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/types.h>
65 #include <sys/systm.h>
66 #include <sys/counter.h>
67 #include <sys/module.h>
68 #include <sys/errno.h>
69 #include <sys/param.h>
70 #include <sys/kernel.h>
71 #include <sys/conf.h>
72 #include <sys/uio.h>
73 #include <sys/malloc.h>
74 #include <sys/queue.h>
75 #include <sys/lock.h>
76 #include <sys/sx.h>
77 #include <sys/mutex.h>
78 #include <sys/proc.h>
79 #include <sys/vnode.h>
80 #include <sys/namei.h>
81 #include <sys/mount.h>
82 #include <sys/sysctl.h>
83 #include <sys/fcntl.h>
84 #include <sys/priv.h>
85 #include <sys/buf.h>
86 #include <security/mac/mac_framework.h>
87 #include <vm/vm.h>
88 #include <vm/vm_extern.h>
89
90 #include "fuse.h"
91 #include "fuse_node.h"
92 #include "fuse_internal.h"
93 #include "fuse_io.h"
94 #include "fuse_ipc.h"
95
96 SDT_PROVIDER_DECLARE(fusefs);
97 /*
98 * Fuse trace probe:
99 * arg0: verbosity. Higher numbers give more verbose messages
100 * arg1: Textual message
101 */
102 SDT_PROBE_DEFINE2(fusefs, , node, trace, "int", "char*");
103
104 MALLOC_DEFINE(M_FUSEVN, "fuse_vnode", "fuse vnode private data");
105
106 static int sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS);
107
108 static counter_u64_t fuse_node_count;
109
110 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, node_count, CTLFLAG_RD,
111 &fuse_node_count, "Count of FUSE vnodes");
112
113 int fuse_data_cache_mode = FUSE_CACHE_WT;
114
115 /*
116 * OBSOLETE
117 * This sysctl is no longer needed as of fuse protocol 7.23. Now, individual
118 * servers can select the cache behavior they need for each mountpoint:
119 * - writethrough: the default
120 * - writeback: set FUSE_WRITEBACK_CACHE in fuse_init_out.flags
121 * - uncached: set FOPEN_DIRECT_IO for every file
122 * The sysctl is retained primarily due to the enduring popularity of libfuse2,
123 * which is frozen at protocol version 7.19. As of 4-April-2024, 90% of
124 * FreeBSD ports that use libfuse still bind to libfuse2.
125 */
126 SYSCTL_PROC(_vfs_fusefs, OID_AUTO, data_cache_mode,
127 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW,
128 &fuse_data_cache_mode, 0, sysctl_fuse_cache_mode, "I",
129 "Zero: disable caching of FUSE file data; One: write-through caching "
130 "(default); Two: write-back caching (generally unsafe)");
131
132 static int
sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS)133 sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS)
134 {
135 int val, error;
136
137 val = *(int *)arg1;
138 error = sysctl_handle_int(oidp, &val, 0, req);
139 if (error || !req->newptr)
140 return (error);
141
142 switch (val) {
143 case FUSE_CACHE_UC:
144 case FUSE_CACHE_WT:
145 case FUSE_CACHE_WB:
146 *(int *)arg1 = val;
147 break;
148 default:
149 return (EDOM);
150 }
151 return (0);
152 }
153
154 static void
fuse_vnode_init(struct vnode * vp,struct fuse_vnode_data * fvdat,uint64_t nodeid,__enum_uint8 (vtype)vtyp)155 fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat,
156 uint64_t nodeid, __enum_uint8(vtype) vtyp)
157 {
158 fvdat->nid = nodeid;
159 LIST_INIT(&fvdat->handles);
160
161 vattr_null(&fvdat->cached_attrs);
162 fvdat->cached_attrs.va_birthtime.tv_sec = -1;
163 fvdat->cached_attrs.va_birthtime.tv_nsec = 0;
164 fvdat->cached_attrs.va_fsid = VNOVAL;
165 fvdat->cached_attrs.va_gen = 0;
166 fvdat->cached_attrs.va_rdev = NODEV;
167
168 if (nodeid == FUSE_ROOT_ID) {
169 vp->v_vflag |= VV_ROOT;
170 }
171 vp->v_type = vtyp;
172 vp->v_data = fvdat;
173 cluster_init_vn(&fvdat->clusterw);
174 timespecclear(&fvdat->last_local_modify);
175
176 counter_u64_add(fuse_node_count, 1);
177 }
178
179 void
fuse_vnode_destroy(struct vnode * vp)180 fuse_vnode_destroy(struct vnode *vp)
181 {
182 struct fuse_vnode_data *fvdat = vp->v_data;
183
184 vp->v_data = NULL;
185 KASSERT(LIST_EMPTY(&fvdat->handles),
186 ("Destroying fuse vnode with open files!"));
187 free(fvdat, M_FUSEVN);
188
189 counter_u64_add(fuse_node_count, -1);
190 }
191
192 int
fuse_vnode_cmp(struct vnode * vp,void * nidp)193 fuse_vnode_cmp(struct vnode *vp, void *nidp)
194 {
195 return (VTOI(vp) != *((uint64_t *)nidp));
196 }
197
198 SDT_PROBE_DEFINE3(fusefs, , node, stale_vnode, "struct vnode*", "uint8_t",
199 "uint64_t");
200 static int
fuse_vnode_alloc(struct mount * mp,struct thread * td,uint64_t nodeid,__enum_uint8 (vtype)vtyp,struct vnode ** vpp)201 fuse_vnode_alloc(struct mount *mp,
202 struct thread *td,
203 uint64_t nodeid,
204 __enum_uint8(vtype) vtyp,
205 struct vnode **vpp)
206 {
207 struct fuse_data *data;
208 struct fuse_vnode_data *fvdat;
209 struct vnode *vp2;
210 int err = 0;
211
212 data = fuse_get_mpdata(mp);
213 if (vtyp == VNON) {
214 return EINVAL;
215 }
216 *vpp = NULL;
217 err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp,
218 fuse_vnode_cmp, &nodeid);
219 if (err)
220 return (err);
221
222 if (*vpp) {
223 if ((*vpp)->v_type == vtyp) {
224 /* Reuse a vnode that hasn't yet been reclaimed */
225 MPASS((*vpp)->v_data != NULL);
226 MPASS(VTOFUD(*vpp)->nid == nodeid);
227 SDT_PROBE2(fusefs, , node, trace, 1,
228 "vnode taken from hash");
229 return (0);
230 } else {
231 /*
232 * The inode changed types! If we get here, we can't
233 * tell whether the inode's entry cache had expired
234 * yet. So this could be the result of a buggy server,
235 * but more likely the server just reused an inode
236 * number following an entry cache expiration.
237 */
238 SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp,
239 nodeid);
240 fuse_internal_vnode_disappear(*vpp);
241 vgone(*vpp);
242 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
243 }
244 }
245 fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO);
246 switch (vtyp) {
247 case VFIFO:
248 err = getnewvnode("fuse", mp, &fuse_fifoops, vpp);
249 break;
250 default:
251 err = getnewvnode("fuse", mp, &fuse_vnops, vpp);
252 break;
253 }
254 if (err) {
255 free(fvdat, M_FUSEVN);
256 return (err);
257 }
258 lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL);
259 fuse_vnode_init(*vpp, fvdat, nodeid, vtyp);
260 err = insmntque(*vpp, mp);
261 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
262 if (err) {
263 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
264 free(fvdat, M_FUSEVN);
265 *vpp = NULL;
266 return (err);
267 }
268 /* Disallow async reads for fifos because UFS does. I don't know why */
269 if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO)
270 VN_LOCK_ASHARE(*vpp);
271
272 vn_set_state(*vpp, VSTATE_CONSTRUCTED);
273 err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE,
274 td, &vp2, fuse_vnode_cmp, &nodeid);
275 if (err) {
276 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
277 free(fvdat, M_FUSEVN);
278 *vpp = NULL;
279 return (err);
280 }
281 if (vp2 != NULL) {
282 *vpp = vp2;
283 return (0);
284 }
285
286 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
287
288 return (0);
289 }
290
291 int
fuse_vnode_get(struct mount * mp,struct fuse_entry_out * feo,uint64_t nodeid,struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,__enum_uint8 (vtype)vtyp)292 fuse_vnode_get(struct mount *mp,
293 struct fuse_entry_out *feo,
294 uint64_t nodeid,
295 struct vnode *dvp,
296 struct vnode **vpp,
297 struct componentname *cnp,
298 __enum_uint8(vtype) vtyp)
299 {
300 struct thread *td = curthread;
301 /*
302 * feo should only be NULL for the root directory, which (when libfuse
303 * is used) always has generation 0
304 */
305 uint64_t generation = feo ? feo->generation : 0;
306 int err = 0;
307
308 if (dvp != NULL && VTOFUD(dvp)->nid == nodeid) {
309 fuse_warn(fuse_get_mpdata(mp), FSESS_WARN_ILLEGAL_INODE,
310 "Assigned same inode to both parent and child.");
311 return EIO;
312 }
313
314 err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp);
315 if (err) {
316 return err;
317 }
318 if (dvp != NULL) {
319 MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0);
320 MPASS(cnp &&
321 !(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.'));
322 fuse_vnode_setparent(*vpp, dvp);
323 }
324 if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 &&
325 feo != NULL &&
326 (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) {
327 struct timespec timeout;
328
329 ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get");
330 ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get");
331
332 fuse_validity_2_timespec(feo, &timeout);
333 cache_enter_time(dvp, *vpp, cnp, &timeout, NULL);
334 }
335
336 VTOFUD(*vpp)->generation = generation;
337 /*
338 * In userland, libfuse uses cached lookups for dot and dotdot entries,
339 * thus it does not really bump the nlookup counter for forget.
340 * Follow the same semantic and avoid the bump in order to keep
341 * nlookup counters consistent.
342 */
343 if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 &&
344 (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.')))
345 VTOFUD(*vpp)->nlookup++;
346
347 return 0;
348 }
349
350 /*
351 * Called for every fusefs vnode open to initialize the vnode (not
352 * fuse_filehandle) for use
353 */
354 void
fuse_vnode_open(struct vnode * vp,int32_t fuse_open_flags,struct thread * td)355 fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td)
356 {
357 if (vnode_vtype(vp) == VREG)
358 vnode_create_vobject(vp, 0, td);
359 }
360
361 int
fuse_vnode_savesize(struct vnode * vp,struct ucred * cred,pid_t pid)362 fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid)
363 {
364 struct fuse_vnode_data *fvdat = VTOFUD(vp);
365 struct thread *td = curthread;
366 struct fuse_filehandle *fufh = NULL;
367 struct fuse_dispatcher fdi;
368 struct fuse_setattr_in *fsai;
369 int err = 0;
370
371 ASSERT_VOP_ELOCKED(vp, "fuse_io_extend");
372
373 if (fuse_isdeadfs(vp)) {
374 return EBADF;
375 }
376 if (vnode_vtype(vp) == VDIR) {
377 return EISDIR;
378 }
379 if (vfs_isrdonly(vnode_mount(vp))) {
380 return EROFS;
381 }
382 if (cred == NULL) {
383 cred = td->td_ucred;
384 }
385 fdisp_init(&fdi, sizeof(*fsai));
386 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
387 fsai = fdi.indata;
388 fsai->valid = 0;
389
390 /* Truncate to a new value. */
391 MPASS((fvdat->flag & FN_SIZECHANGE) != 0);
392 fsai->size = fvdat->cached_attrs.va_size;
393 fsai->valid |= FATTR_SIZE;
394
395 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
396 if (fufh) {
397 fsai->fh = fufh->fh_id;
398 fsai->valid |= FATTR_FH;
399 }
400 err = fdisp_wait_answ(&fdi);
401 fdisp_destroy(&fdi);
402 if (err == 0) {
403 getnanouptime(&fvdat->last_local_modify);
404 fvdat->flag &= ~FN_SIZECHANGE;
405 }
406
407 return err;
408 }
409
410 /*
411 * Adjust the vnode's size to a new value.
412 *
413 * If the new value came from the server, such as from a FUSE_GETATTR
414 * operation, set `from_server` true. But if it came from a local operation,
415 * such as write(2) or truncate(2), set `from_server` false.
416 */
417 int
fuse_vnode_setsize(struct vnode * vp,off_t newsize,bool from_server)418 fuse_vnode_setsize(struct vnode *vp, off_t newsize, bool from_server)
419 {
420 struct fuse_vnode_data *fvdat = VTOFUD(vp);
421 struct vattr *attrs;
422 off_t oldsize;
423 size_t iosize;
424 struct buf *bp = NULL;
425 int err = 0;
426
427 ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize");
428
429 iosize = fuse_iosize(vp);
430 oldsize = fvdat->cached_attrs.va_size;
431 fvdat->cached_attrs.va_size = newsize;
432 if ((attrs = VTOVA(vp)) != NULL)
433 attrs->va_size = newsize;
434
435 if (newsize < oldsize) {
436 daddr_t lbn;
437
438 err = vtruncbuf(vp, newsize, fuse_iosize(vp));
439 if (err)
440 goto out;
441 if (newsize % iosize == 0)
442 goto out;
443 /*
444 * Zero the contents of the last partial block.
445 * Sure seems like vtruncbuf should do this for us.
446 */
447
448 lbn = newsize / iosize;
449 bp = getblk(vp, lbn, iosize, PCATCH, 0, 0);
450 if (!bp) {
451 err = EINTR;
452 goto out;
453 }
454 if (!(bp->b_flags & B_CACHE))
455 goto out; /* Nothing to do */
456 MPASS(bp->b_flags & B_VMIO);
457 vfs_bio_clrbuf(bp);
458 bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize);
459 } else if (from_server && newsize > oldsize && oldsize != VNOVAL) {
460 /*
461 * The FUSE server changed the file size behind our back. We
462 * should invalidate the entire cache.
463 */
464 daddr_t end_lbn;
465
466 end_lbn = howmany(newsize, iosize);
467 v_inval_buf_range(vp, 0, end_lbn, iosize);
468 }
469 out:
470 if (bp)
471 brelse(bp);
472 vnode_pager_setsize(vp, newsize);
473 return err;
474 }
475
476 /* Get the current, possibly dirty, size of the file */
477 int
fuse_vnode_size(struct vnode * vp,off_t * filesize,struct ucred * cred,struct thread * td)478 fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred,
479 struct thread *td)
480 {
481 struct fuse_vnode_data *fvdat = VTOFUD(vp);
482 int error = 0;
483
484 if (!(fvdat->flag & FN_SIZECHANGE) &&
485 (!fuse_vnode_attr_cache_valid(vp) ||
486 fvdat->cached_attrs.va_size == VNOVAL))
487 error = fuse_internal_do_getattr(vp, NULL, cred, td);
488
489 if (!error)
490 *filesize = fvdat->cached_attrs.va_size;
491
492 return error;
493 }
494
495 void
fuse_vnode_undirty_cached_timestamps(struct vnode * vp,bool atime)496 fuse_vnode_undirty_cached_timestamps(struct vnode *vp, bool atime)
497 {
498 struct fuse_vnode_data *fvdat = VTOFUD(vp);
499
500 fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE);
501 if (atime)
502 fvdat->flag &= ~FN_ATIMECHANGE;
503 }
504
505 /* Update a fuse file's cached timestamps */
506 void
fuse_vnode_update(struct vnode * vp,int flags)507 fuse_vnode_update(struct vnode *vp, int flags)
508 {
509 struct fuse_vnode_data *fvdat = VTOFUD(vp);
510 struct mount *mp = vnode_mount(vp);
511 struct fuse_data *data = fuse_get_mpdata(mp);
512 struct timespec ts;
513
514 vfs_timestamp(&ts);
515
516 if (data->time_gran > 1)
517 ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran);
518
519 if (mp->mnt_flag & MNT_NOATIME)
520 flags &= ~FN_ATIMECHANGE;
521
522 if (flags & FN_ATIMECHANGE)
523 fvdat->cached_attrs.va_atime = ts;
524 if (flags & FN_MTIMECHANGE)
525 fvdat->cached_attrs.va_mtime = ts;
526 if (flags & FN_CTIMECHANGE)
527 fvdat->cached_attrs.va_ctime = ts;
528
529 fvdat->flag |= flags;
530 }
531
532 void
fuse_node_init(void)533 fuse_node_init(void)
534 {
535 fuse_node_count = counter_u64_alloc(M_WAITOK);
536 }
537
538 void
fuse_node_destroy(void)539 fuse_node_destroy(void)
540 {
541 counter_u64_free(fuse_node_count);
542 }
543