puffs.c revision 1.74 1 /* $NetBSD: puffs.c,v 1.74 2007/11/06 15:09:08 pooka Exp $ */
2
3 /*
4 * Copyright (c) 2005, 2006, 2007 Antti Kantee. All Rights Reserved.
5 *
6 * Development of this software was supported by the
7 * Google Summer of Code program and the Ulla Tuominen Foundation.
8 * The Google SoC project was mentored by Bill Studenmund.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
20 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 #if !defined(lint)
34 __RCSID("$NetBSD: puffs.c,v 1.74 2007/11/06 15:09:08 pooka Exp $");
35 #endif /* !lint */
36
37 #include <sys/param.h>
38 #include <sys/mount.h>
39
40 #include <assert.h>
41 #include <err.h>
42 #include <errno.h>
43 #include <fcntl.h>
44 #include <mntopts.h>
45 #include <paths.h>
46 #include <puffs.h>
47 #include <stdio.h>
48 #include <stdlib.h>
49 #include <string.h>
50 #include <syslog.h>
51 #include <unistd.h>
52
53 #include "puffs_priv.h"
54
55 /* Most file systems want this for opts, so just give it to them */
56 const struct mntopt puffsmopts[] = {
57 MOPT_STDOPTS,
58 PUFFSMOPT_STD,
59 MOPT_NULL,
60 };
61
62 #ifdef PUFFS_WITH_THREADS
63 #include <pthread.h>
64 pthread_mutex_t pu_lock = PTHREAD_MUTEX_INITIALIZER;
65 #endif
66
67 #define FILLOP(lower, upper) \
68 do { \
69 if (pops->puffs_node_##lower) \
70 opmask[PUFFS_VN_##upper] = 1; \
71 } while (/*CONSTCOND*/0)
72 static void
73 fillvnopmask(struct puffs_ops *pops, uint8_t *opmask)
74 {
75
76 memset(opmask, 0, PUFFS_VN_MAX);
77
78 FILLOP(create, CREATE);
79 FILLOP(mknod, MKNOD);
80 FILLOP(open, OPEN);
81 FILLOP(close, CLOSE);
82 FILLOP(access, ACCESS);
83 FILLOP(getattr, GETATTR);
84 FILLOP(setattr, SETATTR);
85 FILLOP(poll, POLL); /* XXX: not ready in kernel */
86 FILLOP(mmap, MMAP);
87 FILLOP(fsync, FSYNC);
88 FILLOP(seek, SEEK);
89 FILLOP(remove, REMOVE);
90 FILLOP(link, LINK);
91 FILLOP(rename, RENAME);
92 FILLOP(mkdir, MKDIR);
93 FILLOP(rmdir, RMDIR);
94 FILLOP(symlink, SYMLINK);
95 FILLOP(readdir, READDIR);
96 FILLOP(readlink, READLINK);
97 FILLOP(reclaim, RECLAIM);
98 FILLOP(inactive, INACTIVE);
99 FILLOP(print, PRINT);
100 FILLOP(read, READ);
101 FILLOP(write, WRITE);
102 }
103 #undef FILLOP
104
105 /*ARGSUSED*/
106 static void
107 puffs_defaulterror(struct puffs_usermount *pu, uint8_t type,
108 int error, const char *str, void *cookie)
109 {
110
111 fprintf(stderr, "%s\n", str);
112 abort();
113 }
114
115 int
116 puffs_getselectable(struct puffs_usermount *pu)
117 {
118
119 return pu->pu_fd;
120 }
121
122 int
123 puffs_setblockingmode(struct puffs_usermount *pu, int mode)
124 {
125 int rv, x;
126
127 assert(puffs_getstate(pu) == PUFFS_STATE_RUNNING);
128
129 if (mode != PUFFSDEV_BLOCK && mode != PUFFSDEV_NONBLOCK) {
130 errno = EINVAL;
131 return -1;
132 }
133
134 x = mode;
135 rv = ioctl(pu->pu_fd, FIONBIO, &x);
136
137 if (rv == 0) {
138 if (mode == PUFFSDEV_BLOCK)
139 pu->pu_state &= ~PU_ASYNCFD;
140 else
141 pu->pu_state |= PU_ASYNCFD;
142 }
143
144 return rv;
145 }
146
147 int
148 puffs_getstate(struct puffs_usermount *pu)
149 {
150
151 return pu->pu_state & PU_STATEMASK;
152 }
153
154 void
155 puffs_setstacksize(struct puffs_usermount *pu, size_t ss)
156 {
157 int stackshift;
158
159 assert(puffs_getstate(pu) == PUFFS_STATE_BEFOREMOUNT);
160 stackshift = -1;
161 while (ss) {
162 ss >>= 1;
163 stackshift++;
164 }
165 pu->pu_cc_stackshift = stackshift;
166 assert(1<<stackshift == ss);
167 }
168
169 struct puffs_pathobj *
170 puffs_getrootpathobj(struct puffs_usermount *pu)
171 {
172 struct puffs_node *pnr;
173
174 pnr = pu->pu_pn_root;
175 if (pnr == NULL) {
176 errno = ENOENT;
177 return NULL;
178 }
179
180 return &pnr->pn_po;
181 }
182
183 void
184 puffs_setroot(struct puffs_usermount *pu, struct puffs_node *pn)
185 {
186
187 pu->pu_pn_root = pn;
188 }
189
190 struct puffs_node *
191 puffs_getroot(struct puffs_usermount *pu)
192 {
193
194 return pu->pu_pn_root;
195 }
196
197 void
198 puffs_setrootinfo(struct puffs_usermount *pu, enum vtype vt,
199 vsize_t vsize, dev_t rdev)
200 {
201 struct puffs_kargs *pargs = pu->pu_kargp;
202
203 if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT) {
204 warnx("puffs_setrootinfo: call has effect only "
205 "before mount\n");
206 return;
207 }
208
209 pargs->pa_root_vtype = vt;
210 pargs->pa_root_vsize = vsize;
211 pargs->pa_root_rdev = rdev;
212 }
213
214 void *
215 puffs_getspecific(struct puffs_usermount *pu)
216 {
217
218 return pu->pu_privdata;
219 }
220
221 size_t
222 puffs_getmaxreqlen(struct puffs_usermount *pu)
223 {
224
225 return pu->pu_maxreqlen;
226 }
227
228 void
229 puffs_setmaxreqlen(struct puffs_usermount *pu, size_t reqlen)
230 {
231
232 if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT)
233 warnx("puffs_setmaxreqlen: call has effect only "
234 "before mount\n");
235
236 pu->pu_kargp->pa_maxmsglen = reqlen;
237 }
238
239 void
240 puffs_setfhsize(struct puffs_usermount *pu, size_t fhsize, int flags)
241 {
242
243 if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT)
244 warnx("puffs_setfhsize: call has effect only before mount\n");
245
246 pu->pu_kargp->pa_fhsize = fhsize;
247 pu->pu_kargp->pa_fhflags = flags;
248 }
249
250 void
251 puffs_setncookiehash(struct puffs_usermount *pu, int nhash)
252 {
253
254 if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT)
255 warnx("puffs_setfhsize: call has effect only before mount\n");
256
257 pu->pu_kargp->pa_nhashbuckets = nhash;
258 }
259
260 void
261 puffs_set_pathbuild(struct puffs_usermount *pu, pu_pathbuild_fn fn)
262 {
263
264 pu->pu_pathbuild = fn;
265 }
266
267 void
268 puffs_set_pathtransform(struct puffs_usermount *pu, pu_pathtransform_fn fn)
269 {
270
271 pu->pu_pathtransform = fn;
272 }
273
274 void
275 puffs_set_pathcmp(struct puffs_usermount *pu, pu_pathcmp_fn fn)
276 {
277
278 pu->pu_pathcmp = fn;
279 }
280
281 void
282 puffs_set_pathfree(struct puffs_usermount *pu, pu_pathfree_fn fn)
283 {
284
285 pu->pu_pathfree = fn;
286 }
287
288 void
289 puffs_set_namemod(struct puffs_usermount *pu, pu_namemod_fn fn)
290 {
291
292 pu->pu_namemod = fn;
293 }
294
295 void
296 puffs_set_errnotify(struct puffs_usermount *pu, pu_errnotify_fn fn)
297 {
298
299 pu->pu_errnotify = fn;
300 }
301
302 void
303 puffs_ml_setloopfn(struct puffs_usermount *pu, puffs_ml_loop_fn lfn)
304 {
305
306 pu->pu_ml_lfn = lfn;
307 }
308
309 void
310 puffs_ml_settimeout(struct puffs_usermount *pu, struct timespec *ts)
311 {
312
313 if (ts == NULL) {
314 pu->pu_ml_timep = NULL;
315 } else {
316 pu->pu_ml_timeout = *ts;
317 pu->pu_ml_timep = &pu->pu_ml_timeout;
318 }
319 }
320
321 void
322 puffs_set_prepost(struct puffs_usermount *pu,
323 pu_prepost_fn pre, pu_prepost_fn pst)
324 {
325
326 pu->pu_oppre = pre;
327 pu->pu_oppost = pst;
328 }
329
330 void
331 puffs_setback(struct puffs_cc *pcc, int whatback)
332 {
333 struct puffs_req *preq = pcc->pcc_preq;
334
335 assert(PUFFSOP_OPCLASS(preq->preq_opclass) == PUFFSOP_VN && (
336 preq->preq_optype == PUFFS_VN_OPEN ||
337 preq->preq_optype == PUFFS_VN_MMAP ||
338 preq->preq_optype == PUFFS_VN_REMOVE ||
339 preq->preq_optype == PUFFS_VN_RMDIR ||
340 preq->preq_optype == PUFFS_VN_INACTIVE));
341
342 preq->preq_setbacks |= whatback & PUFFS_SETBACK_MASK;
343 }
344
345 int
346 puffs_mount(struct puffs_usermount *pu, const char *dir, int mntflags,
347 void *cookie)
348 {
349 char rp[MAXPATHLEN];
350 int rv, fd;
351
352 #if 1
353 /* XXXkludgehere */
354 /* kauth doesn't provide this service any longer */
355 if (geteuid() != 0)
356 mntflags |= MNT_NOSUID | MNT_NODEV;
357 #endif
358
359 if (realpath(dir, rp) == NULL)
360 return -1;
361
362 if (strcmp(dir, rp) != 0) {
363 warnx("puffs_mount: \"%s\" is a relative path.", dir);
364 warnx("puffs_mount: using \"%s\" instead.", rp);
365 }
366
367 fd = open(_PATH_PUFFS, O_RDONLY);
368 if (fd == -1) {
369 warnx("puffs_mount: cannot open %s", _PATH_PUFFS);
370 return -1;
371 }
372 if (fd <= 2)
373 warnx("puffs_init: device fd %d (<= 2), sure this is "
374 "what you want?", fd);
375
376 pu->pu_kargp->pa_fd = pu->pu_fd = fd;
377 pu->pu_kargp->pa_root_cookie = cookie;
378 if ((rv = mount(MOUNT_PUFFS, rp, mntflags,
379 pu->pu_kargp, sizeof(struct puffs_kargs))) == -1)
380 goto out;
381 #if 0
382 if ((rv = ioctl(pu->pu_fd, PUFFSREQSIZEOP, &pu->pu_maxreqlen)) == -1)
383 goto out;
384 #endif
385 PU_SETSTATE(pu, PUFFS_STATE_RUNNING);
386
387 out:
388 free(pu->pu_kargp);
389 pu->pu_kargp = NULL;
390
391 return rv;
392 }
393
394 struct puffs_usermount *
395 _puffs_init(int develv, struct puffs_ops *pops, const char *mntfromname,
396 const char *puffsname, void *priv, uint32_t pflags)
397 {
398 struct puffs_usermount *pu;
399 struct puffs_kargs *pargs;
400 int sverrno;
401
402 if (develv != PUFFS_DEVEL_LIBVERSION) {
403 warnx("puffs_init: mounting with lib version %d, need %d",
404 develv, PUFFS_DEVEL_LIBVERSION);
405 errno = EINVAL;
406 return NULL;
407 }
408
409 pu = malloc(sizeof(struct puffs_usermount));
410 if (pu == NULL)
411 goto failfree;
412 memset(pu, 0, sizeof(struct puffs_usermount));
413
414 pargs = pu->pu_kargp = malloc(sizeof(struct puffs_kargs));
415 if (pargs == NULL)
416 goto failfree;
417 memset(pargs, 0, sizeof(struct puffs_kargs));
418
419 pargs->pa_vers = PUFFSDEVELVERS | PUFFSVERSION;
420 pargs->pa_flags = PUFFS_FLAG_KERN(pflags);
421 fillvnopmask(pops, pargs->pa_vnopmask);
422 (void)strlcpy(pargs->pa_typename, puffsname,
423 sizeof(pargs->pa_typename));
424 (void)strlcpy(pargs->pa_mntfromname, mntfromname,
425 sizeof(pargs->pa_mntfromname));
426
427 puffs_zerostatvfs(&pargs->pa_svfsb);
428 pargs->pa_root_cookie = NULL;
429 pargs->pa_root_vtype = VDIR;
430 pargs->pa_root_vsize = 0;
431 pargs->pa_root_rdev = 0;
432 pargs->pa_maxmsglen = 0;
433
434 pu->pu_flags = pflags;
435 pu->pu_ops = *pops;
436 free(pops); /* XXX */
437
438 pu->pu_privdata = priv;
439 pu->pu_cc_stackshift = PUFFS_CC_STACKSHIFT_DEFAULT;
440 LIST_INIT(&pu->pu_pnodelst);
441 LIST_INIT(&pu->pu_framectrl.fb_ios);
442 LIST_INIT(&pu->pu_ccnukelst);
443 TAILQ_INIT(&pu->pu_sched);
444 TAILQ_INIT(&pu->pu_exq);
445
446 /* defaults for some user-settable translation functions */
447 pu->pu_cmap = NULL; /* identity translation */
448
449 pu->pu_pathbuild = puffs_stdpath_buildpath;
450 pu->pu_pathfree = puffs_stdpath_freepath;
451 pu->pu_pathcmp = puffs_stdpath_cmppath;
452 pu->pu_pathtransform = NULL;
453 pu->pu_namemod = NULL;
454
455 pu->pu_errnotify = puffs_defaulterror;
456
457 PU_SETSTATE(pu, PUFFS_STATE_BEFOREMOUNT);
458
459 return pu;
460
461 failfree:
462 /* can't unmount() from here for obvious reasons */
463 sverrno = errno;
464 free(pu);
465 errno = sverrno;
466 return NULL;
467 }
468
469 /*
470 * XXX: there's currently no clean way to request unmount from
471 * within the user server, so be very brutal about it.
472 */
473 /*ARGSUSED1*/
474 int
475 puffs_exit(struct puffs_usermount *pu, int force)
476 {
477 struct puffs_node *pn;
478
479 force = 1; /* currently */
480
481 if (pu->pu_fd)
482 close(pu->pu_fd);
483
484 while ((pn = LIST_FIRST(&pu->pu_pnodelst)) != NULL)
485 puffs_pn_put(pn);
486
487 puffs_framev_exit(pu);
488 if (pu->pu_state & PU_HASKQ)
489 close(pu->pu_kq);
490 free(pu);
491
492 return 0; /* always succesful for now, WILL CHANGE */
493 }
494
495 int
496 puffs_mainloop(struct puffs_usermount *pu)
497 {
498 struct puffs_getreq *pgr = NULL;
499 struct puffs_putreq *ppr = NULL;
500 struct puffs_framectrl *pfctrl = &pu->pu_framectrl;
501 struct puffs_fctrl_io *fio;
502 struct puffs_cc *pcc;
503 struct kevent *curev, *newevs;
504 size_t nchanges;
505 int puffsfd, sverrno;
506 int ndone;
507
508 assert(puffs_getstate(pu) >= PUFFS_STATE_RUNNING);
509
510 pgr = puffs_req_makeget(pu, puffs_getmaxreqlen(pu), 0);
511 if (pgr == NULL)
512 goto out;
513
514 ppr = puffs_req_makeput(pu);
515 if (ppr == NULL)
516 goto out;
517
518 newevs = realloc(pfctrl->evs, (2*pfctrl->nfds+1)*sizeof(struct kevent));
519 if (newevs == NULL)
520 goto out;
521 pfctrl->evs = newevs;
522
523 pu->pu_state |= PU_INLOOP;
524
525 pu->pu_kq = kqueue();
526 if (pu->pu_kq == -1)
527 goto out;
528 pu->pu_state |= PU_HASKQ;
529
530 curev = pfctrl->evs;
531 LIST_FOREACH(fio, &pfctrl->fb_ios, fio_entries) {
532 EV_SET(curev, fio->io_fd, EVFILT_READ, EV_ADD,
533 0, 0, (uintptr_t)fio);
534 curev++;
535 EV_SET(curev, fio->io_fd, EVFILT_WRITE, EV_ADD | EV_DISABLE,
536 0, 0, (uintptr_t)fio);
537 curev++;
538 }
539 puffsfd = puffs_getselectable(pu);
540 EV_SET(curev, puffsfd, EVFILT_READ, EV_ADD, 0, 0, 0);
541 if (kevent(pu->pu_kq, pfctrl->evs, 2*pfctrl->nfds+1,
542 NULL, 0, NULL) == -1)
543 goto out;
544
545 while (puffs_getstate(pu) != PUFFS_STATE_UNMOUNTED) {
546 if (pu->pu_ml_lfn)
547 pu->pu_ml_lfn(pu);
548
549 /*
550 * Do this here, because:
551 * a) loopfunc might generate some results
552 * b) it's still "after" event handling (except for round 1)
553 */
554 if (puffs_req_putput(ppr) == -1)
555 goto out;
556 puffs_req_resetput(ppr);
557
558 /* micro optimization: skip kevent syscall if possible */
559 if (pfctrl->nfds == 0 && pu->pu_ml_timep == NULL
560 && (pu->pu_state & PU_ASYNCFD) == 0) {
561 if (puffs_req_handle(pgr, ppr, 0) == -1)
562 goto out;
563 continue;
564 }
565
566 /* else: do full processing */
567
568 /*
569 * Build list of which to enable/disable in writecheck.
570 * Don't bother worrying about O(n) for now.
571 */
572 nchanges = 0;
573 LIST_FOREACH(fio, &pfctrl->fb_ios, fio_entries) {
574 if (fio->stat & FIO_WRGONE)
575 continue;
576
577 /*
578 * Try to write out everything to avoid the
579 * need for enabling EVFILT_WRITE. The likely
580 * case is that we can fit everything into the
581 * socket buffer.
582 */
583 if (puffs_framev_output(pu, pfctrl, fio, ppr)) {
584 /* need kernel notify? (error condition) */
585 if (puffs_req_putput(ppr) == -1)
586 goto out;
587 puffs_req_resetput(ppr);
588 }
589
590 /* en/disable write checks for kqueue as needed */
591 assert((FIO_EN_WRITE(fio) && FIO_RM_WRITE(fio)) == 0);
592 if (FIO_EN_WRITE(fio)) {
593 EV_SET(&pfctrl->evs[nchanges], fio->io_fd,
594 EVFILT_WRITE, EV_ENABLE, 0, 0,
595 (uintptr_t)fio);
596 fio->stat |= FIO_WR;
597 nchanges++;
598 }
599 if (FIO_RM_WRITE(fio)) {
600 EV_SET(&pfctrl->evs[nchanges], fio->io_fd,
601 EVFILT_WRITE, EV_DISABLE, 0, 0,
602 (uintptr_t)fio);
603 fio->stat &= ~FIO_WR;
604 nchanges++;
605 }
606 assert(nchanges <= pfctrl->nfds);
607 }
608
609 ndone = kevent(pu->pu_kq, pfctrl->evs, nchanges,
610 pfctrl->evs, pfctrl->nfds+1, pu->pu_ml_timep);
611
612 if (ndone == -1) {
613 if (errno != EINTR)
614 goto out;
615 else
616 continue;
617 }
618
619 /* uoptimize */
620 if (ndone == 0)
621 continue;
622
623 /* iterate over the results */
624 for (curev = pfctrl->evs; ndone--; curev++) {
625 int what;
626
627 /* get & possibly dispatch events from kernel */
628 if (curev->ident == puffsfd) {
629 if (puffs_req_handle(pgr, ppr, 0) == -1)
630 goto out;
631 continue;
632 }
633
634 fio = (void *)curev->udata;
635 if (curev->flags & EV_ERROR) {
636 assert(curev->filter == EVFILT_WRITE);
637 fio->stat &= ~FIO_WR;
638
639 /* XXX: how to know if it's a transient error */
640 puffs_framev_writeclose(pu, fio,
641 (int)curev->data);
642 puffs_framev_notify(fio, PUFFS_FBIO_ERROR);
643 continue;
644 }
645
646 what = 0;
647 if (curev->filter == EVFILT_READ) {
648 puffs_framev_input(pu, pfctrl, fio, ppr);
649 what |= PUFFS_FBIO_READ;
650 }
651
652 else if (curev->filter == EVFILT_WRITE) {
653 puffs_framev_output(pu, pfctrl, fio, ppr);
654 what |= PUFFS_FBIO_WRITE;
655 }
656 if (what)
657 puffs_framev_notify(fio, what);
658 }
659
660 /*
661 * Schedule continuations.
662 */
663 while ((pcc = TAILQ_FIRST(&pu->pu_sched)) != NULL) {
664 TAILQ_REMOVE(&pu->pu_sched, pcc, entries);
665 puffs_goto(pcc);
666 }
667
668 /*
669 * Really free fd's now that we don't have references
670 * to them.
671 */
672 while ((fio = LIST_FIRST(&pfctrl->fb_ios_rmlist)) != NULL) {
673 LIST_REMOVE(fio, fio_entries);
674 free(fio);
675 }
676 }
677 errno = 0;
678 puffs_req_putput(ppr);
679
680 out:
681 /* store the real error for a while */
682 sverrno = errno;
683
684 if (ppr)
685 puffs_req_destroyput(ppr);
686 if (pgr)
687 puffs_req_destroyget(pgr);
688
689 errno = sverrno;
690 if (errno)
691 return -1;
692 else
693 return 0;
694 }
695