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: stable/12/sys/fs/fuse/fuse_internal.c 372632 2022-10-19 03:21:04Z asomers $");
65 
66 #include <sys/param.h>
67 #include <sys/systm.h>
68 #include <sys/counter.h>
69 #include <sys/module.h>
70 #include <sys/errno.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/mutex.h>
78 #include <sys/sdt.h>
79 #include <sys/sx.h>
80 #include <sys/proc.h>
81 #include <sys/mount.h>
82 #include <sys/vnode.h>
83 #include <sys/namei.h>
84 #include <sys/stat.h>
85 #include <sys/unistd.h>
86 #include <sys/filedesc.h>
87 #include <sys/file.h>
88 #include <sys/fcntl.h>
89 #include <sys/dirent.h>
90 #include <sys/bio.h>
91 #include <sys/buf.h>
92 #include <sys/sysctl.h>
93 #include <sys/priv.h>
94 
95 #include "fuse.h"
96 #include "fuse_file.h"
97 #include "fuse_internal.h"
98 #include "fuse_io.h"
99 #include "fuse_ipc.h"
100 #include "fuse_node.h"
101 #include "fuse_file.h"
102 
103 SDT_PROVIDER_DECLARE(fusefs);
104 /*
105  * Fuse trace probe:
106  * arg0: verbosity.  Higher numbers give more verbose messages
107  * arg1: Textual message
108  */
109 SDT_PROBE_DEFINE2(fusefs, , internal, trace, "int", "char*");
110 
111 #ifdef ZERO_PAD_INCOMPLETE_BUFS
112 static int isbzero(void *buf, size_t len);
113 
114 #endif
115 
116 counter_u64_t fuse_lookup_cache_hits;
117 counter_u64_t fuse_lookup_cache_misses;
118 
119 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
120     &fuse_lookup_cache_hits, "number of positive cache hits in lookup");
121 
122 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
123     &fuse_lookup_cache_misses, "number of cache misses in lookup");
124 
125 int
fuse_internal_get_cached_vnode(struct mount * mp,ino_t ino,int flags,struct vnode ** vpp)126 fuse_internal_get_cached_vnode(struct mount* mp, ino_t ino, int flags,
127 	struct vnode **vpp)
128 {
129 	struct bintime now;
130 	struct thread *td = curthread;
131 	uint64_t nodeid = ino;
132 	int error;
133 
134 	*vpp = NULL;
135 
136 	error = vfs_hash_get(mp, fuse_vnode_hash(nodeid), flags, td, vpp,
137 	    fuse_vnode_cmp, &nodeid);
138 	if (error)
139 		return error;
140 	/*
141 	 * Check the entry cache timeout.  We have to do this within fusefs
142 	 * instead of by using cache_enter_time/cache_lookup because those
143 	 * routines are only intended to work with pathnames, not inodes
144 	 */
145 	if (*vpp != NULL) {
146 		getbinuptime(&now);
147 		if (bintime_cmp(&(VTOFUD(*vpp)->entry_cache_timeout), &now, >)){
148 			counter_u64_add(fuse_lookup_cache_hits, 1);
149 			return 0;
150 		} else {
151 			/* Entry cache timeout */
152 			counter_u64_add(fuse_lookup_cache_misses, 1);
153 			cache_purge(*vpp);
154 			vput(*vpp);
155 			*vpp = NULL;
156 		}
157 	}
158 	return 0;
159 }
160 
161 SDT_PROBE_DEFINE0(fusefs, , internal, access_vadmin);
162 /* Synchronously send a FUSE_ACCESS operation */
163 int
fuse_internal_access(struct vnode * vp,accmode_t mode,struct thread * td,struct ucred * cred)164 fuse_internal_access(struct vnode *vp,
165     accmode_t mode,
166     struct thread *td,
167     struct ucred *cred)
168 {
169 	int err = 0;
170 	uint32_t mask = F_OK;
171 	int dataflags;
172 	int vtype;
173 	struct mount *mp;
174 	struct fuse_dispatcher fdi;
175 	struct fuse_access_in *fai;
176 	struct fuse_data *data;
177 
178 	mp = vnode_mount(vp);
179 	vtype = vnode_vtype(vp);
180 
181 	data = fuse_get_mpdata(mp);
182 	dataflags = data->dataflags;
183 
184 	if (mode == 0)
185 		return 0;
186 
187 	if (mode & VMODIFY_PERMS && vfs_isrdonly(mp)) {
188 		switch (vp->v_type) {
189 		case VDIR:
190 			/* FALLTHROUGH */
191 		case VLNK:
192 			/* FALLTHROUGH */
193 		case VREG:
194 			return EROFS;
195 		default:
196 			break;
197 		}
198 	}
199 
200 	/* Unless explicitly permitted, deny everyone except the fs owner. */
201 	if (!(dataflags & FSESS_DAEMON_CAN_SPY)) {
202 		if (fuse_match_cred(data->daemoncred, cred))
203 			return EPERM;
204 	}
205 
206 	if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
207 		struct vattr va;
208 
209 		fuse_internal_getattr(vp, &va, cred, td);
210 		return vaccess(vp->v_type, va.va_mode, va.va_uid,
211 		    va.va_gid, mode, cred, NULL);
212 	}
213 
214 	if (mode & VADMIN) {
215 		/*
216 		 * The FUSE protocol doesn't have an equivalent of VADMIN, so
217 		 * it's a bug if we ever reach this point with that bit set.
218 		 */
219 		SDT_PROBE0(fusefs, , internal, access_vadmin);
220 	}
221 
222 	if (!fsess_isimpl(mp, FUSE_ACCESS))
223 		return 0;
224 
225 	if ((mode & (VWRITE | VAPPEND)) != 0)
226 		mask |= W_OK;
227 	if ((mode & VREAD) != 0)
228 		mask |= R_OK;
229 	if ((mode & VEXEC) != 0)
230 		mask |= X_OK;
231 
232 	fdisp_init(&fdi, sizeof(*fai));
233 	fdisp_make_vp(&fdi, FUSE_ACCESS, vp, td, cred);
234 
235 	fai = fdi.indata;
236 	fai->mask = mask;
237 
238 	err = fdisp_wait_answ(&fdi);
239 	fdisp_destroy(&fdi);
240 
241 	if (err == ENOSYS) {
242 		fsess_set_notimpl(mp, FUSE_ACCESS);
243 		err = 0;
244 	}
245 	return err;
246 }
247 
248 /*
249  * Cache FUSE attributes from attr, in attribute cache associated with vnode
250  * 'vp'.  Optionally, if argument 'vap' is not NULL, store a copy of the
251  * converted attributes there as well.
252  *
253  * If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do
254  * return the result to the caller).
255  */
256 void
fuse_internal_cache_attrs(struct vnode * vp,struct fuse_attr * attr,uint64_t attr_valid,uint32_t attr_valid_nsec,struct vattr * vap,bool from_server)257 fuse_internal_cache_attrs(struct vnode *vp, struct fuse_attr *attr,
258 	uint64_t attr_valid, uint32_t attr_valid_nsec, struct vattr *vap,
259 	bool from_server)
260 {
261 	struct mount *mp;
262 	struct fuse_vnode_data *fvdat;
263 	struct fuse_data *data;
264 	struct vattr *vp_cache_at;
265 
266 	mp = vnode_mount(vp);
267 	fvdat = VTOFUD(vp);
268 	data = fuse_get_mpdata(mp);
269 
270 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs");
271 
272 	fuse_validity_2_bintime(attr_valid, attr_valid_nsec,
273 		&fvdat->attr_cache_timeout);
274 
275 	if (vnode_isreg(vp) &&
276 	    fvdat->cached_attrs.va_size != VNOVAL &&
277 	    attr->size != fvdat->cached_attrs.va_size)
278 	{
279 		if ( data->cache_mode == FUSE_CACHE_WB &&
280 		    fvdat->flag & FN_SIZECHANGE)
281 		{
282 			const char *msg;
283 
284 			/*
285 			 * The server changed the file's size even though we're
286 			 * using writeback cacheing and and we have outstanding
287 			 * dirty writes!  That's a server bug.
288 			 */
289 			if (fuse_libabi_geq(data, 7, 23)) {
290 				msg = "writeback cache incoherent!."
291 				    "To prevent data corruption, disable "
292 				    "the writeback cache according to your "
293 				    "FUSE server's documentation.";
294 			} else {
295 				msg = "writeback cache incoherent!."
296 				    "To prevent data corruption, disable "
297 				    "the writeback cache by setting "
298 				    "vfs.fusefs.data_cache_mode to 0 or 1.";
299 			}
300 			fuse_warn(data, FSESS_WARN_WB_CACHE_INCOHERENT, msg);
301 		}
302 		if (fuse_vnode_attr_cache_valid(vp) &&
303 		    data->cache_mode != FUSE_CACHE_UC)
304 		{
305 			/*
306 			 * The server changed the file's size even though we
307 			 * have it cached and our cache has not yet expired.
308 			 * That's a bug.
309 			 */
310 			fuse_warn(data, FSESS_WARN_CACHE_INCOHERENT,
311 			    "cache incoherent!  "
312 			    "To prevent "
313 			    "data corruption, disable the data cache "
314 			    "by mounting with -o direct_io, or as "
315 			    "directed otherwise by your FUSE server's "
316 			    "documentation.");
317 		}
318 	}
319 
320 	/* Fix our buffers if the filesize changed without us knowing */
321 	if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) {
322 		(void)fuse_vnode_setsize(vp, attr->size, from_server);
323 		fvdat->cached_attrs.va_size = attr->size;
324 	}
325 
326 	if (attr_valid > 0 || attr_valid_nsec > 0)
327 		vp_cache_at = &(fvdat->cached_attrs);
328 	else if (vap != NULL)
329 		vp_cache_at = vap;
330 	else
331 		return;
332 
333 	vattr_null(vp_cache_at);
334 	vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0];
335 	vp_cache_at->va_fileid = attr->ino;
336 	vp_cache_at->va_mode = attr->mode & ~S_IFMT;
337 	vp_cache_at->va_nlink     = attr->nlink;
338 	vp_cache_at->va_uid       = attr->uid;
339 	vp_cache_at->va_gid       = attr->gid;
340 	vp_cache_at->va_rdev      = attr->rdev;
341 	vp_cache_at->va_size      = attr->size;
342 	/* XXX on i386, seconds are truncated to 32 bits */
343 	vp_cache_at->va_atime.tv_sec  = attr->atime;
344 	vp_cache_at->va_atime.tv_nsec = attr->atimensec;
345 	vp_cache_at->va_mtime.tv_sec  = attr->mtime;
346 	vp_cache_at->va_mtime.tv_nsec = attr->mtimensec;
347 	vp_cache_at->va_ctime.tv_sec  = attr->ctime;
348 	vp_cache_at->va_ctime.tv_nsec = attr->ctimensec;
349 	if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0)
350 		vp_cache_at->va_blocksize = attr->blksize;
351 	else
352 		vp_cache_at->va_blocksize = PAGE_SIZE;
353 	vp_cache_at->va_type = IFTOVT(attr->mode);
354 	vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE;
355 	vp_cache_at->va_flags = 0;
356 
357 	if (vap != vp_cache_at && vap != NULL)
358 		memcpy(vap, vp_cache_at, sizeof(*vap));
359 }
360 
361 
362 /* fsync */
363 
364 int
fuse_internal_fsync_callback(struct fuse_ticket * tick,struct uio * uio)365 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio)
366 {
367 	if (tick->tk_aw_ohead.error == ENOSYS) {
368 		fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick));
369 	}
370 	return 0;
371 }
372 
373 int
fuse_internal_fsync(struct vnode * vp,struct thread * td,int waitfor,bool datasync)374 fuse_internal_fsync(struct vnode *vp,
375     struct thread *td,
376     int waitfor,
377     bool datasync)
378 {
379 	struct fuse_fsync_in *ffsi = NULL;
380 	struct fuse_dispatcher fdi;
381 	struct fuse_filehandle *fufh;
382 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
383 	struct mount *mp = vnode_mount(vp);
384 	int op = FUSE_FSYNC;
385 	int err = 0;
386 
387 	if (!fsess_isimpl(vnode_mount(vp),
388 	    (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
389 		return 0;
390 	}
391 	if (vnode_isdir(vp))
392 		op = FUSE_FSYNCDIR;
393 
394 	if (!fsess_isimpl(mp, op))
395 		return 0;
396 
397 	fdisp_init(&fdi, sizeof(*ffsi));
398 	/*
399 	 * fsync every open file handle for this file, because we can't be sure
400 	 * which file handle the caller is really referring to.
401 	 */
402 	LIST_FOREACH(fufh, &fvdat->handles, next) {
403 		fdi.iosize = sizeof(*ffsi);
404 		if (ffsi == NULL)
405 			fdisp_make_vp(&fdi, op, vp, td, NULL);
406 		else
407 			fdisp_refresh_vp(&fdi, op, vp, td, NULL);
408 		ffsi = fdi.indata;
409 		ffsi->fh = fufh->fh_id;
410 		ffsi->fsync_flags = 0;
411 
412 		if (datasync)
413 			ffsi->fsync_flags = 1;
414 
415 		if (waitfor == MNT_WAIT) {
416 			err = fdisp_wait_answ(&fdi);
417 		} else {
418 			fuse_insert_callback(fdi.tick,
419 				fuse_internal_fsync_callback);
420 			fuse_insert_message(fdi.tick, false);
421 		}
422 		if (err == ENOSYS) {
423 			/* ENOSYS means "success, and don't call again" */
424 			fsess_set_notimpl(mp, op);
425 			err = 0;
426 			break;
427 		}
428 	}
429 	fdisp_destroy(&fdi);
430 
431 	return err;
432 }
433 
434 /* Asynchronous invalidation */
435 SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry,
436 	"struct vnode*", "struct fuse_notify_inval_entry_out*", "char*");
437 int
fuse_internal_invalidate_entry(struct mount * mp,struct uio * uio)438 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio)
439 {
440 	struct fuse_notify_inval_entry_out fnieo;
441 	struct componentname cn;
442 	struct vnode *dvp, *vp;
443 	char name[PATH_MAX];
444 	int err;
445 
446 	if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0)
447 		return (err);
448 
449 	if (fnieo.namelen >= sizeof(name))
450 		return (EINVAL);
451 
452 	if ((err = uiomove(name, fnieo.namelen, uio)) != 0)
453 		return (err);
454 	name[fnieo.namelen] = '\0';
455 	/* fusefs does not cache "." or ".." entries */
456 	if (strncmp(name, ".", sizeof(".")) == 0 ||
457 	    strncmp(name, "..", sizeof("..")) == 0)
458 		return (0);
459 
460 	if (fnieo.parent == FUSE_ROOT_ID)
461 		err = VFS_ROOT(mp, LK_SHARED, &dvp);
462 	else
463 		err = fuse_internal_get_cached_vnode( mp, fnieo.parent,
464 			LK_SHARED, &dvp);
465 	SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name);
466 	/*
467 	 * If dvp is not in the cache, then it must've been reclaimed.  And
468 	 * since fuse_vnop_reclaim does a cache_purge, name's entry must've
469 	 * been invalidated already.  So we can safely return if dvp == NULL
470 	 */
471 	if (err != 0 || dvp == NULL)
472 		return (err);
473 	/*
474 	 * XXX we can't check dvp's generation because the FUSE invalidate
475 	 * entry message doesn't include it.  Worse case is that we invalidate
476 	 * an entry that didn't need to be invalidated.
477 	 */
478 
479 	cn.cn_nameiop = LOOKUP;
480 	cn.cn_flags = 0;	/* !MAKEENTRY means free cached entry */
481 	cn.cn_thread = curthread;
482 	cn.cn_cred = curthread->td_ucred;
483 	cn.cn_lkflags = LK_SHARED;
484 	cn.cn_pnbuf = NULL;
485 	cn.cn_nameptr = name;
486 	cn.cn_namelen = fnieo.namelen;
487 	err = cache_lookup(dvp, &vp, &cn, NULL, NULL);
488 	MPASS(err == 0);
489 	fuse_vnode_clear_attr_cache(dvp);
490 	vput(dvp);
491 	return (0);
492 }
493 
494 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode,
495 	"struct vnode*", "struct fuse_notify_inval_inode_out *");
496 int
fuse_internal_invalidate_inode(struct mount * mp,struct uio * uio)497 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio)
498 {
499 	struct fuse_notify_inval_inode_out fniio;
500 	struct vnode *vp;
501 	int err;
502 
503 	if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0)
504 		return (err);
505 
506 	if (fniio.ino == FUSE_ROOT_ID)
507 		err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp);
508 	else
509 		err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED,
510 			&vp);
511 	SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio);
512 	if (err != 0 || vp == NULL)
513 		return (err);
514 	/*
515 	 * XXX we can't check vp's generation because the FUSE invalidate
516 	 * entry message doesn't include it.  Worse case is that we invalidate
517 	 * an inode that didn't need to be invalidated.
518 	 */
519 
520 	/*
521 	 * Flush and invalidate buffers if off >= 0.  Technically we only need
522 	 * to flush and invalidate the range of offsets [off, off + len), but
523 	 * for simplicity's sake we do everything.
524 	 */
525 	if (fniio.off >= 0)
526 		fuse_io_invalbuf(vp, curthread);
527 	fuse_vnode_clear_attr_cache(vp);
528 	vput(vp);
529 	return (0);
530 }
531 
532 /* mknod */
533 int
fuse_internal_mknod(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct vattr * vap)534 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp,
535 	struct componentname *cnp, struct vattr *vap)
536 {
537 	struct fuse_data *data;
538 	struct fuse_mknod_in fmni;
539 	size_t insize;
540 
541 	data = fuse_get_mpdata(dvp->v_mount);
542 
543 	fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
544 	fmni.rdev = vap->va_rdev;
545 	if (fuse_libabi_geq(data, 7, 12)) {
546 		insize = sizeof(fmni);
547 		fmni.umask = curthread->td_proc->p_fd->fd_cmask;
548 	} else {
549 		insize = FUSE_COMPAT_MKNOD_IN_SIZE;
550 	}
551 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
552 	    insize, vap->va_type));
553 }
554 
555 /* readdir */
556 
557 int
fuse_internal_readdir(struct vnode * vp,struct uio * uio,off_t startoff,struct fuse_filehandle * fufh,struct fuse_iov * cookediov,int * ncookies,u_long * cookies)558 fuse_internal_readdir(struct vnode *vp,
559     struct uio *uio,
560     off_t startoff,
561     struct fuse_filehandle *fufh,
562     struct fuse_iov *cookediov,
563     int *ncookies,
564     u_long *cookies)
565 {
566 	int err = 0;
567 	struct fuse_dispatcher fdi;
568 	struct fuse_read_in *fri = NULL;
569 	int fnd_start;
570 
571 	if (uio_resid(uio) == 0)
572 		return 0;
573 	fdisp_init(&fdi, 0);
574 
575 	/*
576 	 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p
577 	 * I/O).
578 	 */
579 
580 	/*
581 	 * fnd_start is set non-zero once the offset in the directory gets
582 	 * to the startoff.  This is done because directories must be read
583 	 * from the beginning (offset == 0) when fuse_vnop_readdir() needs
584 	 * to do an open of the directory.
585 	 * If it is not set non-zero here, it will be set non-zero in
586 	 * fuse_internal_readdir_processdata() when uio_offset == startoff.
587 	 */
588 	fnd_start = 0;
589 	if (uio->uio_offset == startoff)
590 		fnd_start = 1;
591 	while (uio_resid(uio) > 0) {
592 		fdi.iosize = sizeof(*fri);
593 		if (fri == NULL)
594 			fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
595 		else
596 			fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
597 
598 		fri = fdi.indata;
599 		fri->fh = fufh->fh_id;
600 		fri->offset = uio_offset(uio);
601 		fri->size = MIN(uio->uio_resid,
602 		    fuse_get_mpdata(vp->v_mount)->max_read);
603 
604 		if ((err = fdisp_wait_answ(&fdi)))
605 			break;
606 		if ((err = fuse_internal_readdir_processdata(uio, startoff,
607 		    &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov,
608 		    ncookies, &cookies)))
609 			break;
610 	}
611 
612 	fdisp_destroy(&fdi);
613 	return ((err == -1) ? 0 : err);
614 }
615 
616 /*
617  * Return -1 to indicate that this readdir is finished, 0 if it copied
618  * all the directory data read in and it may be possible to read more
619  * and greater than 0 for a failure.
620  */
621 int
fuse_internal_readdir_processdata(struct uio * uio,off_t startoff,int * fnd_start,size_t reqsize,void * buf,size_t bufsize,struct fuse_iov * cookediov,int * ncookies,u_long ** cookiesp)622 fuse_internal_readdir_processdata(struct uio *uio,
623     off_t startoff,
624     int *fnd_start,
625     size_t reqsize,
626     void *buf,
627     size_t bufsize,
628     struct fuse_iov *cookediov,
629     int *ncookies,
630     u_long **cookiesp)
631 {
632 	int err = 0;
633 	int oreclen;
634 	size_t freclen;
635 
636 	struct dirent *de;
637 	struct fuse_dirent *fudge;
638 	u_long *cookies;
639 
640 	cookies = *cookiesp;
641 	if (bufsize < FUSE_NAME_OFFSET)
642 		return -1;
643 	for (;;) {
644 		if (bufsize < FUSE_NAME_OFFSET) {
645 			err = -1;
646 			break;
647 		}
648 		fudge = (struct fuse_dirent *)buf;
649 		freclen = FUSE_DIRENT_SIZE(fudge);
650 
651 		if (bufsize < freclen) {
652 			/*
653 			 * This indicates a partial directory entry at the
654 			 * end of the directory data.
655 			 */
656 			err = -1;
657 			break;
658 		}
659 #ifdef ZERO_PAD_INCOMPLETE_BUFS
660 		if (isbzero(buf, FUSE_NAME_OFFSET)) {
661 			err = -1;
662 			break;
663 		}
664 #endif
665 
666 		if (!fudge->namelen || fudge->namelen > MAXNAMLEN) {
667 			err = EINVAL;
668 			break;
669 		}
670 		oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *)
671 					    &fudge->namelen);
672 
673 		if (oreclen > uio_resid(uio)) {
674 			/* Out of space for the dir so we are done. */
675 			err = -1;
676 			break;
677 		}
678 		/*
679 		 * Don't start to copy the directory entries out until
680 		 * the requested offset in the directory is found.
681 		 */
682 		if (*fnd_start != 0) {
683 			fiov_adjust(cookediov, oreclen);
684 			bzero(cookediov->base, oreclen);
685 
686 			de = (struct dirent *)cookediov->base;
687 			de->d_fileno = fudge->ino;
688 			de->d_off = fudge->off;
689 			de->d_reclen = oreclen;
690 			de->d_type = fudge->type;
691 			de->d_namlen = fudge->namelen;
692 			memcpy((char *)cookediov->base + sizeof(struct dirent) -
693 			       MAXNAMLEN - 1,
694 			       (char *)buf + FUSE_NAME_OFFSET, fudge->namelen);
695 			dirent_terminate(de);
696 
697 			err = uiomove(cookediov->base, cookediov->len, uio);
698 			if (err)
699 				break;
700 			if (cookies != NULL) {
701 				if (*ncookies == 0) {
702 					err = -1;
703 					break;
704 				}
705 				*cookies = fudge->off;
706 				cookies++;
707 				(*ncookies)--;
708 			}
709 		} else if (startoff == fudge->off)
710 			*fnd_start = 1;
711 		buf = (char *)buf + freclen;
712 		bufsize -= freclen;
713 		uio_setoffset(uio, fudge->off);
714 	}
715 	*cookiesp = cookies;
716 
717 	return err;
718 }
719 
720 /* remove */
721 
722 int
fuse_internal_remove(struct vnode * dvp,struct vnode * vp,struct componentname * cnp,enum fuse_opcode op)723 fuse_internal_remove(struct vnode *dvp,
724     struct vnode *vp,
725     struct componentname *cnp,
726     enum fuse_opcode op)
727 {
728 	struct fuse_dispatcher fdi;
729 	nlink_t nlink;
730 	int err = 0;
731 
732 	fdisp_init(&fdi, cnp->cn_namelen + 1);
733 	fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred);
734 
735 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
736 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
737 
738 	err = fdisp_wait_answ(&fdi);
739 	fdisp_destroy(&fdi);
740 
741 	if (err)
742 		return (err);
743 
744 	/*
745 	 * Access the cached nlink even if the attr cached has expired.  If
746 	 * it's inaccurate, the worst that will happen is:
747 	 * 1) We'll recycle the vnode even though the file has another link we
748 	 *    don't know about, costing a bit of cpu time, or
749 	 * 2) We won't recycle the vnode even though all of its links are gone.
750 	 *    It will linger around until vnlru reclaims it, costing a bit of
751 	 *    temporary memory.
752 	 */
753 	nlink = VTOFUD(vp)->cached_attrs.va_nlink--;
754 
755 	/*
756 	 * Purge the parent's attribute cache because the daemon
757 	 * should've updated its mtime and ctime.
758 	 */
759 	fuse_vnode_clear_attr_cache(dvp);
760 
761 	/* NB: nlink could be zero if it was never cached */
762 	if (nlink <= 1 || vnode_vtype(vp) == VDIR) {
763 		fuse_internal_vnode_disappear(vp);
764 	} else {
765 		cache_purge(vp);
766 		fuse_vnode_update(vp, FN_CTIMECHANGE);
767 	}
768 
769 	return err;
770 }
771 
772 /* rename */
773 
774 int
fuse_internal_rename(struct vnode * fdvp,struct componentname * fcnp,struct vnode * tdvp,struct componentname * tcnp)775 fuse_internal_rename(struct vnode *fdvp,
776     struct componentname *fcnp,
777     struct vnode *tdvp,
778     struct componentname *tcnp)
779 {
780 	struct fuse_dispatcher fdi;
781 	struct fuse_rename_in *fri;
782 	int err = 0;
783 
784 	fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2);
785 	fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred);
786 
787 	fri = fdi.indata;
788 	fri->newdir = VTOI(tdvp);
789 	memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr,
790 	    fcnp->cn_namelen);
791 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0';
792 	memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1,
793 	    tcnp->cn_nameptr, tcnp->cn_namelen);
794 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen +
795 	    tcnp->cn_namelen + 1] = '\0';
796 
797 	err = fdisp_wait_answ(&fdi);
798 	fdisp_destroy(&fdi);
799 	return err;
800 }
801 
802 /* strategy */
803 
804 /* entity creation */
805 
806 void
fuse_internal_newentry_makerequest(struct mount * mp,uint64_t dnid,struct componentname * cnp,enum fuse_opcode op,void * buf,size_t bufsize,struct fuse_dispatcher * fdip)807 fuse_internal_newentry_makerequest(struct mount *mp,
808     uint64_t dnid,
809     struct componentname *cnp,
810     enum fuse_opcode op,
811     void *buf,
812     size_t bufsize,
813     struct fuse_dispatcher *fdip)
814 {
815 	fdip->iosize = bufsize + cnp->cn_namelen + 1;
816 
817 	fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred);
818 	memcpy(fdip->indata, buf, bufsize);
819 	memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen);
820 	((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0';
821 }
822 
823 int
fuse_internal_newentry_core(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,enum vtype vtyp,struct fuse_dispatcher * fdip)824 fuse_internal_newentry_core(struct vnode *dvp,
825     struct vnode **vpp,
826     struct componentname *cnp,
827     enum vtype vtyp,
828     struct fuse_dispatcher *fdip)
829 {
830 	int err = 0;
831 	struct fuse_entry_out *feo;
832 	struct mount *mp = vnode_mount(dvp);
833 
834 	if ((err = fdisp_wait_answ(fdip))) {
835 		return err;
836 	}
837 	feo = fdip->answ;
838 
839 	if ((err = fuse_internal_checkentry(feo, vtyp))) {
840 		return err;
841 	}
842 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp);
843 	if (err) {
844 		fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred,
845 		    feo->nodeid, 1);
846 		return err;
847 	}
848 
849 	/*
850 	 * Purge the parent's attribute cache because the daemon should've
851 	 * updated its mtime and ctime
852 	 */
853 	fuse_vnode_clear_attr_cache(dvp);
854 
855 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
856 		feo->attr_valid_nsec, NULL, true);
857 
858 	return err;
859 }
860 
861 int
fuse_internal_newentry(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,enum fuse_opcode op,void * buf,size_t bufsize,enum vtype vtype)862 fuse_internal_newentry(struct vnode *dvp,
863     struct vnode **vpp,
864     struct componentname *cnp,
865     enum fuse_opcode op,
866     void *buf,
867     size_t bufsize,
868     enum vtype vtype)
869 {
870 	int err;
871 	struct fuse_dispatcher fdi;
872 	struct mount *mp = vnode_mount(dvp);
873 
874 	fdisp_init(&fdi, 0);
875 	fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf,
876 	    bufsize, &fdi);
877 	err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi);
878 	fdisp_destroy(&fdi);
879 
880 	return err;
881 }
882 
883 /* entity destruction */
884 
885 int
fuse_internal_forget_callback(struct fuse_ticket * ftick,struct uio * uio)886 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio)
887 {
888 	fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL,
889 	    ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1);
890 
891 	return 0;
892 }
893 
894 void
fuse_internal_forget_send(struct mount * mp,struct thread * td,struct ucred * cred,uint64_t nodeid,uint64_t nlookup)895 fuse_internal_forget_send(struct mount *mp,
896     struct thread *td,
897     struct ucred *cred,
898     uint64_t nodeid,
899     uint64_t nlookup)
900 {
901 
902 	struct fuse_dispatcher fdi;
903 	struct fuse_forget_in *ffi;
904 
905 	/*
906          * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu",
907          *         (long long unsigned) nodeid));
908          */
909 
910 	fdisp_init(&fdi, sizeof(*ffi));
911 	fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred);
912 
913 	ffi = fdi.indata;
914 	ffi->nlookup = nlookup;
915 
916 	fuse_insert_message(fdi.tick, false);
917 	fdisp_destroy(&fdi);
918 }
919 
920 /* Fetch the vnode's attributes from the daemon*/
921 int
fuse_internal_do_getattr(struct vnode * vp,struct vattr * vap,struct ucred * cred,struct thread * td)922 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap,
923 	struct ucred *cred, struct thread *td)
924 {
925 	struct fuse_dispatcher fdi;
926 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
927 	struct fuse_getattr_in *fgai;
928 	struct fuse_attr_out *fao;
929 	off_t old_filesize = fvdat->cached_attrs.va_size;
930 	struct timespec old_atime = fvdat->cached_attrs.va_atime;
931 	struct timespec old_ctime = fvdat->cached_attrs.va_ctime;
932 	struct timespec old_mtime = fvdat->cached_attrs.va_mtime;
933 	enum vtype vtyp;
934 	int err;
935 
936 	ASSERT_VOP_LOCKED(vp, __func__);
937 
938 	fdisp_init(&fdi, sizeof(*fgai));
939 	fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred);
940 	fgai = fdi.indata;
941 	/*
942 	 * We could look up a file handle and set it in fgai->fh, but that
943 	 * involves extra runtime work and I'm unaware of any file systems that
944 	 * care.
945 	 */
946 	fgai->getattr_flags = 0;
947 	if ((err = fdisp_wait_answ(&fdi))) {
948 		if (err == ENOENT)
949 			fuse_internal_vnode_disappear(vp);
950 		goto out;
951 	}
952 
953 	fao = (struct fuse_attr_out *)fdi.answ;
954 	vtyp = IFTOVT(fao->attr.mode);
955 	if (fvdat->flag & FN_SIZECHANGE)
956 		fao->attr.size = old_filesize;
957 	if (fvdat->flag & FN_ATIMECHANGE) {
958 		fao->attr.atime = old_atime.tv_sec;
959 		fao->attr.atimensec = old_atime.tv_nsec;
960 	}
961 	if (fvdat->flag & FN_CTIMECHANGE) {
962 		fao->attr.ctime = old_ctime.tv_sec;
963 		fao->attr.ctimensec = old_ctime.tv_nsec;
964 	}
965 	if (fvdat->flag & FN_MTIMECHANGE) {
966 		fao->attr.mtime = old_mtime.tv_sec;
967 		fao->attr.mtimensec = old_mtime.tv_nsec;
968 	}
969 	fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
970 		fao->attr_valid_nsec, vap, true);
971 	if (vtyp != vnode_vtype(vp)) {
972 		fuse_internal_vnode_disappear(vp);
973 		err = ENOENT;
974 	}
975 
976 out:
977 	fdisp_destroy(&fdi);
978 	return err;
979 }
980 
981 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */
982 int
fuse_internal_getattr(struct vnode * vp,struct vattr * vap,struct ucred * cred,struct thread * td)983 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred,
984 	struct thread *td)
985 {
986 	struct vattr *attrs;
987 
988 	if ((attrs = VTOVA(vp)) != NULL) {
989 		*vap = *attrs;	/* struct copy */
990 		return 0;
991 	}
992 
993 	return fuse_internal_do_getattr(vp, vap, cred, td);
994 }
995 
996 void
fuse_internal_vnode_disappear(struct vnode * vp)997 fuse_internal_vnode_disappear(struct vnode *vp)
998 {
999 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1000 
1001 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
1002 	fvdat->flag |= FN_REVOKED;
1003 	cache_purge(vp);
1004 }
1005 
1006 /* fuse start/stop */
1007 
1008 SDT_PROBE_DEFINE2(fusefs, , internal, init_done,
1009 	"struct fuse_data*", "struct fuse_init_out*");
1010 int
fuse_internal_init_callback(struct fuse_ticket * tick,struct uio * uio)1011 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio)
1012 {
1013 	int err = 0;
1014 	struct fuse_data *data = tick->tk_data;
1015 	struct fuse_init_out *fiio = NULL;
1016 
1017 	if ((err = tick->tk_aw_ohead.error)) {
1018 		goto out;
1019 	}
1020 	if ((err = fticket_pull(tick, uio))) {
1021 		goto out;
1022 	}
1023 	fiio = fticket_resp(tick)->base;
1024 
1025 	data->fuse_libabi_major = fiio->major;
1026 	data->fuse_libabi_minor = fiio->minor;
1027 	if (!fuse_libabi_geq(data, 7, 4)) {
1028 		/*
1029 		 * With a little work we could support servers as old as 7.1.
1030 		 * But there would be little payoff.
1031 		 */
1032 		SDT_PROBE2(fusefs, , internal, trace, 1,
1033 			"userpace version too low");
1034 		err = EPROTONOSUPPORT;
1035 		goto out;
1036 	}
1037 
1038 	if (fuse_libabi_geq(data, 7, 5)) {
1039 		if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) ||
1040 		    fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) {
1041 			data->max_write = fiio->max_write;
1042 			if (fiio->flags & FUSE_ASYNC_READ)
1043 				data->dataflags |= FSESS_ASYNC_READ;
1044 			if (fiio->flags & FUSE_POSIX_LOCKS)
1045 				data->dataflags |= FSESS_POSIX_LOCKS;
1046 			if (fiio->flags & FUSE_EXPORT_SUPPORT)
1047 				data->dataflags |= FSESS_EXPORT_SUPPORT;
1048 			/*
1049 			 * Don't bother to check FUSE_BIG_WRITES, because it's
1050 			 * redundant with max_write
1051 			 */
1052 			/*
1053 			 * max_background and congestion_threshold are not
1054 			 * implemented
1055 			 */
1056 		} else {
1057 			err = EINVAL;
1058 		}
1059 	} else {
1060 		/* Old fixed values */
1061 		data->max_write = 4096;
1062 	}
1063 
1064 	if (fuse_libabi_geq(data, 7, 6))
1065 		data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf;
1066 
1067 	if (!fuse_libabi_geq(data, 7, 7))
1068 		fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
1069 
1070 	if (!fuse_libabi_geq(data, 7, 8)) {
1071 		fsess_set_notimpl(data->mp, FUSE_BMAP);
1072 		fsess_set_notimpl(data->mp, FUSE_DESTROY);
1073 	}
1074 
1075 	if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 &&
1076 	    fiio->time_gran <= 1000000000)
1077 		data->time_gran = fiio->time_gran;
1078 	else
1079 		data->time_gran = 1;
1080 
1081 	if (!fuse_libabi_geq(data, 7, 23))
1082 		data->cache_mode = fuse_data_cache_mode;
1083 	else if (fiio->flags & FUSE_WRITEBACK_CACHE)
1084 		data->cache_mode = FUSE_CACHE_WB;
1085 	else
1086 		data->cache_mode = FUSE_CACHE_WT;
1087 
1088 out:
1089 	if (err) {
1090 		fdata_set_dead(data);
1091 	}
1092 	FUSE_LOCK();
1093 	data->dataflags |= FSESS_INITED;
1094 	SDT_PROBE2(fusefs, , internal, init_done, data, fiio);
1095 	wakeup(&data->ticketer);
1096 	FUSE_UNLOCK();
1097 
1098 	return 0;
1099 }
1100 
1101 void
fuse_internal_send_init(struct fuse_data * data,struct thread * td)1102 fuse_internal_send_init(struct fuse_data *data, struct thread *td)
1103 {
1104 	struct fuse_init_in *fiii;
1105 	struct fuse_dispatcher fdi;
1106 
1107 	fdisp_init(&fdi, sizeof(*fiii));
1108 	fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL);
1109 	fiii = fdi.indata;
1110 	fiii->major = FUSE_KERNEL_VERSION;
1111 	fiii->minor = FUSE_KERNEL_MINOR_VERSION;
1112 	/*
1113 	 * fusefs currently reads ahead no more than one cache block at a time.
1114 	 * See fuse_read_biobackend
1115 	 */
1116 	fiii->max_readahead = maxbcachebuf;
1117 	/*
1118 	 * Unsupported features:
1119 	 * FUSE_FILE_OPS: No known FUSE server or client supports it
1120 	 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it
1121 	 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even
1122 	 *	when default ACLs are in use.
1123 	 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD
1124 	 *	doesn't have splice(2).
1125 	 * FUSE_FLOCK_LOCKS: not yet implemented
1126 	 * FUSE_HAS_IOCTL_DIR: not yet implemented
1127 	 * FUSE_AUTO_INVAL_DATA: not yet implemented
1128 	 * FUSE_DO_READDIRPLUS: not yet implemented
1129 	 * FUSE_READDIRPLUS_AUTO: not yet implemented
1130 	 * FUSE_ASYNC_DIO: not yet implemented
1131 	 * FUSE_NO_OPEN_SUPPORT: not yet implemented
1132 	 */
1133 	fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT
1134 		| FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE;
1135 
1136 	fuse_insert_callback(fdi.tick, fuse_internal_init_callback);
1137 	fuse_insert_message(fdi.tick, false);
1138 	fdisp_destroy(&fdi);
1139 }
1140 
1141 /*
1142  * Send a FUSE_SETATTR operation with no permissions checks.  If cred is NULL,
1143  * send the request with root credentials
1144  */
fuse_internal_setattr(struct vnode * vp,struct vattr * vap,struct thread * td,struct ucred * cred)1145 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap,
1146 	struct thread *td, struct ucred *cred)
1147 {
1148 	struct fuse_vnode_data *fvdat;
1149 	struct fuse_dispatcher fdi;
1150 	struct fuse_setattr_in *fsai;
1151 	struct mount *mp;
1152 	pid_t pid = td->td_proc->p_pid;
1153 	struct fuse_data *data;
1154 	int dataflags;
1155 	int err = 0;
1156 	enum vtype vtyp;
1157 	int sizechanged = -1;
1158 	uint64_t newsize = 0;
1159 
1160 	ASSERT_VOP_ELOCKED(vp, __func__);
1161 
1162 	mp = vnode_mount(vp);
1163 	fvdat = VTOFUD(vp);
1164 	data = fuse_get_mpdata(mp);
1165 	dataflags = data->dataflags;
1166 
1167 	fdisp_init(&fdi, sizeof(*fsai));
1168 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1169 	if (!cred) {
1170 		fdi.finh->uid = 0;
1171 		fdi.finh->gid = 0;
1172 	}
1173 	fsai = fdi.indata;
1174 	fsai->valid = 0;
1175 
1176 	if (vap->va_uid != (uid_t)VNOVAL) {
1177 		fsai->uid = vap->va_uid;
1178 		fsai->valid |= FATTR_UID;
1179 	}
1180 	if (vap->va_gid != (gid_t)VNOVAL) {
1181 		fsai->gid = vap->va_gid;
1182 		fsai->valid |= FATTR_GID;
1183 	}
1184 	if (vap->va_size != VNOVAL) {
1185 		struct fuse_filehandle *fufh = NULL;
1186 
1187 		/*Truncate to a new value. */
1188 		fsai->size = vap->va_size;
1189 		sizechanged = 1;
1190 		newsize = vap->va_size;
1191 		fsai->valid |= FATTR_SIZE;
1192 
1193 		fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1194 		if (fufh) {
1195 			fsai->fh = fufh->fh_id;
1196 			fsai->valid |= FATTR_FH;
1197 		}
1198 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1199 	}
1200 	if (vap->va_atime.tv_sec != VNOVAL) {
1201 		fsai->atime = vap->va_atime.tv_sec;
1202 		fsai->atimensec = vap->va_atime.tv_nsec;
1203 		fsai->valid |= FATTR_ATIME;
1204 		if (vap->va_vaflags & VA_UTIMES_NULL)
1205 			fsai->valid |= FATTR_ATIME_NOW;
1206 	} else if (fvdat->flag & FN_ATIMECHANGE) {
1207 		fsai->atime = fvdat->cached_attrs.va_atime.tv_sec;
1208 		fsai->atimensec = fvdat->cached_attrs.va_atime.tv_nsec;
1209 		fsai->valid |= FATTR_ATIME;
1210 	}
1211 	if (vap->va_mtime.tv_sec != VNOVAL) {
1212 		fsai->mtime = vap->va_mtime.tv_sec;
1213 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1214 		fsai->valid |= FATTR_MTIME;
1215 		if (vap->va_vaflags & VA_UTIMES_NULL)
1216 			fsai->valid |= FATTR_MTIME_NOW;
1217 	} else if (fvdat->flag & FN_MTIMECHANGE) {
1218 		fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec;
1219 		fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
1220 		fsai->valid |= FATTR_MTIME;
1221 	}
1222 	if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) {
1223 		fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec;
1224 		fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
1225 		fsai->valid |= FATTR_CTIME;
1226 	}
1227 	if (vap->va_mode != (mode_t)VNOVAL) {
1228 		fsai->mode = vap->va_mode & ALLPERMS;
1229 		fsai->valid |= FATTR_MODE;
1230 	}
1231 	if (!fsai->valid) {
1232 		goto out;
1233 	}
1234 
1235 	if ((err = fdisp_wait_answ(&fdi)))
1236 		goto out;
1237 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1238 
1239 	if (vnode_vtype(vp) != vtyp) {
1240 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1241 			SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! "
1242 				"vnode_vtype is VNON and vtype isn't.");
1243 		} else {
1244 			/*
1245 	                 * STALE vnode, ditch
1246 	                 *
1247 			 * The vnode has changed its type "behind our back".
1248 			 * This probably means that the file got deleted and
1249 			 * recreated on the server, with the same inode.
1250 			 * There's nothing really we can do, so let us just
1251 			 * return ENOENT.  After all, the entry must not have
1252 			 * existed in the recent past.  If the user tries
1253 			 * again, it will work.
1254 	                 */
1255 			fuse_internal_vnode_disappear(vp);
1256 			err = ENOENT;
1257 		}
1258 	}
1259 	if (err == 0) {
1260 		struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1261 		fuse_vnode_undirty_cached_timestamps(vp, true);
1262 		fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1263 			fao->attr_valid_nsec, NULL, false);
1264 		getnanouptime(&fvdat->last_local_modify);
1265 	}
1266 
1267 out:
1268 	fdisp_destroy(&fdi);
1269 	return err;
1270 }
1271 
1272 #ifdef ZERO_PAD_INCOMPLETE_BUFS
1273 static int
isbzero(void * buf,size_t len)1274 isbzero(void *buf, size_t len)
1275 {
1276 	int i;
1277 
1278 	for (i = 0; i < len; i++) {
1279 		if (((char *)buf)[i])
1280 			return (0);
1281 	}
1282 
1283 	return (1);
1284 }
1285 
1286 #endif
1287 
1288 void
fuse_internal_init(void)1289 fuse_internal_init(void)
1290 {
1291 	fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK);
1292 	fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK);
1293 }
1294 
1295 void
fuse_internal_destroy(void)1296 fuse_internal_destroy(void)
1297 {
1298 	counter_u64_free(fuse_lookup_cache_hits);
1299 	counter_u64_free(fuse_lookup_cache_misses);
1300 }
1301