroft * 1. Redistributions of source code must retain the above copyright
11 1.1 mycroft * notice, this list of conditions and the following disclaimer.
12 1.1 mycroft * 2. Redistributions in binary form must reproduce the above copyright
13 1.1 mycroft * notice, this list of conditions and the following disclaimer in the
14 1.1 mycroft * documentation and/or other materials provided with the distribution.
15 1.60 agc * 3. Neither the name of the University nor the names of its contributors
16 1.1 mycroft * may be used to endorse or promote products derived from this software
17 1.1 mycroft * without specific prior written permission.
18 1.1 mycroft *
19 1.1 mycroft * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 1.1 mycroft * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 1.1 mycroft * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 1.1 mycroft * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 1.1 mycroft * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 1.1 mycroft * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 1.1 mycroft * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 1.1 mycroft * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 1.1 mycroft * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 1.1 mycroft * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 1.1 mycroft * SUCH DAMAGE.
30 1.1 mycroft *
31 1.18 fvdl * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95
32 1.1 mycroft */
33 1.46 lukem
34 1.46 lukem #include <sys/cdefs.h>
35 1.71 mycroft __KERNEL_RCSID(0, "$NetBSD: ffs_inode.c,v 1.71 2004/08/15 19:01:16 mycroft Exp $");
36 1.17 mrg
37 1.41 mrg #if defined(_KERNEL_OPT)
38 1.25 thorpej #include "opt_ffs.h"
39 1.20 scottr #include "opt_quota.h"
40 1.21 scottr #endif
41 1.1 mycroft
42 1.1 mycroft #include <sys/param.h>
43 1.1 mycroft #include <sys/systm.h>
44 1.1 mycroft #include <sys/mount.h>
45 1.1 mycroft #include <sys/proc.h>
46 1.1 mycroft #include <sys/file.h>
47 1.1 mycroft #include <sys/buf.h>
48 1.1 mycroft #include <sys/vnode.h>
49 1.1 mycroft #include <sys/kernel.h>
50 1.1 mycroft #include <sys/malloc.h>
51 1.1 mycroft #include <sys/trace.h>
52 1.1 mycroft #include <sys/resourcevar.h>
53 1.16 mrg
54 1.1 mycroft #include <ufs/ufs/quota.h>
55 1.1 mycroft #include <ufs/ufs/inode.h>
56 1.1 mycroft #include <ufs/ufs/ufsmount.h>
57 1.1 mycroft #include <ufs/ufs/ufs_extern.h>
58 1.19 bouyer #include <ufs/ufs/ufs_bswap.h>
59 1.1 mycroft
60 1.1 mycroft #include <ufs/ffs/fs.h>
61 1.1 mycroft #include <ufs/ffs/ffs_extern.h>
62 1.1 mycroft
63 1.53 fvdl static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t,
64 1.55 fvdl daddr_t, int, int64_t *));
65 1.1 mycroft
66 1.1 mycroft /*
67 1.13 tls * Update the access, modified, and inode change times as specified
68 1.13 tls * by the IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.
69 1.13 tls * The IN_MODIFIED flag is used to specify that the inode needs to be
70 1.13 tls * updated but that the times have already been set. The access
71 1.13 tls * and modified times are taken from the second and third parameters;
72 1.13 tls * the inode change time is always taken from the current time. If
73 1.31 perseant * UPDATE_WAIT flag is set, or UPDATE_DIROP is set and we are not doing
74 1.31 perseant * softupdates, then wait for the disk write of the inode to complete.
75 1.1 mycroft */
76 1.13 tls
77 1.1 mycroft int
78 1.9 christos ffs_update(v)
79 1.9 christos void *v;
80 1.9 christos {
81 1.1 mycroft struct vop_update_args /* {
82 1.1 mycroft struct vnode *a_vp;
83 1.10 mycroft struct timespec *a_access;
84 1.10 mycroft struct timespec *a_modify;
85 1.31 perseant int a_flags;
86 1.9 christos } */ *ap = v;
87 1.30 augustss struct fs *fs;
88 1.1 mycroft struct buf *bp;
89 1.1 mycroft struct inode *ip;
90 1.1 mycroft int error;
91 1.11 mycroft struct timespec ts;
92 1.24 thorpej caddr_t cp;
93 1.34 mycroft int waitfor, flags;
94 1.1 mycroft
95 1.11 mycroft if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
96 1.11 mycroft return (0);
97 1.1 mycroft ip = VTOI(ap->a_vp);
98 1.11 mycroft TIMEVAL_TO_TIMESPEC(&time, &ts);
99 1.26 mycroft FFS_ITIMES(ip,
100 1.26 mycroft ap->a_access ? ap->a_access : &ts,
101 1.26 mycroft ap->a_modify ? ap->a_modify : &ts, &ts);
102 1.65 mycroft if (ap->a_flags & UPDATE_CLOSE)
103 1.65 mycroft flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
104 1.65 mycroft else
105 1.65 mycroft flags = ip->i_flag & IN_MODIFIED;
106 1.34 mycroft if (flags == 0)
107 1.1 mycroft return (0);
108 1.1 mycroft fs = ip->i_fs;
109 1.31 perseant
110 1.35 mycroft if ((flags & IN_MODIFIED) != 0 &&
111 1.35 mycroft (ap->a_vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
112 1.35 mycroft waitfor = ap->a_flags & UPDATE_WAIT;
113 1.35 mycroft if ((ap->a_flags & UPDATE_DIROP) && !DOINGSOFTDEP(ap->a_vp))
114 1.35 mycroft waitfor |= UPDATE_WAIT;
115 1.35 mycroft } else
116 1.35 mycroft waitfor = 0;
117 1.31 perseant
118 1.1 mycroft /*
119 1.1 mycroft * Ensure that uid and gid are correct. This is a temporary
120 1.1 mycroft * fix until fsck has been changed to do the update.
121 1.1 mycroft */
122 1.55 fvdl if (fs->fs_magic == FS_UFS1_MAGIC && /* XXX */
123 1.55 fvdl fs->fs_old_inodefmt < FS_44INODEFMT) { /* XXX */
124 1.55 fvdl ip->i_ffs1_ouid = ip->i_uid; /* XXX */
125 1.55 fvdl ip->i_ffs1_ogid = ip->i_gid; /* XXX */
126 1.18 fvdl } /* XXX */
127 1.9 christos error = bread(ip->i_devvp,
128 1.9 christos fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
129 1.9 christos (int)fs->fs_bsize, NOCRED, &bp);
130 1.9 christos if (error) {
131 1.1 mycroft brelse(bp);
132 1.1 mycroft return (error);
133 1.1 mycroft }
134 1.35 mycroft ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED);
135 1.29 fvdl if (DOINGSOFTDEP(ap->a_vp))
136 1.31 perseant softdep_update_inodeblock(ip, bp, waitfor);
137 1.55 fvdl else if (ip->i_ffs_effnlink != ip->i_nlink)
138 1.29 fvdl panic("ffs_update: bad link cnt");
139 1.55 fvdl if (fs->fs_magic == FS_UFS1_MAGIC) {
140 1.55 fvdl cp = (caddr_t)bp->b_data +
141 1.55 fvdl (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
142 1.23 christos #ifdef FFS_EI
143 1.55 fvdl if (UFS_FSNEEDSWAP(fs))
144 1.55 fvdl ffs_dinode1_swap(ip->i_din.ffs1_din,
145 1.55 fvdl (struct ufs1_dinode *)cp);
146 1.55 fvdl else
147 1.55 fvdl #endif
148 1.55 fvdl memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
149 1.55 fvdl } else {
150 1.55 fvdl cp = (caddr_t)bp->b_data +
151 1.55 fvdl (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
152 1.55 fvdl #ifdef FFS_EI
153 1.55 fvdl if (UFS_FSNEEDSWAP(fs))
154 1.55 fvdl ffs_dinode2_swap(ip->i_din.ffs2_din,
155 1.55 fvdl (struct ufs2_dinode *)cp);
156 1.55 fvdl else
157 1.23 christos #endif
158 1.55 fvdl memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
159 1.55 fvdl }
160 1.35 mycroft if (waitfor) {
161 1.1 mycroft return (bwrite(bp));
162 1.29 fvdl } else {
163 1.1 mycroft bdwrite(bp);
164 1.1 mycroft return (0);
165 1.1 mycroft }
166 1.1 mycroft }
167 1.1 mycroft
168 1.1 mycroft #define SINGLE 0 /* index of single indirect block */
169 1.1 mycroft #define DOUBLE 1 /* index of double indirect block */
170 1.1 mycroft #define TRIPLE 2 /* index of triple indirect block */
171 1.1 mycroft /*
172 1.1 mycroft * Truncate the inode oip to at most length size, freeing the
173 1.1 mycroft * disk blocks.
174 1.1 mycroft */
175 1.9 christos int
176 1.9 christos ffs_truncate(v)
177 1.9 christos void *v;
178 1.9 christos {
179 1.1 mycroft struct vop_truncate_args /* {
180 1.1 mycroft struct vnode *a_vp;
181 1.1 mycroft off_t a_length;
182 1.1 mycroft int a_flags;
183 1.1 mycroft struct ucred *a_cred;
184 1.59 fvdl struct proc *a_p;
185 1.9 christos } */ *ap = v;
186 1.30 augustss struct vnode *ovp = ap->a_vp;
187 1.43 chs struct genfs_node *gp = VTOG(ovp);
188 1.53 fvdl daddr_t lastblock;
189 1.68 mycroft struct inode *oip = VTOI(ovp);
190 1.53 fvdl daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
191 1.64 hannken daddr_t blks[NDADDR + NIADDR];
192 1.1 mycroft off_t length = ap->a_length;
193 1.30 augustss struct fs *fs;
194 1.1 mycroft int offset, size, level;
195 1.55 fvdl int64_t count, blocksreleased = 0;
196 1.55 fvdl int i, ioflag, aflag, nblocks;
197 1.38 chs int error, allerror = 0;
198 1.1 mycroft off_t osize;
199 1.67 mycroft int sync;
200 1.68 mycroft struct ufsmount *ump = oip->i_ump;
201 1.1 mycroft
202 1.2 pk if (length < 0)
203 1.3 mycroft return (EINVAL);
204 1.68 mycroft
205 1.1 mycroft if (ovp->v_type == VLNK &&
206 1.68 mycroft (oip->i_size < ump->um_maxsymlinklen ||
207 1.68 mycroft (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
208 1.38 chs KDASSERT(length == 0);
209 1.55 fvdl memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
210 1.55 fvdl oip->i_size = 0;
211 1.57 kristerw DIP_ASSIGN(oip, size, 0);
212 1.1 mycroft oip->i_flag |= IN_CHANGE | IN_UPDATE;
213 1.66 mycroft return (VOP_UPDATE(ovp, NULL, NULL, 0));
214 1.1 mycroft }
215 1.55 fvdl if (oip->i_size == length) {
216 1.1 mycroft oip->i_flag |= IN_CHANGE | IN_UPDATE;
217 1.27 mycroft return (VOP_UPDATE(ovp, NULL, NULL, 0));
218 1.1 mycroft }
219 1.1 mycroft #ifdef QUOTA
220 1.9 christos if ((error = getinoquota(oip)) != 0)
221 1.1 mycroft return (error);
222 1.1 mycroft #endif
223 1.1 mycroft fs = oip->i_fs;
224 1.68 mycroft if (length > ump->um_maxfilesize)
225 1.38 chs return (EFBIG);
226 1.38 chs
227 1.63 hannken if ((oip->i_flags & SF_SNAPSHOT) != 0)
228 1.63 hannken ffs_snapremove(ovp);
229 1.63 hannken
230 1.55 fvdl osize = oip->i_size;
231 1.48 chs ioflag = ap->a_flags;
232 1.50 chs aflag = ioflag & IO_SYNC ? B_SYNC : 0;
233 1.29 fvdl
234 1.38 chs /*
235 1.38 chs * Lengthen the size of the file. We must ensure that the
236 1.38 chs * last byte of the file is allocated. Since the smallest
237 1.38 chs * value of osize is 0, length will be at least 1.
238 1.38 chs */
239 1.38 chs
240 1.38 chs if (osize < length) {
241 1.48 chs if (lblkno(fs, osize) < NDADDR &&
242 1.48 chs lblkno(fs, osize) != lblkno(fs, length) &&
243 1.48 chs blkroundup(fs, osize) != osize) {
244 1.69 mycroft off_t eob;
245 1.69 mycroft
246 1.69 mycroft eob = blkroundup(fs, osize);
247 1.69 mycroft error = ufs_balloc_range(ovp, osize, eob - osize,
248 1.69 mycroft ap->a_cred, aflag);
249 1.68 mycroft if (error)
250 1.48 chs return error;
251 1.48 chs if (ioflag & IO_SYNC) {
252 1.69 mycroft ovp->v_size = eob;
253 1.48 chs simple_lock(&ovp->v_interlock);
254 1.49 chs VOP_PUTPAGES(ovp,
255 1.69 mycroft trunc_page(osize & fs->fs_bmask),
256 1.69 mycroft round_page(eob), PGO_CLEANIT | PGO_SYNCIO);
257 1.48 chs }
258 1.48 chs }
259 1.45 chs error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
260 1.48 chs aflag);
261 1.45 chs if (error) {
262 1.48 chs (void) VOP_TRUNCATE(ovp, osize, ioflag & IO_SYNC,
263 1.59 fvdl ap->a_cred, ap->a_p);
264 1.68 mycroft return (error);
265 1.45 chs }
266 1.43 chs uvm_vnp_setsize(ovp, length);
267 1.38 chs oip->i_flag |= IN_CHANGE | IN_UPDATE;
268 1.55 fvdl KASSERT(ovp->v_size == oip->i_size);
269 1.68 mycroft return (VOP_UPDATE(ovp, NULL, NULL, 0));
270 1.38 chs }
271 1.38 chs
272 1.38 chs /*
273 1.38 chs * When truncating a regular file down to a non-block-aligned size,
274 1.38 chs * we must zero the part of last block which is past the new EOF.
275 1.38 chs * We must synchronously flush the zeroed pages to disk
276 1.38 chs * since the new pages will be invalidated as soon as we
277 1.38 chs * inform the VM system of the new, smaller size.
278 1.48 chs * We must do this before acquiring the GLOCK, since fetching
279 1.38 chs * the pages will acquire the GLOCK internally.
280 1.38 chs * So there is a window where another thread could see a whole
281 1.38 chs * zeroed page past EOF, but that's life.
282 1.38 chs */
283 1.38 chs
284 1.38 chs offset = blkoff(fs, length);
285 1.70 mycroft if (ovp->v_type == VREG && offset != 0 && osize > length) {
286 1.70 mycroft daddr_t lbn;
287 1.38 chs voff_t eoz;
288 1.38 chs
289 1.50 chs error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
290 1.50 chs aflag);
291 1.67 mycroft if (error)
292 1.50 chs return error;
293 1.70 mycroft lbn = lblkno(fs, length);
294 1.70 mycroft size = blksize(fs, oip, lbn);
295 1.70 mycroft eoz = MIN(lblktosize(fs, lbn) + size, osize);
296 1.38 chs uvm_vnp_zerorange(ovp, length, eoz - length);
297 1.70 mycroft if (round_page(eoz) > round_page(length)) {
298 1.70 mycroft simple_lock(&ovp->v_interlock);
299 1.70 mycroft error = VOP_PUTPAGES(ovp, round_page(length),
300 1.70 mycroft round_page(eoz),
301 1.71 mycroft PGO_CLEANIT | PGO_DEACTIVATE |
302 1.71 mycroft ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
303 1.70 mycroft if (error)
304 1.70 mycroft return error;
305 1.70 mycroft }
306 1.38 chs }
307 1.38 chs
308 1.43 chs lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
309 1.38 chs
310 1.29 fvdl if (DOINGSOFTDEP(ovp)) {
311 1.29 fvdl if (length > 0) {
312 1.29 fvdl /*
313 1.29 fvdl * If a file is only partially truncated, then
314 1.29 fvdl * we have to clean up the data structures
315 1.29 fvdl * describing the allocation past the truncation
316 1.29 fvdl * point. Finding and deallocating those structures
317 1.29 fvdl * is a lot of work. Since partial truncation occurs
318 1.29 fvdl * rarely, we solve the problem by syncing the file
319 1.29 fvdl * so that it will have no data structures left.
320 1.29 fvdl */
321 1.29 fvdl if ((error = VOP_FSYNC(ovp, ap->a_cred, FSYNC_WAIT,
322 1.59 fvdl 0, 0, ap->a_p)) != 0) {
323 1.43 chs lockmgr(&gp->g_glock, LK_RELEASE, NULL);
324 1.29 fvdl return (error);
325 1.52 simonb }
326 1.51 fvdl if (oip->i_flag & IN_SPACECOUNTED)
327 1.55 fvdl fs->fs_pendingblocks -= DIP(oip, blocks);
328 1.29 fvdl } else {
329 1.40 augustss uvm_vnp_setsize(ovp, length);
330 1.29 fvdl #ifdef QUOTA
331 1.55 fvdl (void) chkdq(oip, -DIP(oip, blocks), NOCRED, 0);
332 1.29 fvdl #endif
333 1.55 fvdl softdep_setup_freeblocks(oip, length, 0);
334 1.59 fvdl (void) vinvalbuf(ovp, 0, ap->a_cred, ap->a_p, 0, 0);
335 1.43 chs lockmgr(&gp->g_glock, LK_RELEASE, NULL);
336 1.29 fvdl oip->i_flag |= IN_CHANGE | IN_UPDATE;
337 1.29 fvdl return (VOP_UPDATE(ovp, NULL, NULL, 0));
338 1.29 fvdl }
339 1.29 fvdl }
340 1.55 fvdl oip->i_size = length;
341 1.57 kristerw DIP_ASSIGN(oip, size, length);
342 1.18 fvdl uvm_vnp_setsize(ovp, length);
343 1.1 mycroft /*
344 1.1 mycroft * Calculate index into inode's block list of
345 1.1 mycroft * last direct and indirect blocks (if any)
346 1.1 mycroft * which we want to keep. Lastblock is -1 when
347 1.1 mycroft * the file is truncated to 0.
348 1.1 mycroft */
349 1.1 mycroft lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
350 1.1 mycroft lastiblock[SINGLE] = lastblock - NDADDR;
351 1.1 mycroft lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
352 1.1 mycroft lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
353 1.1 mycroft nblocks = btodb(fs->fs_bsize);
354 1.1 mycroft /*
355 1.1 mycroft * Update file and block pointers on disk before we start freeing
356 1.1 mycroft * blocks. If we crash before free'ing blocks below, the blocks
357 1.1 mycroft * will be returned to the free list. lastiblock values are also
358 1.1 mycroft * normalized to -1 for calls to ffs_indirtrunc below.
359 1.1 mycroft */
360 1.67 mycroft sync = 0;
361 1.55 fvdl for (level = TRIPLE; level >= SINGLE; level--) {
362 1.64 hannken blks[NDADDR + level] = DIP(oip, ib[level]);
363 1.67 mycroft if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
364 1.67 mycroft sync = 1;
365 1.57 kristerw DIP_ASSIGN(oip, ib[level], 0);
366 1.1 mycroft lastiblock[level] = -1;
367 1.1 mycroft }
368 1.55 fvdl }
369 1.55 fvdl for (i = 0; i < NDADDR; i++) {
370 1.64 hannken blks[i] = DIP(oip, db[i]);
371 1.67 mycroft if (i > lastblock && blks[i] != 0) {
372 1.67 mycroft sync = 1;
373 1.57 kristerw DIP_ASSIGN(oip, db[i], 0);
374 1.67 mycroft }
375 1.67 mycroft }
376 1.68 mycroft oip->i_flag |= IN_CHANGE | IN_UPDATE;
377 1.67 mycroft if (sync) {
378 1.67 mycroft error = VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT);
379 1.67 mycroft if (error && !allerror)
380 1.67 mycroft allerror = error;
381 1.55 fvdl }
382 1.32 mycroft
383 1.1 mycroft /*
384 1.1 mycroft * Having written the new inode to disk, save its new configuration
385 1.1 mycroft * and put back the old block pointers long enough to process them.
386 1.1 mycroft * Note that we save the new block configuration so we can check it
387 1.1 mycroft * when we are done.
388 1.1 mycroft */
389 1.55 fvdl for (i = 0; i < NDADDR; i++) {
390 1.64 hannken bn = DIP(oip, db[i]);
391 1.64 hannken DIP_ASSIGN(oip, db[i], blks[i]);
392 1.64 hannken blks[i] = bn;
393 1.55 fvdl }
394 1.55 fvdl for (i = 0; i < NIADDR; i++) {
395 1.64 hannken bn = DIP(oip, ib[i]);
396 1.64 hannken DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
397 1.64 hannken blks[NDADDR + i] = bn;
398 1.55 fvdl }
399 1.55 fvdl
400 1.55 fvdl oip->i_size = osize;
401 1.57 kristerw DIP_ASSIGN(oip, size, osize);
402 1.32 mycroft error = vtruncbuf(ovp, lastblock + 1, 0, 0);
403 1.32 mycroft if (error && !allerror)
404 1.32 mycroft allerror = error;
405 1.1 mycroft
406 1.1 mycroft /*
407 1.1 mycroft * Indirect blocks first.
408 1.1 mycroft */
409 1.1 mycroft indir_lbn[SINGLE] = -NDADDR;
410 1.1 mycroft indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
411 1.1 mycroft indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
412 1.1 mycroft for (level = TRIPLE; level >= SINGLE; level--) {
413 1.55 fvdl if (oip->i_ump->um_fstype == UFS1)
414 1.55 fvdl bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
415 1.55 fvdl else
416 1.55 fvdl bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
417 1.1 mycroft if (bn != 0) {
418 1.1 mycroft error = ffs_indirtrunc(oip, indir_lbn[level],
419 1.1 mycroft fsbtodb(fs, bn), lastiblock[level], level, &count);
420 1.1 mycroft if (error)
421 1.1 mycroft allerror = error;
422 1.1 mycroft blocksreleased += count;
423 1.1 mycroft if (lastiblock[level] < 0) {
424 1.57 kristerw DIP_ASSIGN(oip, ib[level], 0);
425 1.63 hannken ffs_blkfree(fs, oip->i_devvp, bn, fs->fs_bsize,
426 1.63 hannken oip->i_number);
427 1.1 mycroft blocksreleased += nblocks;
428 1.1 mycroft }
429 1.1 mycroft }
430 1.1 mycroft if (lastiblock[level] >= 0)
431 1.1 mycroft goto done;
432 1.1 mycroft }
433 1.1 mycroft
434 1.1 mycroft /*
435 1.1 mycroft * All whole direct blocks or frags.
436 1.1 mycroft */
437 1.1 mycroft for (i = NDADDR - 1; i > lastblock; i--) {
438 1.30 augustss long bsize;
439 1.1 mycroft
440 1.55 fvdl if (oip->i_ump->um_fstype == UFS1)
441 1.55 fvdl bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
442 1.55 fvdl else
443 1.55 fvdl bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
444 1.1 mycroft if (bn == 0)
445 1.1 mycroft continue;
446 1.57 kristerw DIP_ASSIGN(oip, db[i], 0);
447 1.1 mycroft bsize = blksize(fs, oip, i);
448 1.63 hannken ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
449 1.1 mycroft blocksreleased += btodb(bsize);
450 1.1 mycroft }
451 1.1 mycroft if (lastblock < 0)
452 1.1 mycroft goto done;
453 1.1 mycroft
454 1.1 mycroft /*
455 1.1 mycroft * Finally, look for a change in size of the
456 1.1 mycroft * last direct block; release any frags.
457 1.1 mycroft */
458 1.55 fvdl if (oip->i_ump->um_fstype == UFS1)
459 1.55 fvdl bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
460 1.55 fvdl else
461 1.55 fvdl bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
462 1.1 mycroft if (bn != 0) {
463 1.1 mycroft long oldspace, newspace;
464 1.1 mycroft
465 1.1 mycroft /*
466 1.1 mycroft * Calculate amount of space we're giving
467 1.1 mycroft * back as old block size minus new block size.
468 1.1 mycroft */
469 1.1 mycroft oldspace = blksize(fs, oip, lastblock);
470 1.55 fvdl oip->i_size = length;
471 1.57 kristerw DIP_ASSIGN(oip, size, length);
472 1.1 mycroft newspace = blksize(fs, oip, lastblock);
473 1.1 mycroft if (newspace == 0)
474 1.1 mycroft panic("itrunc: newspace");
475 1.1 mycroft if (oldspace - newspace > 0) {
476 1.1 mycroft /*
477 1.1 mycroft * Block number of space to be free'd is
478 1.1 mycroft * the old block # plus the number of frags
479 1.1 mycroft * required for the storage we're keeping.
480 1.1 mycroft */
481 1.1 mycroft bn += numfrags(fs, newspace);
482 1.63 hannken ffs_blkfree(fs, oip->i_devvp, bn, oldspace - newspace,
483 1.63 hannken oip->i_number);
484 1.1 mycroft blocksreleased += btodb(oldspace - newspace);
485 1.1 mycroft }
486 1.1 mycroft }
487 1.32 mycroft
488 1.1 mycroft done:
489 1.1 mycroft #ifdef DIAGNOSTIC
490 1.1 mycroft for (level = SINGLE; level <= TRIPLE; level++)
491 1.64 hannken if (blks[NDADDR + level] != DIP(oip, ib[level]))
492 1.1 mycroft panic("itrunc1");
493 1.1 mycroft for (i = 0; i < NDADDR; i++)
494 1.64 hannken if (blks[i] != DIP(oip, db[i]))
495 1.1 mycroft panic("itrunc2");
496 1.1 mycroft if (length == 0 &&
497 1.32 mycroft (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
498 1.1 mycroft panic("itrunc3");
499 1.1 mycroft #endif /* DIAGNOSTIC */
500 1.1 mycroft /*
501 1.1 mycroft * Put back the real size.
502 1.1 mycroft */
503 1.55 fvdl oip->i_size = length;
504 1.57 kristerw DIP_ASSIGN(oip, size, length);
505 1.57 kristerw DIP_ADD(oip, blocks, -blocksreleased);
506 1.43 chs lockmgr(&gp->g_glock, LK_RELEASE, NULL);
507 1.1 mycroft oip->i_flag |= IN_CHANGE;
508 1.1 mycroft #ifdef QUOTA
509 1.1 mycroft (void) chkdq(oip, -blocksreleased, NOCRED, 0);
510 1.1 mycroft #endif
511 1.55 fvdl KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
512 1.1 mycroft return (allerror);
513 1.1 mycroft }
514 1.1 mycroft
515 1.1 mycroft /*
516 1.1 mycroft * Release blocks associated with the inode ip and stored in the indirect
517 1.1 mycroft * block bn. Blocks are free'd in LIFO order up to (but not including)
518 1.1 mycroft * lastbn. If level is greater than SINGLE, the block is an indirect block
519 1.1 mycroft * and recursive calls to indirtrunc must be used to cleanse other indirect
520 1.1 mycroft * blocks.
521 1.1 mycroft *
522 1.1 mycroft * NB: triple indirect blocks are untested.
523 1.1 mycroft */
524 1.1 mycroft static int
525 1.1 mycroft ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
526 1.30 augustss struct inode *ip;
527 1.53 fvdl daddr_t lbn, lastbn;
528 1.53 fvdl daddr_t dbn;
529 1.1 mycroft int level;
530 1.55 fvdl int64_t *countp;
531 1.1 mycroft {
532 1.30 augustss int i;
533 1.1 mycroft struct buf *bp;
534 1.30 augustss struct fs *fs = ip->i_fs;
535 1.55 fvdl int32_t *bap1 = NULL;
536 1.55 fvdl int64_t *bap2 = NULL;
537 1.1 mycroft struct vnode *vp;
538 1.53 fvdl daddr_t nb, nlbn, last;
539 1.55 fvdl char *copy = NULL;
540 1.55 fvdl int64_t blkcount, factor, blocksreleased = 0;
541 1.55 fvdl int nblocks;
542 1.1 mycroft int error = 0, allerror = 0;
543 1.55 fvdl #ifdef FFS_EI
544 1.55 fvdl const int needswap = UFS_FSNEEDSWAP(fs);
545 1.55 fvdl #endif
546 1.55 fvdl #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
547 1.55 fvdl ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
548 1.57 kristerw #define BAP_ASSIGN(ip, i, value) \
549 1.57 kristerw do { \
550 1.57 kristerw if ((ip)->i_ump->um_fstype == UFS1) \
551 1.57 kristerw bap1[i] = (value); \
552 1.57 kristerw else \
553 1.57 kristerw bap2[i] = (value); \
554 1.57 kristerw } while(0)
555 1.1 mycroft
556 1.1 mycroft /*
557 1.1 mycroft * Calculate index in current block of last
558 1.1 mycroft * block to be kept. -1 indicates the entire
559 1.1 mycroft * block so we need not calculate the index.
560 1.1 mycroft */
561 1.1 mycroft factor = 1;
562 1.1 mycroft for (i = SINGLE; i < level; i++)
563 1.1 mycroft factor *= NINDIR(fs);
564 1.1 mycroft last = lastbn;
565 1.1 mycroft if (lastbn > 0)
566 1.1 mycroft last /= factor;
567 1.1 mycroft nblocks = btodb(fs->fs_bsize);
568 1.1 mycroft /*
569 1.1 mycroft * Get buffer of block pointers, zero those entries corresponding
570 1.1 mycroft * to blocks to be free'd, and update on disk copy first. Since
571 1.1 mycroft * double(triple) indirect before single(double) indirect, calls
572 1.1 mycroft * to bmap on these blocks will fail. However, we already have
573 1.1 mycroft * the on disk address, so we have to set the b_blkno field
574 1.1 mycroft * explicitly instead of letting bread do everything for us.
575 1.1 mycroft */
576 1.1 mycroft vp = ITOV(ip);
577 1.1 mycroft bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
578 1.1 mycroft if (bp->b_flags & (B_DONE | B_DELWRI)) {
579 1.1 mycroft /* Braces must be here in case trace evaluates to nothing. */
580 1.1 mycroft trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
581 1.1 mycroft } else {
582 1.1 mycroft trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
583 1.1 mycroft curproc->p_stats->p_ru.ru_inblock++; /* pay for read */
584 1.1 mycroft bp->b_flags |= B_READ;
585 1.1 mycroft if (bp->b_bcount > bp->b_bufsize)
586 1.1 mycroft panic("ffs_indirtrunc: bad buffer size");
587 1.1 mycroft bp->b_blkno = dbn;
588 1.61 yamt BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
589 1.62 hannken VOP_STRATEGY(vp, bp);
590 1.1 mycroft error = biowait(bp);
591 1.1 mycroft }
592 1.1 mycroft if (error) {
593 1.1 mycroft brelse(bp);
594 1.1 mycroft *countp = 0;
595 1.1 mycroft return (error);
596 1.1 mycroft }
597 1.1 mycroft
598 1.55 fvdl if (ip->i_ump->um_fstype == UFS1)
599 1.55 fvdl bap1 = (int32_t *)bp->b_data;
600 1.55 fvdl else
601 1.55 fvdl bap2 = (int64_t *)bp->b_data;
602 1.33 mycroft if (lastbn >= 0) {
603 1.55 fvdl copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
604 1.55 fvdl memcpy((caddr_t)copy, bp->b_data, (u_int)fs->fs_bsize);
605 1.55 fvdl for (i = last + 1; i < NINDIR(fs); i++)
606 1.57 kristerw BAP_ASSIGN(ip, i, 0);
607 1.12 thorpej error = bwrite(bp);
608 1.12 thorpej if (error)
609 1.12 thorpej allerror = error;
610 1.55 fvdl if (ip->i_ump->um_fstype == UFS1)
611 1.55 fvdl bap1 = (int32_t *)copy;
612 1.55 fvdl else
613 1.55 fvdl bap2 = (int64_t *)copy;
614 1.12 thorpej }
615 1.1 mycroft
616 1.1 mycroft /*
617 1.1 mycroft * Recursively free totally unused blocks.
618 1.1 mycroft */
619 1.1 mycroft for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
620 1.1 mycroft i--, nlbn += factor) {
621 1.55 fvdl nb = RBAP(ip, i);
622 1.1 mycroft if (nb == 0)
623 1.1 mycroft continue;
624 1.1 mycroft if (level > SINGLE) {
625 1.9 christos error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
626 1.53 fvdl (daddr_t)-1, level - 1,
627 1.9 christos &blkcount);
628 1.9 christos if (error)
629 1.1 mycroft allerror = error;
630 1.1 mycroft blocksreleased += blkcount;
631 1.1 mycroft }
632 1.63 hannken ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize, ip->i_number);
633 1.1 mycroft blocksreleased += nblocks;
634 1.1 mycroft }
635 1.1 mycroft
636 1.1 mycroft /*
637 1.1 mycroft * Recursively free last partial block.
638 1.1 mycroft */
639 1.1 mycroft if (level > SINGLE && lastbn >= 0) {
640 1.1 mycroft last = lastbn % factor;
641 1.55 fvdl nb = RBAP(ip, i);
642 1.1 mycroft if (nb != 0) {
643 1.9 christos error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
644 1.9 christos last, level - 1, &blkcount);
645 1.9 christos if (error)
646 1.1 mycroft allerror = error;
647 1.1 mycroft blocksreleased += blkcount;
648 1.1 mycroft }
649 1.1 mycroft }
650 1.12 thorpej
651 1.12 thorpej if (copy != NULL) {
652 1.12 thorpej FREE(copy, M_TEMP);
653 1.12 thorpej } else {
654 1.12 thorpej bp->b_flags |= B_INVAL;
655 1.12 thorpej brelse(bp);
656 1.12 thorpej }
657 1.12 thorpej
658 1.1 mycroft *countp = blocksreleased;
659 1.1 mycroft return (allerror);
660 1.1 mycroft }
661