xref: /freebsd/sys/fs/udf/udf_vnops.c (revision c96ae1968a6ab7056427a739bce81bf07447c2d4)
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/conf.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 #include <sys/endian.h>
48 
49 #include <vm/uma.h>
50 
51 #include <fs/udf/ecma167-udf.h>
52 #include <fs/udf/osta.h>
53 #include <fs/udf/udf.h>
54 #include <fs/udf/udf_mount.h>
55 
56 extern struct iconv_functions *udf_iconv;
57 
58 static vop_access_t	udf_access;
59 static vop_getattr_t	udf_getattr;
60 static vop_open_t	udf_open;
61 static vop_ioctl_t	udf_ioctl;
62 static vop_pathconf_t	udf_pathconf;
63 static vop_read_t	udf_read;
64 static vop_readdir_t	udf_readdir;
65 static vop_readlink_t	udf_readlink;
66 static vop_strategy_t	udf_strategy;
67 static vop_bmap_t	udf_bmap;
68 static vop_cachedlookup_t	udf_lookup;
69 static vop_reclaim_t	udf_reclaim;
70 static int udf_readatoffset(struct udf_node *node, int *size, off_t offset,
71     struct buf **bp, uint8_t **data);
72 static int udf_bmap_internal(struct udf_node *node, off_t offset,
73     daddr_t *sector, uint32_t *max_size);
74 
75 static struct vop_vector udf_vnodeops = {
76 	.vop_default =		&default_vnodeops,
77 
78 	.vop_access =		udf_access,
79 	.vop_bmap =		udf_bmap,
80 	.vop_cachedlookup =	udf_lookup,
81 	.vop_getattr =		udf_getattr,
82 	.vop_ioctl =		udf_ioctl,
83 	.vop_lookup =		vfs_cache_lookup,
84 	.vop_open =		udf_open,
85 	.vop_pathconf =		udf_pathconf,
86 	.vop_read =		udf_read,
87 	.vop_readdir =		udf_readdir,
88 	.vop_readlink =		udf_readlink,
89 	.vop_reclaim =		udf_reclaim,
90 	.vop_strategy =		udf_strategy,
91 };
92 
93 MALLOC_DEFINE(M_UDFFID, "udf_fid", "UDF FileId structure");
94 MALLOC_DEFINE(M_UDFDS, "udf_ds", "UDF Dirstream structure");
95 
96 #define UDF_INVALID_BMAP	-1
97 
98 int
99 udf_allocv(struct mount *mp, struct vnode **vpp, struct thread *td)
100 {
101 	int error;
102 	struct vnode *vp;
103 
104 	error = getnewvnode("udf", mp, &udf_vnodeops, &vp);
105 	if (error) {
106 		printf("udf_allocv: failed to allocate new vnode\n");
107 		return (error);
108 	}
109 
110 	*vpp = vp;
111 	return (0);
112 }
113 
114 /* Convert file entry permission (5 bits per owner/group/user) to a mode_t */
115 static mode_t
116 udf_permtomode(struct udf_node *node)
117 {
118 	uint32_t perm;
119 	uint16_t flags;
120 	mode_t mode;
121 
122 	perm = le32toh(node->fentry->perm);
123 	flags = le16toh(node->fentry->icbtag.flags);
124 
125 	mode = perm & UDF_FENTRY_PERM_USER_MASK;
126 	mode |= ((perm & UDF_FENTRY_PERM_GRP_MASK) >> 2);
127 	mode |= ((perm & UDF_FENTRY_PERM_OWNER_MASK) >> 4);
128 	mode |= ((flags & UDF_ICB_TAG_FLAGS_STICKY) << 4);
129 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETGID) << 6);
130 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETUID) << 8);
131 
132 	return (mode);
133 }
134 
135 static int
136 udf_access(struct vop_access_args *a)
137 {
138 	struct vnode *vp;
139 	struct udf_node *node;
140 	mode_t a_mode, mode;
141 
142 	vp = a->a_vp;
143 	node = VTON(vp);
144 	a_mode = a->a_mode;
145 
146 	if (a_mode & VWRITE) {
147 		switch (vp->v_type) {
148 		case VDIR:
149 		case VLNK:
150 		case VREG:
151 			return (EROFS);
152 			/* NOT REACHED */
153 		default:
154 			break;
155 		}
156 	}
157 
158 	mode = udf_permtomode(node);
159 
160 	return (vaccess(vp->v_type, mode, node->fentry->uid, node->fentry->gid,
161 	    a_mode, a->a_cred, NULL));
162 }
163 
164 static int
165 udf_open(struct vop_open_args *ap) {
166 	struct udf_node *np = VTON(ap->a_vp);
167 	off_t fsize;
168 
169 	fsize = le64toh(np->fentry->inf_len);
170 	vnode_create_vobject(ap->a_vp, fsize, ap->a_td);
171 	return 0;
172 }
173 
174 static int mon_lens[2][12] = {
175 	{31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31},
176 	{31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31}
177 };
178 
179 static int
180 udf_isaleapyear(int year)
181 {
182 	int i;
183 
184 	i = (year % 4) ? 0 : 1;
185 	i &= (year % 100) ? 1 : 0;
186 	i |= (year % 400) ? 0 : 1;
187 
188 	return i;
189 }
190 
191 /*
192  * XXX This is just a rough hack.  Daylight savings isn't calculated and tv_nsec
193  * is ignored.
194  * Timezone calculation compliments of Julian Elischer <julian@elischer.org>.
195  */
196 static void
197 udf_timetotimespec(struct timestamp *time, struct timespec *t)
198 {
199 	int i, lpyear, daysinyear, year;
200 	union {
201 		uint16_t	u_tz_offset;
202 		int16_t		s_tz_offset;
203 	} tz;
204 
205 	t->tv_nsec = 0;
206 
207 	/* DirectCD seems to like using bogus year values */
208 	year = le16toh(time->year);
209 	if (year < 1970) {
210 		t->tv_sec = 0;
211 		return;
212 	}
213 
214 	/* Calculate the time and day */
215 	t->tv_sec = time->second;
216 	t->tv_sec += time->minute * 60;
217 	t->tv_sec += time->hour * 3600;
218 	t->tv_sec += time->day * 3600 * 24;
219 
220 	/* Calculate the month */
221 	lpyear = udf_isaleapyear(year);
222 	for (i = 1; i < time->month; i++)
223 		t->tv_sec += mon_lens[lpyear][i] * 3600 * 24;
224 
225 	/* Speed up the calculation */
226 	if (year > 1979)
227 		t->tv_sec += 315532800;
228 	if (year > 1989)
229 		t->tv_sec += 315619200;
230 	if (year > 1999)
231 		t->tv_sec += 315532800;
232 	for (i = 2000; i < year; i++) {
233 		daysinyear = udf_isaleapyear(i) + 365 ;
234 		t->tv_sec += daysinyear * 3600 * 24;
235 	}
236 
237 	/*
238 	 * Calculate the time zone.  The timezone is 12 bit signed 2's
239 	 * complement, so we gotta do some extra magic to handle it right.
240 	 */
241 	tz.u_tz_offset = le16toh(time->type_tz);
242 	tz.u_tz_offset &= 0x0fff;
243 	if (tz.u_tz_offset & 0x0800)
244 		tz.u_tz_offset |= 0xf000;	/* extend the sign to 16 bits */
245 	if ((time->type_tz & 0x1000) && (tz.s_tz_offset != -2047))
246 		t->tv_sec -= tz.s_tz_offset * 60;
247 
248 	return;
249 }
250 
251 static int
252 udf_getattr(struct vop_getattr_args *a)
253 {
254 	struct vnode *vp;
255 	struct udf_node *node;
256 	struct vattr *vap;
257 	struct file_entry *fentry;
258 	struct timespec ts;
259 
260 	ts.tv_sec = 0;
261 
262 	vp = a->a_vp;
263 	vap = a->a_vap;
264 	node = VTON(vp);
265 	fentry = node->fentry;
266 
267 	vap->va_fsid = dev2udev(node->udfmp->im_dev);
268 	vap->va_fileid = node->hash_id;
269 	vap->va_mode = udf_permtomode(node);
270 	vap->va_nlink = le16toh(fentry->link_cnt);
271 	/*
272 	 * XXX The spec says that -1 is valid for uid/gid and indicates an
273 	 * invalid uid/gid.  How should this be represented?
274 	 */
275 	vap->va_uid = (le32toh(fentry->uid) == -1) ? 0 : le32toh(fentry->uid);
276 	vap->va_gid = (le32toh(fentry->gid) == -1) ? 0 : le32toh(fentry->gid);
277 	udf_timetotimespec(&fentry->atime, &vap->va_atime);
278 	udf_timetotimespec(&fentry->mtime, &vap->va_mtime);
279 	vap->va_ctime = vap->va_mtime; /* XXX Stored as an Extended Attribute */
280 	vap->va_rdev = 0; /* XXX */
281 	if (vp->v_type & VDIR) {
282 		/*
283 		 * Directories that are recorded within their ICB will show
284 		 * as having 0 blocks recorded.  Since tradition dictates
285 		 * that directories consume at least one logical block,
286 		 * make it appear so.
287 		 */
288 		if (fentry->logblks_rec != 0) {
289 			vap->va_size =
290 			    le64toh(fentry->logblks_rec) * node->udfmp->bsize;
291 		} else {
292 			vap->va_size = node->udfmp->bsize;
293 		}
294 	} else {
295 		vap->va_size = le64toh(fentry->inf_len);
296 	}
297 	vap->va_flags = 0;
298 	vap->va_gen = 1;
299 	vap->va_blocksize = node->udfmp->bsize;
300 	vap->va_bytes = le64toh(fentry->inf_len);
301 	vap->va_type = vp->v_type;
302 	vap->va_filerev = 0; /* XXX */
303 	return (0);
304 }
305 
306 /*
307  * File specific ioctls.
308  */
309 static int
310 udf_ioctl(struct vop_ioctl_args *a)
311 {
312 	printf("%s called\n", __func__);
313 	return (ENOTTY);
314 }
315 
316 /*
317  * I'm not sure that this has much value in a read-only filesystem, but
318  * cd9660 has it too.
319  */
320 static int
321 udf_pathconf(struct vop_pathconf_args *a)
322 {
323 
324 	switch (a->a_name) {
325 	case _PC_LINK_MAX:
326 		*a->a_retval = 65535;
327 		return (0);
328 	case _PC_NAME_MAX:
329 		*a->a_retval = NAME_MAX;
330 		return (0);
331 	case _PC_PATH_MAX:
332 		*a->a_retval = PATH_MAX;
333 		return (0);
334 	case _PC_NO_TRUNC:
335 		*a->a_retval = 1;
336 		return (0);
337 	default:
338 		return (EINVAL);
339 	}
340 }
341 
342 #define lblkno(udfmp, loc)	((loc) >> (udfmp)->bshift)
343 #define blkoff(udfmp, loc)	((loc) & (udfmp)->bmask)
344 #define lblktosize(imp, blk)	((blk) << (udfmp)->bshift)
345 
346 static int
347 udf_read(struct vop_read_args *ap)
348 {
349 	struct vnode *vp = ap->a_vp;
350 	struct uio *uio = ap->a_uio;
351 	struct udf_node *node = VTON(vp);
352 	struct udf_mnt *udfmp;
353 	struct buf *bp;
354 	daddr_t lbn, rablock;
355 	off_t diff, fsize;
356 	int error = 0;
357 	long size, n, on;
358 
359 	if (uio->uio_resid == 0)
360 		return (0);
361 	if (uio->uio_offset < 0)
362 		return (EINVAL);
363 	fsize = le64toh(node->fentry->inf_len);
364 	udfmp = node->udfmp;
365 	do {
366 		lbn = lblkno(udfmp, uio->uio_offset);
367 		on = blkoff(udfmp, uio->uio_offset);
368 		n = min((u_int)(udfmp->bsize - on),
369 			uio->uio_resid);
370 		diff = fsize - uio->uio_offset;
371 		if (diff <= 0)
372 			return (0);
373 		if (diff < n)
374 			n = diff;
375 		size = udfmp->bsize;
376 		rablock = lbn + 1;
377 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
378 			if (lblktosize(udfmp, rablock) < fsize) {
379 				error = cluster_read(vp, fsize, lbn, size, NOCRED,
380 					uio->uio_resid, (ap->a_ioflag >> 16), &bp);
381 			} else {
382 				error = bread(vp, lbn, size, NOCRED, &bp);
383 			}
384 		} else {
385 			error = bread(vp, lbn, size, NOCRED, &bp);
386 		}
387 		n = min(n, size - bp->b_resid);
388 		if (error) {
389 			brelse(bp);
390 			return (error);
391 		}
392 
393 		error = uiomove(bp->b_data + on, (int)n, uio);
394 		brelse(bp);
395 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
396 	return (error);
397 }
398 
399 /*
400  * Call the OSTA routines to translate the name from a CS0 dstring to a
401  * 16-bit Unicode String.  Hooks need to be placed in here to translate from
402  * Unicode to the encoding that the kernel/user expects.  Return the length
403  * of the translated string.
404  */
405 static int
406 udf_transname(char *cs0string, char *destname, int len, struct udf_mnt *udfmp)
407 {
408 	unicode_t *transname;
409 	char *unibuf, *unip;
410 	int i, destlen;
411 	ssize_t unilen = 0;
412 	size_t destleft = MAXNAMLEN;
413 
414 	/* Convert 16-bit Unicode to destname */
415 	if (udfmp->im_flags & UDFMNT_KICONV && udf_iconv) {
416 		/* allocate a buffer big enough to hold an 8->16 bit expansion */
417 		unibuf = uma_zalloc(udf_zone_trans, M_WAITOK);
418 		unip = unibuf;
419 		if ((unilen = (ssize_t)udf_UncompressUnicodeByte(len, cs0string, unibuf)) == -1) {
420 			printf("udf: Unicode translation failed\n");
421 			uma_zfree(udf_zone_trans, unibuf);
422 			return 0;
423 		}
424 
425 		while (unilen > 0 && destleft > 0) {
426 			udf_iconv->conv(udfmp->im_d2l, (const char **)&unibuf,
427 				(size_t *)&unilen, (char **)&destname, &destleft);
428 			/* Unconverted character found */
429 			if (unilen > 0 && destleft > 0) {
430 				*destname++ = '?';
431 				destleft--;
432 				unibuf += 2;
433 				unilen -= 2;
434 			}
435 		}
436 		uma_zfree(udf_zone_trans, unip);
437 		*destname = '\0';
438 		destlen = MAXNAMLEN - (int)destleft;
439 	} else {
440 		/* allocate a buffer big enough to hold an 8->16 bit expansion */
441 		transname = uma_zalloc(udf_zone_trans, M_WAITOK);
442 
443 		if ((unilen = (ssize_t)udf_UncompressUnicode(len, cs0string, transname)) == -1) {
444 			printf("udf: Unicode translation failed\n");
445 			uma_zfree(udf_zone_trans, transname);
446 			return 0;
447 		}
448 
449 		for (i = 0; i < unilen ; i++) {
450 			if (transname[i] & 0xff00) {
451 				destname[i] = '.';	/* Fudge the 16bit chars */
452 			} else {
453 				destname[i] = transname[i] & 0xff;
454 			}
455 		}
456 		uma_zfree(udf_zone_trans, transname);
457 		destname[unilen] = 0;
458 		destlen = (int)unilen;
459 	}
460 
461 	return (destlen);
462 }
463 
464 /*
465  * Compare a CS0 dstring with a name passed in from the VFS layer.  Return
466  * 0 on a successful match, nonzero otherwise.  Unicode work may need to be done
467  * here also.
468  */
469 static int
470 udf_cmpname(char *cs0string, char *cmpname, int cs0len, int cmplen, struct udf_mnt *udfmp)
471 {
472 	char *transname;
473 	int error = 0;
474 
475 	/* This is overkill, but not worth creating a new zone */
476 	transname = uma_zalloc(udf_zone_trans, M_WAITOK);
477 
478 	cs0len = udf_transname(cs0string, transname, cs0len, udfmp);
479 
480 	/* Easy check.  If they aren't the same length, they aren't equal */
481 	if ((cs0len == 0) || (cs0len != cmplen))
482 		error = -1;
483 	else
484 		error = bcmp(transname, cmpname, cmplen);
485 
486 	uma_zfree(udf_zone_trans, transname);
487 	return (error);
488 }
489 
490 struct udf_uiodir {
491 	struct dirent *dirent;
492 	u_long *cookies;
493 	int ncookies;
494 	int acookies;
495 	int eofflag;
496 };
497 
498 static int
499 udf_uiodir(struct udf_uiodir *uiodir, int de_size, struct uio *uio, long cookie)
500 {
501 	if (uiodir->cookies != NULL) {
502 		if (++uiodir->acookies > uiodir->ncookies) {
503 			uiodir->eofflag = 0;
504 			return (-1);
505 		}
506 		*uiodir->cookies++ = cookie;
507 	}
508 
509 	if (uio->uio_resid < de_size) {
510 		uiodir->eofflag = 0;
511 		return (-1);
512 	}
513 
514 	return (uiomove(uiodir->dirent, de_size, uio));
515 }
516 
517 static struct udf_dirstream *
518 udf_opendir(struct udf_node *node, int offset, int fsize, struct udf_mnt *udfmp)
519 {
520 	struct udf_dirstream *ds;
521 
522 	ds = uma_zalloc(udf_zone_ds, M_WAITOK | M_ZERO);
523 
524 	ds->node = node;
525 	ds->offset = offset;
526 	ds->udfmp = udfmp;
527 	ds->fsize = fsize;
528 
529 	return (ds);
530 }
531 
532 static struct fileid_desc *
533 udf_getfid(struct udf_dirstream *ds)
534 {
535 	struct fileid_desc *fid;
536 	int error, frag_size = 0, total_fid_size;
537 
538 	/* End of directory? */
539 	if (ds->offset + ds->off >= ds->fsize) {
540 		ds->error = 0;
541 		return (NULL);
542 	}
543 
544 	/* Grab the first extent of the directory */
545 	if (ds->off == 0) {
546 		ds->size = 0;
547 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
548 		    &ds->bp, &ds->data);
549 		if (error) {
550 			ds->error = error;
551 			if (ds->bp != NULL)
552 				brelse(ds->bp);
553 			return (NULL);
554 		}
555 	}
556 
557 	/*
558 	 * Clean up from a previous fragmented FID.
559 	 * XXX Is this the right place for this?
560 	 */
561 	if (ds->fid_fragment && ds->buf != NULL) {
562 		ds->fid_fragment = 0;
563 		FREE(ds->buf, M_UDFFID);
564 	}
565 
566 	fid = (struct fileid_desc*)&ds->data[ds->off];
567 
568 	/*
569 	 * Check to see if the fid is fragmented. The first test
570 	 * ensures that we don't wander off the end of the buffer
571 	 * looking for the l_iu and l_fi fields.
572 	 */
573 	if (ds->off + UDF_FID_SIZE > ds->size ||
574 	    ds->off + le16toh(fid->l_iu) + fid->l_fi + UDF_FID_SIZE > ds->size){
575 
576 		/* Copy what we have of the fid into a buffer */
577 		frag_size = ds->size - ds->off;
578 		if (frag_size >= ds->udfmp->bsize) {
579 			printf("udf: invalid FID fragment\n");
580 			ds->error = EINVAL;
581 			return (NULL);
582 		}
583 
584 		/*
585 		 * File ID descriptors can only be at most one
586 		 * logical sector in size.
587 		 */
588 		MALLOC(ds->buf, uint8_t*, ds->udfmp->bsize, M_UDFFID,
589 		     M_WAITOK | M_ZERO);
590 		bcopy(fid, ds->buf, frag_size);
591 
592 		/* Reduce all of the casting magic */
593 		fid = (struct fileid_desc*)ds->buf;
594 
595 		if (ds->bp != NULL)
596 			brelse(ds->bp);
597 
598 		/* Fetch the next allocation */
599 		ds->offset += ds->size;
600 		ds->size = 0;
601 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
602 		    &ds->bp, &ds->data);
603 		if (error) {
604 			ds->error = error;
605 			return (NULL);
606 		}
607 
608 		/*
609 		 * If the fragment was so small that we didn't get
610 		 * the l_iu and l_fi fields, copy those in.
611 		 */
612 		if (frag_size < UDF_FID_SIZE)
613 			bcopy(ds->data, &ds->buf[frag_size],
614 			    UDF_FID_SIZE - frag_size);
615 
616 		/*
617 		 * Now that we have enough of the fid to work with,
618 		 * copy in the rest of the fid from the new
619 		 * allocation.
620 		 */
621 		total_fid_size = UDF_FID_SIZE + le16toh(fid->l_iu) + fid->l_fi;
622 		if (total_fid_size > ds->udfmp->bsize) {
623 			printf("udf: invalid FID\n");
624 			ds->error = EIO;
625 			return (NULL);
626 		}
627 		bcopy(ds->data, &ds->buf[frag_size],
628 		    total_fid_size - frag_size);
629 
630 		ds->fid_fragment = 1;
631 	} else {
632 		total_fid_size = le16toh(fid->l_iu) + fid->l_fi + UDF_FID_SIZE;
633 	}
634 
635 	/*
636 	 * Update the offset. Align on a 4 byte boundary because the
637 	 * UDF spec says so.
638 	 */
639 	ds->this_off = ds->off;
640 	if (!ds->fid_fragment) {
641 		ds->off += (total_fid_size + 3) & ~0x03;
642 	} else {
643 		ds->off = (total_fid_size - frag_size + 3) & ~0x03;
644 	}
645 
646 	return (fid);
647 }
648 
649 static void
650 udf_closedir(struct udf_dirstream *ds)
651 {
652 
653 	if (ds->bp != NULL)
654 		brelse(ds->bp);
655 
656 	if (ds->fid_fragment && ds->buf != NULL)
657 		FREE(ds->buf, M_UDFFID);
658 
659 	uma_zfree(udf_zone_ds, ds);
660 }
661 
662 static int
663 udf_readdir(struct vop_readdir_args *a)
664 {
665 	struct vnode *vp;
666 	struct uio *uio;
667 	struct dirent dir;
668 	struct udf_node *node;
669 	struct udf_mnt *udfmp;
670 	struct fileid_desc *fid;
671 	struct udf_uiodir uiodir;
672 	struct udf_dirstream *ds;
673 	u_long *cookies = NULL;
674 	int ncookies;
675 	int error = 0;
676 
677 	vp = a->a_vp;
678 	uio = a->a_uio;
679 	node = VTON(vp);
680 	udfmp = node->udfmp;
681 	uiodir.eofflag = 1;
682 
683 	if (a->a_ncookies != NULL) {
684 		/*
685 		 * Guess how many entries are needed.  If we run out, this
686 		 * function will be called again and thing will pick up were
687 		 * it left off.
688 		 */
689 		ncookies = uio->uio_resid / 8;
690 		MALLOC(cookies, u_long *, sizeof(u_long) * ncookies,
691 		    M_TEMP, M_WAITOK);
692 		if (cookies == NULL)
693 			return (ENOMEM);
694 		uiodir.ncookies = ncookies;
695 		uiodir.cookies = cookies;
696 		uiodir.acookies = 0;
697 	} else {
698 		uiodir.cookies = NULL;
699 	}
700 
701 	/*
702 	 * Iterate through the file id descriptors.  Give the parent dir
703 	 * entry special attention.
704 	 */
705 	ds = udf_opendir(node, uio->uio_offset, le64toh(node->fentry->inf_len),
706 	    node->udfmp);
707 
708 	while ((fid = udf_getfid(ds)) != NULL) {
709 
710 		/* XXX Should we return an error on a bad fid? */
711 		if (udf_checktag(&fid->tag, TAGID_FID)) {
712 			printf("Invalid FID tag\n");
713 			hexdump(fid, UDF_FID_SIZE, NULL, 0);
714 			error = EIO;
715 			break;
716 		}
717 
718 		/* Is this a deleted file? */
719 		if (fid->file_char & UDF_FILE_CHAR_DEL)
720 			continue;
721 
722 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
723 			/* Do up the '.' and '..' entries.  Dummy values are
724 			 * used for the cookies since the offset here is
725 			 * usually zero, and NFS doesn't like that value
726 			 */
727 			dir.d_fileno = node->hash_id;
728 			dir.d_type = DT_DIR;
729 			dir.d_name[0] = '.';
730 			dir.d_name[1] = '\0';
731 			dir.d_namlen = 1;
732 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
733 			uiodir.dirent = &dir;
734 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, 1);
735 			if (error)
736 				break;
737 
738 			dir.d_fileno = udf_getid(&fid->icb);
739 			dir.d_type = DT_DIR;
740 			dir.d_name[0] = '.';
741 			dir.d_name[1] = '.';
742 			dir.d_name[2] = '\0';
743 			dir.d_namlen = 2;
744 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
745 			uiodir.dirent = &dir;
746 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, 2);
747 		} else {
748 			dir.d_namlen = udf_transname(&fid->data[fid->l_iu],
749 			    &dir.d_name[0], fid->l_fi, udfmp);
750 			dir.d_fileno = udf_getid(&fid->icb);
751 			dir.d_type = (fid->file_char & UDF_FILE_CHAR_DIR) ?
752 			    DT_DIR : DT_UNKNOWN;
753 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
754 			uiodir.dirent = &dir;
755 			error = udf_uiodir(&uiodir, dir.d_reclen, uio,
756 			    ds->this_off);
757 		}
758 		if (error) {
759 			printf("uiomove returned %d\n", error);
760 			break;
761 		}
762 
763 	}
764 
765 	/* tell the calling layer whether we need to be called again */
766 	*a->a_eofflag = uiodir.eofflag;
767 	uio->uio_offset = ds->offset + ds->off;
768 
769 	if (!error)
770 		error = ds->error;
771 
772 	udf_closedir(ds);
773 
774 	if (a->a_ncookies != NULL) {
775 		if (error)
776 			FREE(cookies, M_TEMP);
777 		else {
778 			*a->a_ncookies = uiodir.acookies;
779 			*a->a_cookies = cookies;
780 		}
781 	}
782 
783 	return (error);
784 }
785 
786 /* Are there any implementations out there that do soft-links? */
787 static int
788 udf_readlink(struct vop_readlink_args *ap)
789 {
790 	printf("%s called\n", __func__);
791 	return (EOPNOTSUPP);
792 }
793 
794 static int
795 udf_strategy(struct vop_strategy_args *a)
796 {
797 	struct buf *bp;
798 	struct vnode *vp;
799 	struct udf_node *node;
800 	int maxsize;
801 	daddr_t sector;
802 	struct bufobj *bo;
803 	int multiplier;
804 
805 	bp = a->a_bp;
806 	vp = a->a_vp;
807 	node = VTON(vp);
808 
809 	if (bp->b_blkno == bp->b_lblkno) {
810 		/*
811 		 * Files that are embedded in the fentry don't translate well
812 		 * to a block number.  Reject.
813 		 */
814 		if (udf_bmap_internal(node, bp->b_lblkno * node->udfmp->bsize,
815 		    &sector, &maxsize)) {
816 			clrbuf(bp);
817 			bp->b_blkno = -1;
818 		}
819 
820 		/* bmap gives sector numbers, bio works with device blocks */
821 		multiplier = node->udfmp->bsize / DEV_BSIZE;
822 		bp->b_blkno = sector * multiplier;
823 
824 	}
825 	if ((long)bp->b_blkno == -1) {
826 		bufdone(bp);
827 		return (0);
828 	}
829 	bo = node->udfmp->im_bo;
830 	bp->b_iooffset = dbtob(bp->b_blkno);
831 	BO_STRATEGY(bo, bp);
832 	return (0);
833 }
834 
835 static int
836 udf_bmap(struct vop_bmap_args *a)
837 {
838 	struct udf_node *node;
839 	uint32_t max_size;
840 	daddr_t lsector;
841 	int error;
842 
843 	node = VTON(a->a_vp);
844 
845 	if (a->a_bop != NULL)
846 		*a->a_bop = &node->udfmp->im_devvp->v_bufobj;
847 	if (a->a_bnp == NULL)
848 		return (0);
849 	if (a->a_runb)
850 		*a->a_runb = 0;
851 
852 	error = udf_bmap_internal(node, a->a_bn * node->udfmp->bsize, &lsector,
853 	    &max_size);
854 	if (error)
855 		return (error);
856 
857 	/* Translate logical to physical sector number */
858 	*a->a_bnp = lsector << (node->udfmp->bshift - DEV_BSHIFT);
859 
860 	/* Punt on read-ahead for now */
861 	if (a->a_runp)
862 		*a->a_runp = 0;
863 
864 	return (0);
865 }
866 
867 /*
868  * The all powerful VOP_LOOKUP().
869  */
870 static int
871 udf_lookup(struct vop_cachedlookup_args *a)
872 {
873 	struct vnode *dvp;
874 	struct vnode *tdp = NULL;
875 	struct vnode **vpp = a->a_vpp;
876 	struct udf_node *node;
877 	struct udf_mnt *udfmp;
878 	struct fileid_desc *fid = NULL;
879 	struct udf_dirstream *ds;
880 	struct thread *td;
881 	u_long nameiop;
882 	u_long flags;
883 	char *nameptr;
884 	long namelen;
885 	ino_t id = 0;
886 	int offset, error = 0;
887 	int numdirpasses, fsize;
888 
889 	dvp = a->a_dvp;
890 	node = VTON(dvp);
891 	udfmp = node->udfmp;
892 	nameiop = a->a_cnp->cn_nameiop;
893 	flags = a->a_cnp->cn_flags;
894 	nameptr = a->a_cnp->cn_nameptr;
895 	namelen = a->a_cnp->cn_namelen;
896 	fsize = le64toh(node->fentry->inf_len);
897 	td = a->a_cnp->cn_thread;
898 
899 	/*
900 	 * If this is a LOOKUP and we've already partially searched through
901 	 * the directory, pick up where we left off and flag that the
902 	 * directory may need to be searched twice.  For a full description,
903 	 * see /sys/isofs/cd9660/cd9660_lookup.c:cd9660_lookup()
904 	 */
905 	if (nameiop != LOOKUP || node->diroff == 0 || node->diroff > fsize) {
906 		offset = 0;
907 		numdirpasses = 1;
908 	} else {
909 		offset = node->diroff;
910 		numdirpasses = 2;
911 		nchstats.ncs_2passes++;
912 	}
913 
914 lookloop:
915 	ds = udf_opendir(node, offset, fsize, udfmp);
916 
917 	while ((fid = udf_getfid(ds)) != NULL) {
918 
919 		/* XXX Should we return an error on a bad fid? */
920 		if (udf_checktag(&fid->tag, TAGID_FID)) {
921 			printf("udf_lookup: Invalid tag\n");
922 			error = EIO;
923 			break;
924 		}
925 
926 		/* Is this a deleted file? */
927 		if (fid->file_char & UDF_FILE_CHAR_DEL)
928 			continue;
929 
930 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
931 			if (flags & ISDOTDOT) {
932 				id = udf_getid(&fid->icb);
933 				break;
934 			}
935 		} else {
936 			if (!(udf_cmpname(&fid->data[fid->l_iu],
937 			    nameptr, fid->l_fi, namelen, udfmp))) {
938 				id = udf_getid(&fid->icb);
939 				break;
940 			}
941 		}
942 	}
943 
944 	if (!error)
945 		error = ds->error;
946 
947 	/* XXX Bail out here? */
948 	if (error) {
949 		udf_closedir(ds);
950 		return (error);
951 	}
952 
953 	/* Did we have a match? */
954 	if (id) {
955 		if (flags & ISDOTDOT)
956 			VOP_UNLOCK(dvp, 0, a->a_cnp->cn_thread);
957 		error = udf_vget(udfmp->im_mountp, id, LK_EXCLUSIVE, &tdp);
958 		if (flags & ISDOTDOT)
959 			vn_lock(dvp, LK_EXCLUSIVE|LK_RETRY, a->a_cnp->cn_thread);
960 		if (!error) {
961 			/*
962 			 * Remember where this entry was if it's the final
963 			 * component.
964 			 */
965 			if ((flags & ISLASTCN) && nameiop == LOOKUP)
966 				node->diroff = ds->offset + ds->off;
967 			if (numdirpasses == 2)
968 				nchstats.ncs_pass2++;
969 			*vpp = tdp;
970 			/* Put this entry in the cache */
971 			if (flags & MAKEENTRY)
972 				cache_enter(dvp, *vpp, a->a_cnp);
973 		}
974 	} else {
975 		/* Name wasn't found on this pass.  Do another pass? */
976 		if (numdirpasses == 2) {
977 			numdirpasses--;
978 			offset = 0;
979 			udf_closedir(ds);
980 			goto lookloop;
981 		}
982 
983 		/* Enter name into cache as non-existant */
984 		if (flags & MAKEENTRY)
985 			cache_enter(dvp, *vpp, a->a_cnp);
986 
987 		if ((flags & ISLASTCN) &&
988 		    (nameiop == CREATE || nameiop == RENAME)) {
989 			error = EROFS;
990 		} else {
991 			error = ENOENT;
992 		}
993 	}
994 
995 	udf_closedir(ds);
996 	return (error);
997 }
998 
999 static int
1000 udf_reclaim(struct vop_reclaim_args *a)
1001 {
1002 	struct vnode *vp;
1003 	struct udf_node *unode;
1004 
1005 	vp = a->a_vp;
1006 	unode = VTON(vp);
1007 
1008 	/*
1009 	 * Destroy the vm object and flush associated pages.
1010 	 */
1011 	vnode_destroy_vobject(vp);
1012 
1013 	if (unode != NULL) {
1014 		vfs_hash_remove(vp);
1015 
1016 		if (unode->fentry != NULL)
1017 			FREE(unode->fentry, M_UDFFENTRY);
1018 		uma_zfree(udf_zone_node, unode);
1019 		vp->v_data = NULL;
1020 	}
1021 
1022 	return (0);
1023 }
1024 
1025 /*
1026  * Read the block and then set the data pointer to correspond with the
1027  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1028  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1029  * whole extent.
1030  *
1031  * Note that *bp may be assigned error or not.
1032  *
1033  */
1034 static int
1035 udf_readatoffset(struct udf_node *node, int *size, off_t offset,
1036     struct buf **bp, uint8_t **data)
1037 {
1038 	struct udf_mnt *udfmp;
1039 	struct file_entry *fentry = NULL;
1040 	struct buf *bp1;
1041 	uint32_t max_size;
1042 	daddr_t sector;
1043 	int error;
1044 
1045 	udfmp = node->udfmp;
1046 
1047 	*bp = NULL;
1048 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1049 	if (error == UDF_INVALID_BMAP) {
1050 		/*
1051 		 * This error means that the file *data* is stored in the
1052 		 * allocation descriptor field of the file entry.
1053 		 */
1054 		fentry = node->fentry;
1055 		*data = &fentry->data[le32toh(fentry->l_ea)];
1056 		*size = le32toh(fentry->l_ad);
1057 		return (0);
1058 	} else if (error != 0) {
1059 		return (error);
1060 	}
1061 
1062 	/* Adjust the size so that it is within range */
1063 	if (*size == 0 || *size > max_size)
1064 		*size = max_size;
1065 	*size = min(*size, MAXBSIZE);
1066 
1067 	if ((error = udf_readlblks(udfmp, sector, *size, bp))) {
1068 		printf("warning: udf_readlblks returned error %d\n", error);
1069 		/* note: *bp may be non-NULL */
1070 		return (error);
1071 	}
1072 
1073 	bp1 = *bp;
1074 	*data = (uint8_t *)&bp1->b_data[offset % udfmp->bsize];
1075 	return (0);
1076 }
1077 
1078 /*
1079  * Translate a file offset into a logical block and then into a physical
1080  * block.
1081  */
1082 static int
1083 udf_bmap_internal(struct udf_node *node, off_t offset, daddr_t *sector,
1084     uint32_t *max_size)
1085 {
1086 	struct udf_mnt *udfmp;
1087 	struct file_entry *fentry;
1088 	void *icb;
1089 	struct icb_tag *tag;
1090 	uint32_t icblen = 0;
1091 	daddr_t lsector;
1092 	int ad_offset, ad_num = 0;
1093 	int i, p_offset;
1094 
1095 	udfmp = node->udfmp;
1096 	fentry = node->fentry;
1097 	tag = &fentry->icbtag;
1098 
1099 	switch (le16toh(tag->strat_type)) {
1100 	case 4:
1101 		break;
1102 
1103 	case 4096:
1104 		printf("Cannot deal with strategy4096 yet!\n");
1105 		return (ENODEV);
1106 
1107 	default:
1108 		printf("Unknown strategy type %d\n", tag->strat_type);
1109 		return (ENODEV);
1110 	}
1111 
1112 	switch (le16toh(tag->flags) & 0x7) {
1113 	case 0:
1114 		/*
1115 		 * The allocation descriptor field is filled with short_ad's.
1116 		 * If the offset is beyond the current extent, look for the
1117 		 * next extent.
1118 		 */
1119 		do {
1120 			offset -= icblen;
1121 			ad_offset = sizeof(struct short_ad) * ad_num;
1122 			if (ad_offset > le32toh(fentry->l_ad)) {
1123 				printf("File offset out of bounds\n");
1124 				return (EINVAL);
1125 			}
1126 			icb = GETICB(short_ad, fentry,
1127 			    le32toh(fentry->l_ea) + ad_offset);
1128 			icblen = GETICBLEN(short_ad, icb);
1129 			ad_num++;
1130 		} while(offset >= icblen);
1131 
1132 		lsector = (offset  >> udfmp->bshift) +
1133 		    le32toh(((struct short_ad *)(icb))->pos);
1134 
1135 		*max_size = GETICBLEN(short_ad, icb);
1136 
1137 		break;
1138 	case 1:
1139 		/*
1140 		 * The allocation descriptor field is filled with long_ad's
1141 		 * If the offset is beyond the current extent, look for the
1142 		 * next extent.
1143 		 */
1144 		do {
1145 			offset -= icblen;
1146 			ad_offset = sizeof(struct long_ad) * ad_num;
1147 			if (ad_offset > le32toh(fentry->l_ad)) {
1148 				printf("File offset out of bounds\n");
1149 				return (EINVAL);
1150 			}
1151 			icb = GETICB(long_ad, fentry,
1152 			    le32toh(fentry->l_ea) + ad_offset);
1153 			icblen = GETICBLEN(long_ad, icb);
1154 			ad_num++;
1155 		} while(offset >= icblen);
1156 
1157 		lsector = (offset >> udfmp->bshift) +
1158 		    le32toh(((struct long_ad *)(icb))->loc.lb_num);
1159 
1160 		*max_size = GETICBLEN(long_ad, icb);
1161 
1162 		break;
1163 	case 3:
1164 		/*
1165 		 * This type means that the file *data* is stored in the
1166 		 * allocation descriptor field of the file entry.
1167 		 */
1168 		*max_size = 0;
1169 		*sector = node->hash_id + udfmp->part_start;
1170 
1171 		return (UDF_INVALID_BMAP);
1172 	case 2:
1173 		/* DirectCD does not use extended_ad's */
1174 	default:
1175 		printf("Unsupported allocation descriptor %d\n",
1176 		       tag->flags & 0x7);
1177 		return (ENODEV);
1178 	}
1179 
1180 	*sector = lsector + udfmp->part_start;
1181 
1182 	/*
1183 	 * Check the sparing table.  Each entry represents the beginning of
1184 	 * a packet.
1185 	 */
1186 	if (udfmp->s_table != NULL) {
1187 		for (i = 0; i< udfmp->s_table_entries; i++) {
1188 			p_offset =
1189 			    lsector - le32toh(udfmp->s_table->entries[i].org);
1190 			if ((p_offset < udfmp->p_sectors) && (p_offset >= 0)) {
1191 				*sector =
1192 				   le32toh(udfmp->s_table->entries[i].map) +
1193 				    p_offset;
1194 				break;
1195 			}
1196 		}
1197 	}
1198 
1199 	return (0);
1200 }
1201