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 __FBSDID("$FreeBSD$");
65
66 #include <sys/param.h>
67 #include <sys/module.h>
68 #include <sys/systm.h>
69 #include <sys/counter.h>
70 #include <sys/errno.h>
71 #include <sys/kernel.h>
72 #include <sys/conf.h>
73 #include <sys/uio.h>
74 #include <sys/malloc.h>
75 #include <sys/queue.h>
76 #include <sys/lock.h>
77 #include <sys/sx.h>
78 #include <sys/mutex.h>
79 #include <sys/proc.h>
80 #include <sys/mount.h>
81 #include <sys/sdt.h>
82 #include <sys/vnode.h>
83 #include <sys/signalvar.h>
84 #include <sys/syscallsubr.h>
85 #include <sys/sysctl.h>
86 #include <vm/uma.h>
87
88 #include "fuse.h"
89 #include "fuse_node.h"
90 #include "fuse_ipc.h"
91 #include "fuse_internal.h"
92
93 SDT_PROVIDER_DECLARE(fusefs);
94 /*
95 * Fuse trace probe:
96 * arg0: verbosity. Higher numbers give more verbose messages
97 * arg1: Textual message
98 */
99 SDT_PROBE_DEFINE2(fusefs, , ipc, trace, "int", "char*");
100
101 static void fdisp_make_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op,
102 struct fuse_data *data, uint64_t nid, pid_t pid, struct ucred *cred);
103 static void fuse_interrupt_send(struct fuse_ticket *otick, int err);
104 static struct fuse_ticket *fticket_alloc(struct fuse_data *data);
105 static void fticket_refresh(struct fuse_ticket *ftick);
106 static void fticket_destroy(struct fuse_ticket *ftick);
107 static int fticket_wait_answer(struct fuse_ticket *ftick);
108 static inline int
109 fticket_aw_pull_uio(struct fuse_ticket *ftick,
110 struct uio *uio);
111
112 static int fuse_body_audit(struct fuse_ticket *ftick, size_t blen);
113
114 static fuse_handler_t fuse_standard_handler;
115
116 static counter_u64_t fuse_ticket_count;
117 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, ticket_count, CTLFLAG_RD,
118 &fuse_ticket_count, "Number of allocated tickets");
119
120 static long fuse_iov_permanent_bufsize = 1 << 19;
121
122 SYSCTL_LONG(_vfs_fusefs, OID_AUTO, iov_permanent_bufsize, CTLFLAG_RW,
123 &fuse_iov_permanent_bufsize, 0,
124 "limit for permanently stored buffer size for fuse_iovs");
125 static int fuse_iov_credit = 16;
126
127 SYSCTL_INT(_vfs_fusefs, OID_AUTO, iov_credit, CTLFLAG_RW,
128 &fuse_iov_credit, 0,
129 "how many times is an oversized fuse_iov tolerated");
130
131 MALLOC_DEFINE(M_FUSEMSG, "fuse_msgbuf", "fuse message buffer");
132 static uma_zone_t ticket_zone;
133
134 /*
135 * TODO: figure out how to timeout INTERRUPT requests, because the daemon may
136 * leagally never respond
137 */
138 static int
fuse_interrupt_callback(struct fuse_ticket * tick,struct uio * uio)139 fuse_interrupt_callback(struct fuse_ticket *tick, struct uio *uio)
140 {
141 struct fuse_ticket *otick, *x_tick;
142 struct fuse_interrupt_in *fii;
143 struct fuse_data *data = tick->tk_data;
144 bool found = false;
145
146 fii = (struct fuse_interrupt_in*)((char*)tick->tk_ms_fiov.base +
147 sizeof(struct fuse_in_header));
148
149 fuse_lck_mtx_lock(data->aw_mtx);
150 TAILQ_FOREACH_SAFE(otick, &data->aw_head, tk_aw_link, x_tick) {
151 if (otick->tk_unique == fii->unique) {
152 found = true;
153 break;
154 }
155 }
156 fuse_lck_mtx_unlock(data->aw_mtx);
157
158 if (!found) {
159 /* Original is already complete. Just return */
160 return 0;
161 }
162
163 /* Clear the original ticket's interrupt association */
164 otick->irq_unique = 0;
165
166 if (tick->tk_aw_ohead.error == ENOSYS) {
167 fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
168 return 0;
169 } else if (tick->tk_aw_ohead.error == EAGAIN) {
170 /*
171 * There are two reasons we might get this:
172 * 1) the daemon received the INTERRUPT request before the
173 * original, or
174 * 2) the daemon received the INTERRUPT request after it
175 * completed the original request.
176 * In the first case we should re-send the INTERRUPT. In the
177 * second, we should ignore it.
178 */
179 /* Resend */
180 fuse_interrupt_send(otick, EINTR);
181 return 0;
182 } else {
183 /* Illegal FUSE_INTERRUPT response */
184 return EINVAL;
185 }
186 }
187
188 /* Interrupt the operation otick. Return err as its error code */
189 void
fuse_interrupt_send(struct fuse_ticket * otick,int err)190 fuse_interrupt_send(struct fuse_ticket *otick, int err)
191 {
192 struct fuse_dispatcher fdi;
193 struct fuse_interrupt_in *fii;
194 struct fuse_in_header *ftick_hdr;
195 struct fuse_data *data = otick->tk_data;
196 struct fuse_ticket *tick, *xtick;
197 struct ucred reused_creds;
198 gid_t reused_groups[1];
199
200 if (otick->irq_unique == 0) {
201 /*
202 * If the daemon hasn't yet received otick, then we can answer
203 * it ourselves and return.
204 */
205 fuse_lck_mtx_lock(data->ms_mtx);
206 STAILQ_FOREACH_SAFE(tick, &otick->tk_data->ms_head, tk_ms_link,
207 xtick) {
208 if (tick == otick) {
209 STAILQ_REMOVE(&otick->tk_data->ms_head, tick,
210 fuse_ticket, tk_ms_link);
211 otick->tk_data->ms_count--;
212 otick->tk_ms_link.stqe_next = NULL;
213 fuse_lck_mtx_unlock(data->ms_mtx);
214
215 fuse_lck_mtx_lock(otick->tk_aw_mtx);
216 if (!fticket_answered(otick)) {
217 fticket_set_answered(otick);
218 otick->tk_aw_errno = err;
219 wakeup(otick);
220 }
221 fuse_lck_mtx_unlock(otick->tk_aw_mtx);
222
223 fuse_ticket_drop(tick);
224 return;
225 }
226 }
227 fuse_lck_mtx_unlock(data->ms_mtx);
228
229 /*
230 * If the fuse daemon doesn't support interrupts, then there's
231 * nothing more that we can do
232 */
233 if (!fsess_isimpl(data->mp, FUSE_INTERRUPT))
234 return;
235
236 /*
237 * If the fuse daemon has already received otick, then we must
238 * send FUSE_INTERRUPT.
239 */
240 ftick_hdr = fticket_in_header(otick);
241 reused_creds.cr_uid = ftick_hdr->uid;
242 reused_groups[0] = ftick_hdr->gid;
243 reused_creds.cr_groups = reused_groups;
244 fdisp_init(&fdi, sizeof(*fii));
245 fdisp_make_pid(&fdi, FUSE_INTERRUPT, data, ftick_hdr->nodeid,
246 ftick_hdr->pid, &reused_creds);
247
248 fii = fdi.indata;
249 fii->unique = otick->tk_unique;
250 fuse_insert_callback(fdi.tick, fuse_interrupt_callback);
251
252 otick->irq_unique = fdi.tick->tk_unique;
253 /* Interrupt ops should be delivered ASAP */
254 fuse_insert_message(fdi.tick, true);
255 fdisp_destroy(&fdi);
256 } else {
257 /* This ticket has already been interrupted */
258 }
259 }
260
261 void
fiov_init(struct fuse_iov * fiov,size_t size)262 fiov_init(struct fuse_iov *fiov, size_t size)
263 {
264 uint32_t msize = FU_AT_LEAST(size);
265
266 fiov->len = 0;
267
268 fiov->base = malloc(msize, M_FUSEMSG, M_WAITOK | M_ZERO);
269
270 fiov->allocated_size = msize;
271 fiov->credit = fuse_iov_credit;
272 }
273
274 void
fiov_teardown(struct fuse_iov * fiov)275 fiov_teardown(struct fuse_iov *fiov)
276 {
277 MPASS(fiov->base != NULL);
278 free(fiov->base, M_FUSEMSG);
279 }
280
281 void
fiov_adjust(struct fuse_iov * fiov,size_t size)282 fiov_adjust(struct fuse_iov *fiov, size_t size)
283 {
284 if (fiov->allocated_size < size ||
285 (fuse_iov_permanent_bufsize >= 0 &&
286 fiov->allocated_size - size > fuse_iov_permanent_bufsize &&
287 --fiov->credit < 0)) {
288
289 fiov->base = realloc(fiov->base, FU_AT_LEAST(size), M_FUSEMSG,
290 M_WAITOK | M_ZERO);
291 if (!fiov->base) {
292 panic("FUSE: realloc failed");
293 }
294 fiov->allocated_size = FU_AT_LEAST(size);
295 fiov->credit = fuse_iov_credit;
296 /* Clear data buffer after reallocation */
297 bzero(fiov->base, size);
298 } else if (size > fiov->len) {
299 /* Clear newly extended portion of data buffer */
300 bzero((char*)fiov->base + fiov->len, size - fiov->len);
301 }
302 fiov->len = size;
303 }
304
305 /* Resize the fiov if needed, and clear it's buffer */
306 void
fiov_refresh(struct fuse_iov * fiov)307 fiov_refresh(struct fuse_iov *fiov)
308 {
309 fiov_adjust(fiov, 0);
310 }
311
312 static int
fticket_ctor(void * mem,int size,void * arg,int flags)313 fticket_ctor(void *mem, int size, void *arg, int flags)
314 {
315 struct fuse_ticket *ftick = mem;
316 struct fuse_data *data = arg;
317
318 FUSE_ASSERT_MS_DONE(ftick);
319 FUSE_ASSERT_AW_DONE(ftick);
320
321 ftick->tk_data = data;
322
323 if (ftick->tk_unique != 0)
324 fticket_refresh(ftick);
325
326 /* May be truncated to 32 bits */
327 ftick->tk_unique = atomic_fetchadd_long(&data->ticketer, 1);
328 if (ftick->tk_unique == 0)
329 ftick->tk_unique = atomic_fetchadd_long(&data->ticketer, 1);
330
331 ftick->irq_unique = 0;
332
333 refcount_init(&ftick->tk_refcount, 1);
334 counter_u64_add(fuse_ticket_count, 1);
335
336 return 0;
337 }
338
339 static void
fticket_dtor(void * mem,int size,void * arg)340 fticket_dtor(void *mem, int size, void *arg)
341 {
342 #ifdef INVARIANTS
343 struct fuse_ticket *ftick = mem;
344 #endif
345
346 FUSE_ASSERT_MS_DONE(ftick);
347 FUSE_ASSERT_AW_DONE(ftick);
348
349 counter_u64_add(fuse_ticket_count, -1);
350 }
351
352 static int
fticket_init(void * mem,int size,int flags)353 fticket_init(void *mem, int size, int flags)
354 {
355 struct fuse_ticket *ftick = mem;
356
357 bzero(ftick, sizeof(struct fuse_ticket));
358
359 fiov_init(&ftick->tk_ms_fiov, sizeof(struct fuse_in_header));
360 ftick->tk_ms_type = FT_M_FIOV;
361
362 mtx_init(&ftick->tk_aw_mtx, "fuse answer delivery mutex", NULL, MTX_DEF);
363 fiov_init(&ftick->tk_aw_fiov, 0);
364 ftick->tk_aw_type = FT_A_FIOV;
365
366 return 0;
367 }
368
369 static void
fticket_fini(void * mem,int size)370 fticket_fini(void *mem, int size)
371 {
372 struct fuse_ticket *ftick = mem;
373
374 fiov_teardown(&ftick->tk_ms_fiov);
375 fiov_teardown(&ftick->tk_aw_fiov);
376 mtx_destroy(&ftick->tk_aw_mtx);
377 }
378
379 static inline struct fuse_ticket *
fticket_alloc(struct fuse_data * data)380 fticket_alloc(struct fuse_data *data)
381 {
382 return uma_zalloc_arg(ticket_zone, data, M_WAITOK);
383 }
384
385 static inline void
fticket_destroy(struct fuse_ticket * ftick)386 fticket_destroy(struct fuse_ticket *ftick)
387 {
388 return uma_zfree(ticket_zone, ftick);
389 }
390
391 static inline
392 void
fticket_refresh(struct fuse_ticket * ftick)393 fticket_refresh(struct fuse_ticket *ftick)
394 {
395 FUSE_ASSERT_MS_DONE(ftick);
396 FUSE_ASSERT_AW_DONE(ftick);
397
398 fiov_refresh(&ftick->tk_ms_fiov);
399 ftick->tk_ms_bufdata = NULL;
400 ftick->tk_ms_bufsize = 0;
401 ftick->tk_ms_type = FT_M_FIOV;
402
403 bzero(&ftick->tk_aw_ohead, sizeof(struct fuse_out_header));
404
405 fiov_refresh(&ftick->tk_aw_fiov);
406 ftick->tk_aw_errno = 0;
407 ftick->tk_aw_bufdata = NULL;
408 ftick->tk_aw_bufsize = 0;
409 ftick->tk_aw_type = FT_A_FIOV;
410
411 ftick->tk_flag = 0;
412 }
413
414 /* Prepar the ticket to be reused, but don't clear its data buffers */
415 static inline void
fticket_reset(struct fuse_ticket * ftick)416 fticket_reset(struct fuse_ticket *ftick)
417 {
418 FUSE_ASSERT_MS_DONE(ftick);
419 FUSE_ASSERT_AW_DONE(ftick);
420
421 ftick->tk_ms_bufdata = NULL;
422 ftick->tk_ms_bufsize = 0;
423 ftick->tk_ms_type = FT_M_FIOV;
424
425 bzero(&ftick->tk_aw_ohead, sizeof(struct fuse_out_header));
426
427 ftick->tk_aw_errno = 0;
428 ftick->tk_aw_bufdata = NULL;
429 ftick->tk_aw_bufsize = 0;
430 ftick->tk_aw_type = FT_A_FIOV;
431
432 ftick->tk_flag = 0;
433 }
434
435 static int
fticket_wait_answer(struct fuse_ticket * ftick)436 fticket_wait_answer(struct fuse_ticket *ftick)
437 {
438 struct thread *td = curthread;
439 sigset_t blockedset, oldset;
440 int err = 0, stops_deferred;
441 struct fuse_data *data = ftick->tk_data;
442 bool interrupted = false;
443
444 if (fsess_isimpl(ftick->tk_data->mp, FUSE_INTERRUPT) &&
445 data->dataflags & FSESS_INTR) {
446 SIGEMPTYSET(blockedset);
447 } else {
448 /* Block all signals except (implicitly) SIGKILL */
449 SIGFILLSET(blockedset);
450 }
451 stops_deferred = sigdeferstop(SIGDEFERSTOP_SILENT);
452 kern_sigprocmask(td, SIG_BLOCK, NULL, &oldset, 0);
453
454 fuse_lck_mtx_lock(ftick->tk_aw_mtx);
455
456 retry:
457 if (fticket_answered(ftick)) {
458 goto out;
459 }
460
461 if (fdata_get_dead(data)) {
462 err = ENOTCONN;
463 fticket_set_answered(ftick);
464 goto out;
465 }
466 kern_sigprocmask(td, SIG_BLOCK, &blockedset, NULL, 0);
467 err = msleep(ftick, &ftick->tk_aw_mtx, PCATCH, "fu_ans",
468 data->daemon_timeout * hz);
469 kern_sigprocmask(td, SIG_SETMASK, &oldset, NULL, 0);
470 if (err == EWOULDBLOCK) {
471 SDT_PROBE2(fusefs, , ipc, trace, 3,
472 "fticket_wait_answer: EWOULDBLOCK");
473 #ifdef XXXIP /* die conditionally */
474 if (!fdata_get_dead(data)) {
475 fdata_set_dead(data);
476 }
477 #endif
478 err = ETIMEDOUT;
479 fticket_set_answered(ftick);
480 } else if ((err == EINTR || err == ERESTART)) {
481 /*
482 * Whether we get EINTR or ERESTART depends on whether
483 * SA_RESTART was set by sigaction(2).
484 *
485 * Try to interrupt the operation and wait for an EINTR response
486 * to the original operation. If the file system does not
487 * support FUSE_INTERRUPT, then we'll just wait for it to
488 * complete like normal. If it does support FUSE_INTERRUPT,
489 * then it will either respond EINTR to the original operation,
490 * or EAGAIN to the interrupt.
491 */
492 sigset_t tmpset;
493
494 SDT_PROBE2(fusefs, , ipc, trace, 4,
495 "fticket_wait_answer: interrupt");
496 fuse_lck_mtx_unlock(ftick->tk_aw_mtx);
497 fuse_interrupt_send(ftick, err);
498
499 PROC_LOCK(td->td_proc);
500 mtx_lock(&td->td_proc->p_sigacts->ps_mtx);
501 tmpset = td->td_proc->p_siglist;
502 SIGSETOR(tmpset, td->td_siglist);
503 mtx_unlock(&td->td_proc->p_sigacts->ps_mtx);
504 PROC_UNLOCK(td->td_proc);
505
506 fuse_lck_mtx_lock(ftick->tk_aw_mtx);
507 if (!interrupted && !SIGISMEMBER(tmpset, SIGKILL)) {
508 /*
509 * Block all signals while we wait for an interrupt
510 * response. The protocol doesn't discriminate between
511 * different signals.
512 */
513 SIGFILLSET(blockedset);
514 interrupted = true;
515 goto retry;
516 } else {
517 /*
518 * Return immediately for fatal signals, or if this is
519 * the second interruption. We should only be
520 * interrupted twice if the thread is stopped, for
521 * example during sigexit.
522 */
523 }
524 } else if (err) {
525 SDT_PROBE2(fusefs, , ipc, trace, 6,
526 "fticket_wait_answer: other error");
527 } else {
528 SDT_PROBE2(fusefs, , ipc, trace, 7, "fticket_wait_answer: OK");
529 }
530 out:
531 if (!(err || fticket_answered(ftick))) {
532 SDT_PROBE2(fusefs, , ipc, trace, 1,
533 "FUSE: requester was woken up but still no answer");
534 err = ENXIO;
535 }
536 fuse_lck_mtx_unlock(ftick->tk_aw_mtx);
537 sigallowstop(stops_deferred);
538
539 return err;
540 }
541
542 static inline
543 int
fticket_aw_pull_uio(struct fuse_ticket * ftick,struct uio * uio)544 fticket_aw_pull_uio(struct fuse_ticket *ftick, struct uio *uio)
545 {
546 int err = 0;
547 size_t len = uio_resid(uio);
548
549 if (len) {
550 switch (ftick->tk_aw_type) {
551 case FT_A_FIOV:
552 fiov_adjust(fticket_resp(ftick), len);
553 err = uiomove(fticket_resp(ftick)->base, len, uio);
554 break;
555
556 case FT_A_BUF:
557 ftick->tk_aw_bufsize = len;
558 err = uiomove(ftick->tk_aw_bufdata, len, uio);
559 break;
560
561 default:
562 panic("FUSE: unknown answer type for ticket %p", ftick);
563 }
564 }
565 return err;
566 }
567
568 int
fticket_pull(struct fuse_ticket * ftick,struct uio * uio)569 fticket_pull(struct fuse_ticket *ftick, struct uio *uio)
570 {
571 int err = 0;
572
573 if (ftick->tk_aw_ohead.error) {
574 return 0;
575 }
576 err = fuse_body_audit(ftick, uio_resid(uio));
577 if (!err) {
578 err = fticket_aw_pull_uio(ftick, uio);
579 }
580 return err;
581 }
582
583 struct fuse_data *
fdata_alloc(struct cdev * fdev,struct ucred * cred)584 fdata_alloc(struct cdev *fdev, struct ucred *cred)
585 {
586 struct fuse_data *data;
587
588 data = malloc(sizeof(struct fuse_data), M_FUSEMSG, M_WAITOK | M_ZERO);
589
590 data->fdev = fdev;
591 mtx_init(&data->ms_mtx, "fuse message list mutex", NULL, MTX_DEF);
592 STAILQ_INIT(&data->ms_head);
593 data->ms_count = 0;
594 knlist_init_mtx(&data->ks_rsel.si_note, &data->ms_mtx);
595 mtx_init(&data->aw_mtx, "fuse answer list mutex", NULL, MTX_DEF);
596 TAILQ_INIT(&data->aw_head);
597 data->daemoncred = crhold(cred);
598 data->daemon_timeout = FUSE_DEFAULT_DAEMON_TIMEOUT;
599 sx_init(&data->rename_lock, "fuse rename lock");
600 data->ref = 1;
601
602 return data;
603 }
604
605 void
fdata_trydestroy(struct fuse_data * data)606 fdata_trydestroy(struct fuse_data *data)
607 {
608 data->ref--;
609 MPASS(data->ref >= 0);
610 if (data->ref != 0)
611 return;
612
613 /* Driving off stage all that stuff thrown at device... */
614 sx_destroy(&data->rename_lock);
615 crfree(data->daemoncred);
616 mtx_destroy(&data->aw_mtx);
617 knlist_delete(&data->ks_rsel.si_note, curthread, 0);
618 knlist_destroy(&data->ks_rsel.si_note);
619 mtx_destroy(&data->ms_mtx);
620
621 free(data, M_FUSEMSG);
622 }
623
624 void
fdata_set_dead(struct fuse_data * data)625 fdata_set_dead(struct fuse_data *data)
626 {
627 FUSE_LOCK();
628 if (fdata_get_dead(data)) {
629 FUSE_UNLOCK();
630 return;
631 }
632 fuse_lck_mtx_lock(data->ms_mtx);
633 data->dataflags |= FSESS_DEAD;
634 wakeup_one(data);
635 selwakeuppri(&data->ks_rsel, PZERO + 1);
636 wakeup(&data->ticketer);
637 fuse_lck_mtx_unlock(data->ms_mtx);
638 FUSE_UNLOCK();
639 }
640
641 struct fuse_ticket *
fuse_ticket_fetch(struct fuse_data * data)642 fuse_ticket_fetch(struct fuse_data *data)
643 {
644 int err = 0;
645 struct fuse_ticket *ftick;
646
647 ftick = fticket_alloc(data);
648
649 if (!(data->dataflags & FSESS_INITED)) {
650 /* Sleep until get answer for INIT messsage */
651 FUSE_LOCK();
652 if (!(data->dataflags & FSESS_INITED) && data->ticketer > 2) {
653 err = msleep(&data->ticketer, &fuse_mtx, PCATCH | PDROP,
654 "fu_ini", 0);
655 if (err)
656 fdata_set_dead(data);
657 } else
658 FUSE_UNLOCK();
659 }
660 return ftick;
661 }
662
663 int
fuse_ticket_drop(struct fuse_ticket * ftick)664 fuse_ticket_drop(struct fuse_ticket *ftick)
665 {
666 int die;
667
668 die = refcount_release(&ftick->tk_refcount);
669 if (die)
670 fticket_destroy(ftick);
671
672 return die;
673 }
674
675 void
fuse_insert_callback(struct fuse_ticket * ftick,fuse_handler_t * handler)676 fuse_insert_callback(struct fuse_ticket *ftick, fuse_handler_t * handler)
677 {
678 if (fdata_get_dead(ftick->tk_data)) {
679 return;
680 }
681 ftick->tk_aw_handler = handler;
682
683 fuse_lck_mtx_lock(ftick->tk_data->aw_mtx);
684 fuse_aw_push(ftick);
685 fuse_lck_mtx_unlock(ftick->tk_data->aw_mtx);
686 }
687
688 /*
689 * Insert a new upgoing ticket into the message queue
690 *
691 * If urgent is true, insert at the front of the queue. Otherwise, insert in
692 * FIFO order.
693 */
694 void
fuse_insert_message(struct fuse_ticket * ftick,bool urgent)695 fuse_insert_message(struct fuse_ticket *ftick, bool urgent)
696 {
697 if (ftick->tk_flag & FT_DIRTY) {
698 panic("FUSE: ticket reused without being refreshed");
699 }
700 ftick->tk_flag |= FT_DIRTY;
701
702 if (fdata_get_dead(ftick->tk_data)) {
703 return;
704 }
705 fuse_lck_mtx_lock(ftick->tk_data->ms_mtx);
706 if (urgent)
707 fuse_ms_push_head(ftick);
708 else
709 fuse_ms_push(ftick);
710 wakeup_one(ftick->tk_data);
711 selwakeuppri(&ftick->tk_data->ks_rsel, PZERO + 1);
712 KNOTE_LOCKED(&ftick->tk_data->ks_rsel.si_note, 0);
713 fuse_lck_mtx_unlock(ftick->tk_data->ms_mtx);
714 }
715
716 static int
fuse_body_audit(struct fuse_ticket * ftick,size_t blen)717 fuse_body_audit(struct fuse_ticket *ftick, size_t blen)
718 {
719 int err = 0;
720 enum fuse_opcode opcode;
721
722 opcode = fticket_opcode(ftick);
723
724 switch (opcode) {
725 case FUSE_BMAP:
726 err = (blen == sizeof(struct fuse_bmap_out)) ? 0 : EINVAL;
727 break;
728
729 case FUSE_LINK:
730 case FUSE_LOOKUP:
731 case FUSE_MKDIR:
732 case FUSE_MKNOD:
733 case FUSE_SYMLINK:
734 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) {
735 err = (blen == sizeof(struct fuse_entry_out)) ?
736 0 : EINVAL;
737 } else {
738 err = (blen == FUSE_COMPAT_ENTRY_OUT_SIZE) ? 0 : EINVAL;
739 }
740 break;
741
742 case FUSE_FORGET:
743 panic("FUSE: a handler has been intalled for FUSE_FORGET");
744 break;
745
746 case FUSE_GETATTR:
747 case FUSE_SETATTR:
748 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) {
749 err = (blen == sizeof(struct fuse_attr_out)) ?
750 0 : EINVAL;
751 } else {
752 err = (blen == FUSE_COMPAT_ATTR_OUT_SIZE) ? 0 : EINVAL;
753 }
754 break;
755
756 case FUSE_READLINK:
757 err = (PAGE_SIZE >= blen) ? 0 : EINVAL;
758 break;
759
760 case FUSE_UNLINK:
761 err = (blen == 0) ? 0 : EINVAL;
762 break;
763
764 case FUSE_RMDIR:
765 err = (blen == 0) ? 0 : EINVAL;
766 break;
767
768 case FUSE_RENAME:
769 err = (blen == 0) ? 0 : EINVAL;
770 break;
771
772 case FUSE_OPEN:
773 err = (blen == sizeof(struct fuse_open_out)) ? 0 : EINVAL;
774 break;
775
776 case FUSE_READ:
777 err = (((struct fuse_read_in *)(
778 (char *)ftick->tk_ms_fiov.base +
779 sizeof(struct fuse_in_header)
780 ))->size >= blen) ? 0 : EINVAL;
781 break;
782
783 case FUSE_WRITE:
784 err = (blen == sizeof(struct fuse_write_out)) ? 0 : EINVAL;
785 break;
786
787 case FUSE_STATFS:
788 if (fuse_libabi_geq(ftick->tk_data, 7, 4)) {
789 err = (blen == sizeof(struct fuse_statfs_out)) ?
790 0 : EINVAL;
791 } else {
792 err = (blen == FUSE_COMPAT_STATFS_SIZE) ? 0 : EINVAL;
793 }
794 break;
795
796 case FUSE_RELEASE:
797 err = (blen == 0) ? 0 : EINVAL;
798 break;
799
800 case FUSE_FSYNC:
801 err = (blen == 0) ? 0 : EINVAL;
802 break;
803
804 case FUSE_SETXATTR:
805 err = (blen == 0) ? 0 : EINVAL;
806 break;
807
808 case FUSE_GETXATTR:
809 case FUSE_LISTXATTR:
810 /*
811 * These can have varying response lengths, and 0 length
812 * isn't necessarily invalid.
813 */
814 err = 0;
815 break;
816
817 case FUSE_REMOVEXATTR:
818 err = (blen == 0) ? 0 : EINVAL;
819 break;
820
821 case FUSE_FLUSH:
822 err = (blen == 0) ? 0 : EINVAL;
823 break;
824
825 case FUSE_INIT:
826 if (blen == sizeof(struct fuse_init_out) ||
827 blen == FUSE_COMPAT_INIT_OUT_SIZE ||
828 blen == FUSE_COMPAT_22_INIT_OUT_SIZE) {
829 err = 0;
830 } else {
831 err = EINVAL;
832 }
833 break;
834
835 case FUSE_OPENDIR:
836 err = (blen == sizeof(struct fuse_open_out)) ? 0 : EINVAL;
837 break;
838
839 case FUSE_READDIR:
840 err = (((struct fuse_read_in *)(
841 (char *)ftick->tk_ms_fiov.base +
842 sizeof(struct fuse_in_header)
843 ))->size >= blen) ? 0 : EINVAL;
844 break;
845
846 case FUSE_RELEASEDIR:
847 err = (blen == 0) ? 0 : EINVAL;
848 break;
849
850 case FUSE_FSYNCDIR:
851 err = (blen == 0) ? 0 : EINVAL;
852 break;
853
854 case FUSE_GETLK:
855 err = (blen == sizeof(struct fuse_lk_out)) ? 0 : EINVAL;
856 break;
857
858 case FUSE_SETLK:
859 err = (blen == 0) ? 0 : EINVAL;
860 break;
861
862 case FUSE_SETLKW:
863 err = (blen == 0) ? 0 : EINVAL;
864 break;
865
866 case FUSE_ACCESS:
867 err = (blen == 0) ? 0 : EINVAL;
868 break;
869
870 case FUSE_CREATE:
871 if (fuse_libabi_geq(ftick->tk_data, 7, 9)) {
872 err = (blen == sizeof(struct fuse_entry_out) +
873 sizeof(struct fuse_open_out)) ? 0 : EINVAL;
874 } else {
875 err = (blen == FUSE_COMPAT_ENTRY_OUT_SIZE +
876 sizeof(struct fuse_open_out)) ? 0 : EINVAL;
877 }
878 break;
879
880 case FUSE_DESTROY:
881 err = (blen == 0) ? 0 : EINVAL;
882 break;
883
884 default:
885 panic("FUSE: opcodes out of sync (%d)\n", opcode);
886 }
887
888 return err;
889 }
890
891 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)892 fuse_setup_ihead(struct fuse_in_header *ihead, struct fuse_ticket *ftick,
893 uint64_t nid, enum fuse_opcode op, size_t blen, pid_t pid,
894 struct ucred *cred)
895 {
896 ihead->len = sizeof(*ihead) + blen;
897 ihead->unique = ftick->tk_unique;
898 ihead->nodeid = nid;
899 ihead->opcode = op;
900
901 ihead->pid = pid;
902 ihead->uid = cred->cr_uid;
903 ihead->gid = cred->cr_groups[0];
904 }
905
906 /*
907 * fuse_standard_handler just pulls indata and wakes up pretender.
908 * Doesn't try to interpret data, that's left for the pretender.
909 * Though might do a basic size verification before the pull-in takes place
910 */
911
912 static int
fuse_standard_handler(struct fuse_ticket * ftick,struct uio * uio)913 fuse_standard_handler(struct fuse_ticket *ftick, struct uio *uio)
914 {
915 int err = 0;
916
917 err = fticket_pull(ftick, uio);
918
919 fuse_lck_mtx_lock(ftick->tk_aw_mtx);
920
921 if (!fticket_answered(ftick)) {
922 fticket_set_answered(ftick);
923 ftick->tk_aw_errno = err;
924 wakeup(ftick);
925 }
926 fuse_lck_mtx_unlock(ftick->tk_aw_mtx);
927
928 return err;
929 }
930
931 /*
932 * Reinitialize a dispatcher from a pid and node id, without resizing or
933 * clearing its data buffers
934 */
935 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)936 fdisp_refresh_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op,
937 struct mount *mp, uint64_t nid, pid_t pid, struct ucred *cred)
938 {
939 MPASS(fdip->tick);
940 MPASS2(sizeof(fdip->finh) + fdip->iosize <= fdip->tick->tk_ms_fiov.len,
941 "Must use fdisp_make_pid to increase the size of the fiov");
942 fticket_reset(fdip->tick);
943
944 FUSE_DIMALLOC(&fdip->tick->tk_ms_fiov, fdip->finh,
945 fdip->indata, fdip->iosize);
946
947 fuse_setup_ihead(fdip->finh, fdip->tick, nid, op, fdip->iosize, pid,
948 cred);
949 }
950
951 /* Initialize a dispatcher from a pid and node id */
952 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)953 fdisp_make_pid(struct fuse_dispatcher *fdip, enum fuse_opcode op,
954 struct fuse_data *data, uint64_t nid, pid_t pid, struct ucred *cred)
955 {
956 if (fdip->tick) {
957 fticket_refresh(fdip->tick);
958 } else {
959 fdip->tick = fuse_ticket_fetch(data);
960 }
961
962 /* FUSE_DIMALLOC will bzero the fiovs when it enlarges them */
963 FUSE_DIMALLOC(&fdip->tick->tk_ms_fiov, fdip->finh,
964 fdip->indata, fdip->iosize);
965
966 fuse_setup_ihead(fdip->finh, fdip->tick, nid, op, fdip->iosize, pid, cred);
967 }
968
969 void
fdisp_make(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct mount * mp,uint64_t nid,struct thread * td,struct ucred * cred)970 fdisp_make(struct fuse_dispatcher *fdip, enum fuse_opcode op, struct mount *mp,
971 uint64_t nid, struct thread *td, struct ucred *cred)
972 {
973 struct fuse_data *data = fuse_get_mpdata(mp);
974 RECTIFY_TDCR(td, cred);
975
976 return fdisp_make_pid(fdip, op, data, nid, td->td_proc->p_pid, cred);
977 }
978
979 void
fdisp_make_vp(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct vnode * vp,struct thread * td,struct ucred * cred)980 fdisp_make_vp(struct fuse_dispatcher *fdip, enum fuse_opcode op,
981 struct vnode *vp, struct thread *td, struct ucred *cred)
982 {
983 struct mount *mp = vnode_mount(vp);
984 struct fuse_data *data = fuse_get_mpdata(mp);
985
986 RECTIFY_TDCR(td, cred);
987 return fdisp_make_pid(fdip, op, data, VTOI(vp),
988 td->td_proc->p_pid, cred);
989 }
990
991 /* Refresh a fuse_dispatcher so it can be reused, but don't zero its data */
992 void
fdisp_refresh_vp(struct fuse_dispatcher * fdip,enum fuse_opcode op,struct vnode * vp,struct thread * td,struct ucred * cred)993 fdisp_refresh_vp(struct fuse_dispatcher *fdip, enum fuse_opcode op,
994 struct vnode *vp, struct thread *td, struct ucred *cred)
995 {
996 RECTIFY_TDCR(td, cred);
997 return fdisp_refresh_pid(fdip, op, vnode_mount(vp), VTOI(vp),
998 td->td_proc->p_pid, cred);
999 }
1000
1001 void
fdisp_refresh(struct fuse_dispatcher * fdip)1002 fdisp_refresh(struct fuse_dispatcher *fdip)
1003 {
1004 fticket_refresh(fdip->tick);
1005 }
1006
1007 SDT_PROBE_DEFINE2(fusefs, , ipc, fdisp_wait_answ_error, "char*", "int");
1008
1009 int
fdisp_wait_answ(struct fuse_dispatcher * fdip)1010 fdisp_wait_answ(struct fuse_dispatcher *fdip)
1011 {
1012 int err = 0;
1013
1014 fdip->answ_stat = 0;
1015 fuse_insert_callback(fdip->tick, fuse_standard_handler);
1016 fuse_insert_message(fdip->tick, false);
1017
1018 if ((err = fticket_wait_answer(fdip->tick))) {
1019 fuse_lck_mtx_lock(fdip->tick->tk_aw_mtx);
1020
1021 if (fticket_answered(fdip->tick)) {
1022 /*
1023 * Just between noticing the interrupt and getting here,
1024 * the standard handler has completed his job.
1025 * So we drop the ticket and exit as usual.
1026 */
1027 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1028 "IPC: interrupted, already answered", err);
1029 fuse_lck_mtx_unlock(fdip->tick->tk_aw_mtx);
1030 goto out;
1031 } else {
1032 /*
1033 * So we were faster than the standard handler.
1034 * Then by setting the answered flag we get *him*
1035 * to drop the ticket.
1036 */
1037 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1038 "IPC: interrupted, setting to answered", err);
1039 fticket_set_answered(fdip->tick);
1040 fuse_lck_mtx_unlock(fdip->tick->tk_aw_mtx);
1041 return err;
1042 }
1043 }
1044
1045 if (fdip->tick->tk_aw_errno == ENOTCONN) {
1046 /* The daemon died while we were waiting for a response */
1047 err = ENOTCONN;
1048 goto out;
1049 } else if (fdip->tick->tk_aw_errno) {
1050 /*
1051 * There was some sort of communication error with the daemon
1052 * that the client wouldn't understand.
1053 */
1054 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1055 "IPC: explicit EIO-ing", fdip->tick->tk_aw_errno);
1056 err = EIO;
1057 goto out;
1058 }
1059 if ((err = fdip->tick->tk_aw_ohead.error)) {
1060 SDT_PROBE2(fusefs, , ipc, fdisp_wait_answ_error,
1061 "IPC: setting status", fdip->tick->tk_aw_ohead.error);
1062 /*
1063 * This means a "proper" fuse syscall error.
1064 * We record this value so the caller will
1065 * be able to know it's not a boring messaging
1066 * failure, if she wishes so (and if not, she can
1067 * just simply propagate the return value of this routine).
1068 * [XXX Maybe a bitflag would do the job too,
1069 * if other flags needed, this will be converted thusly.]
1070 */
1071 fdip->answ_stat = err;
1072 goto out;
1073 }
1074 fdip->answ = fticket_resp(fdip->tick)->base;
1075 fdip->iosize = fticket_resp(fdip->tick)->len;
1076
1077 return 0;
1078
1079 out:
1080 return err;
1081 }
1082
1083 void
fuse_ipc_init(void)1084 fuse_ipc_init(void)
1085 {
1086 ticket_zone = uma_zcreate("fuse_ticket", sizeof(struct fuse_ticket),
1087 fticket_ctor, fticket_dtor, fticket_init, fticket_fini,
1088 UMA_ALIGN_PTR, 0);
1089 fuse_ticket_count = counter_u64_alloc(M_WAITOK);
1090 }
1091
1092 void
fuse_ipc_destroy(void)1093 fuse_ipc_destroy(void)
1094 {
1095 counter_u64_free(fuse_ticket_count);
1096 uma_zdestroy(ticket_zone);
1097 }
1098