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
53/*
54 * Convert a component of a pathname into a pointer to a locked inode.
55 * This is a very central and rather complicated routine.
56 * If the filesystem is not maintained in a strict tree hierarchy,
57 * this can result in a deadlock situation (see comments in code below).
58 *
59 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
60 * whether the name is to be looked up, created, renamed, or deleted.
61 * When CREATE, RENAME, or DELETE is specified, information usable in
62 * creating, renaming, or deleting a directory entry may be calculated.
63 * If flag has LOCKPARENT or'ed into it and the target of the pathname
64 * exists, lookup returns both the target and its parent directory locked.
65 * When creating or renaming and LOCKPARENT is specified, the target may
66 * not be ".".  When deleting and LOCKPARENT is specified, the target may
67 * be "."., but the caller must check to ensure it does an vrele and iput
68 * instead of two iputs.
69 *
70 * Overall outline of ufs_lookup:
71 *
72 *	search for name in directory, to found or notfound
73 * notfound:
74 *	if creating, return locked directory, leaving info on available slots
75 *	else return error
76 * found:
77 *	if at end of path and deleting, return information to allow delete
78 *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
79 *	  inode and return info to allow rewrite
80 *	if not at end, add name to cache; if at end and neither creating
81 *	  nor deleting, add name to cache
82 *
83 * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
84 */
85int
86cd9660_lookup(ap)
87	struct vop_cachedlookup_args /* {
88		struct vnode *a_dvp;
89		struct vnode **a_vpp;
90		struct componentname *a_cnp;
91	} */ *ap;
92{
93	struct vnode *vdp;		/* vnode for directory being searched */
94	struct iso_node *dp;		/* inode for directory being searched */
95	struct iso_mnt *imp;		/* filesystem that directory is in */
96	struct buf *bp;			/* a buffer of directory entries */
97	struct iso_directory_record *ep;/* the current directory entry */
98	struct iso_directory_record *ep2;/* copy of current directory entry */
99	int entryoffsetinblock;		/* offset of ep in bp's buffer */
100	int saveoffset = 0;		/* offset of last directory entry in dir */
101	doff_t i_diroff;		/* cached i_diroff value. */
102	doff_t i_offset;		/* cached i_offset value. */
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 error;
109	ino_t ino, i_ino;
110	int ltype, reclen;
111	u_short namelen;
112	int isoflags;
113	char altname[NAME_MAX];
114	int res;
115	int assoc, len;
116	char *name;
117	struct mount *mp;
118	struct vnode **vpp = ap->a_vpp;
119	struct componentname *cnp = ap->a_cnp;
120	int flags = cnp->cn_flags;
121	int nameiop = cnp->cn_nameiop;
122
123	ep2 = ep = NULL;
124	bp = NULL;
125	*vpp = NULL;
126	vdp = ap->a_dvp;
127	dp = VTOI(vdp);
128	imp = dp->i_mnt;
129
130	/*
131	 * We now have a segment name to search for, and a directory to search.
132	 */
133	ino = reclen = 0;
134	i_diroff = dp->i_diroff;
135	len = cnp->cn_namelen;
136	name = cnp->cn_nameptr;
137
138	/*
139	 * A leading `=' means, we are looking for an associated file
140	 */
141	if ((assoc = (imp->iso_ftype != ISO_FTYPE_RRIP && *name == ASSOCCHAR)))
142	{
143		len--;
144		name++;
145	}
146
147	/*
148	 * If there is cached information on a previous search of
149	 * this directory, pick up where we last left off.
150	 * We cache only lookups as these are the most common
151	 * and have the greatest payoff. Caching CREATE has little
152	 * benefit as it usually must search the entire directory
153	 * to determine that the entry does not exist. Caching the
154	 * location of the last DELETE or RENAME has not reduced
155	 * profiling time and hence has been removed in the interest
156	 * of simplicity.
157	 */
158	bmask = imp->im_bmask;
159	if (nameiop != LOOKUP || i_diroff == 0 || i_diroff > dp->i_size) {
160		entryoffsetinblock = 0;
161		i_offset = 0;
162		numdirpasses = 1;
163	} else {
164		i_offset = i_diroff;
165		if ((entryoffsetinblock = i_offset & bmask) &&
166		    (error = cd9660_blkatoff(vdp, (off_t)i_offset, NULL, &bp)))
167				return (error);
168		numdirpasses = 2;
169		nchstats.ncs_2passes++;
170	}
171	endsearch = dp->i_size;
172
173searchloop:
174	while (i_offset < endsearch) {
175		/*
176		 * If offset is on a block boundary,
177		 * read the next directory block.
178		 * Release previous if it exists.
179		 */
180		if ((i_offset & bmask) == 0) {
181			if (bp != NULL)
182				brelse(bp);
183			if ((error =
184			    cd9660_blkatoff(vdp, (off_t)i_offset, NULL, &bp)) != 0)
185				return (error);
186			entryoffsetinblock = 0;
187		}
188		/*
189		 * Get pointer to next entry.
190		 */
191		ep = (struct iso_directory_record *)
192			((char *)bp->b_data + entryoffsetinblock);
193
194		reclen = isonum_711(ep->length);
195		if (reclen == 0) {
196			/* skip to next block, if any */
197			i_offset =
198			    (i_offset & ~bmask) + imp->logical_block_size;
199			continue;
200		}
201
202		if (reclen < ISO_DIRECTORY_RECORD_SIZE)
203			/* illegal entry, stop */
204			break;
205
206		if (entryoffsetinblock + reclen > imp->logical_block_size)
207			/* entries are not allowed to cross boundaries */
208			break;
209
210		namelen = isonum_711(ep->name_len);
211		isoflags = isonum_711(imp->iso_ftype == ISO_FTYPE_HIGH_SIERRA?
212				      &ep->date[6]: ep->flags);
213
214		if (reclen < ISO_DIRECTORY_RECORD_SIZE + namelen)
215			/* illegal entry, stop */
216			break;
217
218		/*
219		 * Check for a name match.
220		 */
221		switch (imp->iso_ftype) {
222		default:
223			if (!(isoflags & 4) == !assoc) {
224				if ((len == 1
225				     && *name == '.')
226				    || (flags & ISDOTDOT)) {
227					if (namelen == 1
228					    && ep->name[0] == ((flags & ISDOTDOT) ? 1 : 0)) {
229						/*
230						 * Save directory entry's inode number and
231						 * release directory buffer.
232						 */
233						i_ino = isodirino(ep, imp);
234						goto found;
235					}
236					if (namelen != 1
237					    || ep->name[0] != 0)
238						goto notfound;
239				} else if (!(res = isofncmp(name, len,
240							    ep->name, namelen,
241							    imp->joliet_level,
242							    imp->im_flags,
243							    imp->im_d2l,
244							    imp->im_l2d))) {
245					if (isoflags & 2)
246						ino = isodirino(ep, imp);
247					else
248						ino = dbtob(bp->b_blkno)
249							+ entryoffsetinblock;
250					saveoffset = i_offset;
251				} else if (ino)
252					goto foundino;
253#ifdef	NOSORTBUG	/* On some CDs directory entries are not sorted correctly */
254				else if (res < 0)
255					goto notfound;
256				else if (res > 0 && numdirpasses == 2)
257					numdirpasses++;
258#endif
259			}
260			break;
261		case ISO_FTYPE_RRIP:
262			if (isonum_711(ep->flags)&2)
263				ino = isodirino(ep, imp);
264			else
265				ino = dbtob(bp->b_blkno) + entryoffsetinblock;
266			i_ino = ino;
267			cd9660_rrip_getname(ep, altname, &namelen, &i_ino, imp);
268			if (namelen == cnp->cn_namelen
269			    && !bcmp(name,altname,namelen))
270				goto found;
271			ino = 0;
272			break;
273		}
274		i_offset += reclen;
275		entryoffsetinblock += reclen;
276	}
277	if (ino) {
278foundino:
279		i_ino = ino;
280		if (saveoffset != i_offset) {
281			if (lblkno(imp, i_offset) !=
282			    lblkno(imp, saveoffset)) {
283				if (bp != NULL)
284					brelse(bp);
285				if ((error = cd9660_blkatoff(vdp,
286				    (off_t)saveoffset, NULL, &bp)) != 0)
287					return (error);
288			}
289			entryoffsetinblock = saveoffset & bmask;
290			ep = (struct iso_directory_record *)
291				((char *)bp->b_data + entryoffsetinblock);
292			reclen = isonum_711(ep->length);
293			i_offset = saveoffset;
294		}
295		goto found;
296	}
297notfound:
298	/*
299	 * If we started in the middle of the directory and failed
300	 * to find our target, we must check the beginning as well.
301	 */
302	if (numdirpasses == 2) {
303		numdirpasses--;
304		i_offset = 0;
305		endsearch = i_diroff;
306		goto searchloop;
307	}
308	if (bp != NULL)
309		brelse(bp);
310
311	/*
312	 * Insert name into cache (as non-existent) if appropriate.
313	 */
314	if (cnp->cn_flags & MAKEENTRY)
315		cache_enter(vdp, *vpp, cnp);
316	if (nameiop == CREATE || nameiop == RENAME)
317		return (EROFS);
318	return (ENOENT);
319
320found:
321	if (numdirpasses == 2)
322		nchstats.ncs_pass2++;
323
324	/*
325	 * Found component in pathname.
326	 * If the final component of path name, save information
327	 * in the cache as to where the entry was found.
328	 */
329	if ((flags & ISLASTCN) && nameiop == LOOKUP)
330		dp->i_diroff = i_offset;
331
332	/*
333	 * Step through the translation in the name.  We do not `vput' the
334	 * directory because we may need it again if a symbolic link
335	 * is relative to the current directory.  Instead we save it
336	 * unlocked as "pdp".  We must get the target inode before unlocking
337	 * the directory to insure that the inode will not be removed
338	 * before we get it.  We prevent deadlock by always fetching
339	 * inodes from the root, moving down the directory tree. Thus
340	 * when following backward pointers ".." we must unlock the
341	 * parent directory before getting the requested directory.
342	 * There is a potential race condition here if both the current
343	 * and parent directories are removed before the `vget' for the
344	 * inode associated with ".." returns.  We hope that this occurs
345	 * infrequently since we cannot avoid this race condition without
346	 * implementing a sophisticated deadlock detection algorithm.
347	 * Note also that this simple deadlock detection scheme will not
348	 * work if the filesystem has any hard links other than ".."
349	 * that point backwards in the directory structure.
350	 */
351	pdp = vdp;
352
353	/*
354	 * Make a copy of the directory entry for non "." lookups so
355	 * we can drop the buffer before calling vget() to avoid a
356	 * lock order reversal between the vnode lock and the buffer
357	 * lock.
358	 */
359	if (dp->i_number != i_ino) {
360		ep2 = malloc(reclen, M_TEMP, M_WAITOK);
361		bcopy(ep, ep2, reclen);
362		ep = ep2;
363	}
364	brelse(bp);
365
366	/*
367	 * If ino is different from i_ino,
368	 * it's a relocated directory.
369	 */
370	if (flags & ISDOTDOT) {
371		/*
372		 * Expanded copy of vn_vget_ino() so that we can use
373		 * cd9660_vget_internal().
374		 */
375		mp = pdp->v_mount;
376		ltype = VOP_ISLOCKED(pdp);
377		error = vfs_busy(mp, MBF_NOWAIT);
378		if (error != 0) {
379			vfs_ref(mp);
380			VOP_UNLOCK(pdp, 0);
381			error = vfs_busy(mp, 0);
382			vn_lock(pdp, ltype | LK_RETRY);
383			vfs_rel(mp);
384			if (error)
385				return (ENOENT);
386			if (pdp->v_iflag & VI_DOOMED) {
387				vfs_unbusy(mp);
388				return (ENOENT);
389			}
390		}
391		VOP_UNLOCK(pdp, 0);
392		error = cd9660_vget_internal(vdp->v_mount, i_ino,
393					     cnp->cn_lkflags, &tdp,
394					     i_ino != ino, ep);
395		free(ep2, M_TEMP);
396		vfs_unbusy(mp);
397		vn_lock(pdp, ltype | LK_RETRY);
398		if (pdp->v_iflag & VI_DOOMED) {
399			if (error == 0)
400				vput(tdp);
401			error = ENOENT;
402		}
403		if (error)
404			return (error);
405		*vpp = tdp;
406	} else if (dp->i_number == i_ino) {
407		VREF(vdp);	/* we want ourself, ie "." */
408		/*
409		 * When we lookup "." we still can be asked to lock it
410		 * differently.
411		 */
412		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
413		if (ltype != VOP_ISLOCKED(vdp)) {
414			if (ltype == LK_EXCLUSIVE)
415				vn_lock(vdp, LK_UPGRADE | LK_RETRY);
416			else /* if (ltype == LK_SHARED) */
417				vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
418		}
419		*vpp = vdp;
420	} else {
421		error = cd9660_vget_internal(vdp->v_mount, i_ino,
422					     cnp->cn_lkflags, &tdp,
423					     i_ino != ino, ep);
424		free(ep2, M_TEMP);
425		if (error)
426			return (error);
427		*vpp = tdp;
428	}
429
430	/*
431	 * Insert name into cache if appropriate.
432	 */
433	if (cnp->cn_flags & MAKEENTRY)
434		cache_enter(vdp, *vpp, cnp);
435	return (0);
436}
437
438/*
439 * Return buffer with the contents of block "offset" from the beginning of
440 * directory "ip".  If "res" is non-zero, fill it in with a pointer to the
441 * remaining space in the directory.
442 */
443int
444cd9660_blkatoff(vp, offset, res, bpp)
445	struct vnode *vp;
446	off_t offset;
447	char **res;
448	struct buf **bpp;
449{
450	struct iso_node *ip;
451	struct iso_mnt *imp;
452	struct buf *bp;
453	daddr_t lbn;
454	int bsize, bshift, error;
455
456	ip = VTOI(vp);
457	imp = ip->i_mnt;
458	lbn = lblkno(imp, offset);
459	bsize = blksize(imp, ip, lbn);
460	bshift = imp->im_bshift;
461
462	if ((error = bread(vp, lbn, bsize, NOCRED, &bp)) != 0) {
463		brelse(bp);
464		*bpp = NULL;
465		return (error);
466	}
467
468	/*
469	 * We must BMAP the buffer because the directory code may use b_blkno
470	 * to calculate the inode for certain types of directory entries.
471	 * We could get away with not doing it before we VMIO-backed the
472	 * directories because the buffers would get freed atomically with
473	 * the invalidation of their data.  But with VMIO-backed buffers
474	 * the buffers may be freed and then later reconstituted - and the
475	 * reconstituted buffer will have no knowledge of b_blkno.
476	 */
477	if (bp->b_blkno == bp->b_lblkno) {
478	        bp->b_blkno = (ip->iso_start + bp->b_lblkno) << (bshift - DEV_BSHIFT);
479        }
480
481	if (res)
482		*res = (char *)bp->b_data + blkoff(imp, offset);
483	*bpp = bp;
484	return (0);
485}
486