xref: /dragonfly/sys/vfs/udf/udf_vnops.c (revision 9f3fc534)
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 1.32 2007/11/20 21:03:51 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 vop_ops udf_vnode_vops = {
70 	.vop_default =		vop_defaultop,
71 	.vop_access =		udf_access,
72 	.vop_bmap =		udf_bmap,
73 	.vop_old_lookup =	udf_lookup,
74 	.vop_getattr =		udf_getattr,
75 	.vop_ioctl =		udf_ioctl,
76 	.vop_pathconf =		udf_pathconf,
77 	.vop_read =		udf_read,
78 	.vop_readdir =		udf_readdir,
79 	.vop_readlink =		udf_readlink,
80 	.vop_reclaim =		udf_reclaim,
81 	.vop_strategy =		udf_strategy
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))
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 	LIST_INSERT_HEAD(lh, node, le);
147 	lwkt_reltoken(&hashlock);
148 
149 	return(0);
150 }
151 
152 int
153 udf_hashrem(struct udf_node *node)
154 {
155 	struct udf_mnt *udfmp;
156 	struct udf_hash_lh *lh;
157 	lwkt_tokref hashlock;
158 
159 	udfmp = node->udfmp;
160 
161 	lwkt_gettoken(&hashlock, &udfmp->hash_token);
162 	lh = &udfmp->hashtbl[node->hash_id % udfmp->hashsz];
163 	if (lh == NULL)
164 		panic("hash entry is NULL, node->hash_id= %"PRId64"\n", node->hash_id);
165 	LIST_REMOVE(node, le);
166 	lwkt_reltoken(&hashlock);
167 
168 	return(0);
169 }
170 
171 int
172 udf_allocv(struct mount *mp, struct vnode **vpp)
173 {
174 	int error;
175 	struct vnode *vp;
176 
177 	error = getnewvnode(VT_UDF, mp, &vp, 0, 0);
178 	if (error) {
179 		kprintf("udf_allocv: failed to allocate new vnode\n");
180 		return(error);
181 	}
182 
183 	*vpp = vp;
184 	return(0);
185 }
186 
187 /* Convert file entry permission (5 bits per owner/group/user) to a mode_t */
188 static mode_t
189 udf_permtomode(struct udf_node *node)
190 {
191 	uint32_t perm;
192 	uint32_t flags;
193 	mode_t mode;
194 
195 	perm = node->fentry->perm;
196 	flags = node->fentry->icbtag.flags;
197 
198 	mode = perm & UDF_FENTRY_PERM_USER_MASK;
199 	mode |= ((perm & UDF_FENTRY_PERM_GRP_MASK) >> 2);
200 	mode |= ((perm & UDF_FENTRY_PERM_OWNER_MASK) >> 4);
201 	mode |= ((flags & UDF_ICB_TAG_FLAGS_STICKY) << 4);
202 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETGID) << 6);
203 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETUID) << 8);
204 
205 	return(mode);
206 }
207 
208 static int
209 udf_access(struct vop_access_args *a)
210 {
211 	struct vnode *vp;
212 	struct udf_node *node;
213 	mode_t a_mode, mode, mask;
214 	struct ucred *cred = a->a_cred;
215 	gid_t *gp;
216 	int i;
217 
218 	vp = a->a_vp;
219 	node = VTON(vp);
220 	a_mode = a->a_mode;
221 
222 	if (a_mode & VWRITE) {
223 		switch (vp->v_type) {
224 		case VDIR:
225 		case VLNK:
226 		case VREG:
227 			return(EROFS);
228 			/* NOT REACHED */
229 		default:
230 			break;
231 		}
232 	}
233 
234 	mode = udf_permtomode(node);
235 
236 	if (cred->cr_uid == 0)
237 		return(0);
238 
239 	mask = 0;
240 
241 	/* Otherwise, check the owner. */
242 	if (cred->cr_uid == node->fentry->uid) {
243 		if (a_mode & VEXEC)
244 			mask |= S_IXUSR;
245 		if (a_mode & VREAD)
246 			mask |= S_IRUSR;
247 		if (a_mode & VWRITE)
248 			mask |= S_IWUSR;
249 		return((mode & mask) == mask ? 0 : EACCES);
250 	}
251 
252 	/* Otherwise, check the groups. */
253 	for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
254 		if (node->fentry->gid == *gp) {
255 			if (a_mode & VEXEC)
256 				mask |= S_IXGRP;
257 			if (a_mode & VREAD)
258 				mask |= S_IRGRP;
259 			if (a_mode & VWRITE)
260 				mask |= S_IWGRP;
261 			return((mode & mask) == mask ? 0 : EACCES);
262 		}
263 
264 	/* Otherwise, check everyone else. */
265 	if (a_mode & VEXEC)
266 		mask |= S_IXOTH;
267 	if (a_mode & VREAD)
268 		mask |= S_IROTH;
269 	if (a_mode & VWRITE)
270 		mask |= S_IWOTH;
271 	return((mode & mask) == mask ? 0 : EACCES);
272 }
273 
274 static int mon_lens[2][12] = {
275 	{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31},
276 	{31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31}
277 };
278 
279 static int
280 udf_isaleapyear(int year)
281 {
282 	int i;
283 
284 	i = (year % 4) ? 0 : 1;
285 	i &= (year % 100) ? 1 : 0;
286 	i |= (year % 400) ? 0 : 1;
287 
288 	return(i);
289 }
290 
291 /*
292  * XXX This is just a rough hack.  Daylight savings isn't calculated and tv_nsec
293  * is ignored.
294  * Timezone calculation compliments of Julian Elischer <julian@elischer.org>.
295  */
296 static void
297 udf_timetotimespec(struct timestamp *time, struct timespec *t)
298 {
299 	int i, lpyear, daysinyear;
300 	union {
301 		uint16_t	u_tz_offset;
302 		int16_t		s_tz_offset;
303 	} tz;
304 
305 	t->tv_nsec = 0;
306 
307 	/* DirectCD seems to like using bogus year values */
308 	if (time->year < 1970) {
309 		t->tv_sec = 0;
310 		return;
311 	}
312 
313 	/* Calculate the time and day */
314 	t->tv_sec = time->second;
315 	t->tv_sec += time->minute * 60;
316 	t->tv_sec += time->hour * 3600;
317 	t->tv_sec += time->day * 3600 * 24;
318 
319 	/* Calclulate the month */
320 	lpyear = udf_isaleapyear(time->year);
321 	for (i = 1; i < time->month; i++)
322 		t->tv_sec += mon_lens[lpyear][i] * 3600 * 24;
323 
324 	/* Speed up the calculation */
325 	if (time->year > 1979)
326 		t->tv_sec += 315532800;
327 	if (time->year > 1989)
328 		t->tv_sec += 315619200;
329 	if (time->year > 1999)
330 		t->tv_sec += 315532800;
331 	for (i = 2000; i < time->year; i++) {
332 		daysinyear = udf_isaleapyear(i) + 365 ;
333 		t->tv_sec += daysinyear * 3600 * 24;
334 	}
335 
336 	/*
337 	 * Calculate the time zone.  The timezone is 12 bit signed 2's
338 	 * compliment, so we gotta do some extra magic to handle it right.
339 	 */
340 	tz.u_tz_offset = time->type_tz;
341 	tz.u_tz_offset &= 0x0fff;
342 	if (tz.u_tz_offset & 0x0800)
343 		tz.u_tz_offset |= 0xf000;	/* extend the sign to 16 bits */
344 	if ((time->type_tz & 0x1000) && (tz.s_tz_offset != -2047))
345 		t->tv_sec -= tz.s_tz_offset * 60;
346 
347 	return;
348 }
349 
350 static int
351 udf_getattr(struct vop_getattr_args *a)
352 {
353 	struct vnode *vp;
354 	struct udf_node *node;
355 	struct vattr *vap;
356 	struct file_entry *fentry;
357 	struct timespec ts;
358 
359 	ts.tv_sec = 0;
360 
361 	vp = a->a_vp;
362 	vap = a->a_vap;
363 	node = VTON(vp);
364 	fentry = node->fentry;
365 
366 	vap->va_fsid = dev2udev(node->i_dev);
367 	vap->va_fileid = node->hash_id;
368 	vap->va_mode = udf_permtomode(node);
369 	vap->va_nlink = fentry->link_cnt;
370 	/*
371 	 * XXX The spec says that -1 is valid for uid/gid and indicates an
372 	 * invalid uid/gid.  How should this be represented?
373 	 */
374 	vap->va_uid = (fentry->uid == 0xffffffff) ? 0 : fentry->uid;
375 	vap->va_gid = (fentry->gid == 0xffffffff) ? 0 : fentry->gid;
376 	udf_timetotimespec(&fentry->atime, &vap->va_atime);
377 	udf_timetotimespec(&fentry->mtime, &vap->va_mtime);
378 	vap->va_ctime = vap->va_mtime; /* XXX Stored as an Extended Attribute */
379 	vap->va_rmajor = VNOVAL;
380 	vap->va_rminor = VNOVAL;
381 	if (vp->v_type & VDIR) {
382 		/*
383 		 * Directories that are recorded within their ICB will show
384 		 * as having 0 blocks recorded.  Since tradition dictates
385 		 * that directories consume at least one logical block,
386 		 * make it appear so.
387 		 */
388 		if (fentry->logblks_rec != 0)
389 			vap->va_size = fentry->logblks_rec * node->udfmp->bsize;
390 		else
391 			vap->va_size = node->udfmp->bsize;
392 	} else
393 		vap->va_size = fentry->inf_len;
394 	vap->va_flags = 0;
395 	vap->va_gen = 1;
396 	vap->va_blocksize = node->udfmp->bsize;
397 	vap->va_bytes = fentry->inf_len;
398 	vap->va_type = vp->v_type;
399 	vap->va_filerev = 0; /* XXX */
400 	return(0);
401 }
402 
403 /*
404  * File specific ioctls.  DeCSS candidate?
405  */
406 static int
407 udf_ioctl(struct vop_ioctl_args *a)
408 {
409 	kprintf("%s called\n", __func__);
410 	return(ENOTTY);
411 }
412 
413 /*
414  * I'm not sure that this has much value in a read-only filesystem, but
415  * cd9660 has it too.
416  */
417 static int
418 udf_pathconf(struct vop_pathconf_args *a)
419 {
420 
421 	switch (a->a_name) {
422 	case _PC_LINK_MAX:
423 		*a->a_retval = 65535;
424 		return(0);
425 	case _PC_NAME_MAX:
426 		*a->a_retval = NAME_MAX;
427 		return(0);
428 	case _PC_PATH_MAX:
429 		*a->a_retval = PATH_MAX;
430 		return(0);
431 	case _PC_NO_TRUNC:
432 		*a->a_retval = 1;
433 		return(0);
434 	default:
435 		return(EINVAL);
436 	}
437 }
438 
439 static int
440 udf_read(struct vop_read_args *a)
441 {
442 	struct vnode *vp = a->a_vp;
443 	struct uio *uio = a->a_uio;
444 	struct udf_node *node = VTON(vp);
445 	struct buf *bp;
446 	uint8_t *data;
447 	int error = 0;
448 	int size, fsize, offset;
449 
450 	if (uio->uio_offset < 0)
451 		return(EINVAL);
452 
453 	fsize = node->fentry->inf_len;
454 
455 	while (uio->uio_offset < fsize && uio->uio_resid > 0) {
456 		offset = uio->uio_offset;
457 		size = uio->uio_resid;
458 		error = udf_readatoffset(node, &size, offset, &bp, &data);
459 		if (error == 0)
460 			error = uiomove(data, size, uio);
461 		if (bp != NULL)
462 			brelse(bp);
463 		if (error)
464 			break;
465 	}
466 
467 	return(error);
468 }
469 
470 /*
471  * Call the OSTA routines to translate the name from a CS0 dstring to a
472  * 16-bit Unicode String.  Hooks need to be placed in here to translate from
473  * Unicode to the encoding that the kernel/user expects.  Return the length
474  * of the translated string.
475  */
476 static int
477 udf_transname(char *cs0string, char *destname, int len, struct udf_mnt *udfmp)
478 {
479 	unicode_t *transname;
480 	int i, unilen = 0, destlen;
481 
482 	/* Convert 16-bit Unicode to destname */
483 	/* allocate a buffer big enough to hold an 8->16 bit expansion */
484 	transname = kmalloc(NAME_MAX * sizeof(unicode_t), M_TEMP, M_WAITOK | M_ZERO);
485 
486 	if ((unilen = udf_UncompressUnicode(len, cs0string, transname)) == -1) {
487 		kprintf("udf: Unicode translation failed\n");
488 		kfree(transname, M_TEMP);
489 		return(0);
490 	}
491 
492 	for (i = 0; i < unilen ; i++)
493 		if (transname[i] & 0xff00)
494 			destname[i] = '.';	/* Fudge the 16bit chars */
495 		else
496 			destname[i] = transname[i] & 0xff;
497 	kfree(transname, M_TEMP);
498 	destname[unilen] = 0;
499 	destlen = unilen;
500 
501 	return(destlen);
502 }
503 
504 /*
505  * Compare a CS0 dstring with a name passed in from the VFS layer.  Return
506  * 0 on a successful match, nonzero therwise.  Unicode work may need to be done
507  * here also.
508  */
509 static int
510 udf_cmpname(char *cs0string, char *cmpname, int cs0len, int cmplen, struct udf_mnt *udfmp)
511 {
512 	char *transname;
513 	int error = 0;
514 
515 	/* This is overkill, but not worth creating a new zone */
516 
517 	transname = kmalloc(NAME_MAX * sizeof(unicode_t), M_TEMP,
518 			   M_WAITOK | M_ZERO);
519 
520 	cs0len = udf_transname(cs0string, transname, cs0len, udfmp);
521 
522 	/* Easy check.  If they aren't the same length, they aren't equal */
523 	if ((cs0len == 0) || (cs0len != cmplen))
524 		error = -1;
525 	else
526 		error = bcmp(transname, cmpname, cmplen);
527 
528 	kfree(transname, M_TEMP);
529 	return(error);
530 }
531 
532 struct udf_uiodir {
533 	struct dirent *dirent;
534 	off_t *cookies;
535 	int ncookies;
536 	int acookies;
537 	int eofflag;
538 };
539 
540 static struct udf_dirstream *
541 udf_opendir(struct udf_node *node, int offset, int fsize, struct udf_mnt *udfmp)
542 {
543 	struct udf_dirstream *ds;
544 
545 	ds = kmalloc(sizeof(*ds), M_UDFDS, M_WAITOK | M_ZERO);
546 
547 	ds->node = node;
548 	ds->offset = offset;
549 	ds->udfmp = udfmp;
550 	ds->fsize = fsize;
551 
552 	return(ds);
553 }
554 
555 static struct fileid_desc *
556 udf_getfid(struct udf_dirstream *ds)
557 {
558 	struct fileid_desc *fid;
559 	int error, frag_size = 0, total_fid_size;
560 
561 	/* End of directory? */
562 	if (ds->offset + ds->off >= ds->fsize) {
563 		ds->error = 0;
564 		return(NULL);
565 	}
566 
567 	/* Grab the first extent of the directory */
568 	if (ds->off == 0) {
569 		ds->size = 0;
570 		if (ds->bp != NULL)
571 			brelse(ds->bp);
572 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
573 		    &ds->bp, &ds->data);
574 		if (error) {
575 			ds->error = error;
576 			return(NULL);
577 		}
578 	}
579 
580 	/*
581 	 * Clean up from a previous fragmented FID.
582 	 * XXX Is this the right place for this?
583 	 */
584 	if (ds->fid_fragment && ds->buf != NULL) {
585 		ds->fid_fragment = 0;
586 		kfree(ds->buf, M_UDFFID);
587 	}
588 
589 	fid = (struct fileid_desc*)&ds->data[ds->off];
590 
591 	/*
592 	 * Check to see if the fid is fragmented. The first test
593 	 * ensures that we don't wander off the end of the buffer
594 	 * looking for the l_iu and l_fi fields.
595 	 */
596 	if (ds->off + UDF_FID_SIZE > ds->size ||
597 	    ds->off + fid->l_iu + fid->l_fi + UDF_FID_SIZE > ds->size) {
598 
599 		/* Copy what we have of the fid into a buffer */
600 		frag_size = ds->size - ds->off;
601 		if (frag_size >= ds->udfmp->bsize) {
602 			kprintf("udf: invalid FID fragment\n");
603 			ds->error = EINVAL;
604 			return(NULL);
605 		}
606 
607 		/*
608 		 * File ID descriptors can only be at most one
609 		 * logical sector in size.
610 		 */
611 		ds->buf = kmalloc(ds->udfmp->bsize, M_UDFFID, M_WAITOK | M_ZERO);
612 		bcopy(fid, ds->buf, frag_size);
613 
614 		/* Reduce all of the casting magic */
615 		fid = (struct fileid_desc*)ds->buf;
616 
617 		if (ds->bp != NULL)
618 			brelse(ds->bp);
619 
620 		/* Fetch the next allocation */
621 		ds->offset += ds->size;
622 		ds->size = 0;
623 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
624 		    &ds->bp, &ds->data);
625 		if (error) {
626 			ds->error = error;
627 			return(NULL);
628 		}
629 
630 		/*
631 		 * If the fragment was so small that we didn't get
632 		 * the l_iu and l_fi fields, copy those in.
633 		 */
634 		if (frag_size < UDF_FID_SIZE)
635 			bcopy(ds->data, &ds->buf[frag_size],
636 			    UDF_FID_SIZE - frag_size);
637 
638 		/*
639 		 * Now that we have enough of the fid to work with,
640 		 * copy in the rest of the fid from the new
641 		 * allocation.
642 		 */
643 		total_fid_size = UDF_FID_SIZE + fid->l_iu + fid->l_fi;
644 		if (total_fid_size > ds->udfmp->bsize) {
645 			kprintf("udf: invalid FID\n");
646 			ds->error = EIO;
647 			return(NULL);
648 		}
649 		bcopy(ds->data, &ds->buf[frag_size],
650 		    total_fid_size - frag_size);
651 
652 		ds->fid_fragment = 1;
653 	} else
654 		total_fid_size = fid->l_iu + fid->l_fi + UDF_FID_SIZE;
655 
656 	/*
657 	 * Update the offset. Align on a 4 byte boundary because the
658 	 * UDF spec says so.
659 	 */
660 	ds->this_off = ds->off;
661 	if (!ds->fid_fragment)
662 		ds->off += (total_fid_size + 3) & ~0x03;
663 	else
664 		ds->off = (total_fid_size - frag_size + 3) & ~0x03;
665 
666 	return(fid);
667 }
668 
669 static void
670 udf_closedir(struct udf_dirstream *ds)
671 {
672 
673 	if (ds->bp != NULL)
674 		brelse(ds->bp);
675 
676 	if (ds->fid_fragment && ds->buf != NULL)
677 		kfree(ds->buf, M_UDFFID);
678 
679 	kfree(ds, M_UDFDS);
680 }
681 
682 static int
683 udf_readdir(struct vop_readdir_args *a)
684 {
685 	struct vnode *vp;
686 	struct uio *uio;
687 	struct udf_node *node;
688 	struct udf_mnt *udfmp;
689 	struct fileid_desc *fid;
690 	struct udf_uiodir uiodir;
691 	struct udf_dirstream *ds;
692 	off_t *cookies = NULL;
693 	int ncookies;
694 	int error = 0;
695 	char *name;
696 
697 	vp = a->a_vp;
698 
699 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
700 		return (error);
701 
702 	uio = a->a_uio;
703 	node = VTON(vp);
704 	udfmp = node->udfmp;
705 	uiodir.eofflag = 1;
706 
707 	if (a->a_ncookies != NULL) {
708 		/*
709 		 * Guess how many entries are needed.  If we run out, this
710 		 * function will be called again and thing will pick up were
711 		 * it left off.
712 		 */
713 		ncookies = uio->uio_resid / 8 + 1;
714 		if (ncookies > 1024)
715 			ncookies = 1024;
716 		cookies = kmalloc(sizeof(off_t) * ncookies, M_TEMP, M_WAITOK);
717 		uiodir.ncookies = ncookies;
718 		uiodir.cookies = cookies;
719 		uiodir.acookies = 0;
720 	} else
721 		uiodir.cookies = NULL;
722 
723 	/*
724 	 * Iterate through the file id descriptors.  Give the parent dir
725 	 * entry special attention.
726 	 */
727 	ds = udf_opendir(node, uio->uio_offset, node->fentry->inf_len,
728 			 node->udfmp);
729 
730 	name = kmalloc(NAME_MAX, M_TEMP, M_WAITOK);
731 
732 	while ((fid = udf_getfid(ds)) != NULL) {
733 
734 		/* XXX Should we return an error on a bad fid? */
735 		if (udf_checktag(&fid->tag, TAGID_FID)) {
736 			kprintf("Invalid FID tag\n");
737 			error = EIO;
738 			break;
739 		}
740 
741 		/* Is this a deleted file? */
742 		if (fid->file_char & UDF_FILE_CHAR_DEL)
743 			continue;
744 
745 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
746 			/* Do up the '.' and '..' entries.  Dummy values are
747 			 * used for the cookies since the offset here is
748 			 * usually zero, and NFS doesn't like that value
749 			 */
750 			if (uiodir.cookies != NULL) {
751 				if (++uiodir.acookies > uiodir.ncookies) {
752 					uiodir.eofflag = 0;
753 					break;
754 				}
755 				*uiodir.cookies++ = 1;
756 			}
757 			if (vop_write_dirent(&error, uio, node->hash_id, DT_DIR,
758 					     1, ".")) {
759 				uiodir.eofflag = 0;
760 				break;
761 			}
762 			if (error) {
763 				uiodir.eofflag = 0;
764 				break;
765 			}
766 			if (uiodir.cookies != NULL) {
767 				if (++uiodir.acookies > uiodir.ncookies) {
768 					uiodir.eofflag = 0;
769 					break;
770 				}
771 				*uiodir.cookies++ = 2;
772 			}
773 			if (vop_write_dirent(&error, uio, udf_getid(&fid->icb),
774 					     DT_DIR, 2, "..")) {
775 				uiodir.eofflag = 0;
776 				break;
777 			}
778 			if (error) {
779 				uiodir.eofflag = 0;
780 				break;
781 			}
782 		} else {
783 			uint8_t d_type = (fid->file_char & UDF_FILE_CHAR_DIR) ?
784 			    DT_DIR : DT_UNKNOWN;
785 			uint16_t namelen = udf_transname(&fid->data[fid->l_iu],
786 			    name, fid->l_fi, udfmp);
787 
788 			if (uiodir.cookies != NULL) {
789 				if (++uiodir.acookies > uiodir.ncookies) {
790 					uiodir.eofflag = 0;
791 					break;
792 				}
793 				*uiodir.cookies++ = ds->this_off;
794 			}
795 			if (vop_write_dirent(&error, uio, udf_getid(&fid->icb),
796 					 d_type, namelen, name)) {
797 				uiodir.eofflag = 0;
798 				break;
799 			}
800 			if (error) {
801 				uiodir.eofflag = 0;
802 				break;
803 			}
804 		}
805 		if (error) {
806 			kprintf("uiomove returned %d\n", error);
807 			break;
808 		}
809 
810 	}
811 
812 	kfree(name, M_TEMP);
813 
814 	/* tell the calling layer whether we need to be called again */
815 	*a->a_eofflag = uiodir.eofflag;
816 	uio->uio_offset = ds->offset + ds->off;
817 
818 	if (!error)
819 		error = ds->error;
820 
821 	udf_closedir(ds);
822 
823 	if (a->a_ncookies != NULL) {
824 		if (error)
825 			kfree(cookies, M_TEMP);
826 		else {
827 			*a->a_ncookies = uiodir.acookies;
828 			*a->a_cookies = cookies;
829 		}
830 	}
831 
832 	vn_unlock(vp);
833 	return(error);
834 }
835 
836 /* Are there any implementations out there that do soft-links? */
837 static int
838 udf_readlink(struct vop_readlink_args *ap)
839 {
840 	kprintf("%s called\n", __func__);
841 	return(EOPNOTSUPP);
842 }
843 
844 static int
845 udf_strategy(struct vop_strategy_args *ap)
846 {
847 	struct bio *bio;
848 	struct bio *nbio;
849 	struct buf *bp;
850 	struct vnode *vp;
851 	struct udf_node *node;
852 	int maxsize;
853 	daddr_t dblkno;
854 
855 	bio = ap->a_bio;
856 	bp = bio->bio_buf;
857 	vp = ap->a_vp;
858 	node = VTON(vp);
859 
860 	nbio = push_bio(bio);
861 	if (nbio->bio_offset == NOOFFSET) {
862 		/*
863 		 * Files that are embedded in the fentry don't translate well
864 		 * to a block number.  Reject.
865 		 */
866 		if (udf_bmap_internal(node,
867 				     bio->bio_offset,
868 				     &dblkno, &maxsize)) {
869 			clrbuf(bp);
870 			nbio->bio_offset = NOOFFSET;
871 		} else {
872 			nbio->bio_offset = dbtob(dblkno);
873 		}
874 	}
875 	if (nbio->bio_offset == NOOFFSET) {
876 		/* I/O was never started on nbio, must biodone(bio) */
877 		biodone(bio);
878 		return(0);
879 	}
880 	vn_strategy(node->i_devvp, nbio);
881 	return(0);
882 }
883 
884 static int
885 udf_bmap(struct vop_bmap_args *a)
886 {
887 	struct udf_node *node;
888 	uint32_t max_size;
889 	daddr_t lsector;
890 	int error;
891 
892 	node = VTON(a->a_vp);
893 
894 	if (a->a_doffsetp == NULL)
895 		return(0);
896 
897 	KKASSERT(a->a_loffset % node->udfmp->bsize == 0);
898 
899 	error = udf_bmap_internal(node, a->a_loffset, &lsector, &max_size);
900 	if (error)
901 		return(error);
902 
903 	/* Translate logical to physical sector number */
904 	*a->a_doffsetp = (off_t)lsector << node->udfmp->bshift;
905 
906 	/* Punt on read-ahead for now */
907 	if (a->a_runp)
908 		*a->a_runp = 0;
909 	if (a->a_runb)
910 		*a->a_runb = 0;
911 	return(0);
912 }
913 
914 /*
915  * The all powerful VOP_LOOKUP().
916  */
917 static int
918 udf_lookup(struct vop_old_lookup_args *a)
919 {
920 	struct vnode *dvp;
921 	struct vnode *tdp = NULL;
922 	struct vnode **vpp = a->a_vpp;
923 	struct udf_node *node;
924 	struct udf_mnt *udfmp;
925 	struct fileid_desc *fid = NULL;
926 	struct udf_dirstream *ds;
927 	struct thread *td;
928 	globaldata_t gd = mycpu;
929 	u_long nameiop;
930 	u_long flags;
931 	char *nameptr;
932 	long namelen;
933 	ino_t id = 0;
934 	int offset, error = 0;
935 	int numdirpasses, fsize;
936 
937 	dvp = a->a_dvp;
938 	node = VTON(dvp);
939 	udfmp = node->udfmp;
940 	nameiop = a->a_cnp->cn_nameiop;
941 	flags = a->a_cnp->cn_flags;
942 	nameptr = a->a_cnp->cn_nameptr;
943 	namelen = a->a_cnp->cn_namelen;
944 	fsize = node->fentry->inf_len;
945 	td = a->a_cnp->cn_td;
946 
947 	*vpp = NULL;
948 
949 	/*
950 	 * If this is a LOOKUP and we've already partially searched through
951 	 * the directory, pick up where we left off and flag that the
952 	 * directory may need to be searched twice.  For a full description,
953 	 * see /sys/isofs/cd9660/cd9660_lookup.c:cd9660_lookup()
954 	 */
955 	if (nameiop != NAMEI_LOOKUP || node->diroff == 0 ||
956 	    node->diroff > fsize) {
957 		offset = 0;
958 		numdirpasses = 1;
959 	} else {
960 		offset = node->diroff;
961 		numdirpasses = 2;
962 		gd->gd_nchstats->ncs_2passes++;
963 	}
964 
965 lookloop:
966 	ds = udf_opendir(node, offset, fsize, udfmp);
967 
968 	while ((fid = udf_getfid(ds)) != NULL) {
969 		/* XXX Should we return an error on a bad fid? */
970 		if (udf_checktag(&fid->tag, TAGID_FID)) {
971 			kprintf("udf_lookup: Invalid tag\n");
972 			error = EIO;
973 			break;
974 		}
975 
976 		/* Is this a deleted file? */
977 		if (fid->file_char & UDF_FILE_CHAR_DEL)
978 			continue;
979 
980 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
981 			if (flags & CNP_ISDOTDOT) {
982 				id = udf_getid(&fid->icb);
983 				break;
984 			}
985 		} else {
986 			if (!(udf_cmpname(&fid->data[fid->l_iu],
987 					  nameptr, fid->l_fi, namelen, udfmp))) {
988 				id = udf_getid(&fid->icb);
989 				break;
990 			}
991 		}
992 	}
993 
994 	if (!error)
995 		error = ds->error;
996 
997 	/* XXX Bail out here? */
998 	if (error) {
999 		udf_closedir(ds);
1000 		return (error);
1001 	}
1002 
1003 	/* Did we have a match? */
1004 	if (id) {
1005 		error = udf_vget(udfmp->im_mountp, id, &tdp);
1006 		if (!error) {
1007 			/*
1008 			 * Remember where this entry was if it's the final
1009 			 * component.
1010 			 */
1011 			if (nameiop == NAMEI_LOOKUP)
1012 				node->diroff = ds->offset + ds->off;
1013 			if (numdirpasses == 2)
1014 				gd->gd_nchstats->ncs_pass2++;
1015 			if ((flags & CNP_LOCKPARENT) == 0) {
1016 				a->a_cnp->cn_flags |= CNP_PDIRUNLOCK;
1017 				vn_unlock(dvp);
1018 			}
1019 
1020 			*vpp = tdp;
1021 		}
1022 	} else {
1023 		/* Name wasn't found on this pass.  Do another pass? */
1024 		if (numdirpasses == 2) {
1025 			numdirpasses--;
1026 			offset = 0;
1027 			udf_closedir(ds);
1028 			goto lookloop;
1029 		}
1030 		if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
1031 			error = EROFS;
1032 		} else {
1033 			error = ENOENT;
1034 		}
1035 	}
1036 
1037 	udf_closedir(ds);
1038 	return(error);
1039 }
1040 
1041 static int
1042 udf_reclaim(struct vop_reclaim_args *a)
1043 {
1044 	struct vnode *vp;
1045 	struct udf_node *unode;
1046 
1047 	vp = a->a_vp;
1048 	unode = VTON(vp);
1049 
1050 	if (unode != NULL) {
1051 		udf_hashrem(unode);
1052 		if (unode->i_devvp) {
1053 			vrele(unode->i_devvp);
1054 			unode->i_devvp = 0;
1055 		}
1056 
1057 		if (unode->fentry != NULL)
1058 			kfree(unode->fentry, M_UDFFENTRY);
1059 		kfree(unode, M_UDFNODE);
1060 		vp->v_data = NULL;
1061 	}
1062 
1063 	return(0);
1064 }
1065 
1066 /*
1067  * Read the block and then set the data pointer to correspond with the
1068  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1069  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1070  * whole extent.
1071  *
1072  * Note that *bp may be assigned error or not.
1073  *
1074  * XXX 'size' is limited to the logical block size for now due to problems
1075  * with udf_read()
1076  */
1077 static int
1078 udf_readatoffset(struct udf_node *node, int *size, int offset, struct buf **bp,
1079 		 uint8_t **data)
1080 {
1081 	struct udf_mnt *udfmp;
1082 	struct file_entry *fentry = NULL;
1083 	struct buf *bp1;
1084 	uint32_t max_size;
1085 	daddr_t sector;
1086 	int error;
1087 
1088 	udfmp = node->udfmp;
1089 
1090 	*bp = NULL;
1091 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1092 	if (error == UDF_INVALID_BMAP) {
1093 		/*
1094 		 * This error means that the file *data* is stored in the
1095 		 * allocation descriptor field of the file entry.
1096 		 */
1097 		fentry = node->fentry;
1098 		*data = &fentry->data[fentry->l_ea];
1099 		*size = fentry->l_ad;
1100 		return(0);
1101 	} else if (error != 0) {
1102 		return(error);
1103 	}
1104 
1105 	/* Adjust the size so that it is within range */
1106 	if (*size == 0 || *size > max_size)
1107 		*size = max_size;
1108 	*size = min(*size, MAXBSIZE);
1109 
1110 	if ((error = udf_readlblks(udfmp, sector, *size, bp))) {
1111 		kprintf("warning: udf_readlblks returned error %d\n", error);
1112 		/* note: *bp may be non-NULL */
1113 		return(error);
1114 	}
1115 
1116 	bp1 = *bp;
1117 	*data = (uint8_t *)&bp1->b_data[offset % udfmp->bsize];
1118 	return(0);
1119 }
1120 
1121 /*
1122  * Translate a file offset into a logical block and then into a physical
1123  * block.
1124  */
1125 static int
1126 udf_bmap_internal(struct udf_node *node, uint32_t offset, daddr_t *sector, uint32_t *max_size)
1127 {
1128 	struct udf_mnt *udfmp;
1129 	struct file_entry *fentry;
1130 	void *icb;
1131 	struct icb_tag *tag;
1132 	uint32_t icblen = 0;
1133 	daddr_t lsector;
1134 	int ad_offset, ad_num = 0;
1135 	int i, p_offset;
1136 
1137 	udfmp = node->udfmp;
1138 	fentry = node->fentry;
1139 	tag = &fentry->icbtag;
1140 
1141 	switch (tag->strat_type) {
1142 	case 4:
1143 		break;
1144 
1145 	case 4096:
1146 		kprintf("Cannot deal with strategy4096 yet!\n");
1147 		return(ENODEV);
1148 
1149 	default:
1150 		kprintf("Unknown strategy type %d\n", tag->strat_type);
1151 		return(ENODEV);
1152 	}
1153 
1154 	switch (tag->flags & 0x7) {
1155 	case 0:
1156 		/*
1157 		 * The allocation descriptor field is filled with short_ad's.
1158 		 * If the offset is beyond the current extent, look for the
1159 		 * next extent.
1160 		 */
1161 		do {
1162 			offset -= icblen;
1163 			ad_offset = sizeof(struct short_ad) * ad_num;
1164 			if (ad_offset > fentry->l_ad) {
1165 				kprintf("File offset out of bounds\n");
1166 				return(EINVAL);
1167 			}
1168 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1169 			icblen = GETICBLEN(short_ad, icb);
1170 			ad_num++;
1171 		} while(offset >= icblen);
1172 
1173 		lsector = (offset  >> udfmp->bshift) +
1174 		    ((struct short_ad *)(icb))->pos;
1175 
1176 		*max_size = GETICBLEN(short_ad, icb);
1177 
1178 		break;
1179 	case 1:
1180 		/*
1181 		 * The allocation descriptor field is filled with long_ad's
1182 		 * If the offset is beyond the current extent, look for the
1183 		 * next extent.
1184 		 */
1185 		do {
1186 			offset -= icblen;
1187 			ad_offset = sizeof(struct long_ad) * ad_num;
1188 			if (ad_offset > fentry->l_ad) {
1189 				kprintf("File offset out of bounds\n");
1190 				return(EINVAL);
1191 			}
1192 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1193 			icblen = GETICBLEN(long_ad, icb);
1194 			ad_num++;
1195 		} while(offset >= icblen);
1196 
1197 		lsector = (offset >> udfmp->bshift) +
1198 		    ((struct long_ad *)(icb))->loc.lb_num;
1199 
1200 		*max_size = GETICBLEN(long_ad, icb);
1201 
1202 		break;
1203 	case 3:
1204 		/*
1205 		 * This type means that the file *data* is stored in the
1206 		 * allocation descriptor field of the file entry.
1207 		 */
1208 		*max_size = 0;
1209 		*sector = node->hash_id + udfmp->part_start;
1210 
1211 		return(UDF_INVALID_BMAP);
1212 	case 2:
1213 		/* DirectCD does not use extended_ad's */
1214 	default:
1215 		kprintf("Unsupported allocation descriptor %d\n",
1216 		       tag->flags & 0x7);
1217 		return(ENODEV);
1218 	}
1219 
1220 	*sector = lsector + udfmp->part_start;
1221 
1222 	/*
1223 	 * Check the sparing table.  Each entry represents the beginning of
1224 	 * a packet.
1225 	 */
1226 	if (udfmp->s_table != NULL) {
1227 		for (i = 0; i< udfmp->s_table_entries; i++) {
1228 			p_offset = lsector - udfmp->s_table->entries[i].org;
1229 			if ((p_offset < udfmp->p_sectors) && (p_offset >= 0)) {
1230 				*sector = udfmp->s_table->entries[i].map +
1231 				    p_offset;
1232 				break;
1233 			}
1234 		}
1235 	}
1236 
1237 	return(0);
1238 }
1239