xref: /dflybsd-src/sys/vfs/udf/udf_vnops.c (revision 41871674d0079dec70d55eb824f39d07dc7b3310)
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 1.19 2006/03/24 18:35:34 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 <machine/inttypes.h>
49 
50 #include <vfs/udf/ecma167-udf.h>
51 #include <vfs/udf/osta.h>
52 #include <vfs/udf/udf.h>
53 #include <vfs/udf/udf_mount.h>
54 
55 static int udf_access(struct vop_access_args *);
56 static int udf_getattr(struct vop_getattr_args *);
57 static int udf_ioctl(struct vop_ioctl_args *);
58 static int udf_pathconf(struct vop_pathconf_args *);
59 static int udf_read(struct vop_read_args *);
60 static int udf_readdir(struct vop_readdir_args *);
61 static int udf_readlink(struct vop_readlink_args *ap);
62 static int udf_strategy(struct vop_strategy_args *);
63 static int udf_bmap(struct vop_bmap_args *);
64 static int udf_lookup(struct vop_old_lookup_args *);
65 static int udf_reclaim(struct vop_reclaim_args *);
66 static int udf_readatoffset(struct udf_node *, int *, int, struct buf **, uint8_t **);
67 static int udf_bmap_internal(struct udf_node *, uint32_t, daddr_t *, uint32_t *);
68 
69 struct vnodeopv_entry_desc udf_vnodeop_entries[] = {
70 	{ &vop_default_desc,		vop_defaultop },
71 	{ &vop_access_desc,		(vnodeopv_entry_t) udf_access },
72 	{ &vop_bmap_desc,		(vnodeopv_entry_t) udf_bmap },
73 	{ &vop_old_lookup_desc,		(vnodeopv_entry_t) udf_lookup },
74 	{ &vop_getattr_desc,		(vnodeopv_entry_t) udf_getattr },
75 	{ &vop_ioctl_desc,		(vnodeopv_entry_t) udf_ioctl },
76 	{ &vop_pathconf_desc,		(vnodeopv_entry_t) udf_pathconf },
77 	{ &vop_read_desc,		(vnodeopv_entry_t) udf_read },
78 	{ &vop_readdir_desc,		(vnodeopv_entry_t) udf_readdir },
79 	{ &vop_readlink_desc,		(vnodeopv_entry_t) udf_readlink },
80 	{ &vop_reclaim_desc,		(vnodeopv_entry_t) udf_reclaim },
81 	{ &vop_strategy_desc,		(vnodeopv_entry_t) udf_strategy },
82 	{ NULL, NULL }
83 };
84 
85 MALLOC_DEFINE(M_UDFFID, "UDF FID", "UDF FileId structure");
86 MALLOC_DEFINE(M_UDFDS, "UDF DS", "UDF Dirstream structure");
87 
88 #define UDF_INVALID_BMAP	-1
89 
90 /* Look up a udf_node based on the ino_t passed in and return it's vnode */
91 int
92 udf_hashlookup(struct udf_mnt *udfmp, ino_t id, struct vnode **vpp)
93 {
94 	struct udf_node *node;
95 	struct udf_hash_lh *lh;
96 	struct vnode *vp;
97 	lwkt_tokref hashlock;
98 
99 	*vpp = NULL;
100 
101 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
102 loop:
103 	lh = &udfmp->hashtbl[id % udfmp->hashsz];
104 	if (lh == NULL) {
105 		lwkt_reltoken(&hashlock);
106 		return(ENOENT);
107 	}
108 	LIST_FOREACH(node, lh, le) {
109 		if (node->hash_id != id)
110 			continue;
111 		vp = node->i_vnode;
112 		if (vget(vp, LK_EXCLUSIVE, curthread))
113 			goto loop;
114 		/*
115 		 * We must check to see if the inode has been ripped
116 		 * out from under us after blocking.
117 		 */
118 		lh = &udfmp->hashtbl[id % udfmp->hashsz];
119 		LIST_FOREACH(node, lh, le) {
120 			if (node->hash_id == id)
121 				break;
122 		}
123 		if (node == NULL || vp != node->i_vnode) {
124 			vput(vp);
125 			goto loop;
126 		}
127 		lwkt_reltoken(&hashlock);
128 		*vpp = vp;
129 		return(0);
130 	}
131 
132 	lwkt_reltoken(&hashlock);
133 	return(0);
134 }
135 
136 int
137 udf_hashins(struct udf_node *node)
138 {
139 	struct udf_mnt *udfmp;
140 	struct udf_hash_lh *lh;
141 	lwkt_tokref hashlock;
142 
143 	udfmp = node->udfmp;
144 
145 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
146 	lh = &udfmp->hashtbl[node->hash_id % udfmp->hashsz];
147 	if (lh == NULL)
148 		LIST_INIT(lh);
149 	LIST_INSERT_HEAD(lh, node, le);
150 	lwkt_reltoken(&hashlock);
151 
152 	return(0);
153 }
154 
155 int
156 udf_hashrem(struct udf_node *node)
157 {
158 	struct udf_mnt *udfmp;
159 	struct udf_hash_lh *lh;
160 	lwkt_tokref hashlock;
161 
162 	udfmp = node->udfmp;
163 
164 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
165 	lh = &udfmp->hashtbl[node->hash_id % udfmp->hashsz];
166 	if (lh == NULL)
167 		panic("hash entry is NULL, node->hash_id= %"PRId64"\n", node->hash_id);
168 	LIST_REMOVE(node, le);
169 	lwkt_reltoken(&hashlock);
170 
171 	return(0);
172 }
173 
174 int
175 udf_allocv(struct mount *mp, struct vnode **vpp)
176 {
177 	int error;
178 	struct vnode *vp;
179 
180 	error = getnewvnode(VT_UDF, mp, &vp, 0, 0);
181 	if (error) {
182 		printf("udf_allocv: failed to allocate new vnode\n");
183 		return(error);
184 	}
185 
186 	*vpp = vp;
187 	return(0);
188 }
189 
190 /* Convert file entry permission (5 bits per owner/group/user) to a mode_t */
191 static mode_t
192 udf_permtomode(struct udf_node *node)
193 {
194 	uint32_t perm;
195 	uint32_t flags;
196 	mode_t mode;
197 
198 	perm = node->fentry->perm;
199 	flags = node->fentry->icbtag.flags;
200 
201 	mode = perm & UDF_FENTRY_PERM_USER_MASK;
202 	mode |= ((perm & UDF_FENTRY_PERM_GRP_MASK) >> 2);
203 	mode |= ((perm & UDF_FENTRY_PERM_OWNER_MASK) >> 4);
204 	mode |= ((flags & UDF_ICB_TAG_FLAGS_STICKY) << 4);
205 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETGID) << 6);
206 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETUID) << 8);
207 
208 	return(mode);
209 }
210 
211 static int
212 udf_access(struct vop_access_args *a)
213 {
214 	struct vnode *vp;
215 	struct udf_node *node;
216 	mode_t a_mode, mode, mask;
217 	struct ucred *cred = a->a_cred;
218 	gid_t *gp;
219 	int i;
220 
221 	vp = a->a_vp;
222 	node = VTON(vp);
223 	a_mode = a->a_mode;
224 
225 	if (a_mode & VWRITE) {
226 		switch (vp->v_type) {
227 		case VDIR:
228 		case VLNK:
229 		case VREG:
230 			return(EROFS);
231 			/* NOT REACHED */
232 		default:
233 			break;
234 		}
235 	}
236 
237 	mode = udf_permtomode(node);
238 
239 	if (cred->cr_uid == 0)
240 		return(0);
241 
242 	mask = 0;
243 
244 	/* Otherwise, check the owner. */
245 	if (cred->cr_uid == node->fentry->uid) {
246 		if (a_mode & VEXEC)
247 			mask |= S_IXUSR;
248 		if (a_mode & VREAD)
249 			mask |= S_IRUSR;
250 		if (a_mode & VWRITE)
251 			mask |= S_IWUSR;
252 		return((mode & mask) == mask ? 0 : EACCES);
253 	}
254 
255 	/* Otherwise, check the groups. */
256 	for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
257 		if (node->fentry->gid == *gp) {
258 			if (a_mode & VEXEC)
259 				mask |= S_IXGRP;
260 			if (a_mode & VREAD)
261 				mask |= S_IRGRP;
262 			if (a_mode & VWRITE)
263 				mask |= S_IWGRP;
264 			return((mode & mask) == mask ? 0 : EACCES);
265 		}
266 
267 	/* Otherwise, check everyone else. */
268 	if (a_mode & VEXEC)
269 		mask |= S_IXOTH;
270 	if (a_mode & VREAD)
271 		mask |= S_IROTH;
272 	if (a_mode & VWRITE)
273 		mask |= S_IWOTH;
274 	return((mode & mask) == mask ? 0 : EACCES);
275 }
276 
277 static int mon_lens[2][12] = {
278 	{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31},
279 	{31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31}
280 };
281 
282 static int
283 udf_isaleapyear(int year)
284 {
285 	int i;
286 
287 	i = (year % 4) ? 0 : 1;
288 	i &= (year % 100) ? 1 : 0;
289 	i |= (year % 400) ? 0 : 1;
290 
291 	return(i);
292 }
293 
294 /*
295  * XXX This is just a rough hack.  Daylight savings isn't calculated and tv_nsec
296  * is ignored.
297  * Timezone calculation compliments of Julian Elischer <julian@elischer.org>.
298  */
299 static void
300 udf_timetotimespec(struct timestamp *time, struct timespec *t)
301 {
302 	int i, lpyear, daysinyear;
303 	union {
304 		uint16_t	u_tz_offset;
305 		int16_t		s_tz_offset;
306 	} tz;
307 
308 	t->tv_nsec = 0;
309 
310 	/* DirectCD seems to like using bogus year values */
311 	if (time->year < 1970) {
312 		t->tv_sec = 0;
313 		return;
314 	}
315 
316 	/* Calculate the time and day */
317 	t->tv_sec = time->second;
318 	t->tv_sec += time->minute * 60;
319 	t->tv_sec += time->hour * 3600;
320 	t->tv_sec += time->day * 3600 * 24;
321 
322 	/* Calclulate the month */
323 	lpyear = udf_isaleapyear(time->year);
324 	for (i = 1; i < time->month; i++)
325 		t->tv_sec += mon_lens[lpyear][i] * 3600 * 24;
326 
327 	/* Speed up the calculation */
328 	if (time->year > 1979)
329 		t->tv_sec += 315532800;
330 	if (time->year > 1989)
331 		t->tv_sec += 315619200;
332 	if (time->year > 1999)
333 		t->tv_sec += 315532800;
334 	for (i = 2000; i < time->year; i++) {
335 		daysinyear = udf_isaleapyear(i) + 365 ;
336 		t->tv_sec += daysinyear * 3600 * 24;
337 	}
338 
339 	/*
340 	 * Calculate the time zone.  The timezone is 12 bit signed 2's
341 	 * compliment, so we gotta do some extra magic to handle it right.
342 	 */
343 	tz.u_tz_offset = time->type_tz;
344 	tz.u_tz_offset &= 0x0fff;
345 	if (tz.u_tz_offset & 0x0800)
346 		tz.u_tz_offset |= 0xf000;	/* extend the sign to 16 bits */
347 	if ((time->type_tz & 0x1000) && (tz.s_tz_offset != -2047))
348 		t->tv_sec -= tz.s_tz_offset * 60;
349 
350 	return;
351 }
352 
353 static int
354 udf_getattr(struct vop_getattr_args *a)
355 {
356 	struct vnode *vp;
357 	struct udf_node *node;
358 	struct vattr *vap;
359 	struct file_entry *fentry;
360 	struct timespec ts;
361 
362 	ts.tv_sec = 0;
363 
364 	vp = a->a_vp;
365 	vap = a->a_vap;
366 	node = VTON(vp);
367 	fentry = node->fentry;
368 
369 	vap->va_fsid = dev2udev(node->i_dev);
370 	vap->va_fileid = node->hash_id;
371 	vap->va_mode = udf_permtomode(node);
372 	vap->va_nlink = fentry->link_cnt;
373 	/*
374 	 * XXX The spec says that -1 is valid for uid/gid and indicates an
375 	 * invalid uid/gid.  How should this be represented?
376 	 */
377 	vap->va_uid = (fentry->uid == 0xffffffff) ? 0 : fentry->uid;
378 	vap->va_gid = (fentry->gid == 0xffffffff) ? 0 : fentry->gid;
379 	udf_timetotimespec(&fentry->atime, &vap->va_atime);
380 	udf_timetotimespec(&fentry->mtime, &vap->va_mtime);
381 	vap->va_ctime = vap->va_mtime; /* XXX Stored as an Extended Attribute */
382 	vap->va_rdev = 0; /* XXX */
383 	if (vp->v_type & VDIR) {
384 		/*
385 		 * Directories that are recorded within their ICB will show
386 		 * as having 0 blocks recorded.  Since tradition dictates
387 		 * that directories consume at least one logical block,
388 		 * make it appear so.
389 		 */
390 		if (fentry->logblks_rec != 0)
391 			vap->va_size = fentry->logblks_rec * node->udfmp->bsize;
392 		else
393 			vap->va_size = node->udfmp->bsize;
394 	} else
395 		vap->va_size = fentry->inf_len;
396 	vap->va_flags = 0;
397 	vap->va_gen = 1;
398 	vap->va_blocksize = node->udfmp->bsize;
399 	vap->va_bytes = fentry->inf_len;
400 	vap->va_type = vp->v_type;
401 	vap->va_filerev = 0; /* XXX */
402 	return(0);
403 }
404 
405 /*
406  * File specific ioctls.  DeCSS candidate?
407  */
408 static int
409 udf_ioctl(struct vop_ioctl_args *a)
410 {
411 	printf("%s called\n", __func__);
412 	return(ENOTTY);
413 }
414 
415 /*
416  * I'm not sure that this has much value in a read-only filesystem, but
417  * cd9660 has it too.
418  */
419 static int
420 udf_pathconf(struct vop_pathconf_args *a)
421 {
422 
423 	switch (a->a_name) {
424 	case _PC_LINK_MAX:
425 		*a->a_retval = 65535;
426 		return(0);
427 	case _PC_NAME_MAX:
428 		*a->a_retval = NAME_MAX;
429 		return(0);
430 	case _PC_PATH_MAX:
431 		*a->a_retval = PATH_MAX;
432 		return(0);
433 	case _PC_NO_TRUNC:
434 		*a->a_retval = 1;
435 		return(0);
436 	default:
437 		return(EINVAL);
438 	}
439 }
440 
441 static int
442 udf_read(struct vop_read_args *a)
443 {
444 	struct vnode *vp = a->a_vp;
445 	struct uio *uio = a->a_uio;
446 	struct udf_node *node = VTON(vp);
447 	struct buf *bp;
448 	uint8_t *data;
449 	int error = 0;
450 	int size, fsize, offset;
451 
452 	if (uio->uio_offset < 0)
453 		return(EINVAL);
454 
455 	fsize = node->fentry->inf_len;
456 
457 	while (uio->uio_offset < fsize && uio->uio_resid > 0) {
458 		offset = uio->uio_offset;
459 		size = uio->uio_resid;
460 		error = udf_readatoffset(node, &size, offset, &bp, &data);
461 		if (error == 0)
462 			error = uiomove(data, size, uio);
463 		if (bp != NULL)
464 			brelse(bp);
465 		if (error)
466 			break;
467 	}
468 
469 	return(error);
470 }
471 
472 /*
473  * Call the OSTA routines to translate the name from a CS0 dstring to a
474  * 16-bit Unicode String.  Hooks need to be placed in here to translate from
475  * Unicode to the encoding that the kernel/user expects.  Return the length
476  * of the translated string.
477  */
478 static int
479 udf_transname(char *cs0string, char *destname, int len, struct udf_mnt *udfmp)
480 {
481 	unicode_t *transname;
482 	int i, unilen = 0, destlen;
483 
484 	/* Convert 16-bit Unicode to destname */
485 	/* allocate a buffer big enough to hold an 8->16 bit expansion */
486 	transname = malloc(NAME_MAX * sizeof(unicode_t), M_TEMP, M_WAITOK | M_ZERO);
487 
488 	if ((unilen = udf_UncompressUnicode(len, cs0string, transname)) == -1) {
489 		printf("udf: Unicode translation failed\n");
490 		free(transname, M_TEMP);
491 		return(0);
492 	}
493 
494 	for (i = 0; i < unilen ; i++)
495 		if (transname[i] & 0xff00)
496 			destname[i] = '.';	/* Fudge the 16bit chars */
497 		else
498 			destname[i] = transname[i] & 0xff;
499 	free(transname, M_TEMP);
500 	destname[unilen] = 0;
501 	destlen = unilen;
502 
503 	return(destlen);
504 }
505 
506 /*
507  * Compare a CS0 dstring with a name passed in from the VFS layer.  Return
508  * 0 on a successful match, nonzero therwise.  Unicode work may need to be done
509  * here also.
510  */
511 static int
512 udf_cmpname(char *cs0string, char *cmpname, int cs0len, int cmplen, struct udf_mnt *udfmp)
513 {
514 	char *transname;
515 	int error = 0;
516 
517 	/* This is overkill, but not worth creating a new zone */
518 
519 	transname = malloc(NAME_MAX * sizeof(unicode_t), M_TEMP,
520 			   M_WAITOK | M_ZERO);
521 
522 	cs0len = udf_transname(cs0string, transname, cs0len, udfmp);
523 
524 	/* Easy check.  If they aren't the same length, they aren't equal */
525 	if ((cs0len == 0) || (cs0len != cmplen))
526 		error = -1;
527 	else
528 		error = bcmp(transname, cmpname, cmplen);
529 
530 	free(transname, M_TEMP);
531 	return(error);
532 }
533 
534 struct udf_uiodir {
535 	struct dirent *dirent;
536 	u_long *cookies;
537 	int ncookies;
538 	int acookies;
539 	int eofflag;
540 };
541 
542 static struct udf_dirstream *
543 udf_opendir(struct udf_node *node, int offset, int fsize, struct udf_mnt *udfmp)
544 {
545 	struct udf_dirstream *ds;
546 
547 	ds = malloc(sizeof(*ds), M_UDFDS, M_WAITOK | M_ZERO);
548 
549 	ds->node = node;
550 	ds->offset = offset;
551 	ds->udfmp = udfmp;
552 	ds->fsize = fsize;
553 
554 	return(ds);
555 }
556 
557 static struct fileid_desc *
558 udf_getfid(struct udf_dirstream *ds)
559 {
560 	struct fileid_desc *fid;
561 	int error, frag_size = 0, total_fid_size;
562 
563 	/* End of directory? */
564 	if (ds->offset + ds->off >= ds->fsize) {
565 		ds->error = 0;
566 		return(NULL);
567 	}
568 
569 	/* Grab the first extent of the directory */
570 	if (ds->off == 0) {
571 		ds->size = 0;
572 		if (ds->bp != NULL)
573 			brelse(ds->bp);
574 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
575 		    &ds->bp, &ds->data);
576 		if (error) {
577 			ds->error = error;
578 			return(NULL);
579 		}
580 	}
581 
582 	/*
583 	 * Clean up from a previous fragmented FID.
584 	 * XXX Is this the right place for this?
585 	 */
586 	if (ds->fid_fragment && ds->buf != NULL) {
587 		ds->fid_fragment = 0;
588 		free(ds->buf, M_UDFFID);
589 	}
590 
591 	fid = (struct fileid_desc*)&ds->data[ds->off];
592 
593 	/*
594 	 * Check to see if the fid is fragmented. The first test
595 	 * ensures that we don't wander off the end of the buffer
596 	 * looking for the l_iu and l_fi fields.
597 	 */
598 	if (ds->off + UDF_FID_SIZE > ds->size ||
599 	    ds->off + fid->l_iu + fid->l_fi + UDF_FID_SIZE > ds->size) {
600 
601 		/* Copy what we have of the fid into a buffer */
602 		frag_size = ds->size - ds->off;
603 		if (frag_size >= ds->udfmp->bsize) {
604 			printf("udf: invalid FID fragment\n");
605 			ds->error = EINVAL;
606 			return(NULL);
607 		}
608 
609 		/*
610 		 * File ID descriptors can only be at most one
611 		 * logical sector in size.
612 		 */
613 		ds->buf = malloc(ds->udfmp->bsize, M_UDFFID, M_WAITOK | M_ZERO);
614 		bcopy(fid, ds->buf, frag_size);
615 
616 		/* Reduce all of the casting magic */
617 		fid = (struct fileid_desc*)ds->buf;
618 
619 		if (ds->bp != NULL)
620 			brelse(ds->bp);
621 
622 		/* Fetch the next allocation */
623 		ds->offset += ds->size;
624 		ds->size = 0;
625 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
626 		    &ds->bp, &ds->data);
627 		if (error) {
628 			ds->error = error;
629 			return(NULL);
630 		}
631 
632 		/*
633 		 * If the fragment was so small that we didn't get
634 		 * the l_iu and l_fi fields, copy those in.
635 		 */
636 		if (frag_size < UDF_FID_SIZE)
637 			bcopy(ds->data, &ds->buf[frag_size],
638 			    UDF_FID_SIZE - frag_size);
639 
640 		/*
641 		 * Now that we have enough of the fid to work with,
642 		 * copy in the rest of the fid from the new
643 		 * allocation.
644 		 */
645 		total_fid_size = UDF_FID_SIZE + fid->l_iu + fid->l_fi;
646 		if (total_fid_size > ds->udfmp->bsize) {
647 			printf("udf: invalid FID\n");
648 			ds->error = EIO;
649 			return(NULL);
650 		}
651 		bcopy(ds->data, &ds->buf[frag_size],
652 		    total_fid_size - frag_size);
653 
654 		ds->fid_fragment = 1;
655 	} else
656 		total_fid_size = fid->l_iu + fid->l_fi + UDF_FID_SIZE;
657 
658 	/*
659 	 * Update the offset. Align on a 4 byte boundary because the
660 	 * UDF spec says so.
661 	 */
662 	ds->this_off = ds->off;
663 	if (!ds->fid_fragment)
664 		ds->off += (total_fid_size + 3) & ~0x03;
665 	else
666 		ds->off = (total_fid_size - frag_size + 3) & ~0x03;
667 
668 	return(fid);
669 }
670 
671 static void
672 udf_closedir(struct udf_dirstream *ds)
673 {
674 
675 	if (ds->bp != NULL)
676 		brelse(ds->bp);
677 
678 	if (ds->fid_fragment && ds->buf != NULL)
679 		free(ds->buf, M_UDFFID);
680 
681 	free(ds, M_UDFDS);
682 }
683 
684 static int
685 udf_readdir(struct vop_readdir_args *a)
686 {
687 	struct vnode *vp;
688 	struct uio *uio;
689 	struct udf_node *node;
690 	struct udf_mnt *udfmp;
691 	struct fileid_desc *fid;
692 	struct udf_uiodir uiodir;
693 	struct udf_dirstream *ds;
694 	u_long *cookies = NULL;
695 	int ncookies;
696 	int error = 0;
697 	char *name;
698 
699 	vp = a->a_vp;
700 	uio = a->a_uio;
701 	node = VTON(vp);
702 	udfmp = node->udfmp;
703 	uiodir.eofflag = 1;
704 
705 	if (a->a_ncookies != NULL) {
706 		/*
707 		 * Guess how many entries are needed.  If we run out, this
708 		 * function will be called again and thing will pick up were
709 		 * it left off.
710 		 */
711 		ncookies = uio->uio_resid / 8;
712 		cookies = malloc(sizeof(u_long) * ncookies, M_TEMP, M_WAITOK);
713 		uiodir.ncookies = ncookies;
714 		uiodir.cookies = cookies;
715 		uiodir.acookies = 0;
716 	} else
717 		uiodir.cookies = NULL;
718 
719 	/*
720 	 * Iterate through the file id descriptors.  Give the parent dir
721 	 * entry special attention.
722 	 */
723 	ds = udf_opendir(node, uio->uio_offset, node->fentry->inf_len,
724 			 node->udfmp);
725 
726 	name = malloc(NAME_MAX, M_TEMP, M_WAITOK);
727 
728 	while ((fid = udf_getfid(ds)) != NULL) {
729 
730 		/* XXX Should we return an error on a bad fid? */
731 		if (udf_checktag(&fid->tag, TAGID_FID)) {
732 			printf("Invalid FID tag\n");
733 			error = EIO;
734 			break;
735 		}
736 
737 		/* Is this a deleted file? */
738 		if (fid->file_char & UDF_FILE_CHAR_DEL)
739 			continue;
740 
741 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
742 			/* Do up the '.' and '..' entries.  Dummy values are
743 			 * used for the cookies since the offset here is
744 			 * usually zero, and NFS doesn't like that value
745 			 */
746 			if (uiodir.cookies != NULL) {
747 				if (++uiodir.acookies > uiodir.ncookies) {
748 					uiodir.eofflag = 0;
749 					break;
750 				}
751 				*uiodir.cookies++ = 1;
752 			}
753 			if (vop_write_dirent(&error, uio, node->hash_id, DT_DIR,
754 					     1, ".")) {
755 				uiodir.eofflag = 0;
756 				break;
757 			}
758 			if (error) {
759 				uiodir.eofflag = 0;
760 				break;
761 			}
762 			if (uiodir.cookies != NULL) {
763 				if (++uiodir.acookies > uiodir.ncookies) {
764 					uiodir.eofflag = 0;
765 					break;
766 				}
767 				*uiodir.cookies++ = 2;
768 			}
769 			if (vop_write_dirent(&error, uio, udf_getid(&fid->icb),
770 					     DT_DIR, 2, "..")) {
771 				uiodir.eofflag = 0;
772 				break;
773 			}
774 			if (error) {
775 				uiodir.eofflag = 0;
776 				break;
777 			}
778 		} else {
779 			uint8_t d_type = (fid->file_char & UDF_FILE_CHAR_DIR) ?
780 			    DT_DIR : DT_UNKNOWN;
781 			uint16_t namelen = udf_transname(&fid->data[fid->l_iu],
782 			    name, fid->l_fi, udfmp);
783 
784 			if (uiodir.cookies != NULL) {
785 				if (++uiodir.acookies > uiodir.ncookies) {
786 					uiodir.eofflag = 0;
787 					break;
788 				}
789 				*uiodir.cookies++ = ds->this_off;
790 			}
791 			if (vop_write_dirent(&error, uio, udf_getid(&fid->icb),
792 					 d_type, namelen, name)) {
793 				uiodir.eofflag = 0;
794 				break;
795 			}
796 			if (error) {
797 				uiodir.eofflag = 0;
798 				break;
799 			}
800 		}
801 		if (error) {
802 			printf("uiomove returned %d\n", error);
803 			break;
804 		}
805 
806 	}
807 
808 	free(name, M_TEMP);
809 
810 	/* tell the calling layer whether we need to be called again */
811 	*a->a_eofflag = uiodir.eofflag;
812 	uio->uio_offset = ds->offset + ds->off;
813 
814 	if (!error)
815 		error = ds->error;
816 
817 	udf_closedir(ds);
818 
819 	if (a->a_ncookies != NULL) {
820 		if (error)
821 			free(cookies, M_TEMP);
822 		else {
823 			*a->a_ncookies = uiodir.acookies;
824 			*a->a_cookies = cookies;
825 		}
826 	}
827 
828 	return(error);
829 }
830 
831 /* Are there any implementations out there that do soft-links? */
832 static int
833 udf_readlink(struct vop_readlink_args *ap)
834 {
835 	printf("%s called\n", __func__);
836 	return(EOPNOTSUPP);
837 }
838 
839 static int
840 udf_strategy(struct vop_strategy_args *ap)
841 {
842 	struct bio *bio;
843 	struct bio *nbio;
844 	struct buf *bp;
845 	struct vnode *vp;
846 	struct udf_node *node;
847 	int maxsize;
848 	daddr_t dblkno;
849 
850 	bio = ap->a_bio;
851 	bp = bio->bio_buf;
852 	vp = ap->a_vp;
853 	node = VTON(vp);
854 
855 	nbio = push_bio(bio);
856 	if (nbio->bio_offset == NOOFFSET) {
857 		/*
858 		 * Files that are embedded in the fentry don't translate well
859 		 * to a block number.  Reject.
860 		 */
861 		if (udf_bmap_internal(node,
862 				     bio->bio_offset,
863 				     &dblkno, &maxsize)) {
864 			clrbuf(bp);
865 			nbio->bio_offset = NOOFFSET;
866 		} else {
867 			nbio->bio_offset = dbtob(dblkno);
868 		}
869 	}
870 	if (nbio->bio_offset == NOOFFSET) {
871 		/* I/O was never started on nbio, must biodone(bio) */
872 		biodone(bio);
873 		return(0);
874 	}
875 	vn_strategy(node->i_devvp, nbio);
876 	return(0);
877 }
878 
879 static int
880 udf_bmap(struct vop_bmap_args *a)
881 {
882 	struct udf_node *node;
883 	uint32_t max_size;
884 	daddr_t lsector;
885 	int error;
886 
887 	node = VTON(a->a_vp);
888 
889 	if (a->a_vpp != NULL)
890 		*a->a_vpp = node->i_devvp;
891 	if (a->a_doffsetp == NULL)
892 		return(0);
893 
894 	KKASSERT(a->a_loffset % node->udfmp->bsize == 0);
895 
896 	error = udf_bmap_internal(node, a->a_loffset, &lsector, &max_size);
897 	if (error)
898 		return(error);
899 
900 	/* Translate logical to physical sector number */
901 	*a->a_doffsetp = (off_t)lsector << node->udfmp->bshift;
902 
903 	/* Punt on read-ahead for now */
904 	if (a->a_runp)
905 		*a->a_runp = 0;
906 	if (a->a_runb)
907 		*a->a_runb = 0;
908 	return(0);
909 }
910 
911 /*
912  * The all powerful VOP_LOOKUP().
913  */
914 static int
915 udf_lookup(struct vop_old_lookup_args *a)
916 {
917 	struct vnode *dvp;
918 	struct vnode *tdp = NULL;
919 	struct vnode **vpp = a->a_vpp;
920 	struct udf_node *node;
921 	struct udf_mnt *udfmp;
922 	struct fileid_desc *fid = NULL;
923 	struct udf_dirstream *ds;
924 	struct thread *td;
925 	globaldata_t gd = mycpu;
926 	u_long nameiop;
927 	u_long flags;
928 	char *nameptr;
929 	long namelen;
930 	ino_t id = 0;
931 	int offset, error = 0;
932 	int numdirpasses, fsize;
933 
934 	dvp = a->a_dvp;
935 	node = VTON(dvp);
936 	udfmp = node->udfmp;
937 	nameiop = a->a_cnp->cn_nameiop;
938 	flags = a->a_cnp->cn_flags;
939 	nameptr = a->a_cnp->cn_nameptr;
940 	namelen = a->a_cnp->cn_namelen;
941 	fsize = node->fentry->inf_len;
942 	td = a->a_cnp->cn_td;
943 
944 	/*
945 	 * If this is a LOOKUP and we've already partially searched through
946 	 * the directory, pick up where we left off and flag that the
947 	 * directory may need to be searched twice.  For a full description,
948 	 * see /sys/isofs/cd9660/cd9660_lookup.c:cd9660_lookup()
949 	 */
950 	if (nameiop != NAMEI_LOOKUP || node->diroff == 0 ||
951 	    node->diroff > fsize) {
952 		offset = 0;
953 		numdirpasses = 1;
954 	} else {
955 		offset = node->diroff;
956 		numdirpasses = 2;
957 		gd->gd_nchstats->ncs_2passes++;
958 	}
959 
960 lookloop:
961 	ds = udf_opendir(node, offset, fsize, udfmp);
962 
963 	while ((fid = udf_getfid(ds)) != NULL) {
964 		/* XXX Should we return an error on a bad fid? */
965 		if (udf_checktag(&fid->tag, TAGID_FID)) {
966 			printf("udf_lookup: Invalid tag\n");
967 			error = EIO;
968 			break;
969 		}
970 
971 		/* Is this a deleted file? */
972 		if (fid->file_char & UDF_FILE_CHAR_DEL)
973 			continue;
974 
975 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
976 			if (flags & CNP_ISDOTDOT) {
977 				id = udf_getid(&fid->icb);
978 				break;
979 			}
980 		} else {
981 			if (!(udf_cmpname(&fid->data[fid->l_iu],
982 					  nameptr, fid->l_fi, namelen, udfmp))) {
983 				id = udf_getid(&fid->icb);
984 				break;
985 			}
986 		}
987 	}
988 
989 	if (!error)
990 		error = ds->error;
991 
992 	/* XXX Bail out here? */
993 	if (error) {
994 		udf_closedir(ds);
995 		return (error);
996 	}
997 
998 	/* Did we have a match? */
999 	if (id) {
1000 		error = udf_vget(udfmp->im_mountp, id, &tdp);
1001 		if (!error) {
1002 			/*
1003 			 * Remember where this entry was if it's the final
1004 			 * component.
1005 			 */
1006 			if (nameiop == NAMEI_LOOKUP)
1007 				node->diroff = ds->offset + ds->off;
1008 			if (numdirpasses == 2)
1009 				gd->gd_nchstats->ncs_pass2++;
1010 			if ((flags & CNP_LOCKPARENT) == 0) {
1011 				a->a_cnp->cn_flags |= CNP_PDIRUNLOCK;
1012 				VOP_UNLOCK(dvp, 0, td);
1013 			}
1014 
1015 			*vpp = tdp;
1016 		}
1017 	} else {
1018 		/* Name wasn't found on this pass.  Do another pass? */
1019 		if (numdirpasses == 2) {
1020 			numdirpasses--;
1021 			offset = 0;
1022 			udf_closedir(ds);
1023 			goto lookloop;
1024 		}
1025 		if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
1026 			error = EROFS;
1027 		} else {
1028 			error = ENOENT;
1029 		}
1030 	}
1031 
1032 	udf_closedir(ds);
1033 	return(error);
1034 }
1035 
1036 static int
1037 udf_reclaim(struct vop_reclaim_args *a)
1038 {
1039 	struct vnode *vp;
1040 	struct udf_node *unode;
1041 
1042 	vp = a->a_vp;
1043 	unode = VTON(vp);
1044 
1045 	if (unode != NULL) {
1046 		udf_hashrem(unode);
1047 		if (unode->i_devvp) {
1048 			vrele(unode->i_devvp);
1049 			unode->i_devvp = 0;
1050 		}
1051 
1052 		if (unode->fentry != NULL)
1053 			free(unode->fentry, M_UDFFENTRY);
1054 		free(unode, M_UDFNODE);
1055 		vp->v_data = NULL;
1056 	}
1057 
1058 	return(0);
1059 }
1060 
1061 /*
1062  * Read the block and then set the data pointer to correspond with the
1063  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1064  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1065  * whole extent.
1066  *
1067  * Note that *bp may be assigned error or not.
1068  *
1069  * XXX 'size' is limited to the logical block size for now due to problems
1070  * with udf_read()
1071  */
1072 static int
1073 udf_readatoffset(struct udf_node *node, int *size, int offset, struct buf **bp,
1074 		 uint8_t **data)
1075 {
1076 	struct udf_mnt *udfmp;
1077 	struct file_entry *fentry = NULL;
1078 	struct buf *bp1;
1079 	uint32_t max_size;
1080 	daddr_t sector;
1081 	int error;
1082 
1083 	udfmp = node->udfmp;
1084 
1085 	*bp = NULL;
1086 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1087 	if (error == UDF_INVALID_BMAP) {
1088 		/*
1089 		 * This error means that the file *data* is stored in the
1090 		 * allocation descriptor field of the file entry.
1091 		 */
1092 		fentry = node->fentry;
1093 		*data = &fentry->data[fentry->l_ea];
1094 		*size = fentry->l_ad;
1095 		return(0);
1096 	} else if (error != 0) {
1097 		return(error);
1098 	}
1099 
1100 	/* Adjust the size so that it is within range */
1101 	if (*size == 0 || *size > max_size)
1102 		*size = max_size;
1103 	*size = min(*size, MAXBSIZE);
1104 
1105 	if ((error = udf_readlblks(udfmp, sector, *size, bp))) {
1106 		printf("warning: udf_readlblks returned error %d\n", error);
1107 		/* note: *bp may be non-NULL */
1108 		return(error);
1109 	}
1110 
1111 	bp1 = *bp;
1112 	*data = (uint8_t *)&bp1->b_data[offset % udfmp->bsize];
1113 	return(0);
1114 }
1115 
1116 /*
1117  * Translate a file offset into a logical block and then into a physical
1118  * block.
1119  */
1120 static int
1121 udf_bmap_internal(struct udf_node *node, uint32_t offset, daddr_t *sector, uint32_t *max_size)
1122 {
1123 	struct udf_mnt *udfmp;
1124 	struct file_entry *fentry;
1125 	void *icb;
1126 	struct icb_tag *tag;
1127 	uint32_t icblen = 0;
1128 	daddr_t lsector;
1129 	int ad_offset, ad_num = 0;
1130 	int i, p_offset;
1131 
1132 	udfmp = node->udfmp;
1133 	fentry = node->fentry;
1134 	tag = &fentry->icbtag;
1135 
1136 	switch (tag->strat_type) {
1137 	case 4:
1138 		break;
1139 
1140 	case 4096:
1141 		printf("Cannot deal with strategy4096 yet!\n");
1142 		return(ENODEV);
1143 
1144 	default:
1145 		printf("Unknown strategy type %d\n", tag->strat_type);
1146 		return(ENODEV);
1147 	}
1148 
1149 	switch (tag->flags & 0x7) {
1150 	case 0:
1151 		/*
1152 		 * The allocation descriptor field is filled with short_ad's.
1153 		 * If the offset is beyond the current extent, look for the
1154 		 * next extent.
1155 		 */
1156 		do {
1157 			offset -= icblen;
1158 			ad_offset = sizeof(struct short_ad) * ad_num;
1159 			if (ad_offset > fentry->l_ad) {
1160 				printf("File offset out of bounds\n");
1161 				return(EINVAL);
1162 			}
1163 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1164 			icblen = GETICBLEN(short_ad, icb);
1165 			ad_num++;
1166 		} while(offset >= icblen);
1167 
1168 		lsector = (offset  >> udfmp->bshift) +
1169 		    ((struct short_ad *)(icb))->pos;
1170 
1171 		*max_size = GETICBLEN(short_ad, icb);
1172 
1173 		break;
1174 	case 1:
1175 		/*
1176 		 * The allocation descriptor field is filled with long_ad's
1177 		 * If the offset is beyond the current extent, look for the
1178 		 * next extent.
1179 		 */
1180 		do {
1181 			offset -= icblen;
1182 			ad_offset = sizeof(struct long_ad) * ad_num;
1183 			if (ad_offset > fentry->l_ad) {
1184 				printf("File offset out of bounds\n");
1185 				return(EINVAL);
1186 			}
1187 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1188 			icblen = GETICBLEN(long_ad, icb);
1189 			ad_num++;
1190 		} while(offset >= icblen);
1191 
1192 		lsector = (offset >> udfmp->bshift) +
1193 		    ((struct long_ad *)(icb))->loc.lb_num;
1194 
1195 		*max_size = GETICBLEN(long_ad, icb);
1196 
1197 		break;
1198 	case 3:
1199 		/*
1200 		 * This type means that the file *data* is stored in the
1201 		 * allocation descriptor field of the file entry.
1202 		 */
1203 		*max_size = 0;
1204 		*sector = node->hash_id + udfmp->part_start;
1205 
1206 		return(UDF_INVALID_BMAP);
1207 	case 2:
1208 		/* DirectCD does not use extended_ad's */
1209 	default:
1210 		printf("Unsupported allocation descriptor %d\n",
1211 		       tag->flags & 0x7);
1212 		return(ENODEV);
1213 	}
1214 
1215 	*sector = lsector + udfmp->part_start;
1216 
1217 	/*
1218 	 * Check the sparing table.  Each entry represents the beginning of
1219 	 * a packet.
1220 	 */
1221 	if (udfmp->s_table != NULL) {
1222 		for (i = 0; i< udfmp->s_table_entries; i++) {
1223 			p_offset = lsector - udfmp->s_table->entries[i].org;
1224 			if ((p_offset < udfmp->p_sectors) && (p_offset >= 0)) {
1225 				*sector = udfmp->s_table->entries[i].map +
1226 				    p_offset;
1227 				break;
1228 			}
1229 		}
1230 	}
1231 
1232 	return(0);
1233 }
1234