xref: /dragonfly/sys/vfs/udf/udf_vnops.c (revision cc93b0eb)
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 	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= %"PRId64"\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, &vp, 0, 0);
180 	if (error) {
181 		kprintf("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_rmajor = VNOVAL;
382 	vap->va_rminor = VNOVAL;
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 	kprintf("%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 = kmalloc(NAME_MAX * sizeof(unicode_t), M_TEMP, M_WAITOK | M_ZERO);
487 
488 	if ((unilen = udf_UncompressUnicode(len, cs0string, transname)) == -1) {
489 		kprintf("udf: Unicode translation failed\n");
490 		kfree(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 	kfree(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 = kmalloc(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 	kfree(transname, M_TEMP);
531 	return(error);
532 }
533 
534 struct udf_uiodir {
535 	struct dirent *dirent;
536 	off_t *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 = kmalloc(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 		kfree(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 			kprintf("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 = kmalloc(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 			kprintf("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 		kfree(ds->buf, M_UDFFID);
680 
681 	kfree(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 	off_t *cookies = NULL;
695 	int ncookies;
696 	int error = 0;
697 	char *name;
698 
699 	vp = a->a_vp;
700 
701 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
702 		return (error);
703 
704 	uio = a->a_uio;
705 	node = VTON(vp);
706 	udfmp = node->udfmp;
707 	uiodir.eofflag = 1;
708 
709 	if (a->a_ncookies != NULL) {
710 		/*
711 		 * Guess how many entries are needed.  If we run out, this
712 		 * function will be called again and thing will pick up were
713 		 * it left off.
714 		 */
715 		ncookies = uio->uio_resid / 8 + 1;
716 		if (ncookies > 1024)
717 			ncookies = 1024;
718 		cookies = kmalloc(sizeof(off_t) * ncookies, M_TEMP, M_WAITOK);
719 		uiodir.ncookies = ncookies;
720 		uiodir.cookies = cookies;
721 		uiodir.acookies = 0;
722 	} else
723 		uiodir.cookies = NULL;
724 
725 	/*
726 	 * Iterate through the file id descriptors.  Give the parent dir
727 	 * entry special attention.
728 	 */
729 	ds = udf_opendir(node, uio->uio_offset, node->fentry->inf_len,
730 			 node->udfmp);
731 
732 	name = kmalloc(NAME_MAX, M_TEMP, M_WAITOK);
733 
734 	while ((fid = udf_getfid(ds)) != NULL) {
735 
736 		/* XXX Should we return an error on a bad fid? */
737 		if (udf_checktag(&fid->tag, TAGID_FID)) {
738 			kprintf("Invalid FID tag\n");
739 			error = EIO;
740 			break;
741 		}
742 
743 		/* Is this a deleted file? */
744 		if (fid->file_char & UDF_FILE_CHAR_DEL)
745 			continue;
746 
747 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
748 			/* Do up the '.' and '..' entries.  Dummy values are
749 			 * used for the cookies since the offset here is
750 			 * usually zero, and NFS doesn't like that value
751 			 */
752 			if (uiodir.cookies != NULL) {
753 				if (++uiodir.acookies > uiodir.ncookies) {
754 					uiodir.eofflag = 0;
755 					break;
756 				}
757 				*uiodir.cookies++ = 1;
758 			}
759 			if (vop_write_dirent(&error, uio, node->hash_id, DT_DIR,
760 					     1, ".")) {
761 				uiodir.eofflag = 0;
762 				break;
763 			}
764 			if (error) {
765 				uiodir.eofflag = 0;
766 				break;
767 			}
768 			if (uiodir.cookies != NULL) {
769 				if (++uiodir.acookies > uiodir.ncookies) {
770 					uiodir.eofflag = 0;
771 					break;
772 				}
773 				*uiodir.cookies++ = 2;
774 			}
775 			if (vop_write_dirent(&error, uio, udf_getid(&fid->icb),
776 					     DT_DIR, 2, "..")) {
777 				uiodir.eofflag = 0;
778 				break;
779 			}
780 			if (error) {
781 				uiodir.eofflag = 0;
782 				break;
783 			}
784 		} else {
785 			uint8_t d_type = (fid->file_char & UDF_FILE_CHAR_DIR) ?
786 			    DT_DIR : DT_UNKNOWN;
787 			uint16_t namelen = udf_transname(&fid->data[fid->l_iu],
788 			    name, fid->l_fi, udfmp);
789 
790 			if (uiodir.cookies != NULL) {
791 				if (++uiodir.acookies > uiodir.ncookies) {
792 					uiodir.eofflag = 0;
793 					break;
794 				}
795 				*uiodir.cookies++ = ds->this_off;
796 			}
797 			if (vop_write_dirent(&error, uio, udf_getid(&fid->icb),
798 					 d_type, namelen, name)) {
799 				uiodir.eofflag = 0;
800 				break;
801 			}
802 			if (error) {
803 				uiodir.eofflag = 0;
804 				break;
805 			}
806 		}
807 		if (error) {
808 			kprintf("uiomove returned %d\n", error);
809 			break;
810 		}
811 
812 	}
813 
814 	kfree(name, M_TEMP);
815 
816 	/* tell the calling layer whether we need to be called again */
817 	*a->a_eofflag = uiodir.eofflag;
818 	uio->uio_offset = ds->offset + ds->off;
819 
820 	if (!error)
821 		error = ds->error;
822 
823 	udf_closedir(ds);
824 
825 	if (a->a_ncookies != NULL) {
826 		if (error)
827 			kfree(cookies, M_TEMP);
828 		else {
829 			*a->a_ncookies = uiodir.acookies;
830 			*a->a_cookies = cookies;
831 		}
832 	}
833 
834 	vn_unlock(vp);
835 	return(error);
836 }
837 
838 /* Are there any implementations out there that do soft-links? */
839 static int
840 udf_readlink(struct vop_readlink_args *ap)
841 {
842 	kprintf("%s called\n", __func__);
843 	return(EOPNOTSUPP);
844 }
845 
846 static int
847 udf_strategy(struct vop_strategy_args *ap)
848 {
849 	struct bio *bio;
850 	struct bio *nbio;
851 	struct buf *bp;
852 	struct vnode *vp;
853 	struct udf_node *node;
854 	int maxsize;
855 	daddr_t dblkno;
856 
857 	bio = ap->a_bio;
858 	bp = bio->bio_buf;
859 	vp = ap->a_vp;
860 	node = VTON(vp);
861 
862 	nbio = push_bio(bio);
863 	if (nbio->bio_offset == NOOFFSET) {
864 		/*
865 		 * Files that are embedded in the fentry don't translate well
866 		 * to a block number.  Reject.
867 		 */
868 		if (udf_bmap_internal(node,
869 				     bio->bio_offset,
870 				     &dblkno, &maxsize)) {
871 			clrbuf(bp);
872 			nbio->bio_offset = NOOFFSET;
873 		} else {
874 			nbio->bio_offset = dbtob(dblkno);
875 		}
876 	}
877 	if (nbio->bio_offset == NOOFFSET) {
878 		/* I/O was never started on nbio, must biodone(bio) */
879 		biodone(bio);
880 		return(0);
881 	}
882 	vn_strategy(node->i_devvp, nbio);
883 	return(0);
884 }
885 
886 static int
887 udf_bmap(struct vop_bmap_args *a)
888 {
889 	struct udf_node *node;
890 	uint32_t max_size;
891 	daddr_t lsector;
892 	int error;
893 
894 	node = VTON(a->a_vp);
895 
896 	if (a->a_doffsetp == NULL)
897 		return(0);
898 
899 	KKASSERT(a->a_loffset % node->udfmp->bsize == 0);
900 
901 	error = udf_bmap_internal(node, a->a_loffset, &lsector, &max_size);
902 	if (error)
903 		return(error);
904 
905 	/* Translate logical to physical sector number */
906 	*a->a_doffsetp = (off_t)lsector << node->udfmp->bshift;
907 
908 	/* Punt on read-ahead for now */
909 	if (a->a_runp)
910 		*a->a_runp = 0;
911 	if (a->a_runb)
912 		*a->a_runb = 0;
913 	return(0);
914 }
915 
916 /*
917  * The all powerful VOP_LOOKUP().
918  */
919 static int
920 udf_lookup(struct vop_old_lookup_args *a)
921 {
922 	struct vnode *dvp;
923 	struct vnode *tdp = NULL;
924 	struct vnode **vpp = a->a_vpp;
925 	struct udf_node *node;
926 	struct udf_mnt *udfmp;
927 	struct fileid_desc *fid = NULL;
928 	struct udf_dirstream *ds;
929 	struct thread *td;
930 	globaldata_t gd = mycpu;
931 	u_long nameiop;
932 	u_long flags;
933 	char *nameptr;
934 	long namelen;
935 	ino_t id = 0;
936 	int offset, error = 0;
937 	int numdirpasses, fsize;
938 
939 	dvp = a->a_dvp;
940 	node = VTON(dvp);
941 	udfmp = node->udfmp;
942 	nameiop = a->a_cnp->cn_nameiop;
943 	flags = a->a_cnp->cn_flags;
944 	nameptr = a->a_cnp->cn_nameptr;
945 	namelen = a->a_cnp->cn_namelen;
946 	fsize = node->fentry->inf_len;
947 	td = a->a_cnp->cn_td;
948 
949 	*vpp = NULL;
950 
951 	/*
952 	 * If this is a LOOKUP and we've already partially searched through
953 	 * the directory, pick up where we left off and flag that the
954 	 * directory may need to be searched twice.  For a full description,
955 	 * see /sys/isofs/cd9660/cd9660_lookup.c:cd9660_lookup()
956 	 */
957 	if (nameiop != NAMEI_LOOKUP || node->diroff == 0 ||
958 	    node->diroff > fsize) {
959 		offset = 0;
960 		numdirpasses = 1;
961 	} else {
962 		offset = node->diroff;
963 		numdirpasses = 2;
964 		gd->gd_nchstats->ncs_2passes++;
965 	}
966 
967 lookloop:
968 	ds = udf_opendir(node, offset, fsize, udfmp);
969 
970 	while ((fid = udf_getfid(ds)) != NULL) {
971 		/* XXX Should we return an error on a bad fid? */
972 		if (udf_checktag(&fid->tag, TAGID_FID)) {
973 			kprintf("udf_lookup: Invalid tag\n");
974 			error = EIO;
975 			break;
976 		}
977 
978 		/* Is this a deleted file? */
979 		if (fid->file_char & UDF_FILE_CHAR_DEL)
980 			continue;
981 
982 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
983 			if (flags & CNP_ISDOTDOT) {
984 				id = udf_getid(&fid->icb);
985 				break;
986 			}
987 		} else {
988 			if (!(udf_cmpname(&fid->data[fid->l_iu],
989 					  nameptr, fid->l_fi, namelen, udfmp))) {
990 				id = udf_getid(&fid->icb);
991 				break;
992 			}
993 		}
994 	}
995 
996 	if (!error)
997 		error = ds->error;
998 
999 	/* XXX Bail out here? */
1000 	if (error) {
1001 		udf_closedir(ds);
1002 		return (error);
1003 	}
1004 
1005 	/* Did we have a match? */
1006 	if (id) {
1007 		error = udf_vget(udfmp->im_mountp, id, &tdp);
1008 		if (!error) {
1009 			/*
1010 			 * Remember where this entry was if it's the final
1011 			 * component.
1012 			 */
1013 			if (nameiop == NAMEI_LOOKUP)
1014 				node->diroff = ds->offset + ds->off;
1015 			if (numdirpasses == 2)
1016 				gd->gd_nchstats->ncs_pass2++;
1017 			if ((flags & CNP_LOCKPARENT) == 0) {
1018 				a->a_cnp->cn_flags |= CNP_PDIRUNLOCK;
1019 				vn_unlock(dvp);
1020 			}
1021 
1022 			*vpp = tdp;
1023 		}
1024 	} else {
1025 		/* Name wasn't found on this pass.  Do another pass? */
1026 		if (numdirpasses == 2) {
1027 			numdirpasses--;
1028 			offset = 0;
1029 			udf_closedir(ds);
1030 			goto lookloop;
1031 		}
1032 		if (nameiop == NAMEI_CREATE || nameiop == NAMEI_RENAME) {
1033 			error = EROFS;
1034 		} else {
1035 			error = ENOENT;
1036 		}
1037 	}
1038 
1039 	udf_closedir(ds);
1040 	return(error);
1041 }
1042 
1043 static int
1044 udf_reclaim(struct vop_reclaim_args *a)
1045 {
1046 	struct vnode *vp;
1047 	struct udf_node *unode;
1048 
1049 	vp = a->a_vp;
1050 	unode = VTON(vp);
1051 
1052 	if (unode != NULL) {
1053 		udf_hashrem(unode);
1054 		if (unode->i_devvp) {
1055 			vrele(unode->i_devvp);
1056 			unode->i_devvp = 0;
1057 		}
1058 
1059 		if (unode->fentry != NULL)
1060 			kfree(unode->fentry, M_UDFFENTRY);
1061 		kfree(unode, M_UDFNODE);
1062 		vp->v_data = NULL;
1063 	}
1064 
1065 	return(0);
1066 }
1067 
1068 /*
1069  * Read the block and then set the data pointer to correspond with the
1070  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1071  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1072  * whole extent.
1073  *
1074  * Note that *bp may be assigned error or not.
1075  *
1076  * XXX 'size' is limited to the logical block size for now due to problems
1077  * with udf_read()
1078  */
1079 static int
1080 udf_readatoffset(struct udf_node *node, int *size, int offset, struct buf **bp,
1081 		 uint8_t **data)
1082 {
1083 	struct udf_mnt *udfmp;
1084 	struct file_entry *fentry = NULL;
1085 	struct buf *bp1;
1086 	uint32_t max_size;
1087 	daddr_t sector;
1088 	int error;
1089 
1090 	udfmp = node->udfmp;
1091 
1092 	*bp = NULL;
1093 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1094 	if (error == UDF_INVALID_BMAP) {
1095 		/*
1096 		 * This error means that the file *data* is stored in the
1097 		 * allocation descriptor field of the file entry.
1098 		 */
1099 		fentry = node->fentry;
1100 		*data = &fentry->data[fentry->l_ea];
1101 		*size = fentry->l_ad;
1102 		return(0);
1103 	} else if (error != 0) {
1104 		return(error);
1105 	}
1106 
1107 	/* Adjust the size so that it is within range */
1108 	if (*size == 0 || *size > max_size)
1109 		*size = max_size;
1110 	*size = min(*size, MAXBSIZE);
1111 
1112 	if ((error = udf_readlblks(udfmp, sector, *size, bp))) {
1113 		kprintf("warning: udf_readlblks returned error %d\n", error);
1114 		/* note: *bp may be non-NULL */
1115 		return(error);
1116 	}
1117 
1118 	bp1 = *bp;
1119 	*data = (uint8_t *)&bp1->b_data[offset % udfmp->bsize];
1120 	return(0);
1121 }
1122 
1123 /*
1124  * Translate a file offset into a logical block and then into a physical
1125  * block.
1126  */
1127 static int
1128 udf_bmap_internal(struct udf_node *node, uint32_t offset, daddr_t *sector, uint32_t *max_size)
1129 {
1130 	struct udf_mnt *udfmp;
1131 	struct file_entry *fentry;
1132 	void *icb;
1133 	struct icb_tag *tag;
1134 	uint32_t icblen = 0;
1135 	daddr_t lsector;
1136 	int ad_offset, ad_num = 0;
1137 	int i, p_offset;
1138 
1139 	udfmp = node->udfmp;
1140 	fentry = node->fentry;
1141 	tag = &fentry->icbtag;
1142 
1143 	switch (tag->strat_type) {
1144 	case 4:
1145 		break;
1146 
1147 	case 4096:
1148 		kprintf("Cannot deal with strategy4096 yet!\n");
1149 		return(ENODEV);
1150 
1151 	default:
1152 		kprintf("Unknown strategy type %d\n", tag->strat_type);
1153 		return(ENODEV);
1154 	}
1155 
1156 	switch (tag->flags & 0x7) {
1157 	case 0:
1158 		/*
1159 		 * The allocation descriptor field is filled with short_ad's.
1160 		 * If the offset is beyond the current extent, look for the
1161 		 * next extent.
1162 		 */
1163 		do {
1164 			offset -= icblen;
1165 			ad_offset = sizeof(struct short_ad) * ad_num;
1166 			if (ad_offset > fentry->l_ad) {
1167 				kprintf("File offset out of bounds\n");
1168 				return(EINVAL);
1169 			}
1170 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1171 			icblen = GETICBLEN(short_ad, icb);
1172 			ad_num++;
1173 		} while(offset >= icblen);
1174 
1175 		lsector = (offset  >> udfmp->bshift) +
1176 		    ((struct short_ad *)(icb))->pos;
1177 
1178 		*max_size = GETICBLEN(short_ad, icb);
1179 
1180 		break;
1181 	case 1:
1182 		/*
1183 		 * The allocation descriptor field is filled with long_ad's
1184 		 * If the offset is beyond the current extent, look for the
1185 		 * next extent.
1186 		 */
1187 		do {
1188 			offset -= icblen;
1189 			ad_offset = sizeof(struct long_ad) * ad_num;
1190 			if (ad_offset > fentry->l_ad) {
1191 				kprintf("File offset out of bounds\n");
1192 				return(EINVAL);
1193 			}
1194 			icb = GETICB(long_ad, fentry, fentry->l_ea + ad_offset);
1195 			icblen = GETICBLEN(long_ad, icb);
1196 			ad_num++;
1197 		} while(offset >= icblen);
1198 
1199 		lsector = (offset >> udfmp->bshift) +
1200 		    ((struct long_ad *)(icb))->loc.lb_num;
1201 
1202 		*max_size = GETICBLEN(long_ad, icb);
1203 
1204 		break;
1205 	case 3:
1206 		/*
1207 		 * This type means that the file *data* is stored in the
1208 		 * allocation descriptor field of the file entry.
1209 		 */
1210 		*max_size = 0;
1211 		*sector = node->hash_id + udfmp->part_start;
1212 
1213 		return(UDF_INVALID_BMAP);
1214 	case 2:
1215 		/* DirectCD does not use extended_ad's */
1216 	default:
1217 		kprintf("Unsupported allocation descriptor %d\n",
1218 		       tag->flags & 0x7);
1219 		return(ENODEV);
1220 	}
1221 
1222 	*sector = lsector + udfmp->part_start;
1223 
1224 	/*
1225 	 * Check the sparing table.  Each entry represents the beginning of
1226 	 * a packet.
1227 	 */
1228 	if (udfmp->s_table != NULL) {
1229 		for (i = 0; i< udfmp->s_table_entries; i++) {
1230 			p_offset = lsector - udfmp->s_table->entries[i].org;
1231 			if ((p_offset < udfmp->p_sectors) && (p_offset >= 0)) {
1232 				*sector = udfmp->s_table->entries[i].map +
1233 				    p_offset;
1234 				break;
1235 			}
1236 		}
1237 	}
1238 
1239 	return(0);
1240 }
1241