xref: /freebsd/sys/fs/udf/udf_vnops.c (revision 9162f64b58d01ec01481d60b6cdc06ffd8e8c7fc)
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 vop_vptofh_t	udf_vptofh;
71 static int udf_readatoffset(struct udf_node *node, int *size, off_t offset,
72     struct buf **bp, uint8_t **data);
73 static int udf_bmap_internal(struct udf_node *node, off_t offset,
74     daddr_t *sector, uint32_t *max_size);
75 
76 static struct vop_vector udf_vnodeops = {
77 	.vop_default =		&default_vnodeops,
78 
79 	.vop_access =		udf_access,
80 	.vop_bmap =		udf_bmap,
81 	.vop_cachedlookup =	udf_lookup,
82 	.vop_getattr =		udf_getattr,
83 	.vop_ioctl =		udf_ioctl,
84 	.vop_lookup =		vfs_cache_lookup,
85 	.vop_open =		udf_open,
86 	.vop_pathconf =		udf_pathconf,
87 	.vop_read =		udf_read,
88 	.vop_readdir =		udf_readdir,
89 	.vop_readlink =		udf_readlink,
90 	.vop_reclaim =		udf_reclaim,
91 	.vop_strategy =		udf_strategy,
92 	.vop_vptofh =		udf_vptofh,
93 };
94 
95 MALLOC_DEFINE(M_UDFFID, "udf_fid", "UDF FileId structure");
96 MALLOC_DEFINE(M_UDFDS, "udf_ds", "UDF Dirstream structure");
97 
98 #define UDF_INVALID_BMAP	-1
99 
100 int
101 udf_allocv(struct mount *mp, struct vnode **vpp, struct thread *td)
102 {
103 	int error;
104 	struct vnode *vp;
105 
106 	error = getnewvnode("udf", mp, &udf_vnodeops, &vp);
107 	if (error) {
108 		printf("udf_allocv: failed to allocate new vnode\n");
109 		return (error);
110 	}
111 
112 	*vpp = vp;
113 	return (0);
114 }
115 
116 /* Convert file entry permission (5 bits per owner/group/user) to a mode_t */
117 static mode_t
118 udf_permtomode(struct udf_node *node)
119 {
120 	uint32_t perm;
121 	uint16_t flags;
122 	mode_t mode;
123 
124 	perm = le32toh(node->fentry->perm);
125 	flags = le16toh(node->fentry->icbtag.flags);
126 
127 	mode = perm & UDF_FENTRY_PERM_USER_MASK;
128 	mode |= ((perm & UDF_FENTRY_PERM_GRP_MASK) >> 2);
129 	mode |= ((perm & UDF_FENTRY_PERM_OWNER_MASK) >> 4);
130 	mode |= ((flags & UDF_ICB_TAG_FLAGS_STICKY) << 4);
131 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETGID) << 6);
132 	mode |= ((flags & UDF_ICB_TAG_FLAGS_SETUID) << 8);
133 
134 	return (mode);
135 }
136 
137 static int
138 udf_access(struct vop_access_args *a)
139 {
140 	struct vnode *vp;
141 	struct udf_node *node;
142 	accmode_t accmode;
143 	mode_t mode;
144 
145 	vp = a->a_vp;
146 	node = VTON(vp);
147 	accmode = a->a_accmode;
148 
149 	if (accmode & VWRITE) {
150 		switch (vp->v_type) {
151 		case VDIR:
152 		case VLNK:
153 		case VREG:
154 			return (EROFS);
155 			/* NOT REACHED */
156 		default:
157 			break;
158 		}
159 	}
160 
161 	mode = udf_permtomode(node);
162 
163 	return (vaccess(vp->v_type, mode, node->fentry->uid, node->fentry->gid,
164 	    accmode, a->a_cred, NULL));
165 }
166 
167 static int
168 udf_open(struct vop_open_args *ap) {
169 	struct udf_node *np = VTON(ap->a_vp);
170 	off_t fsize;
171 
172 	fsize = le64toh(np->fentry->inf_len);
173 	vnode_create_vobject(ap->a_vp, fsize, ap->a_td);
174 	return 0;
175 }
176 
177 static const int mon_lens[2][12] = {
178 	{0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334},
179 	{0, 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335}
180 };
181 
182 static int
183 udf_isaleapyear(int year)
184 {
185 	int i;
186 
187 	i = (year % 4) ? 0 : 1;
188 	i &= (year % 100) ? 1 : 0;
189 	i |= (year % 400) ? 0 : 1;
190 
191 	return i;
192 }
193 
194 /*
195  * Timezone calculation compliments of Julian Elischer <julian@elischer.org>.
196  */
197 static void
198 udf_timetotimespec(struct timestamp *time, struct timespec *t)
199 {
200 	int i, lpyear, daysinyear, year, startyear;
201 	union {
202 		uint16_t	u_tz_offset;
203 		int16_t		s_tz_offset;
204 	} tz;
205 
206 	/*
207 	 * DirectCD seems to like using bogus year values.
208 	 * Don't trust time->month as it will be used for an array index.
209 	 */
210 	year = le16toh(time->year);
211 	if (year < 1970 || time->month < 1 || time->month > 12) {
212 		t->tv_sec = 0;
213 		t->tv_nsec = 0;
214 		return;
215 	}
216 
217 	/* Calculate the time and day */
218 	t->tv_sec = time->second;
219 	t->tv_sec += time->minute * 60;
220 	t->tv_sec += time->hour * 3600;
221 	t->tv_sec += (time->day - 1) * 3600 * 24;
222 
223 	/* Calculate the month */
224 	lpyear = udf_isaleapyear(year);
225 	t->tv_sec += mon_lens[lpyear][time->month - 1] * 3600 * 24;
226 
227 	/* Speed up the calculation */
228 	startyear = 1970;
229 	if (year > 2009) {
230 		t->tv_sec += 1262304000;
231 		startyear += 40;
232 	} else if (year > 1999) {
233 		t->tv_sec += 946684800;
234 		startyear += 30;
235 	} else if (year > 1989) {
236 		t->tv_sec += 631152000;
237 		startyear += 20;
238 	} else if (year > 1979) {
239 		t->tv_sec += 315532800;
240 		startyear += 10;
241 	}
242 
243 	daysinyear = (year - startyear) * 365;
244 	for (i = startyear; i < year; i++)
245 		daysinyear += udf_isaleapyear(i);
246 	t->tv_sec += daysinyear * 3600 * 24;
247 
248 	/* Calculate microseconds */
249 	t->tv_nsec = time->centisec * 10000 + time->hund_usec * 100 +
250 	    time->usec;
251 
252 	/*
253 	 * Calculate the time zone.  The timezone is 12 bit signed 2's
254 	 * complement, so we gotta do some extra magic to handle it right.
255 	 */
256 	tz.u_tz_offset = le16toh(time->type_tz);
257 	tz.u_tz_offset &= 0x0fff;
258 	if (tz.u_tz_offset & 0x0800)
259 		tz.u_tz_offset |= 0xf000;	/* extend the sign to 16 bits */
260 	if ((le16toh(time->type_tz) & 0x1000) && (tz.s_tz_offset != -2047))
261 		t->tv_sec -= tz.s_tz_offset * 60;
262 
263 	return;
264 }
265 
266 static int
267 udf_getattr(struct vop_getattr_args *a)
268 {
269 	struct vnode *vp;
270 	struct udf_node *node;
271 	struct vattr *vap;
272 	struct file_entry *fentry;
273 	struct timespec ts;
274 
275 	ts.tv_sec = 0;
276 
277 	vp = a->a_vp;
278 	vap = a->a_vap;
279 	node = VTON(vp);
280 	fentry = node->fentry;
281 
282 	vap->va_fsid = dev2udev(node->udfmp->im_dev);
283 	vap->va_fileid = node->hash_id;
284 	vap->va_mode = udf_permtomode(node);
285 	vap->va_nlink = le16toh(fentry->link_cnt);
286 	/*
287 	 * XXX The spec says that -1 is valid for uid/gid and indicates an
288 	 * invalid uid/gid.  How should this be represented?
289 	 */
290 	vap->va_uid = (le32toh(fentry->uid) == -1) ? 0 : le32toh(fentry->uid);
291 	vap->va_gid = (le32toh(fentry->gid) == -1) ? 0 : le32toh(fentry->gid);
292 	udf_timetotimespec(&fentry->atime, &vap->va_atime);
293 	udf_timetotimespec(&fentry->mtime, &vap->va_mtime);
294 	vap->va_ctime = vap->va_mtime; /* XXX Stored as an Extended Attribute */
295 	vap->va_rdev = NODEV;
296 	if (vp->v_type & VDIR) {
297 		/*
298 		 * Directories that are recorded within their ICB will show
299 		 * as having 0 blocks recorded.  Since tradition dictates
300 		 * that directories consume at least one logical block,
301 		 * make it appear so.
302 		 */
303 		if (fentry->logblks_rec != 0) {
304 			vap->va_size =
305 			    le64toh(fentry->logblks_rec) * node->udfmp->bsize;
306 		} else {
307 			vap->va_size = node->udfmp->bsize;
308 		}
309 	} else {
310 		vap->va_size = le64toh(fentry->inf_len);
311 	}
312 	vap->va_flags = 0;
313 	vap->va_gen = 1;
314 	vap->va_blocksize = node->udfmp->bsize;
315 	vap->va_bytes = le64toh(fentry->inf_len);
316 	vap->va_type = vp->v_type;
317 	vap->va_filerev = 0; /* XXX */
318 	return (0);
319 }
320 
321 /*
322  * File specific ioctls.
323  */
324 static int
325 udf_ioctl(struct vop_ioctl_args *a)
326 {
327 	printf("%s called\n", __func__);
328 	return (ENOTTY);
329 }
330 
331 /*
332  * I'm not sure that this has much value in a read-only filesystem, but
333  * cd9660 has it too.
334  */
335 static int
336 udf_pathconf(struct vop_pathconf_args *a)
337 {
338 
339 	switch (a->a_name) {
340 	case _PC_LINK_MAX:
341 		*a->a_retval = 65535;
342 		return (0);
343 	case _PC_NAME_MAX:
344 		*a->a_retval = NAME_MAX;
345 		return (0);
346 	case _PC_PATH_MAX:
347 		*a->a_retval = PATH_MAX;
348 		return (0);
349 	case _PC_NO_TRUNC:
350 		*a->a_retval = 1;
351 		return (0);
352 	default:
353 		return (EINVAL);
354 	}
355 }
356 
357 #define lblkno(udfmp, loc)	((loc) >> (udfmp)->bshift)
358 #define blkoff(udfmp, loc)	((loc) & (udfmp)->bmask)
359 #define lblktosize(imp, blk)	((blk) << (udfmp)->bshift)
360 
361 static int
362 udf_read(struct vop_read_args *ap)
363 {
364 	struct vnode *vp = ap->a_vp;
365 	struct uio *uio = ap->a_uio;
366 	struct udf_node *node = VTON(vp);
367 	struct udf_mnt *udfmp;
368 	struct buf *bp;
369 	daddr_t lbn, rablock;
370 	off_t diff, fsize;
371 	int error = 0;
372 	long size, n, on;
373 
374 	if (uio->uio_resid == 0)
375 		return (0);
376 	if (uio->uio_offset < 0)
377 		return (EINVAL);
378 	fsize = le64toh(node->fentry->inf_len);
379 	udfmp = node->udfmp;
380 	do {
381 		lbn = lblkno(udfmp, uio->uio_offset);
382 		on = blkoff(udfmp, uio->uio_offset);
383 		n = min((u_int)(udfmp->bsize - on),
384 			uio->uio_resid);
385 		diff = fsize - uio->uio_offset;
386 		if (diff <= 0)
387 			return (0);
388 		if (diff < n)
389 			n = diff;
390 		size = udfmp->bsize;
391 		rablock = lbn + 1;
392 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
393 			if (lblktosize(udfmp, rablock) < fsize) {
394 				error = cluster_read(vp, fsize, lbn, size, NOCRED,
395 					uio->uio_resid, (ap->a_ioflag >> 16), &bp);
396 			} else {
397 				error = bread(vp, lbn, size, NOCRED, &bp);
398 			}
399 		} else {
400 			error = bread(vp, lbn, size, NOCRED, &bp);
401 		}
402 		n = min(n, size - bp->b_resid);
403 		if (error) {
404 			brelse(bp);
405 			return (error);
406 		}
407 
408 		error = uiomove(bp->b_data + on, (int)n, uio);
409 		brelse(bp);
410 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
411 	return (error);
412 }
413 
414 /*
415  * Call the OSTA routines to translate the name from a CS0 dstring to a
416  * 16-bit Unicode String.  Hooks need to be placed in here to translate from
417  * Unicode to the encoding that the kernel/user expects.  Return the length
418  * of the translated string.
419  */
420 static int
421 udf_transname(char *cs0string, char *destname, int len, struct udf_mnt *udfmp)
422 {
423 	unicode_t *transname;
424 	char *unibuf, *unip;
425 	int i, destlen;
426 	ssize_t unilen = 0;
427 	size_t destleft = MAXNAMLEN;
428 
429 	/* Convert 16-bit Unicode to destname */
430 	if (udfmp->im_flags & UDFMNT_KICONV && udf_iconv) {
431 		/* allocate a buffer big enough to hold an 8->16 bit expansion */
432 		unibuf = uma_zalloc(udf_zone_trans, M_WAITOK);
433 		unip = unibuf;
434 		if ((unilen = (ssize_t)udf_UncompressUnicodeByte(len, cs0string, unibuf)) == -1) {
435 			printf("udf: Unicode translation failed\n");
436 			uma_zfree(udf_zone_trans, unibuf);
437 			return 0;
438 		}
439 
440 		while (unilen > 0 && destleft > 0) {
441 			udf_iconv->conv(udfmp->im_d2l, (const char **)&unibuf,
442 				(size_t *)&unilen, (char **)&destname, &destleft);
443 			/* Unconverted character found */
444 			if (unilen > 0 && destleft > 0) {
445 				*destname++ = '?';
446 				destleft--;
447 				unibuf += 2;
448 				unilen -= 2;
449 			}
450 		}
451 		uma_zfree(udf_zone_trans, unip);
452 		*destname = '\0';
453 		destlen = MAXNAMLEN - (int)destleft;
454 	} else {
455 		/* allocate a buffer big enough to hold an 8->16 bit expansion */
456 		transname = uma_zalloc(udf_zone_trans, M_WAITOK);
457 
458 		if ((unilen = (ssize_t)udf_UncompressUnicode(len, cs0string, transname)) == -1) {
459 			printf("udf: Unicode translation failed\n");
460 			uma_zfree(udf_zone_trans, transname);
461 			return 0;
462 		}
463 
464 		for (i = 0; i < unilen ; i++) {
465 			if (transname[i] & 0xff00) {
466 				destname[i] = '.';	/* Fudge the 16bit chars */
467 			} else {
468 				destname[i] = transname[i] & 0xff;
469 			}
470 		}
471 		uma_zfree(udf_zone_trans, transname);
472 		destname[unilen] = 0;
473 		destlen = (int)unilen;
474 	}
475 
476 	return (destlen);
477 }
478 
479 /*
480  * Compare a CS0 dstring with a name passed in from the VFS layer.  Return
481  * 0 on a successful match, nonzero otherwise.  Unicode work may need to be done
482  * here also.
483  */
484 static int
485 udf_cmpname(char *cs0string, char *cmpname, int cs0len, int cmplen, struct udf_mnt *udfmp)
486 {
487 	char *transname;
488 	int error = 0;
489 
490 	/* This is overkill, but not worth creating a new zone */
491 	transname = uma_zalloc(udf_zone_trans, M_WAITOK);
492 
493 	cs0len = udf_transname(cs0string, transname, cs0len, udfmp);
494 
495 	/* Easy check.  If they aren't the same length, they aren't equal */
496 	if ((cs0len == 0) || (cs0len != cmplen))
497 		error = -1;
498 	else
499 		error = bcmp(transname, cmpname, cmplen);
500 
501 	uma_zfree(udf_zone_trans, transname);
502 	return (error);
503 }
504 
505 struct udf_uiodir {
506 	struct dirent *dirent;
507 	u_long *cookies;
508 	int ncookies;
509 	int acookies;
510 	int eofflag;
511 };
512 
513 static int
514 udf_uiodir(struct udf_uiodir *uiodir, int de_size, struct uio *uio, long cookie)
515 {
516 	if (uiodir->cookies != NULL) {
517 		if (++uiodir->acookies > uiodir->ncookies) {
518 			uiodir->eofflag = 0;
519 			return (-1);
520 		}
521 		*uiodir->cookies++ = cookie;
522 	}
523 
524 	if (uio->uio_resid < de_size) {
525 		uiodir->eofflag = 0;
526 		return (-1);
527 	}
528 
529 	return (uiomove(uiodir->dirent, de_size, uio));
530 }
531 
532 static struct udf_dirstream *
533 udf_opendir(struct udf_node *node, int offset, int fsize, struct udf_mnt *udfmp)
534 {
535 	struct udf_dirstream *ds;
536 
537 	ds = uma_zalloc(udf_zone_ds, M_WAITOK | M_ZERO);
538 
539 	ds->node = node;
540 	ds->offset = offset;
541 	ds->udfmp = udfmp;
542 	ds->fsize = fsize;
543 
544 	return (ds);
545 }
546 
547 static struct fileid_desc *
548 udf_getfid(struct udf_dirstream *ds)
549 {
550 	struct fileid_desc *fid;
551 	int error, frag_size = 0, total_fid_size;
552 
553 	/* End of directory? */
554 	if (ds->offset + ds->off >= ds->fsize) {
555 		ds->error = 0;
556 		return (NULL);
557 	}
558 
559 	/* Grab the first extent of the directory */
560 	if (ds->off == 0) {
561 		ds->size = 0;
562 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
563 		    &ds->bp, &ds->data);
564 		if (error) {
565 			ds->error = error;
566 			if (ds->bp != NULL)
567 				brelse(ds->bp);
568 			return (NULL);
569 		}
570 	}
571 
572 	/*
573 	 * Clean up from a previous fragmented FID.
574 	 * XXX Is this the right place for this?
575 	 */
576 	if (ds->fid_fragment && ds->buf != NULL) {
577 		ds->fid_fragment = 0;
578 		free(ds->buf, M_UDFFID);
579 	}
580 
581 	fid = (struct fileid_desc*)&ds->data[ds->off];
582 
583 	/*
584 	 * Check to see if the fid is fragmented. The first test
585 	 * ensures that we don't wander off the end of the buffer
586 	 * looking for the l_iu and l_fi fields.
587 	 */
588 	if (ds->off + UDF_FID_SIZE > ds->size ||
589 	    ds->off + le16toh(fid->l_iu) + fid->l_fi + UDF_FID_SIZE > ds->size){
590 
591 		/* Copy what we have of the fid into a buffer */
592 		frag_size = ds->size - ds->off;
593 		if (frag_size >= ds->udfmp->bsize) {
594 			printf("udf: invalid FID fragment\n");
595 			ds->error = EINVAL;
596 			return (NULL);
597 		}
598 
599 		/*
600 		 * File ID descriptors can only be at most one
601 		 * logical sector in size.
602 		 */
603 		ds->buf = malloc(ds->udfmp->bsize, M_UDFFID,
604 		     M_WAITOK | M_ZERO);
605 		bcopy(fid, ds->buf, frag_size);
606 
607 		/* Reduce all of the casting magic */
608 		fid = (struct fileid_desc*)ds->buf;
609 
610 		if (ds->bp != NULL)
611 			brelse(ds->bp);
612 
613 		/* Fetch the next allocation */
614 		ds->offset += ds->size;
615 		ds->size = 0;
616 		error = udf_readatoffset(ds->node, &ds->size, ds->offset,
617 		    &ds->bp, &ds->data);
618 		if (error) {
619 			ds->error = error;
620 			return (NULL);
621 		}
622 
623 		/*
624 		 * If the fragment was so small that we didn't get
625 		 * the l_iu and l_fi fields, copy those in.
626 		 */
627 		if (frag_size < UDF_FID_SIZE)
628 			bcopy(ds->data, &ds->buf[frag_size],
629 			    UDF_FID_SIZE - frag_size);
630 
631 		/*
632 		 * Now that we have enough of the fid to work with,
633 		 * copy in the rest of the fid from the new
634 		 * allocation.
635 		 */
636 		total_fid_size = UDF_FID_SIZE + le16toh(fid->l_iu) + fid->l_fi;
637 		if (total_fid_size > ds->udfmp->bsize) {
638 			printf("udf: invalid FID\n");
639 			ds->error = EIO;
640 			return (NULL);
641 		}
642 		bcopy(ds->data, &ds->buf[frag_size],
643 		    total_fid_size - frag_size);
644 
645 		ds->fid_fragment = 1;
646 	} else {
647 		total_fid_size = le16toh(fid->l_iu) + fid->l_fi + UDF_FID_SIZE;
648 	}
649 
650 	/*
651 	 * Update the offset. Align on a 4 byte boundary because the
652 	 * UDF spec says so.
653 	 */
654 	ds->this_off = ds->off;
655 	if (!ds->fid_fragment) {
656 		ds->off += (total_fid_size + 3) & ~0x03;
657 	} else {
658 		ds->off = (total_fid_size - frag_size + 3) & ~0x03;
659 	}
660 
661 	return (fid);
662 }
663 
664 static void
665 udf_closedir(struct udf_dirstream *ds)
666 {
667 
668 	if (ds->bp != NULL)
669 		brelse(ds->bp);
670 
671 	if (ds->fid_fragment && ds->buf != NULL)
672 		free(ds->buf, M_UDFFID);
673 
674 	uma_zfree(udf_zone_ds, ds);
675 }
676 
677 static int
678 udf_readdir(struct vop_readdir_args *a)
679 {
680 	struct vnode *vp;
681 	struct uio *uio;
682 	struct dirent dir;
683 	struct udf_node *node;
684 	struct udf_mnt *udfmp;
685 	struct fileid_desc *fid;
686 	struct udf_uiodir uiodir;
687 	struct udf_dirstream *ds;
688 	u_long *cookies = NULL;
689 	int ncookies;
690 	int error = 0;
691 
692 	vp = a->a_vp;
693 	uio = a->a_uio;
694 	node = VTON(vp);
695 	udfmp = node->udfmp;
696 	uiodir.eofflag = 1;
697 
698 	if (a->a_ncookies != NULL) {
699 		/*
700 		 * Guess how many entries are needed.  If we run out, this
701 		 * function will be called again and thing will pick up were
702 		 * it left off.
703 		 */
704 		ncookies = uio->uio_resid / 8;
705 		cookies = malloc(sizeof(u_long) * ncookies,
706 		    M_TEMP, M_WAITOK);
707 		if (cookies == NULL)
708 			return (ENOMEM);
709 		uiodir.ncookies = ncookies;
710 		uiodir.cookies = cookies;
711 		uiodir.acookies = 0;
712 	} else {
713 		uiodir.cookies = NULL;
714 	}
715 
716 	/*
717 	 * Iterate through the file id descriptors.  Give the parent dir
718 	 * entry special attention.
719 	 */
720 	ds = udf_opendir(node, uio->uio_offset, le64toh(node->fentry->inf_len),
721 	    node->udfmp);
722 
723 	while ((fid = udf_getfid(ds)) != NULL) {
724 
725 		/* XXX Should we return an error on a bad fid? */
726 		if (udf_checktag(&fid->tag, TAGID_FID)) {
727 			printf("Invalid FID tag\n");
728 			hexdump(fid, UDF_FID_SIZE, NULL, 0);
729 			error = EIO;
730 			break;
731 		}
732 
733 		/* Is this a deleted file? */
734 		if (fid->file_char & UDF_FILE_CHAR_DEL)
735 			continue;
736 
737 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
738 			/* Do up the '.' and '..' entries.  Dummy values are
739 			 * used for the cookies since the offset here is
740 			 * usually zero, and NFS doesn't like that value
741 			 */
742 			dir.d_fileno = node->hash_id;
743 			dir.d_type = DT_DIR;
744 			dir.d_name[0] = '.';
745 			dir.d_name[1] = '\0';
746 			dir.d_namlen = 1;
747 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
748 			uiodir.dirent = &dir;
749 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, 1);
750 			if (error)
751 				break;
752 
753 			dir.d_fileno = udf_getid(&fid->icb);
754 			dir.d_type = DT_DIR;
755 			dir.d_name[0] = '.';
756 			dir.d_name[1] = '.';
757 			dir.d_name[2] = '\0';
758 			dir.d_namlen = 2;
759 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
760 			uiodir.dirent = &dir;
761 			error = udf_uiodir(&uiodir, dir.d_reclen, uio, 2);
762 		} else {
763 			dir.d_namlen = udf_transname(&fid->data[fid->l_iu],
764 			    &dir.d_name[0], fid->l_fi, udfmp);
765 			dir.d_fileno = udf_getid(&fid->icb);
766 			dir.d_type = (fid->file_char & UDF_FILE_CHAR_DIR) ?
767 			    DT_DIR : DT_UNKNOWN;
768 			dir.d_reclen = GENERIC_DIRSIZ(&dir);
769 			uiodir.dirent = &dir;
770 			error = udf_uiodir(&uiodir, dir.d_reclen, uio,
771 			    ds->this_off);
772 		}
773 		if (error) {
774 			printf("uiomove returned %d\n", error);
775 			break;
776 		}
777 
778 	}
779 
780 	/* tell the calling layer whether we need to be called again */
781 	*a->a_eofflag = uiodir.eofflag;
782 	uio->uio_offset = ds->offset + ds->off;
783 
784 	if (!error)
785 		error = ds->error;
786 
787 	udf_closedir(ds);
788 
789 	if (a->a_ncookies != NULL) {
790 		if (error)
791 			free(cookies, M_TEMP);
792 		else {
793 			*a->a_ncookies = uiodir.acookies;
794 			*a->a_cookies = cookies;
795 		}
796 	}
797 
798 	return (error);
799 }
800 
801 /* Are there any implementations out there that do soft-links? */
802 static int
803 udf_readlink(struct vop_readlink_args *ap)
804 {
805 	printf("%s called\n", __func__);
806 	return (EOPNOTSUPP);
807 }
808 
809 static int
810 udf_strategy(struct vop_strategy_args *a)
811 {
812 	struct buf *bp;
813 	struct vnode *vp;
814 	struct udf_node *node;
815 	int maxsize;
816 	daddr_t sector;
817 	struct bufobj *bo;
818 	int multiplier;
819 
820 	bp = a->a_bp;
821 	vp = a->a_vp;
822 	node = VTON(vp);
823 
824 	if (bp->b_blkno == bp->b_lblkno) {
825 		/*
826 		 * Files that are embedded in the fentry don't translate well
827 		 * to a block number.  Reject.
828 		 */
829 		if (udf_bmap_internal(node, bp->b_lblkno * node->udfmp->bsize,
830 		    &sector, &maxsize)) {
831 			clrbuf(bp);
832 			bp->b_blkno = -1;
833 		}
834 
835 		/* bmap gives sector numbers, bio works with device blocks */
836 		multiplier = node->udfmp->bsize / DEV_BSIZE;
837 		bp->b_blkno = sector * multiplier;
838 
839 	}
840 	if ((long)bp->b_blkno == -1) {
841 		bufdone(bp);
842 		return (0);
843 	}
844 	bo = node->udfmp->im_bo;
845 	bp->b_iooffset = dbtob(bp->b_blkno);
846 	BO_STRATEGY(bo, bp);
847 	return (0);
848 }
849 
850 static int
851 udf_bmap(struct vop_bmap_args *a)
852 {
853 	struct udf_node *node;
854 	uint32_t max_size;
855 	daddr_t lsector;
856 	int error;
857 
858 	node = VTON(a->a_vp);
859 
860 	if (a->a_bop != NULL)
861 		*a->a_bop = &node->udfmp->im_devvp->v_bufobj;
862 	if (a->a_bnp == NULL)
863 		return (0);
864 	if (a->a_runb)
865 		*a->a_runb = 0;
866 
867 	error = udf_bmap_internal(node, a->a_bn * node->udfmp->bsize, &lsector,
868 	    &max_size);
869 	if (error)
870 		return (error);
871 
872 	/* Translate logical to physical sector number */
873 	*a->a_bnp = lsector << (node->udfmp->bshift - DEV_BSHIFT);
874 
875 	/* Punt on read-ahead for now */
876 	if (a->a_runp)
877 		*a->a_runp = 0;
878 
879 	return (0);
880 }
881 
882 /*
883  * The all powerful VOP_LOOKUP().
884  */
885 static int
886 udf_lookup(struct vop_cachedlookup_args *a)
887 {
888 	struct vnode *dvp;
889 	struct vnode *tdp = NULL;
890 	struct vnode **vpp = a->a_vpp;
891 	struct udf_node *node;
892 	struct udf_mnt *udfmp;
893 	struct fileid_desc *fid = NULL;
894 	struct udf_dirstream *ds;
895 	struct thread *td;
896 	u_long nameiop;
897 	u_long flags;
898 	char *nameptr;
899 	long namelen;
900 	ino_t id = 0;
901 	int offset, error = 0;
902 	int numdirpasses, fsize;
903 
904 	dvp = a->a_dvp;
905 	node = VTON(dvp);
906 	udfmp = node->udfmp;
907 	nameiop = a->a_cnp->cn_nameiop;
908 	flags = a->a_cnp->cn_flags;
909 	nameptr = a->a_cnp->cn_nameptr;
910 	namelen = a->a_cnp->cn_namelen;
911 	fsize = le64toh(node->fentry->inf_len);
912 	td = a->a_cnp->cn_thread;
913 
914 	/*
915 	 * If this is a LOOKUP and we've already partially searched through
916 	 * the directory, pick up where we left off and flag that the
917 	 * directory may need to be searched twice.  For a full description,
918 	 * see /sys/fs/cd9660/cd9660_lookup.c:cd9660_lookup()
919 	 */
920 	if (nameiop != LOOKUP || node->diroff == 0 || node->diroff > fsize) {
921 		offset = 0;
922 		numdirpasses = 1;
923 	} else {
924 		offset = node->diroff;
925 		numdirpasses = 2;
926 		nchstats.ncs_2passes++;
927 	}
928 
929 lookloop:
930 	ds = udf_opendir(node, offset, fsize, udfmp);
931 
932 	while ((fid = udf_getfid(ds)) != NULL) {
933 
934 		/* XXX Should we return an error on a bad fid? */
935 		if (udf_checktag(&fid->tag, TAGID_FID)) {
936 			printf("udf_lookup: Invalid tag\n");
937 			error = EIO;
938 			break;
939 		}
940 
941 		/* Is this a deleted file? */
942 		if (fid->file_char & UDF_FILE_CHAR_DEL)
943 			continue;
944 
945 		if ((fid->l_fi == 0) && (fid->file_char & UDF_FILE_CHAR_PAR)) {
946 			if (flags & ISDOTDOT) {
947 				id = udf_getid(&fid->icb);
948 				break;
949 			}
950 		} else {
951 			if (!(udf_cmpname(&fid->data[fid->l_iu],
952 			    nameptr, fid->l_fi, namelen, udfmp))) {
953 				id = udf_getid(&fid->icb);
954 				break;
955 			}
956 		}
957 	}
958 
959 	if (!error)
960 		error = ds->error;
961 
962 	/* XXX Bail out here? */
963 	if (error) {
964 		udf_closedir(ds);
965 		return (error);
966 	}
967 
968 	/* Did we have a match? */
969 	if (id) {
970 		if (flags & ISDOTDOT)
971 			VOP_UNLOCK(dvp, 0);
972 		error = udf_vget(udfmp->im_mountp, id, LK_EXCLUSIVE, &tdp);
973 		if (flags & ISDOTDOT)
974 			vn_lock(dvp, LK_EXCLUSIVE|LK_RETRY);
975 		if (!error) {
976 			/*
977 			 * Remember where this entry was if it's the final
978 			 * component.
979 			 */
980 			if ((flags & ISLASTCN) && nameiop == LOOKUP)
981 				node->diroff = ds->offset + ds->off;
982 			if (numdirpasses == 2)
983 				nchstats.ncs_pass2++;
984 			*vpp = tdp;
985 			/* Put this entry in the cache */
986 			if (flags & MAKEENTRY)
987 				cache_enter(dvp, *vpp, a->a_cnp);
988 		}
989 	} else {
990 		/* Name wasn't found on this pass.  Do another pass? */
991 		if (numdirpasses == 2) {
992 			numdirpasses--;
993 			offset = 0;
994 			udf_closedir(ds);
995 			goto lookloop;
996 		}
997 
998 		/* Enter name into cache as non-existant */
999 		if (flags & MAKEENTRY)
1000 			cache_enter(dvp, *vpp, a->a_cnp);
1001 
1002 		if ((flags & ISLASTCN) &&
1003 		    (nameiop == CREATE || nameiop == RENAME)) {
1004 			error = EROFS;
1005 		} else {
1006 			error = ENOENT;
1007 		}
1008 	}
1009 
1010 	udf_closedir(ds);
1011 	return (error);
1012 }
1013 
1014 static int
1015 udf_reclaim(struct vop_reclaim_args *a)
1016 {
1017 	struct vnode *vp;
1018 	struct udf_node *unode;
1019 
1020 	vp = a->a_vp;
1021 	unode = VTON(vp);
1022 
1023 	/*
1024 	 * Destroy the vm object and flush associated pages.
1025 	 */
1026 	vnode_destroy_vobject(vp);
1027 
1028 	if (unode != NULL) {
1029 		vfs_hash_remove(vp);
1030 
1031 		if (unode->fentry != NULL)
1032 			free(unode->fentry, M_UDFFENTRY);
1033 		uma_zfree(udf_zone_node, unode);
1034 		vp->v_data = NULL;
1035 	}
1036 
1037 	return (0);
1038 }
1039 
1040 static int
1041 udf_vptofh(struct vop_vptofh_args *a)
1042 {
1043 	struct udf_node *node;
1044 	struct ifid *ifhp;
1045 
1046 	node = VTON(a->a_vp);
1047 	ifhp = (struct ifid *)a->a_fhp;
1048 	ifhp->ifid_len = sizeof(struct ifid);
1049 	ifhp->ifid_ino = node->hash_id;
1050 
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Read the block and then set the data pointer to correspond with the
1056  * offset passed in.  Only read in at most 'size' bytes, and then set 'size'
1057  * to the number of bytes pointed to.  If 'size' is zero, try to read in a
1058  * whole extent.
1059  *
1060  * Note that *bp may be assigned error or not.
1061  *
1062  */
1063 static int
1064 udf_readatoffset(struct udf_node *node, int *size, off_t offset,
1065     struct buf **bp, uint8_t **data)
1066 {
1067 	struct udf_mnt *udfmp;
1068 	struct file_entry *fentry = NULL;
1069 	struct buf *bp1;
1070 	uint32_t max_size;
1071 	daddr_t sector;
1072 	int error;
1073 
1074 	udfmp = node->udfmp;
1075 
1076 	*bp = NULL;
1077 	error = udf_bmap_internal(node, offset, &sector, &max_size);
1078 	if (error == UDF_INVALID_BMAP) {
1079 		/*
1080 		 * This error means that the file *data* is stored in the
1081 		 * allocation descriptor field of the file entry.
1082 		 */
1083 		fentry = node->fentry;
1084 		*data = &fentry->data[le32toh(fentry->l_ea)];
1085 		*size = le32toh(fentry->l_ad);
1086 		return (0);
1087 	} else if (error != 0) {
1088 		return (error);
1089 	}
1090 
1091 	/* Adjust the size so that it is within range */
1092 	if (*size == 0 || *size > max_size)
1093 		*size = max_size;
1094 	*size = min(*size, MAXBSIZE);
1095 
1096 	if ((error = udf_readlblks(udfmp, sector, *size + (offset & udfmp->bmask), bp))) {
1097 		printf("warning: udf_readlblks returned error %d\n", error);
1098 		/* note: *bp may be non-NULL */
1099 		return (error);
1100 	}
1101 
1102 	bp1 = *bp;
1103 	*data = (uint8_t *)&bp1->b_data[offset & udfmp->bmask];
1104 	return (0);
1105 }
1106 
1107 /*
1108  * Translate a file offset into a logical block and then into a physical
1109  * block.
1110  * max_size - maximum number of bytes that can be read starting from given
1111  * offset, rather than beginning of calculated sector number
1112  */
1113 static int
1114 udf_bmap_internal(struct udf_node *node, off_t offset, daddr_t *sector,
1115     uint32_t *max_size)
1116 {
1117 	struct udf_mnt *udfmp;
1118 	struct file_entry *fentry;
1119 	void *icb;
1120 	struct icb_tag *tag;
1121 	uint32_t icblen = 0;
1122 	daddr_t lsector;
1123 	int ad_offset, ad_num = 0;
1124 	int i, p_offset;
1125 
1126 	udfmp = node->udfmp;
1127 	fentry = node->fentry;
1128 	tag = &fentry->icbtag;
1129 
1130 	switch (le16toh(tag->strat_type)) {
1131 	case 4:
1132 		break;
1133 
1134 	case 4096:
1135 		printf("Cannot deal with strategy4096 yet!\n");
1136 		return (ENODEV);
1137 
1138 	default:
1139 		printf("Unknown strategy type %d\n", tag->strat_type);
1140 		return (ENODEV);
1141 	}
1142 
1143 	switch (le16toh(tag->flags) & 0x7) {
1144 	case 0:
1145 		/*
1146 		 * The allocation descriptor field is filled with short_ad's.
1147 		 * If the offset is beyond the current extent, look for the
1148 		 * next extent.
1149 		 */
1150 		do {
1151 			offset -= icblen;
1152 			ad_offset = sizeof(struct short_ad) * ad_num;
1153 			if (ad_offset > le32toh(fentry->l_ad)) {
1154 				printf("File offset out of bounds\n");
1155 				return (EINVAL);
1156 			}
1157 			icb = GETICB(short_ad, fentry,
1158 			    le32toh(fentry->l_ea) + ad_offset);
1159 			icblen = GETICBLEN(short_ad, icb);
1160 			ad_num++;
1161 		} while(offset >= icblen);
1162 
1163 		lsector = (offset  >> udfmp->bshift) +
1164 		    le32toh(((struct short_ad *)(icb))->pos);
1165 
1166 		*max_size = icblen - offset;
1167 
1168 		break;
1169 	case 1:
1170 		/*
1171 		 * The allocation descriptor field is filled with long_ad's
1172 		 * If the offset is beyond the current extent, look for the
1173 		 * next extent.
1174 		 */
1175 		do {
1176 			offset -= icblen;
1177 			ad_offset = sizeof(struct long_ad) * ad_num;
1178 			if (ad_offset > le32toh(fentry->l_ad)) {
1179 				printf("File offset out of bounds\n");
1180 				return (EINVAL);
1181 			}
1182 			icb = GETICB(long_ad, fentry,
1183 			    le32toh(fentry->l_ea) + ad_offset);
1184 			icblen = GETICBLEN(long_ad, icb);
1185 			ad_num++;
1186 		} while(offset >= icblen);
1187 
1188 		lsector = (offset >> udfmp->bshift) +
1189 		    le32toh(((struct long_ad *)(icb))->loc.lb_num);
1190 
1191 		*max_size = icblen - offset;
1192 
1193 		break;
1194 	case 3:
1195 		/*
1196 		 * This type means that the file *data* is stored in the
1197 		 * allocation descriptor field of the file entry.
1198 		 */
1199 		*max_size = 0;
1200 		*sector = node->hash_id + udfmp->part_start;
1201 
1202 		return (UDF_INVALID_BMAP);
1203 	case 2:
1204 		/* DirectCD does not use extended_ad's */
1205 	default:
1206 		printf("Unsupported allocation descriptor %d\n",
1207 		       tag->flags & 0x7);
1208 		return (ENODEV);
1209 	}
1210 
1211 	*sector = lsector + udfmp->part_start;
1212 
1213 	/*
1214 	 * Check the sparing table.  Each entry represents the beginning of
1215 	 * a packet.
1216 	 */
1217 	if (udfmp->s_table != NULL) {
1218 		for (i = 0; i< udfmp->s_table_entries; i++) {
1219 			p_offset =
1220 			    lsector - le32toh(udfmp->s_table->entries[i].org);
1221 			if ((p_offset < udfmp->p_sectors) && (p_offset >= 0)) {
1222 				*sector =
1223 				   le32toh(udfmp->s_table->entries[i].map) +
1224 				    p_offset;
1225 				break;
1226 			}
1227 		}
1228 	}
1229 
1230 	return (0);
1231 }
1232