2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Google Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
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
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.
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.
33 * Copyright (C) 2005 Csaba Henk.
34 * All rights reserved.
36 * Copyright (c) 2019 The FreeBSD Foundation
38 * Portions of this software were developed by BFF Storage Systems, LLC under
39 * sponsorship from the FreeBSD Foundation.
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
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.
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
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
66 #include <sys/types.h>
67 #include <sys/module.h>
68 #include <sys/systm.h>
69 #include <sys/errno.h>
70 #include <sys/param.h>
71 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/queue.h>
78 #include <sys/mutex.h>
80 #include <sys/mount.h>
83 #include <sys/fcntl.h>
84 #include <sys/sysctl.h>
86 #include <sys/selinfo.h>
89 #include "fuse_internal.h"
92 SDT_PROVIDER_DECLARE(fusefs);
95 * arg0: verbosity. Higher numbers give more verbose messages
96 * arg1: Textual message
98 SDT_PROBE_DEFINE2(fusefs, , device, trace, "int", "char*");
100 static struct cdev *fuse_dev;
102 static d_kqfilter_t fuse_device_filter;
103 static d_open_t fuse_device_open;
104 static d_poll_t fuse_device_poll;
105 static d_read_t fuse_device_read;
106 static d_write_t fuse_device_write;
108 static struct cdevsw fuse_device_cdevsw = {
109 .d_kqfilter = fuse_device_filter,
110 .d_open = fuse_device_open,
112 .d_poll = fuse_device_poll,
113 .d_read = fuse_device_read,
114 .d_write = fuse_device_write,
115 .d_version = D_VERSION,
118 static int fuse_device_filt_read(struct knote *kn, long hint);
119 static void fuse_device_filt_detach(struct knote *kn);
121 struct filterops fuse_device_rfiltops = {
123 .f_detach = fuse_device_filt_detach,
124 .f_event = fuse_device_filt_read,
127 /****************************
129 * >>> Fuse device op defs
131 ****************************/
134 fdata_dtor(void *arg)
136 struct fuse_data *fdata;
137 struct fuse_ticket *tick;
143 fdata_set_dead(fdata);
146 fuse_lck_mtx_lock(fdata->aw_mtx);
147 /* wakup poll()ers */
148 selwakeuppri(&fdata->ks_rsel, PZERO + 1);
149 /* Don't let syscall handlers wait in vain */
150 while ((tick = fuse_aw_pop(fdata))) {
151 fuse_lck_mtx_lock(tick->tk_aw_mtx);
152 fticket_set_answered(tick);
153 tick->tk_aw_errno = ENOTCONN;
155 fuse_lck_mtx_unlock(tick->tk_aw_mtx);
156 FUSE_ASSERT_AW_DONE(tick);
157 fuse_ticket_drop(tick);
159 fuse_lck_mtx_unlock(fdata->aw_mtx);
161 /* Cleanup unsent operations */
162 fuse_lck_mtx_lock(fdata->ms_mtx);
163 while ((tick = fuse_ms_pop(fdata))) {
164 fuse_ticket_drop(tick);
166 fuse_lck_mtx_unlock(fdata->ms_mtx);
169 fdata_trydestroy(fdata);
173 fuse_device_filter(struct cdev *dev, struct knote *kn)
175 struct fuse_data *data;
178 error = devfs_get_cdevpriv((void **)&data);
180 /* EVFILT_WRITE is not supported; the device is always ready to write */
181 if (error == 0 && kn->kn_filter == EVFILT_READ) {
182 kn->kn_fop = &fuse_device_rfiltops;
184 knlist_add(&data->ks_rsel.si_note, kn, 0);
186 } else if (error == 0) {
195 fuse_device_filt_detach(struct knote *kn)
197 struct fuse_data *data;
199 data = (struct fuse_data*)kn->kn_hook;
201 knlist_remove(&data->ks_rsel.si_note, kn, 0);
206 fuse_device_filt_read(struct knote *kn, long hint)
208 struct fuse_data *data;
211 data = (struct fuse_data*)kn->kn_hook;
214 mtx_assert(&data->ms_mtx, MA_OWNED);
215 if (fdata_get_dead(data)) {
216 kn->kn_flags |= EV_EOF;
217 kn->kn_fflags = ENODEV;
220 } else if (STAILQ_FIRST(&data->ms_head)) {
221 MPASS(data->ms_count >= 1);
222 kn->kn_data = data->ms_count;
232 * Resources are set up on a per-open basis
235 fuse_device_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
237 struct fuse_data *fdata;
240 SDT_PROBE2(fusefs, , device, trace, 1, "device open");
242 fdata = fdata_alloc(dev, td->td_ucred);
243 error = devfs_set_cdevpriv(fdata, fdata_dtor);
245 fdata_trydestroy(fdata);
247 SDT_PROBE2(fusefs, , device, trace, 1, "device open success");
252 fuse_device_poll(struct cdev *dev, int events, struct thread *td)
254 struct fuse_data *data;
255 int error, revents = 0;
257 error = devfs_get_cdevpriv((void **)&data);
260 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
262 if (events & (POLLIN | POLLRDNORM)) {
263 fuse_lck_mtx_lock(data->ms_mtx);
264 if (fdata_get_dead(data) || STAILQ_FIRST(&data->ms_head))
265 revents |= events & (POLLIN | POLLRDNORM);
267 selrecord(td, &data->ks_rsel);
268 fuse_lck_mtx_unlock(data->ms_mtx);
270 if (events & (POLLOUT | POLLWRNORM)) {
271 revents |= events & (POLLOUT | POLLWRNORM);
277 * fuse_device_read hangs on the queue of VFS messages.
278 * When it's notified that there is a new one, it picks that and
279 * passes up to the daemon
282 fuse_device_read(struct cdev *dev, struct uio *uio, int ioflag)
285 struct fuse_data *data;
286 struct fuse_ticket *tick;
287 void *buf[] = {NULL, NULL, NULL};
291 SDT_PROBE2(fusefs, , device, trace, 1, "fuse device read");
293 err = devfs_get_cdevpriv((void **)&data);
297 fuse_lck_mtx_lock(data->ms_mtx);
299 if (fdata_get_dead(data)) {
300 SDT_PROBE2(fusefs, , device, trace, 2,
301 "we know early on that reader should be kicked so we "
302 "don't wait for news");
303 fuse_lck_mtx_unlock(data->ms_mtx);
306 if (!(tick = fuse_ms_pop(data))) {
307 /* check if we may block */
308 if (ioflag & O_NONBLOCK) {
309 /* get outa here soon */
310 fuse_lck_mtx_unlock(data->ms_mtx);
313 err = msleep(data, &data->ms_mtx, PCATCH, "fu_msg", 0);
315 fuse_lck_mtx_unlock(data->ms_mtx);
316 return (fdata_get_dead(data) ? ENODEV : err);
318 tick = fuse_ms_pop(data);
323 * We can get here if fuse daemon suddenly terminates,
324 * eg, by being hit by a SIGKILL
325 * -- and some other cases, too, tho not totally clear, when
326 * (cv_signal/wakeup_one signals the whole process ?)
328 SDT_PROBE2(fusefs, , device, trace, 1, "no message on thread");
331 fuse_lck_mtx_unlock(data->ms_mtx);
333 if (fdata_get_dead(data)) {
335 * somebody somewhere -- eg., umount routine --
336 * wants this liaison finished off
338 SDT_PROBE2(fusefs, , device, trace, 2,
339 "reader is to be sacked");
341 SDT_PROBE2(fusefs, , device, trace, 2, "weird -- "
342 "\"kick\" is set tho there is message");
343 FUSE_ASSERT_MS_DONE(tick);
344 fuse_ticket_drop(tick);
346 return (ENODEV); /* This should make the daemon get off
349 SDT_PROBE2(fusefs, , device, trace, 1,
350 "fuse device read message successfully");
352 KASSERT(tick->tk_ms_bufdata || tick->tk_ms_bufsize == 0,
353 ("non-null buf pointer with positive size"));
355 switch (tick->tk_ms_type) {
357 buf[0] = tick->tk_ms_fiov.base;
358 buflen[0] = tick->tk_ms_fiov.len;
361 buf[0] = tick->tk_ms_fiov.base;
362 buflen[0] = tick->tk_ms_fiov.len;
363 buf[1] = tick->tk_ms_bufdata;
364 buflen[1] = tick->tk_ms_bufsize;
367 panic("unknown message type for fuse_ticket %p", tick);
370 for (i = 0; buf[i]; i++) {
372 * Why not ban mercilessly stupid daemons who can't keep up
373 * with us? (There is no much use of a partial read here...)
376 * XXX note that in such cases Linux FUSE throws EIO at the
377 * syscall invoker and stands back to the message queue. The
378 * rationale should be made clear (and possibly adopt that
379 * behaviour). Keeping the current scheme at least makes
380 * fallacy as loud as possible...
382 if (uio->uio_resid < buflen[i]) {
383 fdata_set_dead(data);
384 SDT_PROBE2(fusefs, , device, trace, 2,
385 "daemon is stupid, kick it off...");
389 err = uiomove(buf[i], buflen[i], uio);
394 FUSE_ASSERT_MS_DONE(tick);
395 fuse_ticket_drop(tick);
401 fuse_ohead_audit(struct fuse_out_header *ohead, struct uio *uio)
403 if (uio->uio_resid + sizeof(struct fuse_out_header) != ohead->len) {
404 SDT_PROBE2(fusefs, , device, trace, 1,
405 "Format error: body size "
406 "differs from size claimed by header");
409 if (uio->uio_resid && ohead->unique != 0 && ohead->error) {
410 SDT_PROBE2(fusefs, , device, trace, 1,
411 "Format error: non zero error but message had a body");
418 SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_notify,
419 "struct fuse_out_header*");
420 SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_missing_ticket,
422 SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_found,
423 "struct fuse_ticket*");
425 * fuse_device_write first reads the header sent by the daemon.
426 * If that's OK, looks up ticket/callback node by the unique id seen in header.
427 * If the callback node contains a handler function, the uio is passed over
431 fuse_device_write(struct cdev *dev, struct uio *uio, int ioflag)
433 struct fuse_out_header ohead;
435 struct fuse_data *data;
437 struct fuse_ticket *tick, *itick, *x_tick;
440 err = devfs_get_cdevpriv((void **)&data);
445 if (uio->uio_resid < sizeof(struct fuse_out_header)) {
446 SDT_PROBE2(fusefs, , device, trace, 1,
447 "fuse_device_write got less than a header!");
448 fdata_set_dead(data);
451 if ((err = uiomove(&ohead, sizeof(struct fuse_out_header), uio)) != 0)
455 * We check header information (which is redundant) and compare it
456 * with what we see. If we see some inconsistency we discard the
457 * whole answer and proceed on as if it had never existed. In
458 * particular, no pretender will be woken up, regardless the
459 * "unique" value in the header.
461 if ((err = fuse_ohead_audit(&ohead, uio))) {
462 fdata_set_dead(data);
465 /* Pass stuff over to callback if there is one installed */
467 /* Looking for ticket with the unique id of header */
468 fuse_lck_mtx_lock(data->aw_mtx);
469 TAILQ_FOREACH_SAFE(tick, &data->aw_head, tk_aw_link,
471 if (tick->tk_unique == ohead.unique) {
472 SDT_PROBE1(fusefs, , device, fuse_device_write_found,
475 fuse_aw_remove(tick);
479 if (found && tick->irq_unique > 0) {
481 * Discard the FUSE_INTERRUPT ticket that tried to interrupt
484 TAILQ_FOREACH_SAFE(itick, &data->aw_head, tk_aw_link,
486 if (itick->tk_unique == tick->irq_unique) {
487 fuse_aw_remove(itick);
488 fuse_ticket_drop(itick);
492 tick->irq_unique = 0;
494 fuse_lck_mtx_unlock(data->aw_mtx);
497 if (tick->tk_aw_handler) {
499 * We found a callback with proper handler. In this
500 * case the out header will be 0wnd by the callback,
501 * so the fun of freeing that is left for her.
502 * (Then, by all chance, she'll just get that's done
503 * via ticket_drop(), so no manual mucking
506 SDT_PROBE2(fusefs, , device, trace, 1,
507 "pass ticket to a callback");
508 /* Sanitize the linuxism of negative errnos */
510 memcpy(&tick->tk_aw_ohead, &ohead, sizeof(ohead));
511 err = tick->tk_aw_handler(tick, uio);
513 /* pretender doesn't wanna do anything with answer */
514 SDT_PROBE2(fusefs, , device, trace, 1,
515 "stuff devalidated, so we drop it");
519 * As aw_mtx was not held during the callback execution the
520 * ticket may have been inserted again. However, this is safe
521 * because fuse_ticket_drop() will deal with refcount anyway.
523 fuse_ticket_drop(tick);
524 } else if (ohead.unique == 0){
525 /* unique == 0 means asynchronous notification */
526 SDT_PROBE1(fusefs, , device, fuse_device_write_notify, &ohead);
527 switch (ohead.error) {
528 case FUSE_NOTIFY_INVAL_ENTRY:
529 err = fuse_internal_invalidate_entry(mp, uio);
531 case FUSE_NOTIFY_INVAL_INODE:
532 err = fuse_internal_invalidate_inode(mp, uio);
534 case FUSE_NOTIFY_RETRIEVE:
535 case FUSE_NOTIFY_STORE:
537 * Unimplemented. I don't know of any file systems
538 * that use them, and the protocol isn't sound anyway,
539 * since the notification messages don't include the
540 * inode's generation number. Without that, it's
541 * possible to manipulate the cache of the wrong vnode.
542 * Finally, it's not defined what this message should
543 * do for a file with dirty cache.
545 case FUSE_NOTIFY_POLL:
546 /* Unimplemented. See comments in fuse_vnops */
548 /* Not implemented */
552 /* no callback at all! */
553 SDT_PROBE1(fusefs, , device, fuse_device_write_missing_ticket,
555 if (ohead.error == -EAGAIN) {
557 * This was probably a response to a FUSE_INTERRUPT
558 * operation whose original operation is already
559 * complete. We can't store FUSE_INTERRUPT tickets
560 * indefinitely because their responses are optional.
561 * So we delete them when the original operation
562 * completes. And sadly the fuse_header_out doesn't
563 * identify the opcode, so we have to guess.
575 fuse_device_init(void)
578 fuse_dev = make_dev(&fuse_device_cdevsw, 0, UID_ROOT, GID_OPERATOR,
579 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH, "fuse");
580 if (fuse_dev == NULL)
586 fuse_device_destroy(void)
589 MPASS(fuse_dev != NULL);
590 destroy_dev(fuse_dev);