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, CTLTYPE_INT|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
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
155 fuse_vnode_init(struct vnode *vp, struct fuse_vnode_data *fvdat,
156 uint64_t nodeid, enum vtype vtyp)
157 {
158 fvdat->nid = nodeid;
159 LIST_INIT(&fvdat->handles);
160 vattr_null(&fvdat->cached_attrs);
161 if (nodeid == FUSE_ROOT_ID) {
162 vp->v_vflag |= VV_ROOT;
163 }
164 vp->v_type = vtyp;
165 vp->v_data = fvdat;
166 timespecclear(&fvdat->last_local_modify);
167
168 counter_u64_add(fuse_node_count, 1);
169 }
170
171 void
172 fuse_vnode_destroy(struct vnode *vp)
173 {
174 struct fuse_vnode_data *fvdat = vp->v_data;
175
176 vp->v_data = NULL;
177 KASSERT(LIST_EMPTY(&fvdat->handles),
178 ("Destroying fuse vnode with open files!"));
179 free(fvdat, M_FUSEVN);
180
181 counter_u64_add(fuse_node_count, -1);
182 }
183
184 int
185 fuse_vnode_cmp(struct vnode *vp, void *nidp)
186 {
187 return (VTOI(vp) != *((uint64_t *)nidp));
188 }
189
190 SDT_PROBE_DEFINE3(fusefs, , node, stale_vnode, "struct vnode*", "enum vtype",
191 "uint64_t");
192 static int
193 fuse_vnode_alloc(struct mount *mp,
194 struct thread *td,
195 uint64_t nodeid,
196 enum vtype vtyp,
197 struct vnode **vpp)
198 {
199 struct fuse_data *data;
200 struct fuse_vnode_data *fvdat;
201 struct vnode *vp2;
202 int err = 0;
203
204 data = fuse_get_mpdata(mp);
205 if (vtyp == VNON) {
206 return EINVAL;
207 }
208 *vpp = NULL;
209 err = vfs_hash_get(mp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE, td, vpp,
210 fuse_vnode_cmp, &nodeid);
211 if (err)
212 return (err);
213
214 if (*vpp) {
215 if ((*vpp)->v_type == vtyp) {
216 /* Reuse a vnode that hasn't yet been reclaimed */
217 MPASS((*vpp)->v_data != NULL);
218 MPASS(VTOFUD(*vpp)->nid == nodeid);
219 SDT_PROBE2(fusefs, , node, trace, 1,
220 "vnode taken from hash");
221 return (0);
222 } else {
223 /*
224 * The inode changed types! If we get here, we can't
225 * tell whether the inode's entry cache had expired
226 * yet. So this could be the result of a buggy server,
227 * but more likely the server just reused an inode
228 * number following an entry cache expiration.
229 */
230 SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp,
231 nodeid);
232 fuse_internal_vnode_disappear(*vpp);
233 vgone(*vpp);
234 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
235 }
236 }
237 fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO);
238 switch (vtyp) {
239 case VFIFO:
240 err = getnewvnode("fuse", mp, &fuse_fifoops, vpp);
241 break;
242 default:
243 err = getnewvnode("fuse", mp, &fuse_vnops, vpp);
244 break;
245 }
246 if (err) {
247 free(fvdat, M_FUSEVN);
248 return (err);
249 }
250 lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL);
251 fuse_vnode_init(*vpp, fvdat, nodeid, vtyp);
252 err = insmntque(*vpp, mp);
253 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
254 if (err) {
255 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
256 free(fvdat, M_FUSEVN);
257 *vpp = NULL;
258 return (err);
259 }
260 /* Disallow async reads for fifos because UFS does. I don't know why */
261 if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO)
262 VN_LOCK_ASHARE(*vpp);
263
264 err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE,
265 td, &vp2, fuse_vnode_cmp, &nodeid);
266 if (err) {
267 lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
268 free(fvdat, M_FUSEVN);
269 *vpp = NULL;
270 return (err);
271 }
272 if (vp2 != NULL) {
273 *vpp = vp2;
274 return (0);
275 }
276
277 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
278
279 return (0);
280 }
281
282 int
283 fuse_vnode_get(struct mount *mp,
284 struct fuse_entry_out *feo,
285 uint64_t nodeid,
286 struct vnode *dvp,
287 struct vnode **vpp,
288 struct componentname *cnp,
289 enum vtype vtyp)
290 {
291 struct thread *td = (cnp != NULL ? cnp->cn_thread : curthread);
292 /*
293 * feo should only be NULL for the root directory, which (when libfuse
294 * is used) always has generation 0
295 */
296 uint64_t generation = feo ? feo->generation : 0;
297 int err = 0;
298
299 if (dvp != NULL && VTOFUD(dvp)->nid == nodeid) {
300 fuse_warn(fuse_get_mpdata(mp), FSESS_WARN_ILLEGAL_INODE,
301 "Assigned same inode to both parent and child.");
302 return EIO;
303 }
304
305 err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp);
306 if (err) {
307 return err;
308 }
309 if (dvp != NULL) {
310 MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0);
311 MPASS(cnp &&
312 !(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.'));
313 fuse_vnode_setparent(*vpp, dvp);
314 }
315 if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 &&
316 feo != NULL &&
317 (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) {
318 struct timespec timeout;
319
320 ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get");
321 ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get");
322
323 fuse_validity_2_timespec(feo, &timeout);
324 cache_enter_time(dvp, *vpp, cnp, &timeout, NULL);
325 }
326
327 VTOFUD(*vpp)->generation = generation;
328 /*
329 * In userland, libfuse uses cached lookups for dot and dotdot entries,
330 * thus it does not really bump the nlookup counter for forget.
331 * Follow the same semantic and avoid the bump in order to keep
332 * nlookup counters consistent.
333 */
334 if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 &&
335 (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.')))
336 VTOFUD(*vpp)->nlookup++;
337
338 return 0;
339 }
340
341 /*
342 * Called for every fusefs vnode open to initialize the vnode (not
343 * fuse_filehandle) for use
344 */
345 void
346 fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td)
347 {
348 if (vnode_vtype(vp) == VREG)
349 vnode_create_vobject(vp, 0, td);
350 }
351
352 int
353 fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid)
354 {
355 struct fuse_vnode_data *fvdat = VTOFUD(vp);
356 struct thread *td = curthread;
357 struct fuse_filehandle *fufh = NULL;
358 struct fuse_dispatcher fdi;
359 struct fuse_setattr_in *fsai;
360 int err = 0;
361
362 ASSERT_VOP_ELOCKED(vp, "fuse_io_extend");
363
364 if (fuse_isdeadfs(vp)) {
365 return EBADF;
366 }
367 if (vnode_vtype(vp) == VDIR) {
368 return EISDIR;
369 }
370 if (vfs_isrdonly(vnode_mount(vp))) {
371 return EROFS;
372 }
373 if (cred == NULL) {
374 cred = td->td_ucred;
375 }
376 fdisp_init(&fdi, sizeof(*fsai));
377 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
378 fsai = fdi.indata;
379 fsai->valid = 0;
380
381 /* Truncate to a new value. */
382 MPASS((fvdat->flag & FN_SIZECHANGE) != 0);
383 fsai->size = fvdat->cached_attrs.va_size;
384 fsai->valid |= FATTR_SIZE;
385
386 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
387 if (fufh) {
388 fsai->fh = fufh->fh_id;
389 fsai->valid |= FATTR_FH;
390 }
391 err = fdisp_wait_answ(&fdi);
392 fdisp_destroy(&fdi);
393 if (err == 0) {
394 getnanouptime(&fvdat->last_local_modify);
395 fvdat->flag &= ~FN_SIZECHANGE;
396 }
397
398 return err;
399 }
400
401 /*
402 * Adjust the vnode's size to a new value.
403 *
404 * If the new value came from the server, such as from a FUSE_GETATTR
405 * operation, set `from_server` true. But if it came from a local operation,
406 * such as write(2) or truncate(2), set `from_server` false.
407 */
408 int
409 fuse_vnode_setsize(struct vnode *vp, off_t newsize, bool from_server)
410 {
411 struct fuse_vnode_data *fvdat = VTOFUD(vp);
412 struct vattr *attrs;
413 off_t oldsize;
414 size_t iosize;
415 struct buf *bp = NULL;
416 int err = 0;
417
418 ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize");
419
420 iosize = fuse_iosize(vp);
421 oldsize = fvdat->cached_attrs.va_size;
422 fvdat->cached_attrs.va_size = newsize;
423 if ((attrs = VTOVA(vp)) != NULL)
424 attrs->va_size = newsize;
425
426 if (newsize < oldsize) {
427 daddr_t lbn;
428
429 err = vtruncbuf(vp, newsize, fuse_iosize(vp));
430 if (err)
431 goto out;
432 if (newsize % iosize == 0)
433 goto out;
434 /*
435 * Zero the contents of the last partial block.
436 * Sure seems like vtruncbuf should do this for us.
437 */
438
439 lbn = newsize / iosize;
440 bp = getblk(vp, lbn, iosize, PCATCH, 0, 0);
441 if (!bp) {
442 err = EINTR;
443 goto out;
444 }
445 if (!(bp->b_flags & B_CACHE))
446 goto out; /* Nothing to do */
447 MPASS(bp->b_flags & B_VMIO);
448 vfs_bio_clrbuf(bp);
449 bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize);
450 } else if (from_server && newsize > oldsize && oldsize != VNOVAL) {
451 /*
452 * The FUSE server changed the file size behind our back. We
453 * should invalidate the entire cache.
454 */
455 daddr_t left_lbn, end_lbn;
456
457 left_lbn = oldsize / iosize;
458 end_lbn = howmany(newsize, iosize);
459 v_inval_buf_range(vp, 0, end_lbn, iosize);
460 }
461 out:
462 if (bp)
463 brelse(bp);
464 vnode_pager_setsize(vp, newsize);
465 return err;
466 }
467
468 /* Get the current, possibly dirty, size of the file */
469 int
470 fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred,
471 struct thread *td)
472 {
473 struct fuse_vnode_data *fvdat = VTOFUD(vp);
474 int error = 0;
475
476 if (!(fvdat->flag & FN_SIZECHANGE) &&
477 (!fuse_vnode_attr_cache_valid(vp) ||
478 fvdat->cached_attrs.va_size == VNOVAL))
479 error = fuse_internal_do_getattr(vp, NULL, cred, td);
480
481 if (!error)
482 *filesize = fvdat->cached_attrs.va_size;
483
484 return error;
485 }
486
487 void
488 fuse_vnode_undirty_cached_timestamps(struct vnode *vp, bool atime)
489 {
490 struct fuse_vnode_data *fvdat = VTOFUD(vp);
491
492 fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE);
493 if (atime)
494 fvdat->flag &= ~FN_ATIMECHANGE;
495 }
496
497 /* Update a fuse file's cached timestamps */
498 void
499 fuse_vnode_update(struct vnode *vp, int flags)
500 {
501 struct fuse_vnode_data *fvdat = VTOFUD(vp);
502 struct mount *mp = vnode_mount(vp);
503 struct fuse_data *data = fuse_get_mpdata(mp);
504 struct timespec ts;
505
506 vfs_timestamp(&ts);
507
508 if (data->time_gran > 1)
509 ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran);
510
511 if (mp->mnt_flag & MNT_NOATIME)
512 flags &= ~FN_ATIMECHANGE;
513
514 if (flags & FN_ATIMECHANGE)
515 fvdat->cached_attrs.va_atime = ts;
516 if (flags & FN_MTIMECHANGE)
517 fvdat->cached_attrs.va_mtime = ts;
518 if (flags & FN_CTIMECHANGE)
519 fvdat->cached_attrs.va_ctime = ts;
520
521 fvdat->flag |= flags;
522 }
523
524 void
525 fuse_node_init(void)
526 {
527 fuse_node_count = counter_u64_alloc(M_WAITOK);
528 }
529
530 void
531 fuse_node_destroy(void)
532 {
533 counter_u64_free(fuse_node_count);
534 }
Cache object: 4f24ffe9cf65c4e502b792b702cbb237
|