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/types.h>
67 #include <sys/systm.h>
68 #include <sys/counter.h>
69 #include <sys/module.h>
70 #include <sys/errno.h>
71 #include <sys/param.h>
72 #include <sys/kernel.h>
73 #include <sys/conf.h>
74 #include <sys/uio.h>
75 #include <sys/malloc.h>
76 #include <sys/queue.h>
77 #include <sys/lock.h>
78 #include <sys/sx.h>
79 #include <sys/mutex.h>
80 #include <sys/proc.h>
81 #include <sys/vnode.h>
82 #include <sys/namei.h>
83 #include <sys/mount.h>
84 #include <sys/sysctl.h>
85 #include <sys/fcntl.h>
86 #include <sys/priv.h>
87 #include <sys/buf.h>
88 #include <security/mac/mac_framework.h>
89 #include <vm/vm.h>
90 #include <vm/vm_extern.h>
91
92 #include "fuse.h"
93 #include "fuse_node.h"
94 #include "fuse_internal.h"
95 #include "fuse_io.h"
96 #include "fuse_ipc.h"
97
98 SDT_PROVIDER_DECLARE(fusefs);
99 /*
100 * Fuse trace probe:
101 * arg0: verbosity. Higher numbers give more verbose messages
102 * arg1: Textual message
103 */
104 SDT_PROBE_DEFINE2(fusefs, , node, trace, "int", "char*");
105
106 MALLOC_DEFINE(M_FUSEVN, "fuse_vnode", "fuse vnode private data");
107
108 static int sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS);
109
110 static counter_u64_t fuse_node_count;
111
112 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, node_count, CTLFLAG_RD,
113 &fuse_node_count, "Count of FUSE vnodes");
114
115 int fuse_data_cache_mode = FUSE_CACHE_WT;
116
117 /*
118 * DEPRECATED
119 * This sysctl is no longer needed as of fuse protocol 7.23. Individual
120 * servers can select the cache behavior they need for each mountpoint:
121 * - writethrough: the default
122 * - writeback: set FUSE_WRITEBACK_CACHE in fuse_init_out.flags
123 * - uncached: set FOPEN_DIRECT_IO for every file
124 * The sysctl is retained primarily for use by jails supporting older FUSE
125 * protocols. It may be removed entirely once FreeBSD 11.3 and 12.0 are EOL.
126 */
127 SYSCTL_PROC(_vfs_fusefs, OID_AUTO, data_cache_mode,
128 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW,
129 &fuse_data_cache_mode, 0, sysctl_fuse_cache_mode, "I",
130 "Zero: disable caching of FUSE file data; One: write-through caching "
131 "(default); Two: write-back caching (generally unsafe)");
132
133 static int
sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS)134 sysctl_fuse_cache_mode(SYSCTL_HANDLER_ARGS)
135 {
136 int val, error;
137
138 val = *(int *)arg1;
139 error = sysctl_handle_int(oidp, &val, 0, req);
140 if (error || !req->newptr)
141 return (error);
142
143 switch (val) {
144 case FUSE_CACHE_UC:
145 case FUSE_CACHE_WT:
146 case FUSE_CACHE_WB:
147 *(int *)arg1 = val;
148 break;
149 default:
150 return (EDOM);
151 }
152 return (0);
153 }
154
155 static void
fuse_vnode_init(struct vnode * vp,struct fuse_vnode_data * fvdat,uint64_t nodeid,enum vtype vtyp)156 fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat,
157 uint64_t nodeid, enum vtype vtyp)
158 {
159 fvdat->nid = nodeid;
160 LIST_INIT(&fvdat->handles);
161 vattr_null(&fvdat->cached_attrs);
162 if (nodeid == FUSE_ROOT_ID) {
163 vp->v_vflag |= VV_ROOT;
164 }
165 vp->v_type = vtyp;
166 vp->v_data = fvdat;
167 timespecclear(&fvdat->last_local_modify);
168
169 counter_u64_add(fuse_node_count, 1);
170 }
171
172 void
fuse_vnode_destroy(struct vnode * vp)173 fuse_vnode_destroy(struct vnode *vp)
174 {
175 struct fuse_vnode_data *fvdat = vp->v_data;
176
177 vp->v_data = NULL;
178 KASSERT(LIST_EMPTY(&fvdat->handles),
179 ("Destroying fuse vnode with open files!"));
180 free(fvdat, M_FUSEVN);
181
182 counter_u64_add(fuse_node_count, -1);
183 }
184
185 int
fuse_vnode_cmp(struct vnode * vp,void * nidp)186 fuse_vnode_cmp(struct vnode *vp, void *nidp)
187 {
188 return (VTOI(vp) != *((uint64_t *)nidp));
189 }
190
191 SDT_PROBE_DEFINE3(fusefs, , node, stale_vnode, "struct vnode*", "enum vtype",
192 "uint64_t");
193 static int
fuse_vnode_alloc(struct mount * mp,struct thread * td,uint64_t nodeid,enum vtype vtyp,struct vnode ** vpp)194 fuse_vnode_alloc(struct mount *mp,
195 struct thread *td,
196 uint64_t nodeid,
197 enum vtype vtyp,
198 struct vnode **vpp)
199 {
200 struct fuse_data *data;
201 struct fuse_vnode_data *fvdat;
202 struct vnode *vp2;
203 int err = 0;
204
205 data = fuse_get_mpdata(mp);
206 if (vtyp == VNON) {
207 return EINVAL;
208 }
209 *vpp = NULL;
210 err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp,
211 fuse_vnode_cmp, &nodeid);
212 if (err)
213 return (err);
214
215 if (*vpp) {
216 if ((*vpp)->v_type == vtyp) {
217 /* Reuse a vnode that hasn't yet been reclaimed */
218 MPASS((*vpp)->v_data != NULL);
219 MPASS(VTOFUD(*vpp)->nid == nodeid);
220 SDT_PROBE2(fusefs, , node, trace, 1,
221 "vnode taken from hash");
222 return (0);
223 } else {
224 /*
225 * The inode changed types! If we get here, we can't
226 * tell whether the inode's entry cache had expired
227 * yet. So this could be the result of a buggy server,
228 * but more likely the server just reused an inode
229 * number following an entry cache expiration.
230 */
231 SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp,
232 nodeid);
233 fuse_internal_vnode_disappear(*vpp);
234 vgone(*vpp);
235 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
236 }
237 }
238 fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO);
239 switch (vtyp) {
240 case VFIFO:
241 err = getnewvnode("fuse", mp, &fuse_fifoops, vpp);
242 break;
243 default:
244 err = getnewvnode("fuse", mp, &fuse_vnops, vpp);
245 break;
246 }
247 if (err) {
248 free(fvdat, M_FUSEVN);
249 return (err);
250 }
251 lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL);
252 fuse_vnode_init(*vpp, fvdat, nodeid, vtyp);
253 err = insmntque(*vpp, mp);
254 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
255 if (err) {
256 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
257 free(fvdat, M_FUSEVN);
258 *vpp = NULL;
259 return (err);
260 }
261 /* Disallow async reads for fifos because UFS does. I don't know why */
262 if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO)
263 VN_LOCK_ASHARE(*vpp);
264
265 err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE,
266 td, &vp2, fuse_vnode_cmp, &nodeid);
267 if (err) {
268 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
269 free(fvdat, M_FUSEVN);
270 *vpp = NULL;
271 return (err);
272 }
273 if (vp2 != NULL) {
274 *vpp = vp2;
275 return (0);
276 }
277
278 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
279
280 return (0);
281 }
282
283 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 vtype vtyp)284 fuse_vnode_get(struct mount *mp,
285 struct fuse_entry_out *feo,
286 uint64_t nodeid,
287 struct vnode *dvp,
288 struct vnode **vpp,
289 struct componentname *cnp,
290 enum vtype vtyp)
291 {
292 struct thread *td = (cnp != NULL ? cnp->cn_thread : curthread);
293 /*
294 * feo should only be NULL for the root directory, which (when libfuse
295 * is used) always has generation 0
296 */
297 uint64_t generation = feo ? feo->generation : 0;
298 int err = 0;
299
300 err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp);
301 if (err) {
302 return err;
303 }
304 if (dvp != NULL) {
305 MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0);
306 MPASS(cnp &&
307 !(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.'));
308 fuse_vnode_setparent(*vpp, dvp);
309 }
310 if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 &&
311 feo != NULL &&
312 (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) {
313 struct timespec timeout;
314
315 ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get");
316 ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get");
317
318 fuse_validity_2_timespec(feo, &timeout);
319 cache_enter_time(dvp, *vpp, cnp, &timeout, NULL);
320 }
321
322 VTOFUD(*vpp)->generation = generation;
323 /*
324 * In userland, libfuse uses cached lookups for dot and dotdot entries,
325 * thus it does not really bump the nlookup counter for forget.
326 * Follow the same semantic and avoid the bump in order to keep
327 * nlookup counters consistent.
328 */
329 if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 &&
330 (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.')))
331 VTOFUD(*vpp)->nlookup++;
332
333 return 0;
334 }
335
336 /*
337 * Called for every fusefs vnode open to initialize the vnode (not
338 * fuse_filehandle) for use
339 */
340 void
fuse_vnode_open(struct vnode * vp,int32_t fuse_open_flags,struct thread * td)341 fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td)
342 {
343 if (vnode_vtype(vp) == VREG)
344 vnode_create_vobject(vp, 0, td);
345 }
346
347 int
fuse_vnode_savesize(struct vnode * vp,struct ucred * cred,pid_t pid)348 fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid)
349 {
350 struct fuse_vnode_data *fvdat = VTOFUD(vp);
351 struct thread *td = curthread;
352 struct fuse_filehandle *fufh = NULL;
353 struct fuse_dispatcher fdi;
354 struct fuse_setattr_in *fsai;
355 int err = 0;
356
357 ASSERT_VOP_ELOCKED(vp, "fuse_io_extend");
358
359 if (fuse_isdeadfs(vp)) {
360 return EBADF;
361 }
362 if (vnode_vtype(vp) == VDIR) {
363 return EISDIR;
364 }
365 if (vfs_isrdonly(vnode_mount(vp))) {
366 return EROFS;
367 }
368 if (cred == NULL) {
369 cred = td->td_ucred;
370 }
371 fdisp_init(&fdi, sizeof(*fsai));
372 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
373 fsai = fdi.indata;
374 fsai->valid = 0;
375
376 /* Truncate to a new value. */
377 MPASS((fvdat->flag & FN_SIZECHANGE) != 0);
378 fsai->size = fvdat->cached_attrs.va_size;
379 fsai->valid |= FATTR_SIZE;
380
381 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
382 if (fufh) {
383 fsai->fh = fufh->fh_id;
384 fsai->valid |= FATTR_FH;
385 }
386 err = fdisp_wait_answ(&fdi);
387 fdisp_destroy(&fdi);
388 if (err == 0) {
389 getnanouptime(&fvdat->last_local_modify);
390 fvdat->flag &= ~FN_SIZECHANGE;
391 }
392
393 return err;
394 }
395
396 /*
397 * Adjust the vnode's size to a new value.
398 *
399 * If the new value came from the server, such as from a FUSE_GETATTR
400 * operation, set `from_server` true. But if it came from a local operation,
401 * such as write(2) or truncate(2), set `from_server` false.
402 */
403 int
fuse_vnode_setsize(struct vnode * vp,off_t newsize,bool from_server)404 fuse_vnode_setsize(struct vnode *vp, off_t newsize, bool from_server)
405 {
406 struct fuse_vnode_data *fvdat = VTOFUD(vp);
407 struct vattr *attrs;
408 off_t oldsize;
409 size_t iosize;
410 struct buf *bp = NULL;
411 int err = 0;
412
413 ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize");
414
415 iosize = fuse_iosize(vp);
416 oldsize = fvdat->cached_attrs.va_size;
417 fvdat->cached_attrs.va_size = newsize;
418 if ((attrs = VTOVA(vp)) != NULL)
419 attrs->va_size = newsize;
420
421 if (newsize < oldsize) {
422 daddr_t lbn;
423
424 err = vtruncbuf(vp, newsize, fuse_iosize(vp));
425 if (err)
426 goto out;
427 if (newsize % iosize == 0)
428 goto out;
429 /*
430 * Zero the contents of the last partial block.
431 * Sure seems like vtruncbuf should do this for us.
432 */
433
434 lbn = newsize / iosize;
435 bp = getblk(vp, lbn, iosize, PCATCH, 0, 0);
436 if (!bp) {
437 err = EINTR;
438 goto out;
439 }
440 if (!(bp->b_flags & B_CACHE))
441 goto out; /* Nothing to do */
442 MPASS(bp->b_flags & B_VMIO);
443 vfs_bio_clrbuf(bp);
444 bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize);
445 } else if (from_server && newsize > oldsize && oldsize != VNOVAL) {
446 /*
447 * The FUSE server changed the file size behind our back. We
448 * should invalidate the entire cache.
449 */
450 daddr_t end_lbn;
451
452 end_lbn = howmany(newsize, iosize);
453 v_inval_buf_range(vp, 0, end_lbn, iosize);
454 }
455 out:
456 if (bp)
457 brelse(bp);
458 vnode_pager_setsize(vp, newsize);
459 return err;
460 }
461
462 /* Get the current, possibly dirty, size of the file */
463 int
fuse_vnode_size(struct vnode * vp,off_t * filesize,struct ucred * cred,struct thread * td)464 fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred,
465 struct thread *td)
466 {
467 struct fuse_vnode_data *fvdat = VTOFUD(vp);
468 int error = 0;
469
470 if (!(fvdat->flag & FN_SIZECHANGE) &&
471 (!fuse_vnode_attr_cache_valid(vp) ||
472 fvdat->cached_attrs.va_size == VNOVAL))
473 error = fuse_internal_do_getattr(vp, NULL, cred, td);
474
475 if (!error)
476 *filesize = fvdat->cached_attrs.va_size;
477
478 return error;
479 }
480
481 void
fuse_vnode_undirty_cached_timestamps(struct vnode * vp,bool atime)482 fuse_vnode_undirty_cached_timestamps(struct vnode *vp, bool atime)
483 {
484 struct fuse_vnode_data *fvdat = VTOFUD(vp);
485
486 fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE);
487 if (atime)
488 fvdat->flag &= ~FN_ATIMECHANGE;
489 }
490
491 /* Update a fuse file's cached timestamps */
492 void
fuse_vnode_update(struct vnode * vp,int flags)493 fuse_vnode_update(struct vnode *vp, int flags)
494 {
495 struct fuse_vnode_data *fvdat = VTOFUD(vp);
496 struct mount *mp = vnode_mount(vp);
497 struct fuse_data *data = fuse_get_mpdata(mp);
498 struct timespec ts;
499
500 vfs_timestamp(&ts);
501
502 if (data->time_gran > 1)
503 ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran);
504
505 if (mp->mnt_flag & MNT_NOATIME)
506 flags &= ~FN_ATIMECHANGE;
507
508 if (flags & FN_ATIMECHANGE)
509 fvdat->cached_attrs.va_atime = ts;
510 if (flags & FN_MTIMECHANGE)
511 fvdat->cached_attrs.va_mtime = ts;
512 if (flags & FN_CTIMECHANGE)
513 fvdat->cached_attrs.va_ctime = ts;
514
515 fvdat->flag |= flags;
516 }
517
518 void
fuse_node_init(void)519 fuse_node_init(void)
520 {
521 fuse_node_count = counter_u64_alloc(M_WAITOK);
522 }
523
524 void
fuse_node_destroy(void)525 fuse_node_destroy(void)
526 {
527 counter_u64_free(fuse_node_count);
528 }
529