1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
9 * met:
10 *
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following disclaimer
15 * in the documentation and/or other materials provided with the
16 * distribution.
17 * * Neither the name of Google Inc. nor the names of its
18 * contributors may be used to endorse or promote products derived from
19 * this software without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 * Copyright (C) 2005 Csaba Henk.
34 * All rights reserved.
35 *
36 * Copyright (c) 2019 The FreeBSD Foundation
37 *
38 * Portions of this software were developed by BFF Storage Systems, LLC under
39 * sponsorship from the FreeBSD Foundation.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
49 *
50 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60 * SUCH DAMAGE.
61 */
62
63 #include <sys/cdefs.h>
64 #include <sys/param.h>
65 #include <sys/module.h>
66 #include <sys/systm.h>
67 #include <sys/counter.h>
68 #include <sys/errno.h>
69 #include <sys/kernel.h>
70 #include <sys/conf.h>
71 #include <sys/uio.h>
72 #include <sys/malloc.h>
73 #include <sys/queue.h>
74 #include <sys/lock.h>
75 #include <sys/sx.h>
76 #include <sys/mutex.h>
77 #include <sys/proc.h>
78 #include <sys/mount.h>
79 #include <sys/sdt.h>
80 #include <sys/vnode.h>
81 #include <sys/signalvar.h>
82 #include <sys/syscallsubr.h>
83 #include <sys/sysctl.h>
84 #include <vm/uma.h>
85
86 #include "fuse.h"
87 #include "fuse_node.h"
88 #include "fuse_ipc.h"
89 #include "fuse_internal.h"
90
91 SDT_PROVIDER_DECLARE(fusefs);
92 /*
93 * Fuse trace probe:
94 * arg0: verbosity. Higher numbers give more verbose messages
95 * arg1: Textual message
96 */
97 SDT_PROBE_DEFINE2(fusefs, , ipc, trace, "int", "char*");
98
99 static void fdisp_make_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op,
100 struct fuse_data *data, uint64_t nid, pid_t pid, struct ucred *cred);
101 static void fuse_interrupt_send(struct fuse_ticket *otick, int err);
102 static struct fuse_ticket *fticket_alloc(struct fuse_data *data);
103 static void fticket_refresh(struct fuse_ticket *ftick);
104 static inline void fticket_reset(struct fuse_ticket *ftick);
105 static void fticket_destroy(struct fuse_ticket *ftick);
106 static int fticket_wait_answer(struct fuse_ticket *ftick);
107 static inline int
108 fticket_aw_pull_uio(struct fuse_ticket *ftick,
109 struct uio *uio);
110
111 static int fuse_body_audit(struct fuse_ticket *ftick, size_t blen);
112
113 static fuse_handler_t fuse_standard_handler;
114
115 static counter_u64_t fuse_ticket_count;
116 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, ticket_count, CTLFLAG_RD,
117 &fuse_ticket_count, "Number of allocated tickets");
118
119 static long fuse_iov_permanent_bufsize = 1 << 19;
120
121 SYSCTL_LONG(_vfs_fusefs, OID_AUTO, iov_permanent_bufsize, CTLFLAG_RW,
122 &fuse_iov_permanent_bufsize, 0,
123 "limit for permanently stored buffer size for fuse_iovs");
124 static int fuse_iov_credit = 16;
125
126 SYSCTL_INT(_vfs_fusefs, OID_AUTO, iov_credit, CTLFLAG_RW,
127 &fuse_iov_credit, 0,
128 "how many times is an oversized fuse_iov tolerated");
129
130 MALLOC_DEFINE(M_FUSEMSG, "fuse_msgbuf", "fuse message buffer");
131 static uma_zone_t ticket_zone;
132
133 /*
134 * TODO: figure out how to timeout INTERRUPT requests, because the daemon may
135 * leagally never respond
136 */
137 static int
fuse_interrupt_callback(struct fuse_ticket * tick,struct uio * uio)138 fuse_interrupt_callback(struct fuse_ticket *tick, struct uio *uio)
139 {
140 struct fuse_ticket *otick, *x_tick;
141 struct fuse_interrupt_in *fii;
142 struct fuse_data *data = tick->tk_data;
143 bool found = false;
144
145 fii = (struct fuse_interrupt_in*)((char*)tick->tk_ms_fiov.base +
146 sizeof(struct fuse_in_header));
147
148 fuse_lck_mtx_lock(data->aw_mtx);
149 TAILQ_FOREACH_SAFE(otick, &data->aw_head, tk_aw_link, x_tick) {
150 if (otick->tk_unique == fii->unique) {
151 found = true;
152 break;
153 }
154 }
155 fuse_lck_mtx_unlock(data->aw_mtx);
156
157 if (!found) {
158 /* Original is already complete. Just return */
159 return 0;
160 }
161
162 /* Clear the original ticket's interrupt association */
163 otick->irq_unique = 0;
164
165 if (tick->tk_aw_ohead.error == ENOSYS) {
166 fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
167 return 0;
168 } else if (tick->tk_aw_ohead.error == EAGAIN) {
169 /*
170 * There are two reasons we might get this:
171 * 1) the daemon received the INTERRUPT request before the
172 * original, or
173 * 2) the daemon received the INTERRUPT request after it
174 * completed the original request.
175 * In the first case we should re-send the INTERRUPT. In the
176 * second, we should ignore it.
177 */
178 /* Resend */
179 fuse_interrupt_send(otick, EINTR);
180 return 0;
181 } else {
182 /* Illegal FUSE_INTERRUPT response */
183 return EINVAL;
184 }
185 }
186
187 /* Interrupt the operation otick. Return err as its error code */
188 void
fuse_interrupt_send(struct fuse_ticket * otick,int err)189 fuse_interrupt_send(struct fuse_ticket *otick, int err)
190 {
191 struct fuse_dispatcher fdi;
192 struct fuse_interrupt_in *fii;
193 struct fuse_in_header *ftick_hdr;
194 struct fuse_data *data = otick->tk_data;
195 struct fuse_ticket *tick, *xtick;
196 struct ucred reused_creds;
197 gid_t reused_groups[1];
198
199 if (otick->irq_unique == 0) {
200 /*
201 * If the daemon hasn't yet received otick, then we can answer
202 * it ourselves and return.
203 */
204 fuse_lck_mtx_lock(data->ms_mtx);
205 STAILQ_FOREACH_SAFE(tick, &otick->tk_data->ms_head, tk_ms_link,
206 xtick) {
207 if (tick == otick) {
208 STAILQ_REMOVE(&otick->tk_data->ms_head, tick,
209 fuse_ticket, tk_ms_link);
210 otick->tk_data->ms_count--;
211 otick->tk_ms_link.stqe_next = NULL;
212 fuse_lck_mtx_unlock(data->ms_mtx);
213
214 fuse_lck_mtx_lock(otick->tk_aw_mtx);
215 if (!fticket_answered(otick)) {
216 fticket_set_answered(otick);
217 otick->tk_aw_errno = err;
218 wakeup(otick);
219 }
220 fuse_lck_mtx_unlock(otick->tk_aw_mtx);
221
222 fuse_ticket_drop(tick);
223 return;
224 }
225 }
226 fuse_lck_mtx_unlock(data->ms_mtx);
227
228 /*
229 * If the fuse daemon doesn't support interrupts, then there's
230 * nothing more that we can do
231 */
232 if (fsess_not_impl(data->mp, FUSE_INTERRUPT))
233 return;
234
235 /*
236 * If the fuse daemon has already received otick, then we must
237 * send FUSE_INTERRUPT.
238 */
239 ftick_hdr = fticket_in_header(otick);
240 reused_creds.cr_uid = ftick_hdr->uid;
241 reused_groups[0] = ftick_hdr->gid;
242 reused_creds.cr_groups = reused_groups;
243 fdisp_init(&fdi, sizeof(*fii));
244 fdisp_make_pid(&fdi, FUSE_INTERRUPT, data, ftick_hdr->nodeid,
245 ftick_hdr->pid, &reused_creds);
246
247 fii = fdi.indata;
248 fii->unique = otick->tk_unique;
249 fuse_insert_callback(fdi.tick, fuse_interrupt_callback);
250
251 otick->irq_unique = fdi.tick->tk_unique;
252 /* Interrupt ops should be delivered ASAP */
253 fuse_insert_message(fdi.tick, true);
254 fdisp_destroy(&fdi);
255 } else {
256 /* This ticket has already been interrupted */
257 }
258 }
259
260 void
fiov_init(struct fuse_iov * fiov,size_t size)261 fiov_init(struct fuse_iov *fiov, size_t size)
262 {
263 uint32_t msize = FU_AT_LEAST(size);
264
265 fiov->len = 0;
266
267 fiov->base = malloc(msize, M_FUSEMSG, M_WAITOK | M_ZERO);
268
269 fiov->allocated_size = msize;
270 fiov->credit = fuse_iov_credit;
271 }
272
273 void
fiov_teardown(struct fuse_iov * fiov)274 fiov_teardown(struct fuse_iov *fiov)
275 {
276 MPASS(fiov->base != NULL);
277 free(fiov->base, M_FUSEMSG);
278 }
279
280 void
fiov_adjust(struct fuse_iov * fiov,size_t size)281 fiov_adjust(struct fuse_iov *fiov, size_t size)
282 {
283 if (fiov->allocated_size < size ||
284 (fuse_iov_permanent_bufsize >= 0 &&
285 fiov->allocated_size - size > fuse_iov_permanent_bufsize &&
286 --fiov->credit < 0)) {
287 fiov->base = realloc(fiov->base, FU_AT_LEAST(size), M_FUSEMSG,
288 M_WAITOK | M_ZERO);
289 if (!fiov->base) {
290 panic("FUSE: realloc failed");
291 }
292 fiov->allocated_size = FU_AT_LEAST(size);
293 fiov->credit = fuse_iov_credit;
294 /* Clear data buffer after reallocation */
295 bzero(fiov->base, size);
296 } else if (size > fiov->len) {
297 /* Clear newly extended portion of data buffer */
298 bzero((char*)fiov->base + fiov->len, size - fiov->len);
299 }
300 fiov->len = size;
301 }
302
303 /* Resize the fiov if needed, and clear it's buffer */
304 void
fiov_refresh(struct fuse_iov * fiov)305 fiov_refresh(struct fuse_iov *fiov)
306 {
307 fiov_adjust(fiov, 0);
308 }
309
310 static int
fticket_ctor(void * mem,int size,void * arg,int flags)311 fticket_ctor(void *mem, int size, void *arg, int flags)
312 {
313 struct fuse_ticket *ftick = mem;
314 struct fuse_data *data = arg;
315
316 FUSE_ASSERT_MS_DONE(ftick);
317 FUSE_ASSERT_AW_DONE(ftick);
318
319 ftick->tk_data = data;
320 ftick->irq_unique = 0;
321 refcount_init(&ftick->tk_refcount, 1);
322 counter_u64_add(fuse_ticket_count, 1);
323
324 fticket_refresh(ftick);
325
326 return 0;
327 }
328
329 static void
fticket_dtor(void * mem,int size,void * arg)330 fticket_dtor(void *mem, int size, void *arg)
331 {
332 #ifdef INVARIANTS
333 struct fuse_ticket *ftick = mem;
334 #endif
335
336 FUSE_ASSERT_MS_DONE(ftick);
337 FUSE_ASSERT_AW_DONE(ftick);
338
339 counter_u64_add(fuse_ticket_count, -1);
340 }
341
342 static int
fticket_init(void * mem,int size,int flags)343 fticket_init(void *mem, int size, int flags)
344 {
345 struct fuse_ticket *ftick = mem;
346
347 bzero(ftick, sizeof(struct fuse_ticket));
348
349 fiov_init(&ftick->tk_ms_fiov, sizeof(struct fuse_in_header));
350
351 mtx_init(&ftick->tk_aw_mtx, "fuse answer delivery mutex", NULL, MTX_DEF);
352 fiov_init(&ftick->tk_aw_fiov, 0);
353
354 return 0;
355 }
356
357 static void
fticket_fini(void * mem,int size)358 fticket_fini(void *mem, int size)
359 {
360 struct fuse_ticket *ftick = mem;
361
362 fiov_teardown(&ftick->tk_ms_fiov);
363 fiov_teardown(&ftick->tk_aw_fiov);
364 mtx_destroy(&ftick->tk_aw_mtx);
365 }
366
367 static inline struct fuse_ticket *
fticket_alloc(struct fuse_data * data)368 fticket_alloc(struct fuse_data *data)
369 {
370 return uma_zalloc_arg(ticket_zone, data, M_WAITOK);
371 }
372
373 static inline void
fticket_destroy(struct fuse_ticket * ftick)374 fticket_destroy(struct fuse_ticket *ftick)
375 {
376 return uma_zfree(ticket_zone, ftick);
377 }
378
379 /* Prepare the ticket to be reused and clear its data buffers */
380 static inline void
fticket_refresh(struct fuse_ticket * ftick)381 fticket_refresh(struct fuse_ticket *ftick)
382 {
383 fticket_reset(ftick);
384
385 fiov_refresh(&ftick->tk_ms_fiov);
386 fiov_refresh(&ftick->tk_aw_fiov);
387 }
388
389 /* Prepare the ticket to be reused, but don't clear its data buffers */
390 static inline void
fticket_reset(struct fuse_ticket * ftick)391 fticket_reset(struct fuse_ticket *ftick)
392 {
393 struct fuse_data *data = ftick->tk_data;
394
395 FUSE_ASSERT_MS_DONE(ftick);
396 FUSE_ASSERT_AW_DONE(ftick);
397
398 bzero(&ftick->tk_aw_ohead, sizeof(struct fuse_out_header));
399
400 ftick->tk_aw_errno = 0;
401 ftick->tk_flag = 0;
402
403 /* May be truncated to 32 bits on LP32 arches */
404 ftick->tk_unique = atomic_fetchadd_long(&data->ticketer, 1);
405 if (ftick->tk_unique == 0)
406 ftick->tk_unique = atomic_fetchadd_long(&data->ticketer, 1);
407 }
408
409 static int
fticket_wait_answer(struct fuse_ticket * ftick)410 fticket_wait_answer(struct fuse_ticket *ftick)
411 {
412 struct thread *td = curthread;
413 sigset_t blockedset, oldset;
414 int err = 0, stops_deferred;
415 struct fuse_data *data = ftick->tk_data;
416 bool interrupted = false;
417
418 if (fsess_maybe_impl(ftick->tk_data->mp, FUSE_INTERRUPT) &&
419 data->dataflags & FSESS_INTR) {
420 SIGEMPTYSET(blockedset);
421 } else {
422 /* Block all signals except (implicitly) SIGKILL */
423 SIGFILLSET(blockedset);
424 }
425 stops_deferred = sigdeferstop(SIGDEFERSTOP_SILENT);
426 kern_sigprocmask(td, SIG_BLOCK, NULL, &oldset, 0);
427
428 fuse_lck_mtx_lock(ftick->tk_aw_mtx);
429
430 retry:
431 if (fticket_answered(ftick)) {
432 goto out;
433 }
434
435 if (fdata_get_dead(data)) {
436 err = ENOTCONN;
437 fticket_set_answered(ftick);
438 goto out;
439 }
440 kern_sigprocmask(td, SIG_BLOCK, &blockedset, NULL, 0);
441 err = msleep(ftick, &ftick->tk_aw_mtx, PCATCH, "fu_ans",
442 data->daemon_timeout * hz);
443 kern_sigprocmask(td, SIG_SETMASK, &oldset, NULL, 0);
444 if (err == EWOULDBLOCK) {
445 SDT_PROBE2(fusefs, , ipc, trace, 3,
446 "fticket_wait_answer: EWOULDBLOCK");
447 #ifdef XXXIP /* die conditionally */
448 if (!fdata_get_dead(data)) {
449 fdata_set_dead(data);
450 }
451 #endif
452 err = ETIMEDOUT;
453 fticket_set_answered(ftick);
454 } else if ((err == EINTR || err == ERESTART)) {
455 /*
456 * Whether we get EINTR or ERESTART depends on whether
457 * SA_RESTART was set by sigaction(2).
458 *
459 * Try to interrupt the operation and wait for an EINTR response
460 * to the original operation. If the file system does not
461 * support FUSE_INTERRUPT, then we'll just wait for it to
462 * complete like normal. If it does support FUSE_INTERRUPT,
463 * then it will either respond EINTR to the original operation,
464 * or EAGAIN to the interrupt.
465 */
466 sigset_t tmpset;
467
468 SDT_PROBE2(fusefs, , ipc, trace, 4,
469 "fticket_wait_answer: interrupt");
470 fuse_lck_mtx_unlock(ftick->tk_aw_mtx);
471 fuse_interrupt_send(ftick, err);
472
473 PROC_LOCK(td->td_proc);
474 mtx_lock(&td->td_proc->p_sigacts->ps_mtx);
475 tmpset = td->td_proc->p_siglist;
476 SIGSETOR(tmpset, td->td_siglist);
477 mtx_unlock(&td->td_proc->p_sigacts->ps_mtx);
478 PROC_UNLOCK(td->td_proc);
479
480 fuse_lck_mtx_lock(ftick->tk_aw_mtx);
481 if (!interrupted && !SIGISMEMBER(tmpset, SIGKILL)) {
482 /*
483 * Block all signals while we wait for an interrupt
484 * response. The protocol doesn't discriminate between
485 * different signals.
486 */
487 SIGFILLSET(blockedset);
488 interrupted = true;
489 goto retry;
490 } else {
491 /*
492 * Return immediately for fatal signals, or if this is
493 * the second interruption. We should only be
494 * interrupted twice if the thread is stopped, for
495 * example during sigexit.
496 */
497 }
498 } else if (err) {
499 SDT_PROBE2(fusefs, , ipc, trace, 6,
500 "fticket_wait_answer: other error");
501 } else {
502 SDT_PROBE2(fusefs, , ipc, trace, 7, "fticket_wait_answer: OK");
503 }
504 out:
505 if (!(err || fticket_answered(ftick))) {
506 SDT_PROBE2(fusefs, , ipc, trace, 1,
507 "FUSE: requester was woken up but still no answer");
508 err = ENXIO;
509 }
510 fuse_lck_mtx_unlock(ftick->tk_aw_mtx);
511 sigallowstop(stops_deferred);
512
513 return err;
514 }
515
516 static inline
517 int
fticket_aw_pull_uio(struct fuse_ticket * ftick,struct uio * uio)518 fticket_aw_pull_uio(struct fuse_ticket *ftick, struct uio *uio)
519 {
520 int err = 0;
521 size_t len = uio_resid(uio);
522
523 if (len) {
524 fiov_adjust(fticket_resp(ftick), len);
525 err = uiomove(fticket_resp(ftick)->base, len, uio);
526 }
527 return err;
528 }
529
530 int
fticket_pull(struct fuse_ticket * ftick,struct uio * uio)531 fticket_pull(struct fuse_ticket *ftick, struct uio *uio)
532 {
533 int err = 0;
534
535 if (ftick->tk_aw_ohead.error) {
536 return 0;
537 }
538 err = fuse_body_audit(ftick, uio_resid(uio));
539 if (!err) {
540 err = fticket_aw_pull_uio(ftick, uio);
541 }
542 return err;
543 }
544
545 struct fuse_data *
fdata_alloc(struct cdev * fdev,struct ucred * cred)546 fdata_alloc(struct cdev *fdev, struct ucred *cred)
547 {
548 struct fuse_data *data;
549
550 data = malloc(sizeof(struct fuse_data), M_FUSEMSG, M_WAITOK | M_ZERO);
551
552 data->fdev = fdev;
553 mtx_init(&data->ms_mtx, "fuse message list mutex", NULL, MTX_DEF);
554 STAILQ_INIT(&data->ms_head);
555 data->ms_count = 0;
556 knlist_init_mtx(&data->ks_rsel.si_note, &data->ms_mtx);
557 mtx_init(&data->aw_mtx, "fuse answer list mutex", NULL, MTX_DEF);
558 TAILQ_INIT(&data->aw_head);
559 data->daemoncred = crhold(cred);
560 data->daemon_timeout = FUSE_DEFAULT_DAEMON_TIMEOUT;
561 sx_init(&data->rename_lock, "fuse rename lock");
562 data->ref = 1;
563
564 return data;
565 }
566
567 void
fdata_trydestroy(struct fuse_data * data)568 fdata_trydestroy(struct fuse_data *data)
569 {
570 data->ref--;
571 MPASS(data->ref >= 0);
572 if (data->ref != 0)
573 return;
574
575 /* Driving off stage all that stuff thrown at device... */
576 sx_destroy(&data->rename_lock);
577 crfree(data->daemoncred);
578 mtx_destroy(&data->aw_mtx);
579 knlist_delete(&data->ks_rsel.si_note, curthread, 0);
580 knlist_destroy(&data->ks_rsel.si_note);
581 mtx_destroy(&data->ms_mtx);
582
583 free(data, M_FUSEMSG);
584 }
585
586 void
fdata_set_dead(struct fuse_data * data)587 fdata_set_dead(struct fuse_data *data)
588 {
589 FUSE_LOCK();
590 if (fdata_get_dead(data)) {
591 FUSE_UNLOCK();
592 return;
593 }
594 fuse_lck_mtx_lock(data->ms_mtx);
595 data->dataflags |= FSESS_DEAD;
596 wakeup_one(data);
597 selwakeuppri(&data->ks_rsel, PZERO + 1);
598 wakeup(&data->ticketer);
599 fuse_lck_mtx_unlock(data->ms_mtx);
600 FUSE_UNLOCK();
601 }
602
603 struct fuse_ticket *
fuse_ticket_fetch(struct fuse_data * data)604 fuse_ticket_fetch(struct fuse_data *data)
605 {
606 int err = 0;
607 struct fuse_ticket *ftick;
608
609 ftick = fticket_alloc(data);
610
611 if (!(data->dataflags & FSESS_INITED)) {
612 /* Sleep until get answer for INIT messsage */
613 FUSE_LOCK();
614 if (!(data->dataflags & FSESS_INITED) && data->ticketer > 2) {
615 err = msleep(&data->ticketer, &fuse_mtx, PCATCH | PDROP,
616 "fu_ini", 0);
617 if (err)
618 fdata_set_dead(data);
619 } else
620 FUSE_UNLOCK();
621 }
622 return ftick;
623 }
624
625 int
fuse_ticket_drop(struct fuse_ticket * ftick)626 fuse_ticket_drop(struct fuse_ticket *ftick)
627 {
628 int die;
629
630 die = refcount_release(&ftick->tk_refcount);
631 if (die)
632 fticket_destroy(ftick);
633
634 return die;
635 }
636
637 void
fuse_insert_callback(struct fuse_ticket * ftick,fuse_handler_t * handler)638 fuse_insert_callback(struct fuse_ticket *ftick, fuse_handler_t * handler)
639 {
640 if (fdata_get_dead(ftick->tk_data)) {
641 return;
642 }
643 ftick->tk_aw_handler = handler;
644
645 fuse_lck_mtx_lock(ftick->tk_data->aw_mtx);
646 fuse_aw_push(ftick);
647 fuse_lck_mtx_unlock(ftick->tk_data->aw_mtx);
648 }
649
650 /*
651 * Insert a new upgoing ticket into the message queue
652 *
653 * If urgent is true, insert at the front of the queue. Otherwise, insert in
654 * FIFO order.
655 */
656 void
fuse_insert_message(struct fuse_ticket * ftick,bool urgent)657 fuse_insert_message(struct fuse_ticket *ftick, bool urgent)
658 {
659 if (ftick->tk_flag & FT_DIRTY) {
660 panic("FUSE: ticket reused without being refreshed");
661 }
662 ftick->tk_flag |= FT_DIRTY;
663
664 if (fdata_get_dead(ftick->tk_data)) {
665 return;
666 }
667 fuse_lck_mtx_lock(ftick->tk_data->ms_mtx);
668 if (urgent)
669 fuse_ms_push_head(ftick);
670 else
671 fuse_ms_push(ftick);
672 wakeup_one(ftick->tk_data);
673 selwakeuppri(&ftick->tk_data->ks_rsel, PZERO + 1);
674 KNOTE_LOCKED(&ftick->tk_data->ks_rsel.si_note, 0);
675 fuse_lck_mtx_unlock(ftick->tk_data->ms_mtx);
676 }
677
678 static int
fuse_body_audit(struct fuse_ticket * ftick,size_t blen)679 fuse_body_audit(struct fuse_ticket *ftick, size_t blen)
680 {
681 int err = 0;
682 enum fuse_opcode opcode;
683
684 opcode = fticket_opcode(ftick);
685
686 switch (opcode) {
687 case FUSE_BMAP:
688 err = (blen == sizeof(struct fuse_bmap_out)) ? 0 : EINVAL;
689 break;
690
691 case FUSE_LINK:
692 case FUSE_LOOKUP:
693 case FUSE_MKDIR:
694 case FUSE_MKNOD:
695 case FUSE_SYMLINK:
696 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) {
697 err = (blen == sizeof(struct fuse_entry_out)) ?
698 0 : EINVAL;
699 } else {
700 err = (blen == FUSE_COMPAT_ENTRY_OUT_SIZE) ? 0 : EINVAL;
701 }
702 break;
703
704 case FUSE_FORGET:
705 panic("FUSE: a handler has been intalled for FUSE_FORGET");
706 break;
707
708 case FUSE_GETATTR:
709 case FUSE_SETATTR:
710 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) {
711 err = (blen == sizeof(struct fuse_attr_out)) ?
712 0 : EINVAL;
713 } else {
714 err = (blen == FUSE_COMPAT_ATTR_OUT_SIZE) ? 0 : EINVAL;
715 }
716 break;
717
718 case FUSE_READLINK:
719 err = (PAGE_SIZE >= blen) ? 0 : EINVAL;
720 break;
721
722 case FUSE_UNLINK:
723 err = (blen == 0) ? 0 : EINVAL;
724 break;
725
726 case FUSE_RMDIR:
727 err = (blen == 0) ? 0 : EINVAL;
728 break;
729
730 case FUSE_RENAME:
731 err = (blen == 0) ? 0 : EINVAL;
732 break;
733
734 case FUSE_OPEN:
735 err = (blen == sizeof(struct fuse_open_out)) ? 0 : EINVAL;
736 break;
737
738 case FUSE_READ:
739 err = (((struct fuse_read_in *)(
740 (char *)ftick->tk_ms_fiov.base +
741 sizeof(struct fuse_in_header)
742 ))->size >= blen) ? 0 : EINVAL;
743 break;
744
745 case FUSE_WRITE:
746 err = (blen == sizeof(struct fuse_write_out)) ? 0 : EINVAL;
747 break;
748
749 case FUSE_STATFS:
750 if (fuse_libabi_geq(ftick->tk_data, 7, 4)) {
751 err = (blen == sizeof(struct fuse_statfs_out)) ?
752 0 : EINVAL;
753 } else {
754 err = (blen == FUSE_COMPAT_STATFS_SIZE) ? 0 : EINVAL;
755 }
756 break;
757
758 case FUSE_RELEASE:
759 err = (blen == 0) ? 0 : EINVAL;
760 break;
761
762 case FUSE_FSYNC:
763 err = (blen == 0) ? 0 : EINVAL;
764 break;
765
766 case FUSE_SETXATTR:
767 err = (blen == 0) ? 0 : EINVAL;
768 break;
769
770 case FUSE_GETXATTR:
771 case FUSE_LISTXATTR:
772 /*
773 * These can have varying response lengths, and 0 length
774 * isn't necessarily invalid.
775 */
776 err = 0;
777 break;
778
779 case FUSE_REMOVEXATTR:
780 err = (blen == 0) ? 0 : EINVAL;
781 break;
782
783 case FUSE_FLUSH:
784 err = (blen == 0) ? 0 : EINVAL;
785 break;
786
787 case FUSE_INIT:
788 if (blen == sizeof(struct fuse_init_out) ||
789 blen == FUSE_COMPAT_INIT_OUT_SIZE ||
790 blen == FUSE_COMPAT_22_INIT_OUT_SIZE) {
791 err = 0;
792 } else {
793 err = EINVAL;
794 }
795 break;
796
797 case FUSE_OPENDIR:
798 err = (blen == sizeof(struct fuse_open_out)) ? 0 : EINVAL;
799 break;
800
801 case FUSE_READDIR:
802 err = (((struct fuse_read_in *)(
803 (char *)ftick->tk_ms_fiov.base +
804 sizeof(struct fuse_in_header)
805 ))->size >= blen) ? 0 : EINVAL;
806 break;
807
808 case FUSE_RELEASEDIR:
809 err = (blen == 0) ? 0 : EINVAL;
810 break;
811
812 case FUSE_FSYNCDIR:
813 err = (blen == 0) ? 0 : EINVAL;
814 break;
815
816 case FUSE_GETLK:
817 err = (blen == sizeof(struct fuse_lk_out)) ? 0 : EINVAL;
818 break;
819
820 case FUSE_SETLK:
821 err = (blen == 0) ? 0 : EINVAL;
822 break;
823
824 case FUSE_SETLKW:
825 err = (blen == 0) ? 0 : EINVAL;
826 break;
827
828 case FUSE_ACCESS:
829 err = (blen == 0) ? 0 : EINVAL;
830 break;
831
832 case FUSE_CREATE:
833 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) {
834 err = (blen == sizeof(struct fuse_entry_out) +
835 sizeof(struct fuse_open_out)) ? 0 : EINVAL;
836 } else {
837 err = (blen == FUSE_COMPAT_ENTRY_OUT_SIZE +
838 sizeof(struct fuse_open_out)) ? 0 : EINVAL;
839 }
840 break;
841
842 case FUSE_DESTROY:
843 err = (blen == 0) ? 0 : EINVAL;
844 break;
845
846 case FUSE_FALLOCATE:
847 err = (blen == 0) ? 0 : EINVAL;
848 break;
849
850 case FUSE_LSEEK:
851 err = (blen == sizeof(struct fuse_lseek_out)) ? 0 : EINVAL;
852 break;
853
854 case FUSE_COPY_FILE_RANGE:
855 err = (blen == sizeof(struct fuse_write_out)) ? 0 : EINVAL;
856 break;
857
858 default:
859 panic("FUSE: opcodes out of sync (%d)\n", opcode);
860 }
861
862 return err;
863 }
864
865 static inline void
fuse_setup_ihead(struct fuse_in_header * ihead,struct fuse_ticket * ftick,uint64_t nid,enum fuse_opcode op,size_t blen,pid_t pid,struct ucred * cred)866 fuse_setup_ihead(struct fuse_in_header *ihead, struct fuse_ticket *ftick,
867 uint64_t nid, enum fuse_opcode op, size_t blen, pid_t pid,
868 struct ucred *cred)
869 {
870 ihead->len = sizeof(*ihead) + blen;
871 ihead->unique = ftick->tk_unique;
872 ihead->nodeid = nid;
873 ihead->opcode = op;
874
875 ihead->pid = pid;
876 ihead->uid = cred->cr_uid;
877 ihead->gid = cred->cr_groups[0];
878 }
879
880 /*
881 * fuse_standard_handler just pulls indata and wakes up pretender.
882 * Doesn't try to interpret data, that's left for the pretender.
883 * Though might do a basic size verification before the pull-in takes place
884 */
885
886 static int
fuse_standard_handler(struct fuse_ticket * ftick,struct uio * uio)887 fuse_standard_handler(struct fuse_ticket *ftick, struct uio *uio)
888 {
889 int err = 0;
890
891 err = fticket_pull(ftick, uio);
892
893 fuse_lck_mtx_lock(ftick->tk_aw_mtx);
894
895 if (!fticket_answered(ftick)) {
896 fticket_set_answered(ftick);
897 ftick->tk_aw_errno = err;
898 wakeup(ftick);
899 }
900 fuse_lck_mtx_unlock(ftick->tk_aw_mtx);
901
902 return err;
903 }
904
905 /*
906 * Reinitialize a dispatcher from a pid and node id, without resizing or
907 * clearing its data buffers
908 */
909 static void
fdisp_refresh_pid(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct mount * mp,uint64_t nid,pid_t pid,struct ucred * cred)910 fdisp_refresh_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op,
911 struct mount *mp, uint64_t nid, pid_t pid, struct ucred *cred)
912 {
913 MPASS(fdip->tick);
914 MPASS2(sizeof(fdip->finh) + fdip->iosize <= fdip->tick->tk_ms_fiov.len,
915 "Must use fdisp_make_pid to increase the size of the fiov");
916 fticket_reset(fdip->tick);
917
918 FUSE_DIMALLOC(&fdip->tick->tk_ms_fiov, fdip->finh,
919 fdip->indata, fdip->iosize);
920
921 fuse_setup_ihead(fdip->finh, fdip->tick, nid, op, fdip->iosize, pid,
922 cred);
923 }
924
925 /* Initialize a dispatcher from a pid and node id */
926 static void
fdisp_make_pid(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct fuse_data * data,uint64_t nid,pid_t pid,struct ucred * cred)927 fdisp_make_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op,
928 struct fuse_data *data, uint64_t nid, pid_t pid, struct ucred *cred)
929 {
930 if (fdip->tick) {
931 fticket_refresh(fdip->tick);
932 } else {
933 fdip->tick = fuse_ticket_fetch(data);
934 }
935
936 /* FUSE_DIMALLOC will bzero the fiovs when it enlarges them */
937 FUSE_DIMALLOC(&fdip->tick->tk_ms_fiov, fdip->finh,
938 fdip->indata, fdip->iosize);
939
940 fuse_setup_ihead(fdip->finh, fdip->tick, nid, op, fdip->iosize, pid, cred);
941 }
942
943 void
fdisp_make(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct mount * mp,uint64_t nid,struct thread * td,struct ucred * cred)944 fdisp_make(struct fuse_dispatcher *fdip, enum fuse_opcode op, struct mount *mp,
945 uint64_t nid, struct thread *td, struct ucred *cred)
946 {
947 struct fuse_data *data = fuse_get_mpdata(mp);
948 RECTIFY_TDCR(td, cred);
949
950 return fdisp_make_pid(fdip, op, data, nid, td->td_proc->p_pid, cred);
951 }
952
953 void
fdisp_make_vp(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct vnode * vp,struct thread * td,struct ucred * cred)954 fdisp_make_vp(struct fuse_dispatcher *fdip, enum fuse_opcode op,
955 struct vnode *vp, struct thread *td, struct ucred *cred)
956 {
957 struct mount *mp = vnode_mount(vp);
958 struct fuse_data *data = fuse_get_mpdata(mp);
959
960 RECTIFY_TDCR(td, cred);
961 return fdisp_make_pid(fdip, op, data, VTOI(vp),
962 td->td_proc->p_pid, cred);
963 }
964
965 /* Refresh a fuse_dispatcher so it can be reused, but don't zero its data */
966 void
fdisp_refresh_vp(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct vnode * vp,struct thread * td,struct ucred * cred)967 fdisp_refresh_vp(struct fuse_dispatcher *fdip, enum fuse_opcode op,
968 struct vnode *vp, struct thread *td, struct ucred *cred)
969 {
970 RECTIFY_TDCR(td, cred);
971 return fdisp_refresh_pid(fdip, op, vnode_mount(vp), VTOI(vp),
972 td->td_proc->p_pid, cred);
973 }
974
975 SDT_PROBE_DEFINE2(fusefs, , ipc, fdisp_wait_answ_error, "char*", "int");
976
977 int
fdisp_wait_answ(struct fuse_dispatcher * fdip)978 fdisp_wait_answ(struct fuse_dispatcher *fdip)
979 {
980 int err = 0;
981
982 fdip->answ_stat = 0;
983 fuse_insert_callback(fdip->tick, fuse_standard_handler);
984 fuse_insert_message(fdip->tick, false);
985
986 if ((err = fticket_wait_answer(fdip->tick))) {
987 fuse_lck_mtx_lock(fdip->tick->tk_aw_mtx);
988
989 if (fticket_answered(fdip->tick)) {
990 /*
991 * Just between noticing the interrupt and getting here,
992 * the standard handler has completed his job.
993 * So we drop the ticket and exit as usual.
994 */
995 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
996 "IPC: interrupted, already answered", err);
997 fuse_lck_mtx_unlock(fdip->tick->tk_aw_mtx);
998 goto out;
999 } else {
1000 /*
1001 * So we were faster than the standard handler.
1002 * Then by setting the answered flag we get *him*
1003 * to drop the ticket.
1004 */
1005 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1006 "IPC: interrupted, setting to answered", err);
1007 fticket_set_answered(fdip->tick);
1008 fuse_lck_mtx_unlock(fdip->tick->tk_aw_mtx);
1009 return err;
1010 }
1011 }
1012
1013 if (fdip->tick->tk_aw_errno == ENOTCONN) {
1014 /* The daemon died while we were waiting for a response */
1015 err = ENOTCONN;
1016 goto out;
1017 } else if (fdip->tick->tk_aw_errno) {
1018 /*
1019 * There was some sort of communication error with the daemon
1020 * that the client wouldn't understand.
1021 */
1022 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1023 "IPC: explicit EIO-ing", fdip->tick->tk_aw_errno);
1024 err = EIO;
1025 goto out;
1026 }
1027 if ((err = fdip->tick->tk_aw_ohead.error)) {
1028 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1029 "IPC: setting status", fdip->tick->tk_aw_ohead.error);
1030 /*
1031 * This means a "proper" fuse syscall error.
1032 * We record this value so the caller will
1033 * be able to know it's not a boring messaging
1034 * failure, if she wishes so (and if not, she can
1035 * just simply propagate the return value of this routine).
1036 * [XXX Maybe a bitflag would do the job too,
1037 * if other flags needed, this will be converted thusly.]
1038 */
1039 fdip->answ_stat = err;
1040 goto out;
1041 }
1042 fdip->answ = fticket_resp(fdip->tick)->base;
1043 fdip->iosize = fticket_resp(fdip->tick)->len;
1044
1045 return 0;
1046
1047 out:
1048 return err;
1049 }
1050
1051 void
fuse_ipc_init(void)1052 fuse_ipc_init(void)
1053 {
1054 ticket_zone = uma_zcreate("fuse_ticket", sizeof(struct fuse_ticket),
1055 fticket_ctor, fticket_dtor, fticket_init, fticket_fini,
1056 UMA_ALIGN_PTR, 0);
1057 fuse_ticket_count = counter_u64_alloc(M_WAITOK);
1058 }
1059
1060 void
fuse_ipc_destroy(void)1061 fuse_ipc_destroy(void)
1062 {
1063 counter_u64_free(fuse_ticket_count);
1064 uma_zdestroy(ticket_zone);
1065 }
1066
1067 SDT_PROBE_DEFINE3(fusefs,, ipc, warn, "struct fuse_data*", "unsigned", "char*");
1068 void
fuse_warn(struct fuse_data * data,unsigned flag,const char * msg)1069 fuse_warn(struct fuse_data *data, unsigned flag, const char *msg)
1070 {
1071 SDT_PROBE3(fusefs, , ipc, warn, data, flag, msg);
1072 if (!(data->dataflags & flag)) {
1073 printf("WARNING: FUSE protocol violation for server mounted at "
1074 "%s: %s "
1075 "This warning will not be repeated.\n",
1076 data->mp->mnt_stat.f_mntonname, msg);
1077 data->dataflags |= flag;
1078 }
1079 }
1080