xref: /netbsd-src/lib/libukfs/ukfs.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: ukfs.c,v 1.47 2009/12/13 20:52:36 pooka Exp $	*/
2 
3 /*
4  * Copyright (c) 2007, 2008, 2009  Antti Kantee.  All Rights Reserved.
5  *
6  * Development of this software was supported by the
7  * Finnish Cultural Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
19  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
20  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
21  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 /*
32  * This library enables access to files systems directly without
33  * involving system calls.
34  */
35 
36 #ifdef __linux__
37 #define _XOPEN_SOURCE 500
38 #define _BSD_SOURCE
39 #define _FILE_OFFSET_BITS 64
40 #endif
41 
42 #include <sys/param.h>
43 #include <sys/queue.h>
44 #include <sys/stat.h>
45 #include <sys/sysctl.h>
46 #include <sys/mount.h>
47 
48 #include <assert.h>
49 #include <dirent.h>
50 #include <dlfcn.h>
51 #include <err.h>
52 #include <errno.h>
53 #include <fcntl.h>
54 #include <pthread.h>
55 #include <stdio.h>
56 #include <stdlib.h>
57 #include <string.h>
58 #include <unistd.h>
59 #include <stdint.h>
60 
61 #include <rump/ukfs.h>
62 
63 #include <rump/rump.h>
64 #include <rump/rump_syscalls.h>
65 #include <rump/rumpuser.h>
66 
67 #include "ukfs_int_disklabel.h"
68 
69 #define UKFS_MODE_DEFAULT 0555
70 
71 struct ukfs {
72 	struct mount *ukfs_mp;
73 	struct vnode *ukfs_rvp;
74 	void *ukfs_specific;
75 
76 	pthread_spinlock_t ukfs_spin;
77 	pid_t ukfs_nextpid;
78 	struct vnode *ukfs_cdir;
79 	int ukfs_devfd;
80 	char *ukfs_devpath;
81 	char *ukfs_mountpath;
82 	struct ukfs_part *ukfs_part;
83 };
84 
85 static int builddirs(const char *, mode_t,
86     int (*mkdirfn)(struct ukfs *, const char *, mode_t), struct ukfs *);
87 
88 struct mount *
89 ukfs_getmp(struct ukfs *ukfs)
90 {
91 
92 	return ukfs->ukfs_mp;
93 }
94 
95 struct vnode *
96 ukfs_getrvp(struct ukfs *ukfs)
97 {
98 	struct vnode *rvp;
99 
100 	rvp = ukfs->ukfs_rvp;
101 	rump_pub_vp_incref(rvp);
102 
103 	return rvp;
104 }
105 
106 void
107 ukfs_setspecific(struct ukfs *ukfs, void *priv)
108 {
109 
110 	ukfs->ukfs_specific = priv;
111 }
112 
113 void *
114 ukfs_getspecific(struct ukfs *ukfs)
115 {
116 
117 	return ukfs->ukfs_specific;
118 }
119 
120 #ifdef DONT_WANT_PTHREAD_LINKAGE
121 #define pthread_spin_lock(a)
122 #define pthread_spin_unlock(a)
123 #define pthread_spin_init(a,b)
124 #define pthread_spin_destroy(a)
125 #endif
126 
127 static pid_t
128 nextpid(struct ukfs *ukfs)
129 {
130 	pid_t npid;
131 
132 	pthread_spin_lock(&ukfs->ukfs_spin);
133 	if (ukfs->ukfs_nextpid == 0)
134 		ukfs->ukfs_nextpid++;
135 	npid = ukfs->ukfs_nextpid++;
136 	pthread_spin_unlock(&ukfs->ukfs_spin);
137 
138 	return npid;
139 }
140 
141 static void
142 precall(struct ukfs *ukfs)
143 {
144 	struct vnode *rvp, *cvp;
145 
146 	rump_pub_lwp_alloc_and_switch(nextpid(ukfs), 1);
147 	rvp = ukfs_getrvp(ukfs);
148 	pthread_spin_lock(&ukfs->ukfs_spin);
149 	cvp = ukfs->ukfs_cdir;
150 	pthread_spin_unlock(&ukfs->ukfs_spin);
151 	rump_pub_rcvp_set(rvp, cvp); /* takes refs */
152 	rump_pub_vp_rele(rvp);
153 }
154 
155 static void
156 postcall(struct ukfs *ukfs)
157 {
158 	struct vnode *rvp;
159 
160 	rvp = ukfs_getrvp(ukfs);
161 	rump_pub_rcvp_set(NULL, rvp);
162 	rump_pub_vp_rele(rvp);
163 	rump_pub_lwp_release(rump_pub_lwp_curlwp());
164 }
165 
166 struct ukfs_part {
167 	pthread_spinlock_t part_lck;
168 	int part_refcount;
169 
170 	int part_type;
171 	char part_labelchar;
172 	off_t part_devoff;
173 	off_t part_devsize;
174 };
175 
176 enum ukfs_parttype { UKFS_PART_NONE, UKFS_PART_DISKLABEL, UKFS_PART_OFFSET };
177 
178 static struct ukfs_part ukfs__part_none = {
179 	.part_type = UKFS_PART_NONE,
180 	.part_devoff = 0,
181 	.part_devsize = RUMP_ETFS_SIZE_ENDOFF,
182 };
183 static struct ukfs_part ukfs__part_na;
184 struct ukfs_part *ukfs_part_none = &ukfs__part_none;
185 struct ukfs_part *ukfs_part_na = &ukfs__part_na;
186 
187 #define PART2LOCKSIZE(len) ((len) == RUMP_ETFS_SIZE_ENDOFF ? 0 : (len))
188 
189 int
190 _ukfs_init(int version)
191 {
192 	int rv;
193 
194 	if (version != UKFS_VERSION) {
195 		printf("incompatible ukfs version, %d vs. %d\n",
196 		    version, UKFS_VERSION);
197 		errno = EPROGMISMATCH;
198 		return -1;
199 	}
200 
201 	if ((rv = rump_init()) != 0) {
202 		errno = rv;
203 		return -1;
204 	}
205 
206 	return 0;
207 }
208 
209 /*ARGSUSED*/
210 static int
211 rumpmkdir(struct ukfs *dummy, const char *path, mode_t mode)
212 {
213 
214 	return rump_sys_mkdir(path, mode);
215 }
216 
217 int
218 ukfs_part_probe(char *devpath, struct ukfs_part **partp)
219 {
220 	struct ukfs_part *part;
221 	char *p;
222 	int error = 0;
223 	int devfd = -1;
224 
225 	if ((p = strstr(devpath, UKFS_PARTITION_SCANMAGIC)) != NULL) {
226 		fprintf(stderr, "ukfs: %%PART is deprecated.  use "
227 		    "%%DISKLABEL instead\n");
228 		errno = ENODEV;
229 		return -1;
230 	}
231 
232 	part = malloc(sizeof(*part));
233 	if (part == NULL) {
234 		errno = ENOMEM;
235 		return -1;
236 	}
237 	if (pthread_spin_init(&part->part_lck, PTHREAD_PROCESS_PRIVATE) == -1) {
238 		error = errno;
239 		free(part);
240 		errno = error;
241 		return -1;
242 	}
243 	part->part_type = UKFS_PART_NONE;
244 	part->part_refcount = 1;
245 
246 	/*
247 	 * Check for magic in pathname:
248 	 *   disklabel: /regularpath%DISKLABEL:labelchar%\0
249 	 *     offsets: /regularpath%OFFSET:start,end%\0
250 	 */
251 #define MAGICADJ_DISKLABEL(p, n) (p+sizeof(UKFS_DISKLABEL_SCANMAGIC)-1+n)
252 	if ((p = strstr(devpath, UKFS_DISKLABEL_SCANMAGIC)) != NULL
253 	    && strlen(p) == UKFS_DISKLABEL_MAGICLEN
254 	    && *(MAGICADJ_DISKLABEL(p,1)) == '%') {
255 		if (*(MAGICADJ_DISKLABEL(p,0)) >= 'a' &&
256 		    *(MAGICADJ_DISKLABEL(p,0)) < 'a' + UKFS_MAXPARTITIONS) {
257 			struct ukfs__disklabel dl;
258 			struct ukfs__partition *pp;
259 			char buf[65536];
260 			char labelchar = *(MAGICADJ_DISKLABEL(p,0));
261 			int partition = labelchar - 'a';
262 
263 			*p = '\0';
264 			devfd = open(devpath, O_RDONLY);
265 			if (devfd == -1) {
266 				error = errno;
267 				goto out;
268 			}
269 
270 			/* Locate the disklabel and find the partition. */
271 			if (pread(devfd, buf, sizeof(buf), 0) == -1) {
272 				error = errno;
273 				goto out;
274 			}
275 
276 			if (ukfs__disklabel_scan(&dl, buf, sizeof(buf)) != 0) {
277 				error = ENOENT;
278 				goto out;
279 			}
280 
281 			if (dl.d_npartitions < partition) {
282 				error = ENOENT;
283 				goto out;
284 			}
285 
286 			pp = &dl.d_partitions[partition];
287 			part->part_type = UKFS_PART_DISKLABEL;
288 			part->part_labelchar = labelchar;
289 			part->part_devoff = pp->p_offset << DEV_BSHIFT;
290 			part->part_devsize = pp->p_size << DEV_BSHIFT;
291 		} else {
292 			error = EINVAL;
293 		}
294 #define MAGICADJ_OFFSET(p, n) (p+sizeof(UKFS_OFFSET_SCANMAGIC)-1+n)
295 	} else if (((p = strstr(devpath, UKFS_OFFSET_SCANMAGIC)) != NULL)
296 	    && (strlen(p) >= UKFS_OFFSET_MINLEN)) {
297 		char *comma, *pers, *ep, *nptr;
298 		u_quad_t val;
299 
300 		comma = strchr(p, ',');
301 		if (comma == NULL) {
302 			error = EINVAL;
303 			goto out;
304 		}
305 		pers = strchr(comma, '%');
306 		if (pers == NULL) {
307 			error = EINVAL;
308 			goto out;
309 		}
310 		*comma = '\0';
311 		*pers = '\0';
312 		*p = '\0';
313 
314 		nptr = MAGICADJ_OFFSET(p,0);
315 		/* check if string is negative */
316 		if (*nptr == '-') {
317 			error = ERANGE;
318 			goto out;
319 		}
320 		val = strtouq(nptr, &ep, 10);
321 		if (val == UQUAD_MAX) {
322 			error = ERANGE;
323 			goto out;
324 		}
325 		if (*ep != '\0') {
326 			error = EADDRNOTAVAIL; /* creative ;) */
327 			goto out;
328 		}
329 		part->part_devoff = val;
330 
331 		/* omstart */
332 
333 		nptr = comma+1;
334 		/* check if string is negative */
335 		if (*nptr == '-') {
336 			error = ERANGE;
337 			goto out;
338 		}
339 		val = strtouq(nptr, &ep, 10);
340 		if (val == UQUAD_MAX) {
341 			error = ERANGE;
342 			goto out;
343 		}
344 		if (*ep != '\0') {
345 			error = EADDRNOTAVAIL; /* creative ;) */
346 			goto out;
347 		}
348 		part->part_devsize = val;
349 		part->part_type = UKFS_PART_OFFSET;
350 	} else {
351 		ukfs_part_release(part);
352 		part = ukfs_part_none;
353 	}
354 
355  out:
356 	if (devfd != -1)
357 		close(devfd);
358 	if (error) {
359 		free(part);
360 		errno = error;
361 	} else {
362 		*partp = part;
363 	}
364 
365 	return error ? -1 : 0;
366 }
367 
368 int
369 ukfs_part_tostring(struct ukfs_part *part, char *str, size_t strsize)
370 {
371 	int rv;
372 
373 	*str = '\0';
374 	/* "pseudo" values */
375 	if (part == ukfs_part_na) {
376 		errno = EINVAL;
377 		return -1;
378 	}
379 	if (part == ukfs_part_none)
380 		return 0;
381 
382 	rv = 0;
383 	switch (part->part_type) {
384 	case UKFS_PART_NONE:
385 		break;
386 
387 	case UKFS_PART_DISKLABEL:
388 		snprintf(str, strsize, "%%DISKLABEL:%c%%",part->part_labelchar);
389 		rv = 1;
390 		break;
391 
392 	case UKFS_PART_OFFSET:
393 		snprintf(str, strsize, "[%llu,%llu]",
394 		    (unsigned long long)part->part_devoff,
395 		    (unsigned long long)(part->part_devoff+part->part_devsize));
396 		rv = 1;
397 		break;
398 	}
399 
400 	return rv;
401 }
402 
403 static void
404 unlockdev(int fd, struct ukfs_part *part)
405 {
406 	struct flock flarg;
407 
408 	if (part == ukfs_part_na)
409 		return;
410 
411 	memset(&flarg, 0, sizeof(flarg));
412 	flarg.l_type = F_UNLCK;
413 	flarg.l_whence = SEEK_SET;
414 	flarg.l_start = part->part_devoff;
415 	flarg.l_len = PART2LOCKSIZE(part->part_devsize);
416 	if (fcntl(fd, F_SETLK, &flarg) == -1)
417 		warn("ukfs: cannot unlock device file");
418 }
419 
420 /*
421  * Open the disk file and flock it.  Also, if we are operation on
422  * an embedded partition, find the partition offset and size from
423  * the disklabel.
424  *
425  * We hard-fail only in two cases:
426  *  1) we failed to get the partition info out (don't know what offset
427  *     to mount from)
428  *  2) we failed to flock the source device (i.e. fcntl() fails,
429  *     not e.g. open() before it)
430  *
431  * Otherwise we let the code proceed to mount and let the file system
432  * throw the proper error.  The only questionable bit is that if we
433  * soft-fail before flock and mount does succeed...
434  *
435  * Returns: -1 error (errno reports error code)
436  *           0 success
437  *
438  * dfdp: -1  device is not open
439  *        n  device is open
440  */
441 static int
442 process_diskdevice(const char *devpath, struct ukfs_part *part, int rdonly,
443 	int *dfdp)
444 {
445 	struct stat sb;
446 	int rv = 0, devfd;
447 
448 	/* defaults */
449 	*dfdp = -1;
450 
451 	devfd = open(devpath, rdonly ? O_RDONLY : O_RDWR);
452 	if (devfd == -1) {
453 		rv = errno;
454 		goto out;
455 	}
456 
457 	if (fstat(devfd, &sb) == -1) {
458 		rv = errno;
459 		goto out;
460 	}
461 
462 	/*
463 	 * We do this only for non-block device since the
464 	 * (NetBSD) kernel allows block device open only once.
465 	 * We also need to close the device for fairly obvious reasons.
466 	 */
467 	if (!S_ISBLK(sb.st_mode)) {
468 		struct flock flarg;
469 
470 		memset(&flarg, 0, sizeof(flarg));
471 		flarg.l_type = rdonly ? F_RDLCK : F_WRLCK;
472 		flarg.l_whence = SEEK_SET;
473 		flarg.l_start = part->part_devoff;
474 		flarg.l_len = PART2LOCKSIZE(part->part_devsize);
475 		if (fcntl(devfd, F_SETLK, &flarg) == -1) {
476 			pid_t holder;
477 			int sverrno;
478 
479 			sverrno = errno;
480 			if (fcntl(devfd, F_GETLK, &flarg) != 1)
481 				holder = flarg.l_pid;
482 			else
483 				holder = -1;
484 			warnx("ukfs_mount: cannot lock device.  held by pid %d",
485 			    holder);
486 			rv = sverrno;
487 			goto out;
488 		}
489 	} else {
490 		close(devfd);
491 		devfd = -1;
492 	}
493 	*dfdp = devfd;
494 
495  out:
496 	if (rv) {
497 		if (devfd != -1)
498 			close(devfd);
499 	}
500 
501 	return rv;
502 }
503 
504 static struct ukfs *
505 doukfsmount(const char *vfsname, const char *devpath, struct ukfs_part *part,
506 	const char *mountpath, int mntflags, void *arg, size_t alen)
507 {
508 	struct ukfs *fs = NULL;
509 	int rv = 0, devfd = -1;
510 	int mounted = 0;
511 	int regged = 0;
512 
513 	pthread_spin_lock(&part->part_lck);
514 	part->part_refcount++;
515 	pthread_spin_unlock(&part->part_lck);
516 	if (part != ukfs_part_na) {
517 		if ((rv = process_diskdevice(devpath, part,
518 		    mntflags & MNT_RDONLY, &devfd)) != 0)
519 			goto out;
520 	}
521 
522 	fs = malloc(sizeof(struct ukfs));
523 	if (fs == NULL) {
524 		rv = ENOMEM;
525 		goto out;
526 	}
527 	memset(fs, 0, sizeof(struct ukfs));
528 
529 	/* create our mountpoint.  this is never removed. */
530 	if (builddirs(mountpath, 0777, rumpmkdir, NULL) == -1) {
531 		if (errno != EEXIST) {
532 			rv = errno;
533 			goto out;
534 		}
535 	}
536 
537 	if (part != ukfs_part_na) {
538 		/* LINTED */
539 		rv = rump_pub_etfs_register_withsize(devpath, devpath,
540 		    RUMP_ETFS_BLK, part->part_devoff, part->part_devsize);
541 		if (rv) {
542 			goto out;
543 		}
544 		regged = 1;
545 	}
546 
547 	rv = rump_sys_mount(vfsname, mountpath, mntflags, arg, alen);
548 	if (rv) {
549 		rv = errno;
550 		goto out;
551 	}
552 	mounted = 1;
553 	rv = rump_pub_vfs_getmp(mountpath, &fs->ukfs_mp);
554 	if (rv) {
555 		goto out;
556 	}
557 	rv = rump_pub_vfs_root(fs->ukfs_mp, &fs->ukfs_rvp, 0);
558 	if (rv) {
559 		goto out;
560 	}
561 
562 	if (regged) {
563 		fs->ukfs_devpath = strdup(devpath);
564 	}
565 	fs->ukfs_mountpath = strdup(mountpath);
566 	fs->ukfs_cdir = ukfs_getrvp(fs);
567 	pthread_spin_init(&fs->ukfs_spin, PTHREAD_PROCESS_SHARED);
568 	fs->ukfs_devfd = devfd;
569 	fs->ukfs_part = part;
570 	assert(rv == 0);
571 
572  out:
573 	if (rv) {
574 		if (fs) {
575 			if (fs->ukfs_rvp)
576 				rump_pub_vp_rele(fs->ukfs_rvp);
577 			free(fs);
578 			fs = NULL;
579 		}
580 		if (mounted)
581 			rump_sys_unmount(mountpath, MNT_FORCE);
582 		if (regged)
583 			rump_pub_etfs_remove(devpath);
584 		if (devfd != -1) {
585 			unlockdev(devfd, part);
586 			close(devfd);
587 		}
588 		ukfs_part_release(part);
589 		errno = rv;
590 	}
591 
592 	return fs;
593 }
594 
595 struct ukfs *
596 ukfs_mount(const char *vfsname, const char *devpath,
597 	const char *mountpath, int mntflags, void *arg, size_t alen)
598 {
599 
600 	return doukfsmount(vfsname, devpath, ukfs_part_na,
601 	    mountpath, mntflags, arg, alen);
602 }
603 
604 struct ukfs *
605 ukfs_mount_disk(const char *vfsname, const char *devpath,
606 	struct ukfs_part *part, const char *mountpath, int mntflags,
607 	void *arg, size_t alen)
608 {
609 
610 	return doukfsmount(vfsname, devpath, part,
611 	    mountpath, mntflags, arg, alen);
612 }
613 
614 int
615 ukfs_release(struct ukfs *fs, int flags)
616 {
617 
618 	if ((flags & UKFS_RELFLAG_NOUNMOUNT) == 0) {
619 		int rv, mntflag, error;
620 
621 		ukfs_chdir(fs, "/");
622 		mntflag = 0;
623 		if (flags & UKFS_RELFLAG_FORCE)
624 			mntflag = MNT_FORCE;
625 		rump_pub_lwp_alloc_and_switch(nextpid(fs), 1);
626 		rump_pub_vp_rele(fs->ukfs_rvp);
627 		fs->ukfs_rvp = NULL;
628 		rv = rump_sys_unmount(fs->ukfs_mountpath, mntflag);
629 		if (rv == -1) {
630 			error = errno;
631 			rump_pub_vfs_root(fs->ukfs_mp, &fs->ukfs_rvp, 0);
632 			rump_pub_lwp_release(rump_pub_lwp_curlwp());
633 			ukfs_chdir(fs, fs->ukfs_mountpath);
634 			errno = error;
635 			return -1;
636 		}
637 		rump_pub_lwp_release(rump_pub_lwp_curlwp());
638 	}
639 
640 	if (fs->ukfs_devpath) {
641 		rump_pub_etfs_remove(fs->ukfs_devpath);
642 		free(fs->ukfs_devpath);
643 	}
644 	free(fs->ukfs_mountpath);
645 
646 	pthread_spin_destroy(&fs->ukfs_spin);
647 	if (fs->ukfs_devfd != -1) {
648 		unlockdev(fs->ukfs_devfd, fs->ukfs_part);
649 		close(fs->ukfs_devfd);
650 	}
651 	ukfs_part_release(fs->ukfs_part);
652 	free(fs);
653 
654 	return 0;
655 }
656 
657 void
658 ukfs_part_release(struct ukfs_part *part)
659 {
660 	int release;
661 
662 	if (part != ukfs_part_none && part != ukfs_part_na) {
663 		pthread_spin_lock(&part->part_lck);
664 		release = --part->part_refcount == 0;
665 		pthread_spin_unlock(&part->part_lck);
666 		if (release) {
667 			pthread_spin_destroy(&part->part_lck);
668 			free(part);
669 		}
670 	}
671 }
672 
673 #define STDCALL(ukfs, thecall)						\
674 	int rv = 0;							\
675 									\
676 	precall(ukfs);							\
677 	rv = thecall;							\
678 	postcall(ukfs);							\
679 	return rv;
680 
681 int
682 ukfs_opendir(struct ukfs *ukfs, const char *dirname, struct ukfs_dircookie **c)
683 {
684 	struct vnode *vp;
685 	int rv;
686 
687 	precall(ukfs);
688 	rv = rump_pub_namei(RUMP_NAMEI_LOOKUP, RUMP_NAMEI_LOCKLEAF, dirname,
689 	    NULL, &vp, NULL);
690 	postcall(ukfs);
691 
692 	if (rv == 0) {
693 		RUMP_VOP_UNLOCK(vp, 0);
694 	} else {
695 		errno = rv;
696 		rv = -1;
697 	}
698 
699 	/*LINTED*/
700 	*c = (struct ukfs_dircookie *)vp;
701 	return rv;
702 }
703 
704 static int
705 getmydents(struct vnode *vp, off_t *off, uint8_t *buf, size_t bufsize)
706 {
707 	struct uio *uio;
708 	size_t resid;
709 	int rv, eofflag;
710 	kauth_cred_t cred;
711 
712 	uio = rump_pub_uio_setup(buf, bufsize, *off, RUMPUIO_READ);
713 	cred = rump_pub_cred_suserget();
714 	rv = RUMP_VOP_READDIR(vp, uio, cred, &eofflag, NULL, NULL);
715 	rump_pub_cred_put(cred);
716 	RUMP_VOP_UNLOCK(vp, 0);
717 	*off = rump_pub_uio_getoff(uio);
718 	resid = rump_pub_uio_free(uio);
719 
720 	if (rv) {
721 		errno = rv;
722 		return -1;
723 	}
724 
725 	/* LINTED: not totally correct return type, but follows syscall */
726 	return bufsize - resid;
727 }
728 
729 /*ARGSUSED*/
730 int
731 ukfs_getdents_cookie(struct ukfs *ukfs, struct ukfs_dircookie *c, off_t *off,
732 	uint8_t *buf, size_t bufsize)
733 {
734 	/*LINTED*/
735 	struct vnode *vp = (struct vnode *)c;
736 
737 	RUMP_VOP_LOCK(vp, RUMP_LK_SHARED);
738 	return getmydents(vp, off, buf, bufsize);
739 }
740 
741 int
742 ukfs_getdents(struct ukfs *ukfs, const char *dirname, off_t *off,
743 	uint8_t *buf, size_t bufsize)
744 {
745 	struct vnode *vp;
746 	int rv;
747 
748 	precall(ukfs);
749 	rv = rump_pub_namei(RUMP_NAMEI_LOOKUP, RUMP_NAMEI_LOCKLEAF, dirname,
750 	    NULL, &vp, NULL);
751 	postcall(ukfs);
752 	if (rv) {
753 		errno = rv;
754 		return -1;
755 	}
756 
757 	rv = getmydents(vp, off, buf, bufsize);
758 	rump_pub_vp_rele(vp);
759 	return rv;
760 }
761 
762 /*ARGSUSED*/
763 int
764 ukfs_closedir(struct ukfs *ukfs, struct ukfs_dircookie *c)
765 {
766 
767 	/*LINTED*/
768 	rump_pub_vp_rele((struct vnode *)c);
769 	return 0;
770 }
771 
772 int
773 ukfs_open(struct ukfs *ukfs, const char *filename, int flags)
774 {
775 	int fd;
776 
777 	precall(ukfs);
778 	fd = rump_sys_open(filename, flags, 0);
779 	postcall(ukfs);
780 	if (fd == -1)
781 		return -1;
782 
783 	return fd;
784 }
785 
786 ssize_t
787 ukfs_read(struct ukfs *ukfs, const char *filename, off_t off,
788 	uint8_t *buf, size_t bufsize)
789 {
790 	int fd;
791 	ssize_t xfer = -1; /* XXXgcc */
792 
793 	precall(ukfs);
794 	fd = rump_sys_open(filename, RUMP_O_RDONLY, 0);
795 	if (fd == -1)
796 		goto out;
797 
798 	xfer = rump_sys_pread(fd, buf, bufsize, off);
799 	rump_sys_close(fd);
800 
801  out:
802 	postcall(ukfs);
803 	if (fd == -1) {
804 		return -1;
805 	}
806 	return xfer;
807 }
808 
809 /*ARGSUSED*/
810 ssize_t
811 ukfs_read_fd(struct ukfs *ukfs, int fd, off_t off, uint8_t *buf, size_t buflen)
812 {
813 
814 	return rump_sys_pread(fd, buf, buflen, off);
815 }
816 
817 ssize_t
818 ukfs_write(struct ukfs *ukfs, const char *filename, off_t off,
819 	uint8_t *buf, size_t bufsize)
820 {
821 	int fd;
822 	ssize_t xfer = -1; /* XXXgcc */
823 
824 	precall(ukfs);
825 	fd = rump_sys_open(filename, RUMP_O_WRONLY, 0);
826 	if (fd == -1)
827 		goto out;
828 
829 	/* write and commit */
830 	xfer = rump_sys_pwrite(fd, buf, bufsize, off);
831 	if (xfer > 0)
832 		rump_sys_fsync(fd);
833 
834 	rump_sys_close(fd);
835 
836  out:
837 	postcall(ukfs);
838 	if (fd == -1) {
839 		return -1;
840 	}
841 	return xfer;
842 }
843 
844 /*ARGSUSED*/
845 ssize_t
846 ukfs_write_fd(struct ukfs *ukfs, int fd, off_t off, uint8_t *buf, size_t buflen,
847 	int dosync)
848 {
849 	ssize_t xfer;
850 
851 	xfer = rump_sys_pwrite(fd, buf, buflen, off);
852 	if (xfer > 0 && dosync)
853 		rump_sys_fsync(fd);
854 
855 	return xfer;
856 }
857 
858 /*ARGSUSED*/
859 int
860 ukfs_close(struct ukfs *ukfs, int fd)
861 {
862 
863 	rump_sys_close(fd);
864 	return 0;
865 }
866 
867 int
868 ukfs_create(struct ukfs *ukfs, const char *filename, mode_t mode)
869 {
870 	int fd;
871 
872 	precall(ukfs);
873 	fd = rump_sys_open(filename, RUMP_O_WRONLY | RUMP_O_CREAT, mode);
874 	if (fd == -1)
875 		return -1;
876 	rump_sys_close(fd);
877 
878 	postcall(ukfs);
879 	return 0;
880 }
881 
882 int
883 ukfs_mknod(struct ukfs *ukfs, const char *path, mode_t mode, dev_t dev)
884 {
885 
886 	STDCALL(ukfs, rump_sys_mknod(path, mode, dev));
887 }
888 
889 int
890 ukfs_mkfifo(struct ukfs *ukfs, const char *path, mode_t mode)
891 {
892 
893 	STDCALL(ukfs, rump_sys_mkfifo(path, mode));
894 }
895 
896 int
897 ukfs_mkdir(struct ukfs *ukfs, const char *filename, mode_t mode)
898 {
899 
900 	STDCALL(ukfs, rump_sys_mkdir(filename, mode));
901 }
902 
903 int
904 ukfs_remove(struct ukfs *ukfs, const char *filename)
905 {
906 
907 	STDCALL(ukfs, rump_sys_unlink(filename));
908 }
909 
910 int
911 ukfs_rmdir(struct ukfs *ukfs, const char *filename)
912 {
913 
914 	STDCALL(ukfs, rump_sys_rmdir(filename));
915 }
916 
917 int
918 ukfs_link(struct ukfs *ukfs, const char *filename, const char *f_create)
919 {
920 
921 	STDCALL(ukfs, rump_sys_link(filename, f_create));
922 }
923 
924 int
925 ukfs_symlink(struct ukfs *ukfs, const char *filename, const char *linkname)
926 {
927 
928 	STDCALL(ukfs, rump_sys_symlink(filename, linkname));
929 }
930 
931 ssize_t
932 ukfs_readlink(struct ukfs *ukfs, const char *filename,
933 	char *linkbuf, size_t buflen)
934 {
935 	ssize_t rv;
936 
937 	precall(ukfs);
938 	rv = rump_sys_readlink(filename, linkbuf, buflen);
939 	postcall(ukfs);
940 	return rv;
941 }
942 
943 int
944 ukfs_rename(struct ukfs *ukfs, const char *from, const char *to)
945 {
946 
947 	STDCALL(ukfs, rump_sys_rename(from, to));
948 }
949 
950 int
951 ukfs_chdir(struct ukfs *ukfs, const char *path)
952 {
953 	struct vnode *newvp, *oldvp;
954 	int rv;
955 
956 	precall(ukfs);
957 	rv = rump_sys_chdir(path);
958 	if (rv == -1)
959 		goto out;
960 
961 	newvp = rump_pub_cdir_get();
962 	pthread_spin_lock(&ukfs->ukfs_spin);
963 	oldvp = ukfs->ukfs_cdir;
964 	ukfs->ukfs_cdir = newvp;
965 	pthread_spin_unlock(&ukfs->ukfs_spin);
966 	if (oldvp)
967 		rump_pub_vp_rele(oldvp);
968 
969  out:
970 	postcall(ukfs);
971 	return rv;
972 }
973 
974 /*
975  * If we want to use post-time_t file systems on pre-time_t hosts,
976  * we must translate the stat structure.  Since we don't currently
977  * have a general method for making compat calls in rump, special-case
978  * this one.
979  *
980  * Note that this does not allow making system calls to older rump
981  * kernels from newer hosts.
982  */
983 #define VERS_TIMECHANGE 599000700
984 
985 static int
986 needcompat(void)
987 {
988 
989 #ifdef __NetBSD__
990 	/*LINTED*/
991 	return __NetBSD_Version__ < VERS_TIMECHANGE
992 	    && rump_pub_getversion() >= VERS_TIMECHANGE;
993 #else
994 	return 0;
995 #endif
996 }
997 
998 int
999 ukfs_stat(struct ukfs *ukfs, const char *filename, struct stat *file_stat)
1000 {
1001 	int rv;
1002 
1003 	precall(ukfs);
1004 	if (needcompat())
1005 		rv = rump_pub_sys___stat30(filename, file_stat);
1006 	else
1007 		rv = rump_sys_stat(filename, file_stat);
1008 	postcall(ukfs);
1009 
1010 	return rv;
1011 }
1012 
1013 int
1014 ukfs_lstat(struct ukfs *ukfs, const char *filename, struct stat *file_stat)
1015 {
1016 	int rv;
1017 
1018 	precall(ukfs);
1019 	if (needcompat())
1020 		rv = rump_pub_sys___lstat30(filename, file_stat);
1021 	else
1022 		rv = rump_sys_lstat(filename, file_stat);
1023 	postcall(ukfs);
1024 
1025 	return rv;
1026 }
1027 
1028 int
1029 ukfs_chmod(struct ukfs *ukfs, const char *filename, mode_t mode)
1030 {
1031 
1032 	STDCALL(ukfs, rump_sys_chmod(filename, mode));
1033 }
1034 
1035 int
1036 ukfs_lchmod(struct ukfs *ukfs, const char *filename, mode_t mode)
1037 {
1038 
1039 	STDCALL(ukfs, rump_sys_lchmod(filename, mode));
1040 }
1041 
1042 int
1043 ukfs_chown(struct ukfs *ukfs, const char *filename, uid_t uid, gid_t gid)
1044 {
1045 
1046 	STDCALL(ukfs, rump_sys_chown(filename, uid, gid));
1047 }
1048 
1049 int
1050 ukfs_lchown(struct ukfs *ukfs, const char *filename, uid_t uid, gid_t gid)
1051 {
1052 
1053 	STDCALL(ukfs, rump_sys_lchown(filename, uid, gid));
1054 }
1055 
1056 int
1057 ukfs_chflags(struct ukfs *ukfs, const char *filename, u_long flags)
1058 {
1059 
1060 	STDCALL(ukfs, rump_sys_chflags(filename, flags));
1061 }
1062 
1063 int
1064 ukfs_lchflags(struct ukfs *ukfs, const char *filename, u_long flags)
1065 {
1066 
1067 	STDCALL(ukfs, rump_sys_lchflags(filename, flags));
1068 }
1069 
1070 int
1071 ukfs_utimes(struct ukfs *ukfs, const char *filename, const struct timeval *tptr)
1072 {
1073 
1074 	STDCALL(ukfs, rump_sys_utimes(filename, tptr));
1075 }
1076 
1077 int
1078 ukfs_lutimes(struct ukfs *ukfs, const char *filename,
1079 	      const struct timeval *tptr)
1080 {
1081 
1082 	STDCALL(ukfs, rump_sys_lutimes(filename, tptr));
1083 }
1084 
1085 /*
1086  * Dynamic module support
1087  */
1088 
1089 /* load one library */
1090 
1091 /*
1092  * XXX: the dlerror stuff isn't really threadsafe, but then again I
1093  * can't protect against other threads calling dl*() outside of ukfs,
1094  * so just live with it being flimsy
1095  */
1096 int
1097 ukfs_modload(const char *fname)
1098 {
1099 	void *handle;
1100 	struct modinfo **mi;
1101 	int error;
1102 
1103 	handle = dlopen(fname, RTLD_LAZY|RTLD_GLOBAL);
1104 	if (handle == NULL) {
1105 		const char *dlmsg = dlerror();
1106 		if (strstr(dlmsg, "Undefined symbol"))
1107 			return 0;
1108 		warnx("dlopen %s failed: %s\n", fname, dlmsg);
1109 		/* XXXerrno */
1110 		return -1;
1111 	}
1112 
1113 	mi = dlsym(handle, "__start_link_set_modules");
1114 	if (mi) {
1115 		error = rump_pub_module_init(*mi, NULL);
1116 		if (error)
1117 			goto errclose;
1118 		return 1;
1119 	}
1120 	error = EINVAL;
1121 
1122  errclose:
1123 	dlclose(handle);
1124 	errno = error;
1125 	return -1;
1126 }
1127 
1128 struct loadfail {
1129 	char *pname;
1130 
1131 	LIST_ENTRY(loadfail) entries;
1132 };
1133 
1134 #define RUMPFSMOD_PREFIX "librumpfs_"
1135 #define RUMPFSMOD_SUFFIX ".so"
1136 
1137 int
1138 ukfs_modload_dir(const char *dir)
1139 {
1140 	char nbuf[MAXPATHLEN+1], *p;
1141 	struct dirent entry, *result;
1142 	DIR *libdir;
1143 	struct loadfail *lf, *nlf;
1144 	int error, nloaded = 0, redo;
1145 	LIST_HEAD(, loadfail) lfs;
1146 
1147 	libdir = opendir(dir);
1148 	if (libdir == NULL)
1149 		return -1;
1150 
1151 	LIST_INIT(&lfs);
1152 	for (;;) {
1153 		if ((error = readdir_r(libdir, &entry, &result)) != 0)
1154 			break;
1155 		if (!result)
1156 			break;
1157 		if (strncmp(result->d_name, RUMPFSMOD_PREFIX,
1158 		    strlen(RUMPFSMOD_PREFIX)) != 0)
1159 			continue;
1160 		if (((p = strstr(result->d_name, RUMPFSMOD_SUFFIX)) == NULL)
1161 		    || strlen(p) != strlen(RUMPFSMOD_SUFFIX))
1162 			continue;
1163 		strlcpy(nbuf, dir, sizeof(nbuf));
1164 		strlcat(nbuf, "/", sizeof(nbuf));
1165 		strlcat(nbuf, result->d_name, sizeof(nbuf));
1166 		switch (ukfs_modload(nbuf)) {
1167 		case 0:
1168 			lf = malloc(sizeof(*lf));
1169 			if (lf == NULL) {
1170 				error = ENOMEM;
1171 				break;
1172 			}
1173 			lf->pname = strdup(nbuf);
1174 			if (lf->pname == NULL) {
1175 				free(lf);
1176 				error = ENOMEM;
1177 				break;
1178 			}
1179 			LIST_INSERT_HEAD(&lfs, lf, entries);
1180 			break;
1181 		case 1:
1182 			nloaded++;
1183 			break;
1184 		default:
1185 			/* ignore errors */
1186 			break;
1187 		}
1188 	}
1189 	closedir(libdir);
1190 	if (error && nloaded != 0)
1191 		error = 0;
1192 
1193 	/*
1194 	 * El-cheapo dependency calculator.  Just try to load the
1195 	 * modules n times in a loop
1196 	 */
1197 	for (redo = 1; redo;) {
1198 		redo = 0;
1199 		nlf = LIST_FIRST(&lfs);
1200 		while ((lf = nlf) != NULL) {
1201 			nlf = LIST_NEXT(lf, entries);
1202 			if (ukfs_modload(lf->pname) == 1) {
1203 				nloaded++;
1204 				redo = 1;
1205 				LIST_REMOVE(lf, entries);
1206 				free(lf->pname);
1207 				free(lf);
1208 			}
1209 		}
1210 	}
1211 
1212 	while ((lf = LIST_FIRST(&lfs)) != NULL) {
1213 		LIST_REMOVE(lf, entries);
1214 		free(lf->pname);
1215 		free(lf);
1216 	}
1217 
1218 	if (error && nloaded == 0) {
1219 		errno = error;
1220 		return -1;
1221 	}
1222 
1223 	return nloaded;
1224 }
1225 
1226 /* XXX: this code uses definitions from NetBSD, needs rumpdefs */
1227 ssize_t
1228 ukfs_vfstypes(char *buf, size_t buflen)
1229 {
1230 	int mib[3];
1231 	struct sysctlnode q, ans[128];
1232 	size_t alen;
1233 	int i;
1234 
1235 	mib[0] = CTL_VFS;
1236 	mib[1] = VFS_GENERIC;
1237 	mib[2] = CTL_QUERY;
1238 	alen = sizeof(ans);
1239 
1240 	memset(&q, 0, sizeof(q));
1241 	q.sysctl_flags = SYSCTL_VERSION;
1242 
1243 	if (rump_sys___sysctl(mib, 3, ans, &alen, &q, sizeof(q)) == -1) {
1244 		return -1;
1245 	}
1246 
1247 	for (i = 0; i < alen/sizeof(ans[0]); i++)
1248 		if (strcmp("fstypes", ans[i].sysctl_name) == 0)
1249 			break;
1250 	if (i == alen/sizeof(ans[0])) {
1251 		errno = ENXIO;
1252 		return -1;
1253 	}
1254 
1255 	mib[0] = CTL_VFS;
1256 	mib[1] = VFS_GENERIC;
1257 	mib[2] = ans[i].sysctl_num;
1258 
1259 	if (rump_sys___sysctl(mib, 3, buf, &buflen, NULL, 0) == -1) {
1260 		return -1;
1261 	}
1262 
1263 	return buflen;
1264 }
1265 
1266 /*
1267  * Utilities
1268  */
1269 static int
1270 builddirs(const char *pathname, mode_t mode,
1271 	int (*mkdirfn)(struct ukfs *, const char *, mode_t), struct ukfs *fs)
1272 {
1273 	char *f1, *f2;
1274 	int rv;
1275 	mode_t mask;
1276 	bool end;
1277 
1278 	/*ukfs_umask((mask = ukfs_umask(0)));*/
1279 	umask((mask = umask(0)));
1280 
1281 	f1 = f2 = strdup(pathname);
1282 	if (f1 == NULL) {
1283 		errno = ENOMEM;
1284 		return -1;
1285 	}
1286 
1287 	end = false;
1288 	for (;;) {
1289 		/* find next component */
1290 		f2 += strspn(f2, "/");
1291 		f2 += strcspn(f2, "/");
1292 		if (*f2 == '\0')
1293 			end = true;
1294 		else
1295 			*f2 = '\0';
1296 
1297 		rv = mkdirfn(fs, f1, mode & ~mask);
1298 		if (errno == EEXIST)
1299 			rv = 0;
1300 
1301 		if (rv == -1 || *f2 != '\0' || end)
1302 			break;
1303 
1304 		*f2 = '/';
1305 	}
1306 
1307 	free(f1);
1308 
1309 	return rv;
1310 }
1311 
1312 int
1313 ukfs_util_builddirs(struct ukfs *ukfs, const char *pathname, mode_t mode)
1314 {
1315 
1316 	return builddirs(pathname, mode, ukfs_mkdir, ukfs);
1317 }
1318