xref: /freebsd/sys/fs/udf/udf_vnops.c (revision c11e094d96120a2e0e726ed9705ae0ec08db49b6)
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  * $FreeBSD$
27  */
28 
29 /* udf_vnops.c */
30 /* Take care of the vnode side of things */
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/namei.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/stat.h>
38 #include <sys/bio.h>
39 #include <sys/buf.h>
40 #include <sys/mount.h>
41 #include <sys/vnode.h>
42 #include <sys/dirent.h>
43 #include <sys/queue.h>
44 #include <sys/unistd.h>
45 
46 #include <vm/uma.h>
47 
48 #include <fs/udf/ecma167-udf.h>
49 #include <fs/udf/osta.h>
50 #include <fs/udf/udf.h>
51 
52 static int udf_access(struct vop_access_args *);
53 static int udf_getattr(struct vop_getattr_args *);
54 static int udf_ioctl(struct vop_ioctl_args *);
55 static int udf_pathconf(struct vop_pathconf_args *);
56 static int udf_read(struct vop_read_args *);
57 static int udf_readdir(struct vop_readdir_args *);
58 static int udf_readlink(struct vop_readlink_args *ap);
59 static int udf_strategy(struct vop_strategy_args *);
60 static int udf_print(struct vop_print_args *);
61 static int udf_bmap(struct vop_bmap_args *);
62 static int udf_lookup(struct vop_cachedlookup_args *);
63 static int udf_reclaim(struct vop_reclaim_args *);
64 static void udf_dumpblock(void *, int) __unused;
65 static int udf_readatoffset(struct udf_node *, int *, int, struct buf **, uint8_t **);
66 static int udf_bmap_internal(struct udf_node *, uint32_t, daddr_t *, uint32_t *);
67 
68 vop_t **udf_vnodeop_p;
69 static struct vnodeopv_entry_desc udf_vnodeop_entries[] = {
70 	{ &vop_default_desc,		(vop_t *) vop_defaultop },
71 	{ &vop_access_desc,		(vop_t *) udf_access },
72 	{ &vop_bmap_desc,		(vop_t *) udf_bmap },
73 	{ &vop_cachedlookup_desc,	(vop_t *) udf_lookup },
74 	{ &vop_getattr_desc,		(vop_t *) udf_getattr },
75 	{ &vop_ioctl_desc,		(vop_t *) udf_ioctl },
76 	{ &vop_islocked_desc,		(vop_t *) vop_stdislocked },
77 	{ &vop_lock_desc,		(vop_t *) vop_stdlock },
78 	{ &vop_lookup_desc,		(vop_t *) vfs_cache_lookup },
79 	{ &vop_pathconf_desc,		(vop_t *) udf_pathconf },
80 	{ &vop_print_desc,		(vop_t *) udf_print },
81 	{ &vop_read_desc,		(vop_t *) udf_read },
82 	{ &vop_readdir_desc,		(vop_t *) udf_readdir },
83 	{ &vop_readlink_desc,		(vop_t *) udf_readlink },
84 	{ &vop_reclaim_desc,		(vop_t *) udf_reclaim },
85 	{ &vop_strategy_desc,		(vop_t *) udf_strategy },
86 	{ &vop_unlock_desc,		(vop_t *) vop_stdunlock },
87 	{ NULL, NULL }
88 };
89 static struct vnodeopv_desc udf_vnodeop_opv_desc =
90 	{ &udf_vnodeop_p, udf_vnodeop_entries };
91 VNODEOP_SET(udf_vnodeop_opv_desc);
92 
93 MALLOC_DEFINE(M_UDFFID, "UDF FID", "UDF FileId structure");
94 
95 /* Look up a udf_node based on the ino_t passed in and return it's vnode */
96 int
97 udf_hashlookup(struct udf_mnt *udfmp, ino_t id, int flags, struct vnode **vpp)
98 {
99 	struct udf_node *node;
100 	int error;
101 
102 	*vpp = NULL;
103 
104 loop:
105 	mtx_lock(&udfmp->hash_mtx);
106 	TAILQ_FOREACH(node, &udfmp->udf_tqh, tq) {
107 		if (node->hash_id == id) {
108 			VI_LOCK(node->i_vnode);
109 			mtx_unlock(&udfmp->hash_mtx);
110 			error = vget(node->i_vnode, flags | LK_INTERLOCK,
111 			    curthread);
112 			if (error == ENOENT)
113 				goto loop;
114 			if (error)
115 				return (error);
116 			*vpp = node->i_vnode;
117 			return (0);
118 		}
119 	}
120 
121 	mtx_unlock(&udfmp->hash_mtx);
122 	return (0);
123 }
124 
125 int
126 udf_hashins(struct udf_node *node)
127 {
128 	struct udf_mnt *udfmp;
129 
130 	udfmp = node->udfmp;
131 
132 	mtx_lock(&udfmp->hash_mtx);
133 	TAILQ_INSERT_TAIL(&udfmp->udf_tqh, node, tq);
134 	mtx_unlock(&udfmp->hash_mtx);
135 	lockmgr(&node->i_vnode->v_lock, LK_EXCLUSIVE, (struct mtx *)0,
136 		curthread);
137 
138 	return (0);
139 }
140 
141 int
142 udf_hashrem(struct udf_node *node)
143 {
144 	struct udf_mnt *udfmp;
145 
146 	udfmp = node->udfmp;
147 
148 	mtx_lock(&udfmp->hash_mtx);
149 	TAILQ_REMOVE(&udfmp->udf_tqh, node, tq);
150 	mtx_unlock(&udfmp->hash_mtx);
151 
152 	return (0);
153 }
154 
155 int
156 udf_allocv(struct mount *mp, struct vnode **vpp, struct thread *td)
157 {
158 	int error;
159 	struct vnode *vp;
160 
161 	error = getnewvnode(VT_UDF, mp, udf_vnodeop_p, &vp);
162 	if (error) {
163 		printf("udf_allocv: failed to allocate new vnode\n");
164 		return (error);
165 	}
166 
167 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
168 	*vpp = vp;
169 	return (0);
170 }
171 
172 /* Convert file entry permission (5 bits per owner/group/user) to a mode_t */
173 static mode_t
174 udf_permtomode(struct udf_node *node)
175 {
176 	uint32_t perm;
177 	uint32_t flags;
178 	mode_t mode;
179 
180 	perm = node->fentry->perm;
181 	flags = node->fentry->icbtag.flags;
182 
183 	mode = perm & UDF_FENTRY_PERM_USER_MASK;
184 	mode |= ((perm & UDF_FENTRY_PERM_GRP_MASK) >> 2);
185 	mode |= ((perm & UDF_FENTRY_PERM_OWNER_MASK) >> 4);
186 	mode |= ((flags & UDF_ICB_TAG_FLAGS_STICKY) << 4);
187 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETGID) << 6);
188 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETUID) << 8);
189 
190 	return (mode);
191 }
192 
193 static int
194 udf_access(struct vop_access_args *a)
195 {
196 	struct vnode *vp;
197 	struct udf_node *node;
198 	mode_t a_mode, mode;
199 
200 	vp = a->a_vp;
201 	node = VTON(vp);
202 	a_mode = a->a_mode;
203 
204 	if (a_mode & VWRITE) {
205 		switch (vp->v_type) {
206 		case VDIR:
207 		case VLNK:
208 		case VREG:
209 			return (EROFS);
210 			/* NOT REACHED */
211 		default:
212 			break;
213 		}
214 	}
215 
216 	mode = udf_permtomode(node);
217 
218 	return (vaccess(vp->v_type, mode, node->fentry->uid, node->fentry->gid,
219 	    a_mode, a->a_cred, NULL));
220 }
221 
222 static int mon_lens[2][12] = {
223 	{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31},
224 	{31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31}
225 };
226 
227 static int
228 udf_isaleapyear(int year)
229 {
230 	int i;
231 
232 	i = (year % 4) ? 0 : 1;
233 	i &= (year % 100) ? 1 : 0;
234 	i |= (year % 400) ? 0 : 1;
235 
236 	return i;
237 }
238 
239 /*
240  * XXX This is just a rough hack.  Daylight savings isn't calculated and tv_nsec
241  * is ignored.
242  * Timezone calculation compliments of Julian Elischer <julian@elischer.org>.
243  */
244 static void
245 udf_timetotimespec(struct timestamp *time, struct timespec *t)
246 {
247 	int i, lpyear, daysinyear;
248 	union {
249 		uint16_t	u_tz_offset;
250 		int16_t		s_tz_offset;
251 	} tz;
252 
253 	t->tv_nsec = 0;
254 
255 	/* DirectCD seems to like using bogus year values */
256 	if (time->year < 1970) {
257 		t->tv_sec = 0;
258 		return;
259 	}
260 
261 	/* Calculate the time and day */
262 	t->tv_sec = time->second;
263 	t->tv_sec += time->minute * 60;
264 	t->tv_sec += time->hour * 3600;
265 	t->tv_sec += time->day * 3600 * 24;
266 
267 	/* Calclulate the month */
268 	lpyear = udf_isaleapyear(time->year);
269 	for (i = 1; i < time->month; i++)
270 		t->tv_sec += mon_lens[lpyear][i] * 3600 * 24;
271 
272 	/* Speed up the calculation */
273 	if (time->year > 1979)
274 		t->tv_sec += 315532800;
275 	if (time->year > 1989)
276 		t->tv_sec += 315619200;
277 	if (time->year > 1999)
278 		t->tv_sec += 315532800;
279 	for (i = 2000; i < time->year; i++) {
280 		daysinyear = udf_isaleapyear(i) + 365 ;
281 		t->tv_sec += daysinyear * 3600 * 24;
282 	}
283 
284 	/*
285 	 * Calculate the time zone.  The timezone is 12 bit signed 2's
286 	 * compliment, so we gotta do some extra magic to handle it right.
287 	 */
288 	tz.u_tz_offset = time->type_tz;
289 	tz.u_tz_offset &= 0x0fff;
290 	if (tz.u_tz_offset & 0x0800)
291 		tz.u_tz_offset |= 0xf000;	/* extend the sign to 16 bits */
292 	if ((time->type_tz & 0x1000) && (tz.s_tz_offset != -2047))
293 		t->tv_sec -= tz.s_tz_offset * 60;
294 
295 	return;
296 }
297 
298 static int
299 udf_getattr(struct vop_getattr_args *a)
300 {
301 	struct vnode *vp;
302 	struct udf_node *node;
303 	struct vattr *vap;
304 	struct file_entry *fentry;
305 	struct timespec ts;
306 
307 	ts.tv_sec = 0;
308 
309 	vp = a->a_vp;
310 	vap = a->a_vap;
311 	node = VTON(vp);
312 	fentry = node->fentry;
313 
314 	vap->va_fsid = dev2udev(node->i_dev);
315 	vap->va_fileid = node->hash_id;
316 	vap->va_mode = udf_permtomode(node);
317 	vap->va_nlink = fentry->link_cnt;
318 	/*
319 	 * XXX The spec says that -1 is valid for uid/gid and indicates an
320 	 * invalid uid/gid.  How should this be represented?
321 	 */
322 	vap->va_uid = (fentry->uid == -1) ? 0 : fentry->uid;
323 	vap->va_gid = (fentry->gid == -1) ? 0 : fentry->gid;
324 	udf_timetotimespec(&fentry->atime, &vap->va_atime);
325 	udf_timetotimespec(&fentry->mtime, &vap->va_mtime);
326 	vap->va_ctime = vap->va_mtime; /* XXX Stored as an Extended Attribute */
327 	vap->va_rdev = 0; /* XXX */
328 	if (vp->v_type & VDIR) {
329 		/*
330 		 * Directories that are recorded within their ICB will show
331 		 * as having 0 blocks recorded.  Since tradition dictates
332 		 * that directories consume at least one logical block,
333 		 * make it appear so.
334 		 */
335 		if (fentry->logblks_rec != 0) {
336 			vap->va_size = fentry->logblks_rec * node->udfmp->bsize;
337 		} else {
338 			vap->va_size = node->udfmp->bsize;
339 		}
340 	} else {
341 		vap->va_size = fentry->inf_len;
342 	}
343 	vap->va_flags = 0;
344 	vap->va_gen = 1;
345 	vap->va_blocksize = node->udfmp->bsize;
346 	vap->va_bytes = fentry->inf_len;
347 	vap->va_type = vp->v_type;
348 	vap->va_filerev = 0; /* XXX */
349 	return (0);
350 }
351 
352 /*
353  * File specific ioctls.  DeCSS candidate?
354  */
355 static int
356 udf_ioctl(struct vop_ioctl_args *a)
357 {
358 	printf("%s called\n", __FUNCTION__);
359 	return (EOPNOTSUPP);
360 }
361 
362 /*
363  * I'm not sure that this has much value in a read-only filesystem, but
364  * cd9660 has it too.
365  */
366 static int
367 udf_pathconf(struct vop_pathconf_args *a)
368 {
369 
370 	switch (a->a_name) {
371 	case _PC_LINK_MAX:
372 		*a->a_retval = 65535;
373 		return (0);
374 	case _PC_NAME_MAX:
375 		*a->a_retval = NAME_MAX;
376 		return (0);
377 	case _PC_PATH_MAX:
378 		*a->a_retval = PATH_MAX;
379 		return (0);
380 	case _PC_NO_TRUNC:
381 		*a->a_retval = 1;
382 		return (0);
383 	default:
384 		return (EINVAL);
385 	}
386 }
387 
388 static int
389 udf_read(struct vop_read_args *a)
390 {
391 	struct vnode *vp = a->a_vp;
392 	struct uio *uio = a->a_uio;
393 	struct udf_node *node = VTON(vp);
394 	struct buf *bp;
395 	uint8_t *data;
396 	int error = 0;
397 	int size, fsize, offset;
398 
399 	if (uio->uio_offset < 0)
400 		return (EINVAL);
401 
402 	fsize = node->fentry->inf_len;
403 
404 	while (uio->uio_offset < fsize && uio->uio_resid > 0) {
405 		offset = uio->uio_offset;
406 		size = uio->uio_resid;
407 		error = udf_readatoffset(node, &size, offset, &bp, &data);
408 		if (error)
409 			return (error);
410 		error = uiomove((caddr_t)data, size, uio);
411 		if (bp != NULL)
412 			brelse(bp);
413 		if (error)
414 			break;
415 	};
416 
417 	return (error);
418 }
419 
420 /* Convienience routine to dump a block in hex */
421 static void
422 udf_dumpblock(void *data, int len)
423 {
424 	int i, j;
425 
426 	for (i = 0; i < len; i++) {
427 		printf("\noffset= %d: ", i);
428 		for (j = 0; j < 8; j++) {
429 			if (i + j == len)
430 				break;
431 			printf("0x%02x ", (uint8_t)((uint8_t*)(data))[i + j]);
432 		}
433 		i += j - 1;
434 	}
435 	printf("\n");
436 }
437 
438 /*
439  * Call the OSTA routines to translate the name from a CS0 dstring to a
440  * 16-bit Unicode String.  Hooks need to be placed in here to translate from
441  * Unicode to the encoding that the kernel/user expects.  For now, compact
442  * the encoding to 8 bits if possible.  Return the length of the translated
443  * string.
444  * XXX This horribly pessimizes the 8bit case
445  */
446 static int
447 udf_transname(char *cs0string, char *destname, int len)
448 {
449 	unicode_t *transname;
450 	int i, unilen = 0;
451 
452 	/* allocate a buffer big enough to hold an 8->16 bit expansion */
453 	transname = uma_zalloc(udf_zone_trans, M_WAITOK);
454 	if (transname == NULL) {
455 		printf("udf: out of memory?\n");
456 		return 0;
457 	}
458 
459 	if ((unilen = udf_UncompressUnicode(len, cs0string, transname)) == -1) {
460 		printf("udf: Unicode translation failed\n");
461 		uma_zfree(udf_zone_trans, transname);
462 		return 0;
463 	}
464 
465 	/* At this point, the name is in 16-bit Unicode.  Compact it down
466  	 * to 8-bit
467 	 */
468 	for (i = 0; i < unilen ; i++) {
469 		if (transname[i] & 0xff00) {
470 			destname[i] = '.';	/* Fudge the 16bit chars */
471 		} else {
472 			destname[i] = transname[i] & 0xff;
473 		}
474 	}
475 
476 	destname[unilen] = 0;
477 	uma_zfree(udf_zone_trans, transname);
478 
479 	return unilen;
480 }
481 
482 /*
483  * Compare a CS0 dstring with a name passed in from the VFS layer.  Return
484  * 0 on a successful match, nonzero therwise.  Unicode work may need to be done
485  * here also.
486  */
487 static int
488 udf_cmpname(char *cs0string, char *cmpname, int cs0len, int cmplen)
489 {
490 	char transname[MAXNAMLEN+1]; /* XXX stack */
491 
492 	if ((cs0len = udf_transname(cs0string, &transname[0], cs0len)) == 0)
493 		return -1;
494 
495 	/* Easy check.  If they aren't the same length, they aren't equal */
496 	if (cs0len != cmplen)
497 		return -1;
498 
499 	return (bcmp(transname, cmpname, cmplen));
500 }
501 
502 struct udf_uiodir {
503 	struct dirent *dirent;
504 	u_long *cookies;
505 	int ncookies;
506 	int acookies;
507 	int eofflag;
508 };
509 
510 static int
511 udf_uiodir(struct udf_uiodir *uiodir, int de_size, struct uio *uio, long cookie)
512 {
513 	if (uiodir->cookies != NULL) {
514 		if (++uiodir->acookies > uiodir->ncookies) {
515 			uiodir->eofflag = 0;
516 			return (-1);
517 		}
518 		*uiodir->cookies++ = cookie;
519 	}
520 
521 	if (uio->uio_resid < de_size) {
522 		uiodir->eofflag = 0;
523 		return (-1);
524 	}
525 
526 	return (uiomove((caddr_t)uiodir->dirent, de_size, uio));
527 }
528 
529 /* Prebuild the . and .. dirents.  d_fileno will need to be filled in */
530 static struct dirent udf_de_dot =
531 	{ 0, sizeof(struct dirent), DT_DIR, 1, "." };
532 static struct dirent udf_de_dotdot =
533 	{ 0, sizeof(struct dirent), DT_DIR, 2, ".." };
534 
535 static int
536 udf_readdir(struct vop_readdir_args *a)
537 {
538 	struct vnode *vp;
539 	struct buf *bp;
540 	struct uio *uio;
541 	struct dirent dir;
542 	struct udf_node *node;
543 	struct udf_mnt *udfmp;
544 	struct fileid_desc *fid;
545 	struct udf_uiodir uiodir;
546 	u_long *cookies = NULL;
547 	uint8_t *data;
548 	int ncookies;
549 	int error = 0, offset, off, size, de_size, fid_size, fsize;
550 	int total_fid_size = 0, frag_size = 0, fid_fragment = 0;
551 
552 	vp = a->a_vp;
553 	uio = a->a_uio;
554 	node = VTON(vp);
555 	udfmp = node->udfmp;
556 	de_size = sizeof(struct dirent);
557 	fid_size = UDF_FID_SIZE;
558 	fsize = node->fentry->inf_len;
559 	uiodir.eofflag = 1;
560 
561 	if (a->a_ncookies != NULL) {
562 		/*
563 		 * Guess how many entries are needed.  If we run out, this
564 		 * function will be called again and thing will pick up were
565 		 * it left off.
566 		 */
567 		ncookies = uio->uio_resid / 8;
568 		MALLOC(cookies, u_long *, sizeof(u_long) * ncookies,
569 		    M_TEMP, M_WAITOK);
570 		if (cookies == NULL)
571 			return (ENOMEM);
572 		uiodir.ncookies = ncookies;
573 		uiodir.cookies = cookies;
574 		uiodir.acookies = 0;
575 	} else {
576 		uiodir.cookies = NULL;
577 	}
578 
579 	/*
580 	 * offset is the absolute offset into the file data. off is the offset
581 	 * into the data, minus the blocks that weren't read because they fell
582 	 * before offset.
583 	 */
584 	offset = uio->uio_offset;
585 	off = 0;
586 
587 	/*
588 	 * Iterate through the file id descriptors.  Give the parent dir
589 	 * entry special attention.  size will be the size of the extent
590 	 * returned in data.  If there is more than one extent, things get
591 	 * ugly.
592 	 */
593 	size = 0;
594 	error = udf_readatoffset(node, &size, offset, &bp, &data);
595 	if (error) {
596 		if (a->a_ncookies != NULL)
597 			FREE(cookies, M_TEMP);
598 		return (error);
599 	}
600 
601 	while (offset + off < fsize) {
602 
603 		fid = (struct fileid_desc*)&data[off];
604 
605 		/*
606 		 * Check to see if the fid is fragmented. The first test
607 		 * ensures that we don't wander off the end of the buffer
608 		 * looking for the l_iu and l_fi fields.
609 		 */
610 		if (off + fid_size > size ||
611 		    off + fid->l_iu + fid->l_fi + fid_size > size) {
612 			struct fileid_desc *fid_buf;
613 			uint8_t *buf;
614 
615 			/* Copy what we have of the fid into a buffer */
616 			frag_size = size - off;
617 			MALLOC(buf, uint8_t*, max(frag_size, fid_size),
618 			    M_UDFFID, M_NOWAIT | M_ZERO);
619 			if (buf == NULL)
620 				panic("No memory?");
621 			bcopy(fid, buf, frag_size);
622 
623 			/* Reduce all of the casting magic */
624 			fid_buf = (struct fileid_desc*)buf;
625 
626 			if (bp != NULL)
627 				brelse(bp);
628 
629 			/* Fetch the next allocation */
630 			offset += size;
631 			size = 0;
632 			error = udf_readatoffset(node, &size, offset, &bp,
633 			    &data);
634 			if (error)
635 				break;
636 
637 			/*
638 			 * If the fragment was so small that we didn't get
639 			 * the l_iu and l_fi fields, copy those in.
640 			 */
641 			if (fid_size > frag_size)
642 				bcopy(data, &buf[frag_size],
643 				    fid_size - frag_size);
644 
645 			/*
646 			 * Now that we have enough of the fid to work with,
647 			 * allocate a new fid, copy the fragment into it,
648 			 * and copy the rest of the fid from the new
649 			 * allocation.
650 			 */
651 			total_fid_size = fid_size + fid_buf->l_iu +
652 			    fid_buf->l_fi;
653 			MALLOC(fid, struct fileid_desc *, total_fid_size,
654 			    M_UDFFID, M_NOWAIT | M_ZERO);
655 			if (fid == NULL) {
656 				if (bp != NULL)
657 					brelse(bp);
658 				error = ENOMEM;
659 				break;
660 			}
661 			bcopy(fid_buf, fid, frag_size);
662 			bcopy(data, &((uint8_t*)(fid))[frag_size],
663 			    total_fid_size - frag_size);
664 
665 			fid_fragment = 1;
666 			FREE(buf, M_UDFFID);
667 		} else {
668 			total_fid_size = fid->l_iu + fid->l_fi + fid_size;
669 		}
670 
671 		/* XXX Should we return an error on a bad fid? */
672 		if (udf_checktag(&fid->tag, TAGID_FID)) {
673 			printf("Invalid FID tag\n");
674 			break;
675 		}
676 
677 		/* Is this a deleted file? */
678 		if (fid->file_char & 0x4)
679 			goto update_offset;
680 
681 		if (fid->l_iu != 0) {
682 			printf("Possibly invalid fid found.\n");
683 			goto update_offset;
684 		}
685 
686 		if ((fid->l_fi == 0) && (fid->file_char & 0x08)) {
687 			/* Do up the '.' and '..' entries.  Dummy values are
688 			 * used for the cookies since the offset here is
689 			 * usually zero, and NFS doesn't like that value
690 			 * XXX Should the magic dirents be locked?
691 			 */
692 			udf_de_dot.d_fileno = node->hash_id;
693 			uiodir.dirent = &udf_de_dot;
694 			error = udf_uiodir(&uiodir, de_size, uio, 1);
695 			if (error)
696 				break;
697 
698 			udf_de_dotdot.d_fileno = udf_getid(&fid->icb);
699 			uiodir.dirent = &udf_de_dotdot;
700 			error = udf_uiodir(&uiodir, de_size, uio, 2);
701 		} else {
702 			dir.d_namlen = udf_transname(&fid->data[fid->l_iu],
703 			    &dir.d_name[0], fid->l_fi);
704 			dir.d_fileno = udf_getid(&fid->icb);
705 			dir.d_type = (fid->file_char & 0x02) ? DT_DIR :
706 			    DT_UNKNOWN;
707 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
708 			uiodir.dirent = &dir;
709 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, off);
710 		}
711 		if (error) {
712 			printf("uiomove returned %d\n", error);
713 			break;
714 		}
715 
716 update_offset:	/*
717 		 * Update the offset. Align on a 4 byte boundary because the
718 		 * UDF spec says so.  If it was a fragmented entry, clean up.
719 		 */
720 		if (fid_fragment) {
721 			off = (total_fid_size - frag_size + 3) & ~0x03;
722 			FREE(fid, M_UDFFID);
723 			fid_fragment = 0;
724 		} else {
725 			off += (total_fid_size + 3) & ~0x03;
726 		}
727 	}
728 
729 	/* tell the calling layer whether we need to be called again */
730 	*a->a_eofflag = uiodir.eofflag;
731 	uio->uio_offset = offset + off;
732 
733 	if (bp != NULL)
734 		brelse(bp);
735 
736 	if (a->a_ncookies != NULL) {
737 		if (error)
738 			free(cookies, M_TEMP);
739 		else {
740 			*a->a_ncookies = uiodir.acookies;
741 			*a->a_cookies = cookies;
742 		}
743 	}
744 
745 	return (error);
746 }
747 
748 /* Are there any implementations out there that do soft-links? */
749 static int
750 udf_readlink(struct vop_readlink_args *ap)
751 {
752 	printf("%s called\n", __FUNCTION__);
753 	return (EOPNOTSUPP);
754 }
755 
756 static int
757 udf_strategy(struct vop_strategy_args *a)
758 {
759 	struct buf *bp;
760 	struct vnode *vp;
761 	struct udf_node *node;
762 	int maxsize;
763 
764 	bp = a->a_bp;
765 	vp = bp->b_vp;
766 	node = VTON(vp);
767 
768 	/* cd9660 has this test reversed, but it seems more logical this way */
769 	if (bp->b_blkno != bp->b_lblkno) {
770 		/*
771 		 * Files that are embedded in the fentry don't translate well
772 		 * to a block number.  Reject.
773 		 */
774 		if (udf_bmap_internal(node, bp->b_lblkno * node->udfmp->bsize,
775 		    &bp->b_lblkno, &maxsize)) {
776 			clrbuf(bp);
777 			bp->b_blkno = -1;
778 		}
779 	}
780 	if ((long)bp->b_blkno == -1) {
781 		bufdone(bp);
782 		return (0);
783 	}
784 	vp = node->i_devvp;
785 	bp->b_dev = vp->v_rdev;
786 	VOP_STRATEGY(vp, bp);
787 	return (0);
788 }
789 
790 static int
791 udf_print(struct vop_print_args *a)
792 {
793 	printf("%s called\n", __FUNCTION__);
794 	return (EOPNOTSUPP);
795 }
796 
797 static int
798 udf_bmap(struct vop_bmap_args *a)
799 {
800 	struct udf_node *node;
801 	uint32_t max_size;
802 	daddr_t lsector;
803 	int error;
804 
805 	node = VTON(a->a_vp);
806 
807 	if (a->a_vpp != NULL)
808 		*a->a_vpp = node->i_devvp;
809 	if (a->a_bnp == NULL)
810 		return (0);
811 	if (a->a_runb)
812 		*a->a_runb = 0;
813 
814 	error = udf_bmap_internal(node, a->a_bn * node->udfmp->bsize, &lsector,
815 	    &max_size);
816 	if (error > 0)
817 		return (error);
818 
819 	/* Translate logical to physical sector number */
820 	*a->a_bnp = lsector << (node->udfmp->bshift - DEV_BSHIFT);
821 
822 	/* Punt on read-ahead for now */
823 	if (a->a_runp)
824 		*a->a_runp = 0;
825 
826 	return (0);
827 }
828 
829 /*
830  * The all powerful VOP_LOOKUP().
831  */
832 static int
833 udf_lookup(struct vop_cachedlookup_args *a)
834 {
835 	struct vnode *dvp;
836 	struct vnode *tdp = NULL;
837 	struct vnode **vpp = a->a_vpp;
838 	struct buf *bp = NULL;
839 	struct udf_node *node;
840 	struct udf_mnt *udfmp;
841 	struct fileid_desc *fid = NULL;
842 	struct thread *td;
843 	u_long nameiop;
844 	u_long flags;
845 	char *nameptr;
846 	long namelen;
847 	ino_t id = 0;
848 	uint8_t *data;
849 	int offset, off, error, size;
850 	int numdirpasses, fid_size, fsize, icb_len;
851 	int total_fid_size = 0, fid_fragment = 0;
852 
853 	dvp = a->a_dvp;
854 	node = VTON(dvp);
855 	udfmp = node->udfmp;
856 	nameiop = a->a_cnp->cn_nameiop;
857 	flags = a->a_cnp->cn_flags;
858 	nameptr = a->a_cnp->cn_nameptr;
859 	namelen = a->a_cnp->cn_namelen;
860 	fid_size = UDF_FID_SIZE;
861 	fsize = node->fentry->inf_len;
862 	icb_len = sizeof(struct long_ad);
863 	td = a->a_cnp->cn_thread;
864 
865 	/*
866 	 * If this is a LOOKUP and we've already partially searched through
867 	 * the directory, pick up where we left off and flag that the
868 	 * directory may need to be searched twice.  For a full description,
869 	 * see /sys/isofs/cd9660/cd9660_lookup.c:cd9660_lookup()
870 	 */
871 	if (nameiop != LOOKUP || node->diroff == 0 || node->diroff > size) {
872 		offset = 0;
873 		numdirpasses = 1;
874 	} else {
875 		offset = node->diroff;
876 		numdirpasses = 2;
877 		nchstats.ncs_2passes++;
878 	}
879 
880 	/*
881 	 * The name lookup algorithm is quite similar to what is in readdir.
882 	 * Can this be broken out and shared?
883 	 */
884 lookloop:
885 	size = 0;
886 	off = 0;
887 	error = udf_readatoffset(node, &size, offset, &bp, &data);
888 	if (error)
889 		return (error);
890 
891 	while (offset + off < fsize) {
892 		fid = (struct fileid_desc*)&data[off];
893 
894 		/*
895 		 * Check to see if the fid is fragmented. The first test
896 		 * ensures that we don't wander off the end of the buffer
897 		 * looking for the l_iu and l_fi fields.
898 		 */
899 		if (off + fid_size > size ||
900 		    off + fid_size + fid->l_iu + fid->l_fi > size) {
901 			struct fileid_desc *fid_buf;
902 			uint8_t *buf;
903 			int frag_size = 0;
904 
905 			/* Copy what we have of the fid into a buffer */
906 			frag_size = size - off;
907 			MALLOC(buf, uint8_t*, max(frag_size, fid_size),
908 			    M_UDFFID, M_NOWAIT | M_ZERO);
909 			if (buf == NULL)
910 				panic("No memory?");
911 			bcopy(fid, buf, frag_size);
912 
913 			/* Reduce all of the casting magic */
914 			fid_buf = (struct fileid_desc*)buf;
915 
916 			if (bp != NULL)
917 				brelse(bp);
918 
919 			/* Fetch the next allocation */
920 			offset += size;
921 			size = 0;
922 			error = udf_readatoffset(node, &size, offset, &bp,
923 			    &data);
924 			if (error)
925 				return (error);
926 
927 			/*
928 			 * If the fragment was so small that we didn't get
929 			 * the l_iu and l_fi fields, copy those in.
930 			 */
931 			if (fid_size > frag_size)
932 				bcopy(data, &buf[frag_size],
933 				    fid_size - frag_size);
934 
935 			/*
936 			 * Now that we have enough of the fid to work with,
937 			 * allocate a new fid, copy the fragment into it,
938 			 * and copy the rest of the fid from the new
939 			 * allocation.
940 			 */
941 			total_fid_size = fid_size + fid_buf->l_iu +
942 			    fid_buf->l_fi;
943 			MALLOC(fid, struct fileid_desc *, total_fid_size,
944 			    M_UDFFID, M_NOWAIT | M_ZERO);
945 			if (fid == NULL) {
946 				if (bp != NULL)
947 					brelse(bp);
948 				return (ENOMEM);
949 			}
950 			bcopy(fid_buf, fid, frag_size);
951 			bcopy(data, &((uint8_t*)(fid))[frag_size],
952 			    total_fid_size - frag_size);
953 
954 			off = (total_fid_size - frag_size + 3) & ~0x03;
955 			fid_fragment = 1;
956 			FREE(buf, M_UDFFID);
957 		} else {
958 			/*
959 			 * Update the offset here to avoid looking at this fid
960 			 * again on a subsequent lookup.
961 			 */
962 			total_fid_size = fid->l_iu + fid->l_fi + fid_size;
963 			off += (total_fid_size + 3) & ~0x03;
964 		}
965 
966 		/* XXX Should we return an error on a bad fid? */
967 		if (udf_checktag(&fid->tag, TAGID_FID))
968 			break;
969 
970 		if ((fid->l_fi == 0) && (fid->file_char & 0x08)) {
971 			if (flags & ISDOTDOT) {
972 				id = udf_getid(&fid->icb);
973 				break;
974 			}
975 		} else {
976 			if (!(udf_cmpname(&fid->data[fid->l_iu],
977 			    nameptr, fid->l_fi, namelen))) {
978 				id = udf_getid(&fid->icb);
979 				break;
980 			}
981 		}
982 
983 		/*
984 		 * If we got this far then this fid isn't what we were
985 		 * looking for.  It's therefore safe to clean up from a
986 		 * fragmented fid.
987 		 */
988 		if (fid_fragment) {
989 			FREE(fid, M_UDFFID);
990 			fid_fragment = 0;
991 		}
992 	}
993 
994 	/* Did we have a match? */
995 	if (id) {
996 		error = udf_vget(udfmp->im_mountp, id, LK_EXCLUSIVE, &tdp);
997 		if (bp != NULL)
998 			brelse(bp);
999 		if (error)
1000 			return (error);
1001 
1002 		/* Remember where this entry was if it's the final component */
1003 		if ((flags & ISLASTCN) && nameiop == LOOKUP)
1004 			node->diroff = offset + off;
1005 		if (numdirpasses == 2)
1006 			nchstats.ncs_pass2++;
1007 		if (!(flags & LOCKPARENT) || !(flags & ISLASTCN)) {
1008 			a->a_cnp->cn_flags |= PDIRUNLOCK;
1009 			VOP_UNLOCK(dvp, 0, td);
1010 		}
1011 
1012 		*vpp = tdp;
1013 
1014 		/* Put this entry in the cache */
1015 		if (flags & MAKEENTRY)
1016 			cache_enter(dvp, *vpp, a->a_cnp);
1017 
1018 		if (fid_fragment)
1019 			FREE(fid, M_UDFFID);
1020 
1021 		return (0);
1022 	}
1023 
1024 	/* Name wasn't found on this pass.  Do another pass? */
1025 	if (numdirpasses == 2) {
1026 		numdirpasses--;
1027 		offset = 0;
1028 		goto lookloop;
1029 	}
1030 
1031 	if (bp != NULL)
1032 		brelse(bp);
1033 
1034 	/* Enter name into cache as non-existant */
1035 	if (flags & MAKEENTRY)
1036 		cache_enter(dvp, *vpp, a->a_cnp);
1037 
1038 	if ((flags & ISLASTCN) && (nameiop == CREATE || nameiop == RENAME))
1039 		return (EROFS);
1040 	return (ENOENT);
1041 
1042 }
1043 
1044 static int
1045 udf_reclaim(struct vop_reclaim_args *a)
1046 {
1047 	struct vnode *vp;
1048 	struct udf_node *unode;
1049 
1050 	vp = a->a_vp;
1051 	unode = VTON(vp);
1052 
1053 	cache_purge(vp);
1054 	if (unode != NULL) {
1055 		udf_hashrem(unode);
1056 		if (unode->i_devvp) {
1057 			vrele(unode->i_devvp);
1058 			unode->i_devvp = 0;
1059 		}
1060 
1061 		if (unode->fentry != NULL)
1062 			FREE(unode->fentry, M_UDFFENTRY);
1063 		lockdestroy(&unode->i_vnode->v_lock);
1064 		uma_zfree(udf_zone_node, unode);
1065 		vp->v_data = NULL;
1066 	}
1067 
1068 	return (0);
1069 }
1070 
1071 /*
1072  * Read the block and then set the data pointer to correspond with the
1073  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1074  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1075  * whole extent.
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, uint8_t **data)
1081 {
1082 	struct udf_mnt *udfmp;
1083 	struct file_entry *fentry = NULL;
1084 	struct buf *bp1;
1085 	uint32_t max_size;
1086 	daddr_t sector;
1087 	int error;
1088 
1089 	udfmp = node->udfmp;
1090 
1091 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1092 	if (error == -1) {
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 		*bp = NULL;
1101 		return (0);
1102 	} else if (error != 0) {
1103 		return (error);
1104 	}
1105 
1106 	/* Adjust the size so that it is within range */
1107 	if (*size == 0 || *size > max_size)
1108 		*size = max_size;
1109 	*size = min(*size, MAXBSIZE);
1110 
1111 	if ((error = udf_readlblks(udfmp, sector, *size, bp))) {
1112 		printf("udf_readlblks returned %d\n", error);
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 		printf("Cannot deal with strategy4096 yet!\n");
1147 		return (ENODEV);
1148 
1149 	default:
1150 		printf("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 				printf("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) - offset;
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 				printf("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) - offset;
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->bsize;
1210 
1211 		return (-1);
1212 	case 2:
1213 		/* DirectCD does not use extended_ad's */
1214 	default:
1215 		printf("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