proc->thread stage 3.5: Add an IO_CORE flag so coda doesn't have to dig
[dragonfly.git] / sys / vfs / isofs / cd9660 / cd9660_lookup.c
CommitLineData
984263bc
MD
1/*-
2 * Copyright (c) 1989, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley
6 * by Pace Willisson (pace@blitz.com). The Rock Ridge Extension
7 * Support code is derived from software contributed to Berkeley
8 * by Atsushi Murai (amurai@spec.co.jp).
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: @(#)ufs_lookup.c 7.33 (Berkeley) 5/19/91
39 *
40 * @(#)cd9660_lookup.c 8.2 (Berkeley) 1/23/94
41 * $FreeBSD: src/sys/isofs/cd9660/cd9660_lookup.c,v 1.23.2.2 2001/11/04 06:19:47 dillon Exp $
1de703da 42 * $DragonFly: src/sys/vfs/isofs/cd9660/cd9660_lookup.c,v 1.2 2003/06/17 04:28:41 dillon Exp $
984263bc
MD
43 */
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/namei.h>
48#include <sys/buf.h>
49#include <sys/vnode.h>
50#include <sys/mount.h>
51
52#include <isofs/cd9660/iso.h>
53#include <isofs/cd9660/cd9660_node.h>
54#include <isofs/cd9660/iso_rrip.h>
55
56/*
57 * Convert a component of a pathname into a pointer to a locked inode.
58 * This is a very central and rather complicated routine.
59 * If the file system is not maintained in a strict tree hierarchy,
60 * this can result in a deadlock situation (see comments in code below).
61 *
62 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
63 * whether the name is to be looked up, created, renamed, or deleted.
64 * When CREATE, RENAME, or DELETE is specified, information usable in
65 * creating, renaming, or deleting a directory entry may be calculated.
66 * If flag has LOCKPARENT or'ed into it and the target of the pathname
67 * exists, lookup returns both the target and its parent directory locked.
68 * When creating or renaming and LOCKPARENT is specified, the target may
69 * not be ".". When deleting and LOCKPARENT is specified, the target may
70 * be "."., but the caller must check to ensure it does an vrele and iput
71 * instead of two iputs.
72 *
73 * Overall outline of ufs_lookup:
74 *
75 * search for name in directory, to found or notfound
76 * notfound:
77 * if creating, return locked directory, leaving info on available slots
78 * else return error
79 * found:
80 * if at end of path and deleting, return information to allow delete
81 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
82 * inode and return info to allow rewrite
83 * if not at end, add name to cache; if at end and neither creating
84 * nor deleting, add name to cache
85 *
86 * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
87 */
88int
89cd9660_lookup(ap)
90 struct vop_cachedlookup_args /* {
91 struct vnode *a_dvp;
92 struct vnode **a_vpp;
93 struct componentname *a_cnp;
94 } */ *ap;
95{
96 register struct vnode *vdp; /* vnode for directory being searched */
97 register struct iso_node *dp; /* inode for directory being searched */
98 register struct iso_mnt *imp; /* file system that directory is in */
99 struct buf *bp; /* a buffer of directory entries */
100 struct iso_directory_record *ep = 0;/* the current directory entry */
101 int entryoffsetinblock; /* offset of ep in bp's buffer */
102 int saveoffset = 0; /* offset of last directory entry in dir */
103 int numdirpasses; /* strategy for directory search */
104 doff_t endsearch; /* offset to end directory search */
105 struct vnode *pdp; /* saved dp during symlink work */
106 struct vnode *tdp; /* returned by cd9660_vget_internal */
107 u_long bmask; /* block offset mask */
108 int lockparent; /* 1 => lockparent flag is set */
109 int wantparent; /* 1 => wantparent or lockparent flag */
110 int error;
111 ino_t ino = 0;
112 int reclen;
113 u_short namelen;
114 int isoflags;
115 char altname[NAME_MAX];
116 int res;
117 int assoc, len;
118 char *name;
119 struct vnode **vpp = ap->a_vpp;
120 struct componentname *cnp = ap->a_cnp;
121 int flags = cnp->cn_flags;
122 int nameiop = cnp->cn_nameiop;
123 struct proc *p = cnp->cn_proc;
124
125 bp = NULL;
126 *vpp = NULL;
127 vdp = ap->a_dvp;
128 dp = VTOI(vdp);
129 imp = dp->i_mnt;
130 lockparent = flags & LOCKPARENT;
131 wantparent = flags & (LOCKPARENT|WANTPARENT);
132 cnp->cn_flags &= ~PDIRUNLOCK;
133
134 /*
135 * We now have a segment name to search for, and a directory to search.
136 */
137
138 len = cnp->cn_namelen;
139 name = cnp->cn_nameptr;
140 /*
141 * A leading `=' means, we are looking for an associated file
142 */
143 if ((assoc = (imp->iso_ftype != ISO_FTYPE_RRIP && *name == ASSOCCHAR)))
144 {
145 len--;
146 name++;
147 }
148
149 /*
150 * If there is cached information on a previous search of
151 * this directory, pick up where we last left off.
152 * We cache only lookups as these are the most common
153 * and have the greatest payoff. Caching CREATE has little
154 * benefit as it usually must search the entire directory
155 * to determine that the entry does not exist. Caching the
156 * location of the last DELETE or RENAME has not reduced
157 * profiling time and hence has been removed in the interest
158 * of simplicity.
159 */
160 bmask = imp->im_bmask;
161 if (nameiop != LOOKUP || dp->i_diroff == 0 ||
162 dp->i_diroff > dp->i_size) {
163 entryoffsetinblock = 0;
164 dp->i_offset = 0;
165 numdirpasses = 1;
166 } else {
167 dp->i_offset = dp->i_diroff;
168 if ((entryoffsetinblock = dp->i_offset & bmask) &&
169 (error = cd9660_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp)))
170 return (error);
171 numdirpasses = 2;
172 nchstats.ncs_2passes++;
173 }
174 endsearch = dp->i_size;
175
176searchloop:
177 while (dp->i_offset < endsearch) {
178 /*
179 * If offset is on a block boundary,
180 * read the next directory block.
181 * Release previous if it exists.
182 */
183 if ((dp->i_offset & bmask) == 0) {
184 if (bp != NULL)
185 brelse(bp);
186 if ((error =
187 cd9660_blkatoff(vdp, (off_t)dp->i_offset, NULL, &bp)) != 0)
188 return (error);
189 entryoffsetinblock = 0;
190 }
191 /*
192 * Get pointer to next entry.
193 */
194 ep = (struct iso_directory_record *)
195 ((char *)bp->b_data + entryoffsetinblock);
196
197 reclen = isonum_711(ep->length);
198 if (reclen == 0) {
199 /* skip to next block, if any */
200 dp->i_offset =
201 (dp->i_offset & ~bmask) + imp->logical_block_size;
202 continue;
203 }
204
205 if (reclen < ISO_DIRECTORY_RECORD_SIZE)
206 /* illegal entry, stop */
207 break;
208
209 if (entryoffsetinblock + reclen > imp->logical_block_size)
210 /* entries are not allowed to cross boundaries */
211 break;
212
213 namelen = isonum_711(ep->name_len);
214 isoflags = isonum_711(imp->iso_ftype == ISO_FTYPE_HIGH_SIERRA?
215 &ep->date[6]: ep->flags);
216
217 if (reclen < ISO_DIRECTORY_RECORD_SIZE + namelen)
218 /* illegal entry, stop */
219 break;
220
221 /*
222 * Check for a name match.
223 */
224 switch (imp->iso_ftype) {
225 default:
226 if (!(isoflags & 4) == !assoc) {
227 if ((len == 1
228 && *name == '.')
229 || (flags & ISDOTDOT)) {
230 if (namelen == 1
231 && ep->name[0] == ((flags & ISDOTDOT) ? 1 : 0)) {
232 /*
233 * Save directory entry's inode number and
234 * release directory buffer.
235 */
236 dp->i_ino = isodirino(ep, imp);
237 goto found;
238 }
239 if (namelen != 1
240 || ep->name[0] != 0)
241 goto notfound;
242 } else if (!(res = isofncmp(name, len, ep->name, namelen, imp->joliet_level))) {
243 if (isoflags & 2)
244 ino = isodirino(ep, imp);
245 else
246 ino = dbtob(bp->b_blkno)
247 + entryoffsetinblock;
248 saveoffset = dp->i_offset;
249 } else if (ino)
250 goto foundino;
251#ifdef NOSORTBUG /* On some CDs directory entries are not sorted correctly */
252 else if (res < 0)
253 goto notfound;
254 else if (res > 0 && numdirpasses == 2)
255 numdirpasses++;
256#endif
257 }
258 break;
259 case ISO_FTYPE_RRIP:
260 if (isonum_711(ep->flags)&2)
261 ino = isodirino(ep, imp);
262 else
263 ino = dbtob(bp->b_blkno) + entryoffsetinblock;
264 dp->i_ino = ino;
265 cd9660_rrip_getname(ep,altname,&namelen,&dp->i_ino,imp);
266 if (namelen == cnp->cn_namelen
267 && !bcmp(name,altname,namelen))
268 goto found;
269 ino = 0;
270 break;
271 }
272 dp->i_offset += reclen;
273 entryoffsetinblock += reclen;
274 }
275 if (ino) {
276foundino:
277 dp->i_ino = ino;
278 if (saveoffset != dp->i_offset) {
279 if (lblkno(imp, dp->i_offset) !=
280 lblkno(imp, saveoffset)) {
281 if (bp != NULL)
282 brelse(bp);
283 if ((error = cd9660_blkatoff(vdp,
284 (off_t)saveoffset, NULL, &bp)) != 0)
285 return (error);
286 }
287 entryoffsetinblock = saveoffset & bmask;
288 ep = (struct iso_directory_record *)
289 ((char *)bp->b_data + entryoffsetinblock);
290 dp->i_offset = saveoffset;
291 }
292 goto found;
293 }
294notfound:
295 /*
296 * If we started in the middle of the directory and failed
297 * to find our target, we must check the beginning as well.
298 */
299 if (numdirpasses == 2) {
300 numdirpasses--;
301 dp->i_offset = 0;
302 endsearch = dp->i_diroff;
303 goto searchloop;
304 }
305 if (bp != NULL)
306 brelse(bp);
307
308 /*
309 * Insert name into cache (as non-existent) if appropriate.
310 */
311 if (cnp->cn_flags & MAKEENTRY)
312 cache_enter(vdp, *vpp, cnp);
313 if (nameiop == CREATE || nameiop == RENAME)
314 return (EROFS);
315 return (ENOENT);
316
317found:
318 if (numdirpasses == 2)
319 nchstats.ncs_pass2++;
320
321 /*
322 * Found component in pathname.
323 * If the final component of path name, save information
324 * in the cache as to where the entry was found.
325 */
326 if ((flags & ISLASTCN) && nameiop == LOOKUP)
327 dp->i_diroff = dp->i_offset;
328
329 /*
330 * Step through the translation in the name. We do not `iput' the
331 * directory because we may need it again if a symbolic link
332 * is relative to the current directory. Instead we save it
333 * unlocked as "pdp". We must get the target inode before unlocking
334 * the directory to insure that the inode will not be removed
335 * before we get it. We prevent deadlock by always fetching
336 * inodes from the root, moving down the directory tree. Thus
337 * when following backward pointers ".." we must unlock the
338 * parent directory before getting the requested directory.
339 * There is a potential race condition here if both the current
340 * and parent directories are removed before the `iget' for the
341 * inode associated with ".." returns. We hope that this occurs
342 * infrequently since we cannot avoid this race condition without
343 * implementing a sophisticated deadlock detection algorithm.
344 * Note also that this simple deadlock detection scheme will not
345 * work if the file system has any hard links other than ".."
346 * that point backwards in the directory structure.
347 */
348 pdp = vdp;
349 /*
350 * If ino is different from dp->i_ino,
351 * it's a relocated directory.
352 */
353 if (flags & ISDOTDOT) {
354 VOP_UNLOCK(pdp, 0, p); /* race to get the inode */
355 error = cd9660_vget_internal(vdp->v_mount, dp->i_ino, &tdp,
356 dp->i_ino != ino, ep);
357 brelse(bp);
358 if (error) {
359 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY, p);
360 return (error);
361 }
362 if (lockparent && (flags & ISLASTCN)) {
363 if ((error = vn_lock(pdp, LK_EXCLUSIVE, p)) != 0) {
364 cnp->cn_flags |= PDIRUNLOCK;
365 vput(tdp);
366 return (error);
367 }
368 } else
369 cnp->cn_flags |= PDIRUNLOCK;
370 *vpp = tdp;
371 } else if (dp->i_number == dp->i_ino) {
372 brelse(bp);
373 VREF(vdp); /* we want ourself, ie "." */
374 *vpp = vdp;
375 } else {
376 error = cd9660_vget_internal(vdp->v_mount, dp->i_ino, &tdp,
377 dp->i_ino != ino, ep);
378 brelse(bp);
379 if (error)
380 return (error);
381 if (!lockparent || !(flags & ISLASTCN)) {
382 cnp->cn_flags |= PDIRUNLOCK;
383 VOP_UNLOCK(pdp, 0, p);
384 }
385 *vpp = tdp;
386 }
387
388 /*
389 * Insert name into cache if appropriate.
390 */
391 if (cnp->cn_flags & MAKEENTRY)
392 cache_enter(vdp, *vpp, cnp);
393 return (0);
394}
395
396/*
397 * Return buffer with the contents of block "offset" from the beginning of
398 * directory "ip". If "res" is non-zero, fill it in with a pointer to the
399 * remaining space in the directory.
400 */
401int
402cd9660_blkatoff(vp, offset, res, bpp)
403 struct vnode *vp;
404 off_t offset;
405 char **res;
406 struct buf **bpp;
407{
408 struct iso_node *ip;
409 register struct iso_mnt *imp;
410 struct buf *bp;
411 daddr_t lbn;
412 int bsize, error;
413
414 ip = VTOI(vp);
415 imp = ip->i_mnt;
416 lbn = lblkno(imp, offset);
417 bsize = blksize(imp, ip, lbn);
418
419 if ((error = bread(vp, lbn, bsize, NOCRED, &bp)) != 0) {
420 brelse(bp);
421 *bpp = NULL;
422 return (error);
423 }
424
425 /*
426 * We must BMAP the buffer because the directory code may use b_blkno
427 * to calculate the inode for certain types of directory entries.
428 * We could get away with not doing it before we VMIO-backed the
429 * directories because the buffers would get freed atomically with
430 * the invalidation of their data. But with VMIO-backed buffers
431 * the buffers may be freed and then later reconstituted - and the
432 * reconstituted buffer will have no knowledge of b_blkno.
433 */
434 if (bp->b_blkno == bp->b_lblkno) {
435 error = VOP_BMAP(vp, bp->b_lblkno, NULL,
436 &bp->b_blkno, NULL, NULL);
437 if (error) {
438 bp->b_error = error;
439 bp->b_flags |= B_ERROR;
440 brelse(bp);
441 *bpp = NULL;
442 return (error);
443 }
444 }
445
446 if (res)
447 *res = (char *)bp->b_data + blkoff(imp, offset);
448 *bpp = bp;
449 return (0);
450}