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 * 4. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 *	from: @(#)ufs_lookup.c	7.33 (Berkeley) 5/19/91
35 *	@(#)cd9660_lookup.c	8.2 (Berkeley) 1/23/94
36 */
37
38#include <sys/cdefs.h>
39__FBSDID("$FreeBSD$");
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/namei.h>
44#include <sys/bio.h>
45#include <sys/buf.h>
46#include <sys/vnode.h>
47#include <sys/mount.h>
48
49#include <fs/cd9660/iso.h>
50#include <fs/cd9660/cd9660_node.h>
51#include <fs/cd9660/iso_rrip.h>
52
53struct cd9660_ino_alloc_arg {
54	ino_t ino;
55	ino_t i_ino;
56	struct iso_directory_record *ep;
57};
58
59static int
60cd9660_ino_alloc(struct mount *mp, void *arg, int lkflags,
61    struct vnode **vpp)
62{
63	struct cd9660_ino_alloc_arg *dd_arg;
64
65	dd_arg = arg;
66	return (cd9660_vget_internal(mp, dd_arg->i_ino, lkflags, vpp,
67	    dd_arg->i_ino != dd_arg->ino, dd_arg->ep));
68}
69
70/*
71 * Convert a component of a pathname into a pointer to a locked inode.
72 * This is a very central and rather complicated routine.
73 * If the filesystem is not maintained in a strict tree hierarchy,
74 * this can result in a deadlock situation (see comments in code below).
75 *
76 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
77 * whether the name is to be looked up, created, renamed, or deleted.
78 * When CREATE, RENAME, or DELETE is specified, information usable in
79 * creating, renaming, or deleting a directory entry may be calculated.
80 * If flag has LOCKPARENT or'ed into it and the target of the pathname
81 * exists, lookup returns both the target and its parent directory locked.
82 * When creating or renaming and LOCKPARENT is specified, the target may
83 * not be ".".  When deleting and LOCKPARENT is specified, the target may
84 * be "."., but the caller must check to ensure it does an vrele and iput
85 * instead of two iputs.
86 *
87 * Overall outline of ufs_lookup:
88 *
89 *	search for name in directory, to found or notfound
90 * notfound:
91 *	if creating, return locked directory, leaving info on available slots
92 *	else return error
93 * found:
94 *	if at end of path and deleting, return information to allow delete
95 *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
96 *	  inode and return info to allow rewrite
97 *	if not at end, add name to cache; if at end and neither creating
98 *	  nor deleting, add name to cache
99 *
100 * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
101 */
102int
103cd9660_lookup(ap)
104	struct vop_cachedlookup_args /* {
105		struct vnode *a_dvp;
106		struct vnode **a_vpp;
107		struct componentname *a_cnp;
108	} */ *ap;
109{
110	struct vnode *vdp;		/* vnode for directory being searched */
111	struct iso_node *dp;		/* inode for directory being searched */
112	struct iso_mnt *imp;		/* filesystem that directory is in */
113	struct buf *bp;			/* a buffer of directory entries */
114	struct iso_directory_record *ep;/* the current directory entry */
115	struct iso_directory_record *ep2;/* copy of current directory entry */
116	int entryoffsetinblock;		/* offset of ep in bp's buffer */
117	int saveoffset = 0;		/* offset of last directory entry in dir */
118	doff_t i_diroff;		/* cached i_diroff value. */
119	doff_t i_offset;		/* cached i_offset value. */
120	int numdirpasses;		/* strategy for directory search */
121	doff_t endsearch;		/* offset to end directory search */
122	struct vnode *pdp;		/* saved dp during symlink work */
123	struct vnode *tdp;		/* returned by cd9660_vget_internal */
124	struct cd9660_ino_alloc_arg dd_arg;
125	u_long bmask;			/* block offset mask */
126	int error;
127	ino_t ino, i_ino;
128	int ltype, reclen;
129	u_short namelen;
130	int isoflags;
131	char altname[NAME_MAX];
132	int res;
133	int assoc, len;
134	char *name;
135	struct vnode **vpp = ap->a_vpp;
136	struct componentname *cnp = ap->a_cnp;
137	int flags = cnp->cn_flags;
138	int nameiop = cnp->cn_nameiop;
139
140	ep2 = ep = NULL;
141	bp = NULL;
142	*vpp = NULL;
143	vdp = ap->a_dvp;
144	dp = VTOI(vdp);
145	imp = dp->i_mnt;
146
147	/*
148	 * We now have a segment name to search for, and a directory to search.
149	 */
150	ino = reclen = 0;
151	i_diroff = dp->i_diroff;
152	len = cnp->cn_namelen;
153	name = cnp->cn_nameptr;
154
155	/*
156	 * A leading `=' means, we are looking for an associated file
157	 */
158	if ((assoc = (imp->iso_ftype != ISO_FTYPE_RRIP && *name == ASSOCCHAR)))
159	{
160		len--;
161		name++;
162	}
163
164	/*
165	 * If there is cached information on a previous search of
166	 * this directory, pick up where we last left off.
167	 * We cache only lookups as these are the most common
168	 * and have the greatest payoff. Caching CREATE has little
169	 * benefit as it usually must search the entire directory
170	 * to determine that the entry does not exist. Caching the
171	 * location of the last DELETE or RENAME has not reduced
172	 * profiling time and hence has been removed in the interest
173	 * of simplicity.
174	 */
175	bmask = imp->im_bmask;
176	if (nameiop != LOOKUP || i_diroff == 0 || i_diroff > dp->i_size) {
177		entryoffsetinblock = 0;
178		i_offset = 0;
179		numdirpasses = 1;
180	} else {
181		i_offset = i_diroff;
182		if ((entryoffsetinblock = i_offset & bmask) &&
183		    (error = cd9660_blkatoff(vdp, (off_t)i_offset, NULL, &bp)))
184				return (error);
185		numdirpasses = 2;
186		nchstats.ncs_2passes++;
187	}
188	endsearch = dp->i_size;
189
190searchloop:
191	while (i_offset < endsearch) {
192		/*
193		 * If offset is on a block boundary,
194		 * read the next directory block.
195		 * Release previous if it exists.
196		 */
197		if ((i_offset & bmask) == 0) {
198			if (bp != NULL)
199				brelse(bp);
200			if ((error =
201			    cd9660_blkatoff(vdp, (off_t)i_offset, NULL, &bp)) != 0)
202				return (error);
203			entryoffsetinblock = 0;
204		}
205		/*
206		 * Get pointer to next entry.
207		 */
208		ep = (struct iso_directory_record *)
209			((char *)bp->b_data + entryoffsetinblock);
210
211		reclen = isonum_711(ep->length);
212		if (reclen == 0) {
213			/* skip to next block, if any */
214			i_offset =
215			    (i_offset & ~bmask) + imp->logical_block_size;
216			continue;
217		}
218
219		if (reclen < ISO_DIRECTORY_RECORD_SIZE)
220			/* illegal entry, stop */
221			break;
222
223		if (entryoffsetinblock + reclen > imp->logical_block_size)
224			/* entries are not allowed to cross boundaries */
225			break;
226
227		namelen = isonum_711(ep->name_len);
228		isoflags = isonum_711(imp->iso_ftype == ISO_FTYPE_HIGH_SIERRA?
229				      &ep->date[6]: ep->flags);
230
231		if (reclen < ISO_DIRECTORY_RECORD_SIZE + namelen)
232			/* illegal entry, stop */
233			break;
234
235		/*
236		 * Check for a name match.
237		 */
238		switch (imp->iso_ftype) {
239		default:
240			if (!(isoflags & 4) == !assoc) {
241				if ((len == 1
242				     && *name == '.')
243				    || (flags & ISDOTDOT)) {
244					if (namelen == 1
245					    && ep->name[0] == ((flags & ISDOTDOT) ? 1 : 0)) {
246						/*
247						 * Save directory entry's inode number and
248						 * release directory buffer.
249						 */
250						i_ino = isodirino(ep, imp);
251						goto found;
252					}
253					if (namelen != 1
254					    || ep->name[0] != 0)
255						goto notfound;
256				} else if (!(res = isofncmp(name, len,
257							    ep->name, namelen,
258							    imp->joliet_level,
259							    imp->im_flags,
260							    imp->im_d2l,
261							    imp->im_l2d))) {
262					if (isoflags & 2)
263						ino = isodirino(ep, imp);
264					else
265						ino = dbtob(bp->b_blkno)
266							+ entryoffsetinblock;
267					saveoffset = i_offset;
268				} else if (ino)
269					goto foundino;
270#ifdef	NOSORTBUG	/* On some CDs directory entries are not sorted correctly */
271				else if (res < 0)
272					goto notfound;
273				else if (res > 0 && numdirpasses == 2)
274					numdirpasses++;
275#endif
276			}
277			break;
278		case ISO_FTYPE_RRIP:
279			if (isonum_711(ep->flags)&2)
280				ino = isodirino(ep, imp);
281			else
282				ino = dbtob(bp->b_blkno) + entryoffsetinblock;
283			i_ino = ino;
284			cd9660_rrip_getname(ep, altname, &namelen, &i_ino, imp);
285			if (namelen == cnp->cn_namelen
286			    && !bcmp(name,altname,namelen))
287				goto found;
288			ino = 0;
289			break;
290		}
291		i_offset += reclen;
292		entryoffsetinblock += reclen;
293	}
294	if (ino) {
295foundino:
296		i_ino = ino;
297		if (saveoffset != i_offset) {
298			if (lblkno(imp, i_offset) !=
299			    lblkno(imp, saveoffset)) {
300				if (bp != NULL)
301					brelse(bp);
302				if ((error = cd9660_blkatoff(vdp,
303				    (off_t)saveoffset, NULL, &bp)) != 0)
304					return (error);
305			}
306			entryoffsetinblock = saveoffset & bmask;
307			ep = (struct iso_directory_record *)
308				((char *)bp->b_data + entryoffsetinblock);
309			reclen = isonum_711(ep->length);
310			i_offset = saveoffset;
311		}
312		goto found;
313	}
314notfound:
315	/*
316	 * If we started in the middle of the directory and failed
317	 * to find our target, we must check the beginning as well.
318	 */
319	if (numdirpasses == 2) {
320		numdirpasses--;
321		i_offset = 0;
322		endsearch = i_diroff;
323		goto searchloop;
324	}
325	if (bp != NULL)
326		brelse(bp);
327
328	/*
329	 * Insert name into cache (as non-existent) if appropriate.
330	 */
331	if (cnp->cn_flags & MAKEENTRY)
332		cache_enter(vdp, *vpp, cnp);
333	if (nameiop == CREATE || nameiop == RENAME)
334		return (EROFS);
335	return (ENOENT);
336
337found:
338	if (numdirpasses == 2)
339		nchstats.ncs_pass2++;
340
341	/*
342	 * Found component in pathname.
343	 * If the final component of path name, save information
344	 * in the cache as to where the entry was found.
345	 */
346	if ((flags & ISLASTCN) && nameiop == LOOKUP)
347		dp->i_diroff = i_offset;
348
349	/*
350	 * Step through the translation in the name.  We do not `vput' the
351	 * directory because we may need it again if a symbolic link
352	 * is relative to the current directory.  Instead we save it
353	 * unlocked as "pdp".  We must get the target inode before unlocking
354	 * the directory to insure that the inode will not be removed
355	 * before we get it.  We prevent deadlock by always fetching
356	 * inodes from the root, moving down the directory tree. Thus
357	 * when following backward pointers ".." we must unlock the
358	 * parent directory before getting the requested directory.
359	 * There is a potential race condition here if both the current
360	 * and parent directories are removed before the `vget' for the
361	 * inode associated with ".." returns.  We hope that this occurs
362	 * infrequently since we cannot avoid this race condition without
363	 * implementing a sophisticated deadlock detection algorithm.
364	 * Note also that this simple deadlock detection scheme will not
365	 * work if the filesystem has any hard links other than ".."
366	 * that point backwards in the directory structure.
367	 */
368	pdp = vdp;
369
370	/*
371	 * Make a copy of the directory entry for non "." lookups so
372	 * we can drop the buffer before calling vget() to avoid a
373	 * lock order reversal between the vnode lock and the buffer
374	 * lock.
375	 */
376	if (dp->i_number != i_ino) {
377		ep2 = malloc(reclen, M_TEMP, M_WAITOK);
378		bcopy(ep, ep2, reclen);
379		ep = ep2;
380	}
381	brelse(bp);
382
383	/*
384	 * If ino is different from i_ino,
385	 * it's a relocated directory.
386	 */
387	if (flags & ISDOTDOT) {
388		dd_arg.ino = ino;
389		dd_arg.i_ino = i_ino;
390		dd_arg.ep = ep;
391		error = vn_vget_ino_gen(pdp, cd9660_ino_alloc, &dd_arg,
392		    cnp->cn_lkflags, &tdp);
393		free(ep2, M_TEMP);
394		if (error != 0)
395			return (error);
396		*vpp = tdp;
397	} else if (dp->i_number == i_ino) {
398		VREF(vdp);	/* we want ourself, ie "." */
399		/*
400		 * When we lookup "." we still can be asked to lock it
401		 * differently.
402		 */
403		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
404		if (ltype != VOP_ISLOCKED(vdp)) {
405			if (ltype == LK_EXCLUSIVE)
406				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
407			else /* if (ltype == LK_SHARED) */
408				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
409		}
410		*vpp = vdp;
411	} else {
412		error = cd9660_vget_internal(vdp->v_mount, i_ino,
413					     cnp->cn_lkflags, &tdp,
414					     i_ino != ino, ep);
415		free(ep2, M_TEMP);
416		if (error)
417			return (error);
418		*vpp = tdp;
419	}
420
421	/*
422	 * Insert name into cache if appropriate.
423	 */
424	if (cnp->cn_flags & MAKEENTRY)
425		cache_enter(vdp, *vpp, cnp);
426	return (0);
427}
428
429/*
430 * Return buffer with the contents of block "offset" from the beginning of
431 * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
432 * remaining space in the directory.
433 */
434int
435cd9660_blkatoff(vp, offset, res, bpp)
436	struct vnode *vp;
437	off_t offset;
438	char **res;
439	struct buf **bpp;
440{
441	struct iso_node *ip;
442	struct iso_mnt *imp;
443	struct buf *bp;
444	daddr_t lbn;
445	int bsize, bshift, error;
446
447	ip = VTOI(vp);
448	imp = ip->i_mnt;
449	lbn = lblkno(imp, offset);
450	bsize = blksize(imp, ip, lbn);
451	bshift = imp->im_bshift;
452
453	if ((error = bread(vp, lbn, bsize, NOCRED, &bp)) != 0) {
454		brelse(bp);
455		*bpp = NULL;
456		return (error);
457	}
458
459	/*
460	 * We must BMAP the buffer because the directory code may use b_blkno
461	 * to calculate the inode for certain types of directory entries.
462	 * We could get away with not doing it before we VMIO-backed the
463	 * directories because the buffers would get freed atomically with
464	 * the invalidation of their data.  But with VMIO-backed buffers
465	 * the buffers may be freed and then later reconstituted - and the
466	 * reconstituted buffer will have no knowledge of b_blkno.
467	 */
468	if (bp->b_blkno == bp->b_lblkno) {
469	        bp->b_blkno = (ip->iso_start + bp->b_lblkno) << (bshift - DEV_BSHIFT);
470        }
471
472	if (res)
473		*res = (char *)bp->b_data + blkoff(imp, offset);
474	*bpp = bp;
475	return (0);
476}
477