xref: /freebsd/sys/fs/fuse/fuse_node.c (revision 3cbb4cc200f8a0ad7ed08233425ea54524a21f1c)
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
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
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 
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 			/*
217 			 * STALE vnode!  This probably indicates a buggy
218 			 * server, but it could also be the result of a race
219 			 * between FUSE_LOOKUP and another client's
220 			 * FUSE_UNLINK/FUSE_CREATE
221 			 */
222 			SDT_PROBE3(fusefs, , node, stale_vnode, *vpp, vtyp,
223 				nodeid);
224 			fuse_internal_vnode_disappear(*vpp);
225 			lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
226 			*vpp = NULL;
227 			return (EAGAIN);
228 		}
229 		MPASS((*vpp)->v_data != NULL);
230 		MPASS(VTOFUD(*vpp)->nid == nodeid);
231 		SDT_PROBE2(fusefs, , node, trace, 1, "vnode taken from hash");
232 		return (0);
233 	}
234 	fvdat = malloc(sizeof(*fvdat), M_FUSEVN, M_WAITOK | M_ZERO);
235 	switch (vtyp) {
236 	case VFIFO:
237 		err = getnewvnode("fuse", mp, &fuse_fifoops, vpp);
238 		break;
239 	default:
240 		err = getnewvnode("fuse", mp, &fuse_vnops, vpp);
241 		break;
242 	}
243 	if (err) {
244 		free(fvdat, M_FUSEVN);
245 		return (err);
246 	}
247 	lockmgr((*vpp)->v_vnlock, LK_EXCLUSIVE, NULL);
248 	fuse_vnode_init(*vpp, fvdat, nodeid, vtyp);
249 	err = insmntque(*vpp, mp);
250 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
251 	if (err) {
252 		lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
253 		free(fvdat, M_FUSEVN);
254 		*vpp = NULL;
255 		return (err);
256 	}
257 	/* Disallow async reads for fifos because UFS does.  I don't know why */
258 	if (data->dataflags & FSESS_ASYNC_READ && vtyp != VFIFO)
259 		VN_LOCK_ASHARE(*vpp);
260 
261 	err = vfs_hash_insert(*vpp, fuse_vnode_hash(nodeid), LK_EXCLUSIVE,
262 	    td, &vp2, fuse_vnode_cmp, &nodeid);
263 	if (err) {
264 		lockmgr((*vpp)->v_vnlock, LK_RELEASE, NULL);
265 		free(fvdat, M_FUSEVN);
266 		*vpp = NULL;
267 		return (err);
268 	}
269 	if (vp2 != NULL) {
270 		*vpp = vp2;
271 		return (0);
272 	}
273 
274 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnode_alloc");
275 
276 	return (0);
277 }
278 
279 int
280 fuse_vnode_get(struct mount *mp,
281     struct fuse_entry_out *feo,
282     uint64_t nodeid,
283     struct vnode *dvp,
284     struct vnode **vpp,
285     struct componentname *cnp,
286     enum vtype vtyp)
287 {
288 	struct thread *td = (cnp != NULL ? cnp->cn_thread : curthread);
289 	/*
290 	 * feo should only be NULL for the root directory, which (when libfuse
291 	 * is used) always has generation 0
292 	 */
293 	uint64_t generation = feo ? feo->generation : 0;
294 	int err = 0;
295 
296 	err = fuse_vnode_alloc(mp, td, nodeid, vtyp, vpp);
297 	if (err) {
298 		return err;
299 	}
300 	if (dvp != NULL) {
301 		MPASS(cnp && (cnp->cn_flags & ISDOTDOT) == 0);
302 		MPASS(cnp &&
303 			!(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.'));
304 		fuse_vnode_setparent(*vpp, dvp);
305 	}
306 	if (dvp != NULL && cnp != NULL && (cnp->cn_flags & MAKEENTRY) != 0 &&
307 	    feo != NULL &&
308 	    (feo->entry_valid != 0 || feo->entry_valid_nsec != 0)) {
309 		struct timespec timeout;
310 
311 		ASSERT_VOP_LOCKED(*vpp, "fuse_vnode_get");
312 		ASSERT_VOP_LOCKED(dvp, "fuse_vnode_get");
313 
314 		fuse_validity_2_timespec(feo, &timeout);
315 		cache_enter_time(dvp, *vpp, cnp, &timeout, NULL);
316 	}
317 
318 	VTOFUD(*vpp)->generation = generation;
319 	/*
320 	 * In userland, libfuse uses cached lookups for dot and dotdot entries,
321 	 * thus it does not really bump the nlookup counter for forget.
322 	 * Follow the same semantic and avoid the bump in order to keep
323 	 * nlookup counters consistent.
324 	 */
325 	if (cnp == NULL || ((cnp->cn_flags & ISDOTDOT) == 0 &&
326 	    (cnp->cn_namelen != 1 || cnp->cn_nameptr[0] != '.')))
327 		VTOFUD(*vpp)->nlookup++;
328 
329 	return 0;
330 }
331 
332 /*
333  * Called for every fusefs vnode open to initialize the vnode (not
334  * fuse_filehandle) for use
335  */
336 void
337 fuse_vnode_open(struct vnode *vp, int32_t fuse_open_flags, struct thread *td)
338 {
339 	if (vnode_vtype(vp) == VREG)
340 		vnode_create_vobject(vp, 0, td);
341 }
342 
343 int
344 fuse_vnode_savesize(struct vnode *vp, struct ucred *cred, pid_t pid)
345 {
346 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
347 	struct thread *td = curthread;
348 	struct fuse_filehandle *fufh = NULL;
349 	struct fuse_dispatcher fdi;
350 	struct fuse_setattr_in *fsai;
351 	int err = 0;
352 
353 	ASSERT_VOP_ELOCKED(vp, "fuse_io_extend");
354 
355 	if (fuse_isdeadfs(vp)) {
356 		return EBADF;
357 	}
358 	if (vnode_vtype(vp) == VDIR) {
359 		return EISDIR;
360 	}
361 	if (vfs_isrdonly(vnode_mount(vp))) {
362 		return EROFS;
363 	}
364 	if (cred == NULL) {
365 		cred = td->td_ucred;
366 	}
367 	fdisp_init(&fdi, sizeof(*fsai));
368 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
369 	fsai = fdi.indata;
370 	fsai->valid = 0;
371 
372 	/* Truncate to a new value. */
373 	MPASS((fvdat->flag & FN_SIZECHANGE) != 0);
374 	fsai->size = fvdat->cached_attrs.va_size;
375 	fsai->valid |= FATTR_SIZE;
376 
377 	fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
378 	if (fufh) {
379 		fsai->fh = fufh->fh_id;
380 		fsai->valid |= FATTR_FH;
381 	}
382 	err = fdisp_wait_answ(&fdi);
383 	fdisp_destroy(&fdi);
384 	if (err == 0)
385 		fvdat->flag &= ~FN_SIZECHANGE;
386 
387 	return err;
388 }
389 
390 /*
391  * Adjust the vnode's size to a new value, such as that provided by
392  * FUSE_GETATTR.
393  */
394 int
395 fuse_vnode_setsize(struct vnode *vp, off_t newsize)
396 {
397 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
398 	struct vattr *attrs;
399 	off_t oldsize;
400 	size_t iosize;
401 	struct buf *bp = NULL;
402 	int err = 0;
403 
404 	ASSERT_VOP_ELOCKED(vp, "fuse_vnode_setsize");
405 
406 	iosize = fuse_iosize(vp);
407 	oldsize = fvdat->cached_attrs.va_size;
408 	fvdat->cached_attrs.va_size = newsize;
409 	if ((attrs = VTOVA(vp)) != NULL)
410 		attrs->va_size = newsize;
411 
412 	if (newsize < oldsize) {
413 		daddr_t lbn;
414 
415 		err = vtruncbuf(vp, newsize, fuse_iosize(vp));
416 		if (err)
417 			goto out;
418 		if (newsize % iosize == 0)
419 			goto out;
420 		/*
421 		 * Zero the contents of the last partial block.
422 		 * Sure seems like vtruncbuf should do this for us.
423 		 */
424 
425 		lbn = newsize / iosize;
426 		bp = getblk(vp, lbn, iosize, PCATCH, 0, 0);
427 		if (!bp) {
428 			err = EINTR;
429 			goto out;
430 		}
431 		if (!(bp->b_flags & B_CACHE))
432 			goto out;	/* Nothing to do */
433 		MPASS(bp->b_flags & B_VMIO);
434 		vfs_bio_clrbuf(bp);
435 		bp->b_dirtyend = MIN(bp->b_dirtyend, newsize - lbn * iosize);
436 	}
437 out:
438 	if (bp)
439 		brelse(bp);
440 	vnode_pager_setsize(vp, newsize);
441 	return err;
442 }
443 
444 /* Get the current, possibly dirty, size of the file */
445 int
446 fuse_vnode_size(struct vnode *vp, off_t *filesize, struct ucred *cred,
447 	struct thread *td)
448 {
449 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
450 	int error = 0;
451 
452 	if (!(fvdat->flag & FN_SIZECHANGE) &&
453 		(!fuse_vnode_attr_cache_valid(vp) ||
454 		  fvdat->cached_attrs.va_size == VNOVAL))
455 		error = fuse_internal_do_getattr(vp, NULL, cred, td);
456 
457 	if (!error)
458 		*filesize = fvdat->cached_attrs.va_size;
459 
460 	return error;
461 }
462 
463 void
464 fuse_vnode_undirty_cached_timestamps(struct vnode *vp)
465 {
466 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
467 
468 	fvdat->flag &= ~(FN_MTIMECHANGE | FN_CTIMECHANGE);
469 }
470 
471 /* Update a fuse file's cached timestamps */
472 void
473 fuse_vnode_update(struct vnode *vp, int flags)
474 {
475 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
476 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
477 	struct timespec ts;
478 
479 	vfs_timestamp(&ts);
480 
481 	if (data->time_gran > 1)
482 		ts.tv_nsec = rounddown(ts.tv_nsec, data->time_gran);
483 
484 	if (flags & FN_MTIMECHANGE)
485 		fvdat->cached_attrs.va_mtime = ts;
486 	if (flags & FN_CTIMECHANGE)
487 		fvdat->cached_attrs.va_ctime = ts;
488 
489 	fvdat->flag |= flags;
490 }
491 
492 void
493 fuse_node_init(void)
494 {
495 	fuse_node_count = counter_u64_alloc(M_WAITOK);
496 }
497 
498 void
499 fuse_node_destroy(void)
500 {
501 	counter_u64_free(fuse_node_count);
502 }
503