xref: /freebsd/sys/fs/fuse/fuse_device.c (revision eda14cbc264d6969b02f2b1994cef11148e914f1)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2007-2009 Google Inc.
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/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>
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/stat.h>
83 #include <sys/fcntl.h>
84 #include <sys/sysctl.h>
85 #include <sys/poll.h>
86 #include <sys/selinfo.h>
87 
88 #include "fuse.h"
89 #include "fuse_internal.h"
90 #include "fuse_ipc.h"
91 
92 SDT_PROVIDER_DECLARE(fusefs);
93 /*
94  * Fuse trace probe:
95  * arg0: verbosity.  Higher numbers give more verbose messages
96  * arg1: Textual message
97  */
98 SDT_PROBE_DEFINE2(fusefs, , device, trace, "int", "char*");
99 
100 static struct cdev *fuse_dev;
101 
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;
107 
108 static struct cdevsw fuse_device_cdevsw = {
109 	.d_kqfilter = fuse_device_filter,
110 	.d_open = fuse_device_open,
111 	.d_name = "fuse",
112 	.d_poll = fuse_device_poll,
113 	.d_read = fuse_device_read,
114 	.d_write = fuse_device_write,
115 	.d_version = D_VERSION,
116 };
117 
118 static int fuse_device_filt_read(struct knote *kn, long hint);
119 static void fuse_device_filt_detach(struct knote *kn);
120 
121 struct filterops fuse_device_rfiltops = {
122 	.f_isfd = 1,
123 	.f_detach = fuse_device_filt_detach,
124 	.f_event = fuse_device_filt_read,
125 };
126 
127 /****************************
128  *
129  * >>> Fuse device op defs
130  *
131  ****************************/
132 
133 static void
134 fdata_dtor(void *arg)
135 {
136 	struct fuse_data *fdata;
137 	struct fuse_ticket *tick;
138 
139 	fdata = arg;
140 	if (fdata == NULL)
141 		return;
142 
143 	fdata_set_dead(fdata);
144 
145 	FUSE_LOCK();
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;
154 		wakeup(tick);
155 		fuse_lck_mtx_unlock(tick->tk_aw_mtx);
156 		FUSE_ASSERT_AW_DONE(tick);
157 		fuse_ticket_drop(tick);
158 	}
159 	fuse_lck_mtx_unlock(fdata->aw_mtx);
160 
161 	/* Cleanup unsent operations */
162 	fuse_lck_mtx_lock(fdata->ms_mtx);
163 	while ((tick = fuse_ms_pop(fdata))) {
164 		fuse_ticket_drop(tick);
165 	}
166 	fuse_lck_mtx_unlock(fdata->ms_mtx);
167 	FUSE_UNLOCK();
168 
169 	fdata_trydestroy(fdata);
170 }
171 
172 static int
173 fuse_device_filter(struct cdev *dev, struct knote *kn)
174 {
175 	struct fuse_data *data;
176 	int error;
177 
178 	error = devfs_get_cdevpriv((void **)&data);
179 
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;
183 		kn->kn_hook = data;
184 		knlist_add(&data->ks_rsel.si_note, kn, 0);
185 		error = 0;
186 	} else if (error == 0) {
187 		error = EINVAL;
188 		kn->kn_data = error;
189 	}
190 
191 	return (error);
192 }
193 
194 static void
195 fuse_device_filt_detach(struct knote *kn)
196 {
197 	struct fuse_data *data;
198 
199 	data = (struct fuse_data*)kn->kn_hook;
200 	MPASS(data != NULL);
201 	knlist_remove(&data->ks_rsel.si_note, kn, 0);
202 	kn->kn_hook = NULL;
203 }
204 
205 static int
206 fuse_device_filt_read(struct knote *kn, long hint)
207 {
208 	struct fuse_data *data;
209 	int ready;
210 
211 	data = (struct fuse_data*)kn->kn_hook;
212 	MPASS(data != NULL);
213 
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;
218 		kn->kn_data = 1;
219 		ready = 1;
220 	} else if (STAILQ_FIRST(&data->ms_head)) {
221 		MPASS(data->ms_count >= 1);
222 		kn->kn_data = data->ms_count;
223 		ready = 1;
224 	} else {
225 		ready = 0;
226 	}
227 
228 	return (ready);
229 }
230 
231 /*
232  * Resources are set up on a per-open basis
233  */
234 static int
235 fuse_device_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
236 {
237 	struct fuse_data *fdata;
238 	int error;
239 
240 	SDT_PROBE2(fusefs, , device, trace, 1, "device open");
241 
242 	fdata = fdata_alloc(dev, td->td_ucred);
243 	error = devfs_set_cdevpriv(fdata, fdata_dtor);
244 	if (error != 0)
245 		fdata_trydestroy(fdata);
246 	else
247 		SDT_PROBE2(fusefs, , device, trace, 1, "device open success");
248 	return (error);
249 }
250 
251 int
252 fuse_device_poll(struct cdev *dev, int events, struct thread *td)
253 {
254 	struct fuse_data *data;
255 	int error, revents = 0;
256 
257 	error = devfs_get_cdevpriv((void **)&data);
258 	if (error != 0)
259 		return (events &
260 		    (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
261 
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);
266 		else
267 			selrecord(td, &data->ks_rsel);
268 		fuse_lck_mtx_unlock(data->ms_mtx);
269 	}
270 	if (events & (POLLOUT | POLLWRNORM)) {
271 		revents |= events & (POLLOUT | POLLWRNORM);
272 	}
273 	return (revents);
274 }
275 
276 /*
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
280  */
281 int
282 fuse_device_read(struct cdev *dev, struct uio *uio, int ioflag)
283 {
284 	int err;
285 	struct fuse_data *data;
286 	struct fuse_ticket *tick;
287 	void *buf[] = {NULL, NULL, NULL};
288 	int buflen[3];
289 	int i;
290 
291 	SDT_PROBE2(fusefs, , device, trace, 1, "fuse device read");
292 
293 	err = devfs_get_cdevpriv((void **)&data);
294 	if (err != 0)
295 		return (err);
296 
297 	fuse_lck_mtx_lock(data->ms_mtx);
298 again:
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);
304 		return (ENODEV);
305 	}
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);
311 			return (EAGAIN);
312 		} else {
313 			err = msleep(data, &data->ms_mtx, PCATCH, "fu_msg", 0);
314 			if (err != 0) {
315 				fuse_lck_mtx_unlock(data->ms_mtx);
316 				return (fdata_get_dead(data) ? ENODEV : err);
317 			}
318 			tick = fuse_ms_pop(data);
319 		}
320 	}
321 	if (!tick) {
322 		/*
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 ?)
327 		 */
328 		SDT_PROBE2(fusefs, , device, trace, 1, "no message on thread");
329 		goto again;
330 	}
331 	fuse_lck_mtx_unlock(data->ms_mtx);
332 
333 	if (fdata_get_dead(data)) {
334 		/*
335 		 * somebody somewhere -- eg., umount routine --
336 		 * wants this liaison finished off
337 		 */
338 		SDT_PROBE2(fusefs, , device, trace, 2,
339 			"reader is to be sacked");
340 		if (tick) {
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);
345 		}
346 		return (ENODEV);	/* This should make the daemon get off
347 					 * of us */
348 	}
349 	SDT_PROBE2(fusefs, , device, trace, 1,
350 		"fuse device read message successfully");
351 
352 	KASSERT(tick->tk_ms_bufdata || tick->tk_ms_bufsize == 0,
353 	    ("non-null buf pointer with positive size"));
354 
355 	switch (tick->tk_ms_type) {
356 	case FT_M_FIOV:
357 		buf[0] = tick->tk_ms_fiov.base;
358 		buflen[0] = tick->tk_ms_fiov.len;
359 		break;
360 	case FT_M_BUF:
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;
365 		break;
366 	default:
367 		panic("unknown message type for fuse_ticket %p", tick);
368 	}
369 
370 	for (i = 0; buf[i]; i++) {
371 		/*
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...)
374 		 */
375 		/*
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...
381 		 */
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...");
386 			err = ENODEV;
387 			break;
388 		}
389 		err = uiomove(buf[i], buflen[i], uio);
390 		if (err)
391 			break;
392 	}
393 
394 	FUSE_ASSERT_MS_DONE(tick);
395 	fuse_ticket_drop(tick);
396 
397 	return (err);
398 }
399 
400 static inline int
401 fuse_ohead_audit(struct fuse_out_header *ohead, struct uio *uio)
402 {
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");
407 		return (EINVAL);
408 	}
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");
412 		return (EINVAL);
413 	}
414 
415 	return (0);
416 }
417 
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,
421 	"uint64_t");
422 SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_found,
423 	"struct fuse_ticket*");
424 /*
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
428  * that.
429  */
430 static int
431 fuse_device_write(struct cdev *dev, struct uio *uio, int ioflag)
432 {
433 	struct fuse_out_header ohead;
434 	int err = 0;
435 	struct fuse_data *data;
436 	struct mount *mp;
437 	struct fuse_ticket *tick, *itick, *x_tick;
438 	int found = 0;
439 
440 	err = devfs_get_cdevpriv((void **)&data);
441 	if (err != 0)
442 		return (err);
443 	mp = data->mp;
444 
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);
449 		return (EINVAL);
450 	}
451 	if ((err = uiomove(&ohead, sizeof(struct fuse_out_header), uio)) != 0)
452 		return (err);
453 
454 	/*
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.
460 	 */
461 	if ((err = fuse_ohead_audit(&ohead, uio))) {
462 		fdata_set_dead(data);
463 		return (err);
464 	}
465 	/* Pass stuff over to callback if there is one installed */
466 
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,
470 	    x_tick) {
471 		if (tick->tk_unique == ohead.unique) {
472 			SDT_PROBE1(fusefs, , device, fuse_device_write_found,
473 				tick);
474 			found = 1;
475 			fuse_aw_remove(tick);
476 			break;
477 		}
478 	}
479 	if (found && tick->irq_unique > 0) {
480 		/*
481 		 * Discard the FUSE_INTERRUPT ticket that tried to interrupt
482 		 * this operation
483 		 */
484 		TAILQ_FOREACH_SAFE(itick, &data->aw_head, tk_aw_link,
485 		    x_tick) {
486 			if (itick->tk_unique == tick->irq_unique) {
487 				fuse_aw_remove(itick);
488 				fuse_ticket_drop(itick);
489 				break;
490 			}
491 		}
492 		tick->irq_unique = 0;
493 	}
494 	fuse_lck_mtx_unlock(data->aw_mtx);
495 
496 	if (found) {
497 		if (tick->tk_aw_handler) {
498 			/*
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
504 			 * around...)
505 			 */
506 			SDT_PROBE2(fusefs, , device, trace, 1,
507 				"pass ticket to a callback");
508 			/* Sanitize the linuxism of negative errnos */
509 			ohead.error *= -1;
510 			memcpy(&tick->tk_aw_ohead, &ohead, sizeof(ohead));
511 			err = tick->tk_aw_handler(tick, uio);
512 		} else {
513 			/* pretender doesn't wanna do anything with answer */
514 			SDT_PROBE2(fusefs, , device, trace, 1,
515 				"stuff devalidated, so we drop it");
516 		}
517 
518 		/*
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.
522 		 */
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);
530 			break;
531 		case FUSE_NOTIFY_INVAL_INODE:
532 			err = fuse_internal_invalidate_inode(mp, uio);
533 			break;
534 		case FUSE_NOTIFY_RETRIEVE:
535 		case FUSE_NOTIFY_STORE:
536 			/*
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.
544 			 */
545 		case FUSE_NOTIFY_POLL:
546 			/* Unimplemented.  See comments in fuse_vnops */
547 		default:
548 			/* Not implemented */
549 			err = ENOSYS;
550 		}
551 	} else {
552 		/* no callback at all! */
553 		SDT_PROBE1(fusefs, , device, fuse_device_write_missing_ticket,
554 			ohead.unique);
555 		if (ohead.error == -EAGAIN) {
556 			/*
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.
564 			 */
565 			err = 0;
566 		} else {
567 			err = EINVAL;
568 		}
569 	}
570 
571 	return (err);
572 }
573 
574 int
575 fuse_device_init(void)
576 {
577 
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)
581 		return (ENOMEM);
582 	return (0);
583 }
584 
585 void
586 fuse_device_destroy(void)
587 {
588 
589 	MPASS(fuse_dev != NULL);
590 	destroy_dev(fuse_dev);
591 }
592