xref: /dflybsd-src/sys/vfs/udf/udf_vnops.c (revision 57fed2afae86702adfd8bc0f2b73e76280fa6847)
1 /*-
2  * Copyright (c) 2001, 2002 Scott Long <scottl@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD: src/sys/fs/udf/udf_vnops.c,v 1.33 2003/12/07 05:04:49 scottl Exp $
27  * $DragonFly: src/sys/vfs/udf/udf_vnops.c,v 1.9 2004/10/12 19:21:10 dillon Exp $
28  */
29 
30 /* udf_vnops.c */
31 /* Take care of the vnode side of things */
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/namei.h>
36 #include <sys/kernel.h>
37 #include <sys/malloc.h>
38 #include <sys/stat.h>
39 #include <sys/module.h>
40 #include <sys/buf.h>
41 #include <sys/iconv.h>
42 #include <sys/mount.h>
43 #include <sys/vnode.h>
44 #include <sys/dirent.h>
45 #include <sys/queue.h>
46 #include <sys/unistd.h>
47 
48 #include <vfs/udf/ecma167-udf.h>
49 #include <vfs/udf/osta.h>
50 #include <vfs/udf/udf.h>
51 #include <vfs/udf/udf_mount.h>
52 
53 static int udf_access(struct vop_access_args *);
54 static int udf_getattr(struct vop_getattr_args *);
55 static int udf_ioctl(struct vop_ioctl_args *);
56 static int udf_pathconf(struct vop_pathconf_args *);
57 static int udf_read(struct vop_read_args *);
58 static int udf_readdir(struct vop_readdir_args *);
59 static int udf_readlink(struct vop_readlink_args *ap);
60 static int udf_strategy(struct vop_strategy_args *);
61 static int udf_bmap(struct vop_bmap_args *);
62 static int udf_lookup(struct vop_cachedlookup_args *);
63 static int udf_reclaim(struct vop_reclaim_args *);
64 static int udf_readatoffset(struct udf_node *, int *, int, struct buf **, uint8_t **);
65 static int udf_bmap_internal(struct udf_node *, uint32_t, daddr_t *, uint32_t *);
66 
67 struct vnodeopv_entry_desc udf_vnodeop_entries[] = {
68 	{ &vop_default_desc,		vop_defaultop },
69 	{ &vop_access_desc,		(void *) udf_access },
70 	{ &vop_bmap_desc,		(void *) udf_bmap },
71 	{ &vop_cachedlookup_desc,	(void *) udf_lookup },
72 	{ &vop_getattr_desc,		(void *) udf_getattr },
73 	{ &vop_ioctl_desc,		(void *) udf_ioctl },
74 	{ &vop_lookup_desc,		(void *) vfs_cache_lookup },
75 	{ &vop_pathconf_desc,		(void *) udf_pathconf },
76 	{ &vop_read_desc,		(void *) udf_read },
77 	{ &vop_readdir_desc,		(void *) udf_readdir },
78 	{ &vop_readlink_desc,		(void *) udf_readlink },
79 	{ &vop_reclaim_desc,		(void *) udf_reclaim },
80 	{ &vop_strategy_desc,		(void *) udf_strategy },
81 	{ NULL, NULL }
82 };
83 
84 MALLOC_DEFINE(M_UDFFID, "UDF FID", "UDF FileId structure");
85 MALLOC_DEFINE(M_UDFDS, "UDF DS", "UDF Dirstream structure");
86 
87 #define UDF_INVALID_BMAP	-1
88 
89 /* Look up a udf_node based on the ino_t passed in and return it's vnode */
90 int
91 udf_hashlookup(struct udf_mnt *udfmp, ino_t id, struct vnode **vpp)
92 {
93 	struct udf_node *node;
94 	struct udf_hash_lh *lh;
95 	struct vnode *vp;
96 	lwkt_tokref hashlock;
97 
98 	*vpp = NULL;
99 
100 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
101 loop:
102 	lh = &udfmp->hashtbl[id % udfmp->hashsz];
103 	if (lh == NULL) {
104 		lwkt_reltoken(&hashlock);
105 		return(ENOENT);
106 	}
107 	LIST_FOREACH(node, lh, le) {
108 		if (node->hash_id != id)
109 			continue;
110 		vp = node->i_vnode;
111 		if (vget(vp, LK_EXCLUSIVE, curthread))
112 			goto loop;
113 		/*
114 		 * We must check to see if the inode has been ripped
115 		 * out from under us after blocking.
116 		 */
117 		lh = &udfmp->hashtbl[id % udfmp->hashsz];
118 		LIST_FOREACH(node, lh, le) {
119 			if (node->hash_id == id)
120 				break;
121 		}
122 		if (node == NULL || vp != node->i_vnode) {
123 			vput(vp);
124 			goto loop;
125 		}
126 		lwkt_reltoken(&hashlock);
127 		*vpp = vp;
128 		return(0);
129 	}
130 
131 	lwkt_reltoken(&hashlock);
132 	return(0);
133 }
134 
135 int
136 udf_hashins(struct udf_node *node)
137 {
138 	struct udf_mnt *udfmp;
139 	struct udf_hash_lh *lh;
140 	lwkt_tokref hashlock;
141 
142 	udfmp = node->udfmp;
143 
144 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
145 	lh = &udfmp->hashtbl[node->hash_id % udfmp->hashsz];
146 	if (lh == NULL)
147 		LIST_INIT(lh);
148 	LIST_INSERT_HEAD(lh, node, le);
149 	lwkt_reltoken(&hashlock);
150 
151 	return(0);
152 }
153 
154 int
155 udf_hashrem(struct udf_node *node)
156 {
157 	struct udf_mnt *udfmp;
158 	struct udf_hash_lh *lh;
159 	lwkt_tokref hashlock;
160 
161 	udfmp = node->udfmp;
162 
163 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
164 	lh = &udfmp->hashtbl[node->hash_id % udfmp->hashsz];
165 	if (lh == NULL)
166 		panic("hash entry is NULL, node->hash_id= %d\n", node->hash_id);
167 	LIST_REMOVE(node, le);
168 	lwkt_reltoken(&hashlock);
169 
170 	return(0);
171 }
172 
173 int
174 udf_allocv(struct mount *mp, struct vnode **vpp)
175 {
176 	int error;
177 	struct vnode *vp;
178 
179 	error = getnewvnode(VT_UDF, mp, mp->mnt_vn_ops, &vp, 0, 0);
180 	if (error) {
181 		printf("udf_allocv: failed to allocate new vnode\n");
182 		return(error);
183 	}
184 
185 	*vpp = vp;
186 	return(0);
187 }
188 
189 /* Convert file entry permission (5 bits per owner/group/user) to a mode_t */
190 static mode_t
191 udf_permtomode(struct udf_node *node)
192 {
193 	uint32_t perm;
194 	uint32_t flags;
195 	mode_t mode;
196 
197 	perm = node->fentry->perm;
198 	flags = node->fentry->icbtag.flags;
199 
200 	mode = perm & UDF_FENTRY_PERM_USER_MASK;
201 	mode |= ((perm & UDF_FENTRY_PERM_GRP_MASK) >> 2);
202 	mode |= ((perm & UDF_FENTRY_PERM_OWNER_MASK) >> 4);
203 	mode |= ((flags & UDF_ICB_TAG_FLAGS_STICKY) << 4);
204 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETGID) << 6);
205 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETUID) << 8);
206 
207 	return(mode);
208 }
209 
210 static int
211 udf_access(struct vop_access_args *a)
212 {
213 	struct vnode *vp;
214 	struct udf_node *node;
215 	mode_t a_mode, mode, mask;
216 	struct ucred *cred = a->a_cred;
217 	gid_t *gp;
218 	int i;
219 
220 	vp = a->a_vp;
221 	node = VTON(vp);
222 	a_mode = a->a_mode;
223 
224 	if (a_mode & VWRITE) {
225 		switch (vp->v_type) {
226 		case VDIR:
227 		case VLNK:
228 		case VREG:
229 			return(EROFS);
230 			/* NOT REACHED */
231 		default:
232 			break;
233 		}
234 	}
235 
236 	mode = udf_permtomode(node);
237 
238 	if (cred->cr_uid == 0)
239 		return(0);
240 
241 	mask = 0;
242 
243 	/* Otherwise, check the owner. */
244 	if (cred->cr_uid == node->fentry->uid) {
245 		if (a_mode & VEXEC)
246 			mask |= S_IXUSR;
247 		if (a_mode & VREAD)
248 			mask |= S_IRUSR;
249 		if (a_mode & VWRITE)
250 			mask |= S_IWUSR;
251 		return((mode & mask) == mask ? 0 : EACCES);
252 	}
253 
254 	/* Otherwise, check the groups. */
255 	for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
256 		if (node->fentry->gid == *gp) {
257 			if (a_mode & VEXEC)
258 				mask |= S_IXGRP;
259 			if (a_mode & VREAD)
260 				mask |= S_IRGRP;
261 			if (a_mode & VWRITE)
262 				mask |= S_IWGRP;
263 			return((mode & mask) == mask ? 0 : EACCES);
264 		}
265 
266 	/* Otherwise, check everyone else. */
267 	if (a_mode & VEXEC)
268 		mask |= S_IXOTH;
269 	if (a_mode & VREAD)
270 		mask |= S_IROTH;
271 	if (a_mode & VWRITE)
272 		mask |= S_IWOTH;
273 	return((mode & mask) == mask ? 0 : EACCES);
274 }
275 
276 static int mon_lens[2][12] = {
277 	{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31},
278 	{31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31}
279 };
280 
281 static int
282 udf_isaleapyear(int year)
283 {
284 	int i;
285 
286 	i = (year % 4) ? 0 : 1;
287 	i &= (year % 100) ? 1 : 0;
288 	i |= (year % 400) ? 0 : 1;
289 
290 	return(i);
291 }
292 
293 /*
294  * XXX This is just a rough hack.  Daylight savings isn't calculated and tv_nsec
295  * is ignored.
296  * Timezone calculation compliments of Julian Elischer <julian@elischer.org>.
297  */
298 static void
299 udf_timetotimespec(struct timestamp *time, struct timespec *t)
300 {
301 	int i, lpyear, daysinyear;
302 	union {
303 		uint16_t	u_tz_offset;
304 		int16_t		s_tz_offset;
305 	} tz;
306 
307 	t->tv_nsec = 0;
308 
309 	/* DirectCD seems to like using bogus year values */
310 	if (time->year < 1970) {
311 		t->tv_sec = 0;
312 		return;
313 	}
314 
315 	/* Calculate the time and day */
316 	t->tv_sec = time->second;
317 	t->tv_sec += time->minute * 60;
318 	t->tv_sec += time->hour * 3600;
319 	t->tv_sec += time->day * 3600 * 24;
320 
321 	/* Calclulate the month */
322 	lpyear = udf_isaleapyear(time->year);
323 	for (i = 1; i < time->month; i++)
324 		t->tv_sec += mon_lens[lpyear][i] * 3600 * 24;
325 
326 	/* Speed up the calculation */
327 	if (time->year > 1979)
328 		t->tv_sec += 315532800;
329 	if (time->year > 1989)
330 		t->tv_sec += 315619200;
331 	if (time->year > 1999)
332 		t->tv_sec += 315532800;
333 	for (i = 2000; i < time->year; i++) {
334 		daysinyear = udf_isaleapyear(i) + 365 ;
335 		t->tv_sec += daysinyear * 3600 * 24;
336 	}
337 
338 	/*
339 	 * Calculate the time zone.  The timezone is 12 bit signed 2's
340 	 * compliment, so we gotta do some extra magic to handle it right.
341 	 */
342 	tz.u_tz_offset = time->type_tz;
343 	tz.u_tz_offset &= 0x0fff;
344 	if (tz.u_tz_offset & 0x0800)
345 		tz.u_tz_offset |= 0xf000;	/* extend the sign to 16 bits */
346 	if ((time->type_tz & 0x1000) && (tz.s_tz_offset != -2047))
347 		t->tv_sec -= tz.s_tz_offset * 60;
348 
349 	return;
350 }
351 
352 static int
353 udf_getattr(struct vop_getattr_args *a)
354 {
355 	struct vnode *vp;
356 	struct udf_node *node;
357 	struct vattr *vap;
358 	struct file_entry *fentry;
359 	struct timespec ts;
360 
361 	ts.tv_sec = 0;
362 
363 	vp = a->a_vp;
364 	vap = a->a_vap;
365 	node = VTON(vp);
366 	fentry = node->fentry;
367 
368 	vap->va_fsid = dev2udev(node->i_dev);
369 	vap->va_fileid = node->hash_id;
370 	vap->va_mode = udf_permtomode(node);
371 	vap->va_nlink = fentry->link_cnt;
372 	/*
373 	 * XXX The spec says that -1 is valid for uid/gid and indicates an
374 	 * invalid uid/gid.  How should this be represented?
375 	 */
376 	vap->va_uid = (fentry->uid == 0xffffffff) ? 0 : fentry->uid;
377 	vap->va_gid = (fentry->gid == 0xffffffff) ? 0 : fentry->gid;
378 	udf_timetotimespec(&fentry->atime, &vap->va_atime);
379 	udf_timetotimespec(&fentry->mtime, &vap->va_mtime);
380 	vap->va_ctime = vap->va_mtime; /* XXX Stored as an Extended Attribute */
381 	vap->va_rdev = 0; /* XXX */
382 	if (vp->v_type & VDIR) {
383 		/*
384 		 * Directories that are recorded within their ICB will show
385 		 * as having 0 blocks recorded.  Since tradition dictates
386 		 * that directories consume at least one logical block,
387 		 * make it appear so.
388 		 */
389 		if (fentry->logblks_rec != 0)
390 			vap->va_size = fentry->logblks_rec * node->udfmp->bsize;
391 		else
392 			vap->va_size = node->udfmp->bsize;
393 	} else
394 		vap->va_size = fentry->inf_len;
395 	vap->va_flags = 0;
396 	vap->va_gen = 1;
397 	vap->va_blocksize = node->udfmp->bsize;
398 	vap->va_bytes = fentry->inf_len;
399 	vap->va_type = vp->v_type;
400 	vap->va_filerev = 0; /* XXX */
401 	return(0);
402 }
403 
404 /*
405  * File specific ioctls.  DeCSS candidate?
406  */
407 static int
408 udf_ioctl(struct vop_ioctl_args *a)
409 {
410 	printf("%s called\n", __FUNCTION__);
411 	return(ENOTTY);
412 }
413 
414 /*
415  * I'm not sure that this has much value in a read-only filesystem, but
416  * cd9660 has it too.
417  */
418 static int
419 udf_pathconf(struct vop_pathconf_args *a)
420 {
421 
422 	switch (a->a_name) {
423 	case _PC_LINK_MAX:
424 		*a->a_retval = 65535;
425 		return(0);
426 	case _PC_NAME_MAX:
427 		*a->a_retval = NAME_MAX;
428 		return(0);
429 	case _PC_PATH_MAX:
430 		*a->a_retval = PATH_MAX;
431 		return(0);
432 	case _PC_NO_TRUNC:
433 		*a->a_retval = 1;
434 		return(0);
435 	default:
436 		return(EINVAL);
437 	}
438 }
439 
440 static int
441 udf_read(struct vop_read_args *a)
442 {
443 	struct vnode *vp = a->a_vp;
444 	struct uio *uio = a->a_uio;
445 	struct udf_node *node = VTON(vp);
446 	struct buf *bp;
447 	uint8_t *data;
448 	int error = 0;
449 	int size, fsize, offset;
450 
451 	if (uio->uio_offset < 0)
452 		return(EINVAL);
453 
454 	fsize = node->fentry->inf_len;
455 
456 	while (uio->uio_offset < fsize && uio->uio_resid > 0) {
457 		offset = uio->uio_offset;
458 		size = uio->uio_resid;
459 		error = udf_readatoffset(node, &size, offset, &bp, &data);
460 		if (error == 0)
461 			error = uiomove(data, size, uio);
462 		if (bp != NULL)
463 			brelse(bp);
464 		if (error)
465 			break;
466 	}
467 
468 	return(error);
469 }
470 
471 /*
472  * Call the OSTA routines to translate the name from a CS0 dstring to a
473  * 16-bit Unicode String.  Hooks need to be placed in here to translate from
474  * Unicode to the encoding that the kernel/user expects.  Return the length
475  * of the translated string.
476  */
477 static int
478 udf_transname(char *cs0string, char *destname, int len, struct udf_mnt *udfmp)
479 {
480 	unicode_t *transname;
481 	int i, unilen = 0, destlen;
482 
483 	/* Convert 16-bit Unicode to destname */
484 	/* allocate a buffer big enough to hold an 8->16 bit expansion */
485 	transname = malloc(MAXNAMLEN * sizeof(unicode_t), M_TEMP, M_WAITOK | M_ZERO);
486 
487 	if ((unilen = udf_UncompressUnicode(len, cs0string, transname)) == -1) {
488 		printf("udf: Unicode translation failed\n");
489 		free(transname, M_TEMP);
490 		return(0);
491 	}
492 
493 	for (i = 0; i < unilen ; i++)
494 		if (transname[i] & 0xff00)
495 			destname[i] = '.';	/* Fudge the 16bit chars */
496 		else
497 			destname[i] = transname[i] & 0xff;
498 	free(transname, M_TEMP);
499 	destname[unilen] = 0;
500 	destlen = unilen;
501 
502 	return(destlen);
503 }
504 
505 /*
506  * Compare a CS0 dstring with a name passed in from the VFS layer.  Return
507  * 0 on a successful match, nonzero therwise.  Unicode work may need to be done
508  * here also.
509  */
510 static int
511 udf_cmpname(char *cs0string, char *cmpname, int cs0len, int cmplen, struct udf_mnt *udfmp)
512 {
513 	char *transname;
514 	int error = 0;
515 
516 	/* This is overkill, but not worth creating a new zone */
517 
518 	transname = malloc(MAXNAMLEN * sizeof(unicode_t), M_TEMP,
519 			   M_WAITOK | M_ZERO);
520 
521 	cs0len = udf_transname(cs0string, transname, cs0len, udfmp);
522 
523 	/* Easy check.  If they aren't the same length, they aren't equal */
524 	if ((cs0len == 0) || (cs0len != cmplen))
525 		error = -1;
526 	else
527 		error = bcmp(transname, cmpname, cmplen);
528 
529 	free(transname, M_TEMP);
530 	return(error);
531 }
532 
533 struct udf_uiodir {
534 	struct dirent *dirent;
535 	u_long *cookies;
536 	int ncookies;
537 	int acookies;
538 	int eofflag;
539 };
540 
541 static int
542 udf_uiodir(struct udf_uiodir *uiodir, int de_size, struct uio *uio, long cookie)
543 {
544 	if (uiodir->cookies != NULL) {
545 		if (++uiodir->acookies > uiodir->ncookies) {
546 			uiodir->eofflag = 0;
547 			return (-1);
548 		}
549 		*uiodir->cookies++ = cookie;
550 	}
551 
552 	if (uio->uio_resid < de_size) {
553 		uiodir->eofflag = 0;
554 		return(-1);
555 	}
556 
557 	return(uiomove((caddr_t)uiodir->dirent, de_size, uio));
558 }
559 
560 static struct udf_dirstream *
561 udf_opendir(struct udf_node *node, int offset, int fsize, struct udf_mnt *udfmp)
562 {
563 	struct udf_dirstream *ds;
564 
565 	ds = malloc(sizeof(*ds), M_UDFDS, M_WAITOK | M_ZERO);
566 
567 	ds->node = node;
568 	ds->offset = offset;
569 	ds->udfmp = udfmp;
570 	ds->fsize = fsize;
571 
572 	return(ds);
573 }
574 
575 static struct fileid_desc *
576 udf_getfid(struct udf_dirstream *ds)
577 {
578 	struct fileid_desc *fid;
579 	int error, frag_size = 0, total_fid_size;
580 
581 	/* End of directory? */
582 	if (ds->offset + ds->off >= ds->fsize) {
583 		ds->error = 0;
584 		return(NULL);
585 	}
586 
587 	/* Grab the first extent of the directory */
588 	if (ds->off == 0) {
589 		ds->size = 0;
590 		if (ds->bp != NULL)
591 			brelse(ds->bp);
592 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
593 		    &ds->bp, &ds->data);
594 		if (error) {
595 			ds->error = error;
596 			return(NULL);
597 		}
598 	}
599 
600 	/*
601 	 * Clean up from a previous fragmented FID.
602 	 * XXX Is this the right place for this?
603 	 */
604 	if (ds->fid_fragment && ds->buf != NULL) {
605 		ds->fid_fragment = 0;
606 		free(ds->buf, M_UDFFID);
607 	}
608 
609 	fid = (struct fileid_desc*)&ds->data[ds->off];
610 
611 	/*
612 	 * Check to see if the fid is fragmented. The first test
613 	 * ensures that we don't wander off the end of the buffer
614 	 * looking for the l_iu and l_fi fields.
615 	 */
616 	if (ds->off + UDF_FID_SIZE > ds->size ||
617 	    ds->off + fid->l_iu + fid->l_fi + UDF_FID_SIZE > ds->size) {
618 
619 		/* Copy what we have of the fid into a buffer */
620 		frag_size = ds->size - ds->off;
621 		if (frag_size >= ds->udfmp->bsize) {
622 			printf("udf: invalid FID fragment\n");
623 			ds->error = EINVAL;
624 			return(NULL);
625 		}
626 
627 		/*
628 		 * File ID descriptors can only be at most one
629 		 * logical sector in size.
630 		 */
631 		ds->buf = malloc(ds->udfmp->bsize, M_UDFFID, M_WAITOK | M_ZERO);
632 		bcopy(fid, ds->buf, frag_size);
633 
634 		/* Reduce all of the casting magic */
635 		fid = (struct fileid_desc*)ds->buf;
636 
637 		if (ds->bp != NULL)
638 			brelse(ds->bp);
639 
640 		/* Fetch the next allocation */
641 		ds->offset += ds->size;
642 		ds->size = 0;
643 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
644 		    &ds->bp, &ds->data);
645 		if (error) {
646 			ds->error = error;
647 			return(NULL);
648 		}
649 
650 		/*
651 		 * If the fragment was so small that we didn't get
652 		 * the l_iu and l_fi fields, copy those in.
653 		 */
654 		if (frag_size < UDF_FID_SIZE)
655 			bcopy(ds->data, &ds->buf[frag_size],
656 			    UDF_FID_SIZE - frag_size);
657 
658 		/*
659 		 * Now that we have enough of the fid to work with,
660 		 * copy in the rest of the fid from the new
661 		 * allocation.
662 		 */
663 		total_fid_size = UDF_FID_SIZE + fid->l_iu + fid->l_fi;
664 		if (total_fid_size > ds->udfmp->bsize) {
665 			printf("udf: invalid FID\n");
666 			ds->error = EIO;
667 			return(NULL);
668 		}
669 		bcopy(ds->data, &ds->buf[frag_size],
670 		    total_fid_size - frag_size);
671 
672 		ds->fid_fragment = 1;
673 	} else
674 		total_fid_size = fid->l_iu + fid->l_fi + UDF_FID_SIZE;
675 
676 	/*
677 	 * Update the offset. Align on a 4 byte boundary because the
678 	 * UDF spec says so.
679 	 */
680 	ds->this_off = ds->off;
681 	if (!ds->fid_fragment)
682 		ds->off += (total_fid_size + 3) & ~0x03;
683 	else
684 		ds->off = (total_fid_size - frag_size + 3) & ~0x03;
685 
686 	return(fid);
687 }
688 
689 static void
690 udf_closedir(struct udf_dirstream *ds)
691 {
692 
693 	if (ds->bp != NULL)
694 		brelse(ds->bp);
695 
696 	if (ds->fid_fragment && ds->buf != NULL)
697 		free(ds->buf, M_UDFFID);
698 
699 	free(ds, M_UDFDS);
700 }
701 
702 static int
703 udf_readdir(struct vop_readdir_args *a)
704 {
705 	struct vnode *vp;
706 	struct uio *uio;
707 	struct dirent dir;
708 	struct udf_node *node;
709 	struct udf_mnt *udfmp;
710 	struct fileid_desc *fid;
711 	struct udf_uiodir uiodir;
712 	struct udf_dirstream *ds;
713 	u_long *cookies = NULL;
714 	int ncookies;
715 	int error = 0;
716 
717 	vp = a->a_vp;
718 	uio = a->a_uio;
719 	node = VTON(vp);
720 	udfmp = node->udfmp;
721 	uiodir.eofflag = 1;
722 
723 	if (a->a_ncookies != NULL) {
724 		/*
725 		 * Guess how many entries are needed.  If we run out, this
726 		 * function will be called again and thing will pick up were
727 		 * it left off.
728 		 */
729 		ncookies = uio->uio_resid / 8;
730 		cookies = malloc(sizeof(u_long) * ncookies, M_TEMP, M_WAITOK);
731 		uiodir.ncookies = ncookies;
732 		uiodir.cookies = cookies;
733 		uiodir.acookies = 0;
734 	} else
735 		uiodir.cookies = NULL;
736 
737 	/*
738 	 * Iterate through the file id descriptors.  Give the parent dir
739 	 * entry special attention.
740 	 */
741 	ds = udf_opendir(node, uio->uio_offset, node->fentry->inf_len,
742 			 node->udfmp);
743 
744 	while ((fid = udf_getfid(ds)) != NULL) {
745 
746 		/* XXX Should we return an error on a bad fid? */
747 		if (udf_checktag(&fid->tag, TAGID_FID)) {
748 			printf("Invalid FID tag\n");
749 			error = EIO;
750 			break;
751 		}
752 
753 		/* Is this a deleted file? */
754 		if (fid->file_char & UDF_FILE_CHAR_DEL)
755 			continue;
756 
757 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
758 			/* Do up the '.' and '..' entries.  Dummy values are
759 			 * used for the cookies since the offset here is
760 			 * usually zero, and NFS doesn't like that value
761 			 */
762 			dir.d_fileno = node->hash_id;
763 			dir.d_type = DT_DIR;
764 			dir.d_name[0] = '.';
765 			dir.d_namlen = 1;
766 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
767 			uiodir.dirent = &dir;
768 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, 1);
769 			if (error)
770 				break;
771 
772 			dir.d_fileno = udf_getid(&fid->icb);
773 			dir.d_type = DT_DIR;
774 			dir.d_name[0] = '.';
775 			dir.d_name[1] = '.';
776 			dir.d_namlen = 2;
777 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
778 			uiodir.dirent = &dir;
779 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, 2);
780 		} else {
781 			dir.d_namlen = udf_transname(&fid->data[fid->l_iu],
782 			    &dir.d_name[0], fid->l_fi, udfmp);
783 			dir.d_fileno = udf_getid(&fid->icb);
784 			dir.d_type = (fid->file_char & UDF_FILE_CHAR_DIR) ?
785 			    DT_DIR : DT_UNKNOWN;
786 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
787 			uiodir.dirent = &dir;
788 			error = udf_uiodir(&uiodir, dir.d_reclen, uio,
789 			    ds->this_off);
790 		}
791 		if (error) {
792 			printf("uiomove returned %d\n", error);
793 			break;
794 		}
795 
796 	}
797 
798 	/* tell the calling layer whether we need to be called again */
799 	*a->a_eofflag = uiodir.eofflag;
800 	uio->uio_offset = ds->offset + ds->off;
801 
802 	if (!error)
803 		error = ds->error;
804 
805 	udf_closedir(ds);
806 
807 	if (a->a_ncookies != NULL) {
808 		if (error)
809 			free(cookies, M_TEMP);
810 		else {
811 			*a->a_ncookies = uiodir.acookies;
812 			*a->a_cookies = cookies;
813 		}
814 	}
815 
816 	return(error);
817 }
818 
819 /* Are there any implementations out there that do soft-links? */
820 static int
821 udf_readlink(struct vop_readlink_args *ap)
822 {
823 	printf("%s called\n", __FUNCTION__);
824 	return(EOPNOTSUPP);
825 }
826 
827 static int
828 udf_strategy(struct vop_strategy_args *a)
829 {
830 	struct buf *bp;
831 	struct vnode *vp;
832 	struct udf_node *node;
833 	int maxsize;
834 
835 	bp = a->a_bp;
836 	vp = bp->b_vp;
837 	node = VTON(vp);
838 
839 	KASSERT(a->a_vp == a->a_bp->b_vp, ("%s(%p != %p)",
840 		__func__, a->a_vp, a->a_bp->b_vp));
841 	/* cd9660 has this test reversed, but it seems more logical this way */
842 	if (bp->b_blkno != bp->b_lblkno) {
843 		/*
844 		 * Files that are embedded in the fentry don't translate well
845 		 * to a block number.  Reject.
846 		 */
847 		if (udf_bmap_internal(node, bp->b_lblkno * node->udfmp->bsize,
848 		    &bp->b_lblkno, &maxsize)) {
849 			clrbuf(bp);
850 			bp->b_blkno = -1;
851 		}
852 	}
853 	if ((long)bp->b_blkno == -1) {
854 		biodone(bp);
855 		return(0);
856 	}
857 	vp = node->i_devvp;
858 	bp->b_dev = vp->v_rdev;
859 	bp->b_offset = dbtob(bp->b_blkno);
860 	VOP_STRATEGY(vp, bp);
861 	return(0);
862 }
863 
864 static int
865 udf_bmap(struct vop_bmap_args *a)
866 {
867 	struct udf_node *node;
868 	uint32_t max_size;
869 	daddr_t lsector;
870 	int error;
871 
872 	node = VTON(a->a_vp);
873 
874 	if (a->a_vpp != NULL)
875 		*a->a_vpp = node->i_devvp;
876 	if (a->a_bnp == NULL)
877 		return(0);
878 	if (a->a_runb)
879 		*a->a_runb = 0;
880 
881 	error = udf_bmap_internal(node, a->a_bn * node->udfmp->bsize, &lsector,
882 				  &max_size);
883 	if (error)
884 		return(error);
885 
886 	/* Translate logical to physical sector number */
887 	*a->a_bnp = lsector << (node->udfmp->bshift - DEV_BSHIFT);
888 
889 	/* Punt on read-ahead for now */
890 	if (a->a_runp)
891 		*a->a_runp = 0;
892 
893 	return(0);
894 }
895 
896 /*
897  * The all powerful VOP_LOOKUP().
898  */
899 static int
900 udf_lookup(struct vop_cachedlookup_args *a)
901 {
902 	struct vnode *dvp;
903 	struct vnode *tdp = NULL;
904 	struct vnode **vpp = a->a_vpp;
905 	struct udf_node *node;
906 	struct udf_mnt *udfmp;
907 	struct fileid_desc *fid = NULL;
908 	struct udf_dirstream *ds;
909 	struct thread *td;
910 	globaldata_t gd = mycpu;
911 	u_long nameiop;
912 	u_long flags;
913 	char *nameptr;
914 	long namelen;
915 	ino_t id = 0;
916 	int offset, error = 0;
917 	int numdirpasses, fsize;
918 
919 	dvp = a->a_dvp;
920 	node = VTON(dvp);
921 	udfmp = node->udfmp;
922 	nameiop = a->a_cnp->cn_nameiop;
923 	flags = a->a_cnp->cn_flags;
924 	nameptr = a->a_cnp->cn_nameptr;
925 	namelen = a->a_cnp->cn_namelen;
926 	fsize = node->fentry->inf_len;
927 	td = a->a_cnp->cn_td;
928 
929 	/*
930 	 * If this is a LOOKUP and we've already partially searched through
931 	 * the directory, pick up where we left off and flag that the
932 	 * directory may need to be searched twice.  For a full description,
933 	 * see /sys/isofs/cd9660/cd9660_lookup.c:cd9660_lookup()
934 	 */
935 	if (nameiop != NAMEI_LOOKUP || node->diroff == 0 ||
936 	    node->diroff > fsize) {
937 		offset = 0;
938 		numdirpasses = 1;
939 	} else {
940 		offset = node->diroff;
941 		numdirpasses = 2;
942 		gd->gd_nchstats->ncs_2passes++;
943 	}
944 
945 lookloop:
946 	ds = udf_opendir(node, offset, fsize, udfmp);
947 
948 	while ((fid = udf_getfid(ds)) != NULL) {
949 		/* XXX Should we return an error on a bad fid? */
950 		if (udf_checktag(&fid->tag, TAGID_FID)) {
951 			printf("udf_lookup: Invalid tag\n");
952 			error = EIO;
953 			break;
954 		}
955 
956 		/* Is this a deleted file? */
957 		if (fid->file_char & UDF_FILE_CHAR_DEL)
958 			continue;
959 
960 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
961 			if (flags & CNP_ISDOTDOT) {
962 				id = udf_getid(&fid->icb);
963 				break;
964 			}
965 		} else {
966 			if (!(udf_cmpname(&fid->data[fid->l_iu],
967 					  nameptr, fid->l_fi, namelen, udfmp))) {
968 				id = udf_getid(&fid->icb);
969 				break;
970 			}
971 		}
972 	}
973 
974 	if (!error)
975 		error = ds->error;
976 
977 	/* XXX Bail out here? */
978 	if (error) {
979 		udf_closedir(ds);
980 		return (error);
981 	}
982 
983 	/* Did we have a match? */
984 	if (id) {
985 		error = udf_vget(udfmp->im_mountp, id, &tdp);
986 		if (!error) {
987 			/*
988 			 * Remember where this entry was if it's the final
989 			 * component.
990 			 */
991 			if ((flags & CNP_ISLASTCN) && nameiop == NAMEI_LOOKUP)
992 				node->diroff = ds->offset + ds->off;
993 			if (numdirpasses == 2)
994 				gd->gd_nchstats->ncs_pass2++;
995 			if (!(flags & CNP_LOCKPARENT) || !(flags & CNP_ISLASTCN)) {
996 				a->a_cnp->cn_flags |= CNP_PDIRUNLOCK;
997 				VOP_UNLOCK(dvp, 0, td);
998 			}
999 
1000 			*vpp = tdp;
1001 
1002 			/* Put this entry in the cache */
1003 			if (flags & CNP_MAKEENTRY)
1004 				cache_enter(dvp, *vpp, a->a_cnp);
1005 		}
1006 	} else {
1007 		/* Name wasn't found on this pass.  Do another pass? */
1008 		if (numdirpasses == 2) {
1009 			numdirpasses--;
1010 			offset = 0;
1011 			udf_closedir(ds);
1012 			goto lookloop;
1013 		}
1014 
1015 		/* Enter name into cache as non-existant */
1016 		if (flags & CNP_MAKEENTRY)
1017 			cache_enter(dvp, *vpp, a->a_cnp);
1018 
1019 		if ((flags & CNP_ISLASTCN) &&
1020 		    (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME)) {
1021 			error = EROFS;
1022 		} else {
1023 			error = ENOENT;
1024 		}
1025 	}
1026 
1027 	udf_closedir(ds);
1028 	return(error);
1029 }
1030 
1031 static int
1032 udf_reclaim(struct vop_reclaim_args *a)
1033 {
1034 	struct vnode *vp;
1035 	struct udf_node *unode;
1036 
1037 	vp = a->a_vp;
1038 	unode = VTON(vp);
1039 
1040 	if (unode != NULL) {
1041 		udf_hashrem(unode);
1042 		if (unode->i_devvp) {
1043 			vrele(unode->i_devvp);
1044 			unode->i_devvp = 0;
1045 		}
1046 
1047 		if (unode->fentry != NULL)
1048 			free(unode->fentry, M_UDFFENTRY);
1049 		free(unode, M_UDFNODE);
1050 		vp->v_data = NULL;
1051 	}
1052 
1053 	return(0);
1054 }
1055 
1056 /*
1057  * Read the block and then set the data pointer to correspond with the
1058  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1059  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1060  * whole extent.
1061  *
1062  * Note that *bp may be assigned error or not.
1063  *
1064  * XXX 'size' is limited to the logical block size for now due to problems
1065  * with udf_read()
1066  */
1067 static int
1068 udf_readatoffset(struct udf_node *node, int *size, int offset, struct buf **bp,
1069 		 uint8_t **data)
1070 {
1071 	struct udf_mnt *udfmp;
1072 	struct file_entry *fentry = NULL;
1073 	struct buf *bp1;
1074 	uint32_t max_size;
1075 	daddr_t sector;
1076 	int error;
1077 
1078 	udfmp = node->udfmp;
1079 
1080 	*bp = NULL;
1081 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1082 	if (error == UDF_INVALID_BMAP) {
1083 		/*
1084 		 * This error means that the file *data* is stored in the
1085 		 * allocation descriptor field of the file entry.
1086 		 */
1087 		fentry = node->fentry;
1088 		*data = &fentry->data[fentry->l_ea];
1089 		*size = fentry->l_ad;
1090 		return(0);
1091 	} else if (error != 0) {
1092 		return(error);
1093 	}
1094 
1095 	/* Adjust the size so that it is within range */
1096 	if (*size == 0 || *size > max_size)
1097 		*size = max_size;
1098 	*size = min(*size, MAXBSIZE);
1099 
1100 	if ((error = udf_readlblks(udfmp, sector, *size, bp))) {
1101 		printf("warning: udf_readlblks returned error %d\n", error);
1102 		/* note: *bp may be non-NULL */
1103 		return(error);
1104 	}
1105 
1106 	bp1 = *bp;
1107 	*data = (uint8_t *)&bp1->b_data[offset % udfmp->bsize];
1108 	return(0);
1109 }
1110 
1111 /*
1112  * Translate a file offset into a logical block and then into a physical
1113  * block.
1114  */
1115 static int
1116 udf_bmap_internal(struct udf_node *node, uint32_t offset, daddr_t *sector, uint32_t *max_size)
1117 {
1118 	struct udf_mnt *udfmp;
1119 	struct file_entry *fentry;
1120 	void *icb;
1121 	struct icb_tag *tag;
1122 	uint32_t icblen = 0;
1123 	daddr_t lsector;
1124 	int ad_offset, ad_num = 0;
1125 	int i, p_offset;
1126 
1127 	udfmp = node->udfmp;
1128 	fentry = node->fentry;
1129 	tag = &fentry->icbtag;
1130 
1131 	switch (tag->strat_type) {
1132 	case 4:
1133 		break;
1134 
1135 	case 4096:
1136 		printf("Cannot deal with strategy4096 yet!\n");
1137 		return(ENODEV);
1138 
1139 	default:
1140 		printf("Unknown strategy type %d\n", tag->strat_type);
1141 		return(ENODEV);
1142 	}
1143 
1144 	switch (tag->flags & 0x7) {
1145 	case 0:
1146 		/*
1147 		 * The allocation descriptor field is filled with short_ad's.
1148 		 * If the offset is beyond the current extent, look for the
1149 		 * next extent.
1150 		 */
1151 		do {
1152 			offset -= icblen;
1153 			ad_offset = sizeof(struct short_ad) * ad_num;
1154 			if (ad_offset > fentry->l_ad) {
1155 				printf("File offset out of bounds\n");
1156 				return(EINVAL);
1157 			}
1158 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1159 			icblen = GETICBLEN(short_ad, icb);
1160 			ad_num++;
1161 		} while(offset >= icblen);
1162 
1163 		lsector = (offset  >> udfmp->bshift) +
1164 		    ((struct short_ad *)(icb))->pos;
1165 
1166 		*max_size = GETICBLEN(short_ad, icb);
1167 
1168 		break;
1169 	case 1:
1170 		/*
1171 		 * The allocation descriptor field is filled with long_ad's
1172 		 * If the offset is beyond the current extent, look for the
1173 		 * next extent.
1174 		 */
1175 		do {
1176 			offset -= icblen;
1177 			ad_offset = sizeof(struct long_ad) * ad_num;
1178 			if (ad_offset > fentry->l_ad) {
1179 				printf("File offset out of bounds\n");
1180 				return(EINVAL);
1181 			}
1182 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1183 			icblen = GETICBLEN(long_ad, icb);
1184 			ad_num++;
1185 		} while(offset >= icblen);
1186 
1187 		lsector = (offset >> udfmp->bshift) +
1188 		    ((struct long_ad *)(icb))->loc.lb_num;
1189 
1190 		*max_size = GETICBLEN(long_ad, icb);
1191 
1192 		break;
1193 	case 3:
1194 		/*
1195 		 * This type means that the file *data* is stored in the
1196 		 * allocation descriptor field of the file entry.
1197 		 */
1198 		*max_size = 0;
1199 		*sector = node->hash_id + udfmp->part_start;
1200 
1201 		return(UDF_INVALID_BMAP);
1202 	case 2:
1203 		/* DirectCD does not use extended_ad's */
1204 	default:
1205 		printf("Unsupported allocation descriptor %d\n",
1206 		       tag->flags & 0x7);
1207 		return(ENODEV);
1208 	}
1209 
1210 	*sector = lsector + udfmp->part_start;
1211 
1212 	/*
1213 	 * Check the sparing table.  Each entry represents the beginning of
1214 	 * a packet.
1215 	 */
1216 	if (udfmp->s_table != NULL) {
1217 		for (i = 0; i< udfmp->s_table_entries; i++) {
1218 			p_offset = lsector - udfmp->s_table->entries[i].org;
1219 			if ((p_offset < udfmp->p_sectors) && (p_offset >= 0)) {
1220 				*sector = udfmp->s_table->entries[i].map +
1221 				    p_offset;
1222 				break;
1223 			}
1224 		}
1225 	}
1226 
1227 	return(0);
1228 }
1229