xref: /freebsd/sys/fs/fuse/fuse_internal.c (revision f61e92ca5a23450bc28169bbdd71d7674df98c19)
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/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
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
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);
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_not_impl(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
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 {
260 	struct mount *mp;
261 	struct fuse_vnode_data *fvdat;
262 	struct fuse_data *data;
263 	struct vattr *vp_cache_at;
264 
265 	mp = vnode_mount(vp);
266 	fvdat = VTOFUD(vp);
267 	data = fuse_get_mpdata(mp);
268 
269 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs");
270 
271 	fuse_validity_2_bintime(attr_valid, attr_valid_nsec,
272 		&fvdat->attr_cache_timeout);
273 
274 	/* Fix our buffers if the filesize changed without us knowing */
275 	if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) {
276 		(void)fuse_vnode_setsize(vp, attr->size);
277 		fvdat->cached_attrs.va_size = attr->size;
278 	}
279 
280 	if (attr_valid > 0 || attr_valid_nsec > 0)
281 		vp_cache_at = &(fvdat->cached_attrs);
282 	else if (vap != NULL)
283 		vp_cache_at = vap;
284 	else
285 		return;
286 
287 	vattr_null(vp_cache_at);
288 	vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0];
289 	vp_cache_at->va_fileid = attr->ino;
290 	vp_cache_at->va_mode = attr->mode & ~S_IFMT;
291 	vp_cache_at->va_nlink     = attr->nlink;
292 	vp_cache_at->va_uid       = attr->uid;
293 	vp_cache_at->va_gid       = attr->gid;
294 	vp_cache_at->va_rdev      = attr->rdev;
295 	vp_cache_at->va_size      = attr->size;
296 	/* XXX on i386, seconds are truncated to 32 bits */
297 	vp_cache_at->va_atime.tv_sec  = attr->atime;
298 	vp_cache_at->va_atime.tv_nsec = attr->atimensec;
299 	vp_cache_at->va_mtime.tv_sec  = attr->mtime;
300 	vp_cache_at->va_mtime.tv_nsec = attr->mtimensec;
301 	vp_cache_at->va_ctime.tv_sec  = attr->ctime;
302 	vp_cache_at->va_ctime.tv_nsec = attr->ctimensec;
303 	if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0)
304 		vp_cache_at->va_blocksize = attr->blksize;
305 	else
306 		vp_cache_at->va_blocksize = PAGE_SIZE;
307 	vp_cache_at->va_type = IFTOVT(attr->mode);
308 	vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE;
309 	vp_cache_at->va_flags = 0;
310 
311 	if (vap != vp_cache_at && vap != NULL)
312 		memcpy(vap, vp_cache_at, sizeof(*vap));
313 }
314 
315 /* fsync */
316 
317 int
318 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio)
319 {
320 	if (tick->tk_aw_ohead.error == ENOSYS) {
321 		fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick));
322 	}
323 	return 0;
324 }
325 
326 int
327 fuse_internal_fsync(struct vnode *vp,
328     struct thread *td,
329     int waitfor,
330     bool datasync)
331 {
332 	struct fuse_fsync_in *ffsi = NULL;
333 	struct fuse_dispatcher fdi;
334 	struct fuse_filehandle *fufh;
335 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
336 	struct mount *mp = vnode_mount(vp);
337 	int op = FUSE_FSYNC;
338 	int err = 0;
339 
340 	if (fsess_not_impl(vnode_mount(vp),
341 	    (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
342 		return 0;
343 	}
344 	if (vnode_isdir(vp))
345 		op = FUSE_FSYNCDIR;
346 
347 	if (fsess_not_impl(mp, op))
348 		return 0;
349 
350 	fdisp_init(&fdi, sizeof(*ffsi));
351 	/*
352 	 * fsync every open file handle for this file, because we can't be sure
353 	 * which file handle the caller is really referring to.
354 	 */
355 	LIST_FOREACH(fufh, &fvdat->handles, next) {
356 		fdi.iosize = sizeof(*ffsi);
357 		if (ffsi == NULL)
358 			fdisp_make_vp(&fdi, op, vp, td, NULL);
359 		else
360 			fdisp_refresh_vp(&fdi, op, vp, td, NULL);
361 		ffsi = fdi.indata;
362 		ffsi->fh = fufh->fh_id;
363 		ffsi->fsync_flags = 0;
364 
365 		if (datasync)
366 			ffsi->fsync_flags = 1;
367 
368 		if (waitfor == MNT_WAIT) {
369 			err = fdisp_wait_answ(&fdi);
370 		} else {
371 			fuse_insert_callback(fdi.tick,
372 				fuse_internal_fsync_callback);
373 			fuse_insert_message(fdi.tick, false);
374 		}
375 		if (err == ENOSYS) {
376 			/* ENOSYS means "success, and don't call again" */
377 			fsess_set_notimpl(mp, op);
378 			err = 0;
379 			break;
380 		}
381 	}
382 	fdisp_destroy(&fdi);
383 
384 	return err;
385 }
386 
387 /* Asynchronous invalidation */
388 SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry,
389 	"struct vnode*", "struct fuse_notify_inval_entry_out*", "char*");
390 int
391 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio)
392 {
393 	struct fuse_notify_inval_entry_out fnieo;
394 	struct componentname cn;
395 	struct vnode *dvp, *vp;
396 	char name[PATH_MAX];
397 	int err;
398 
399 	if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0)
400 		return (err);
401 
402 	if (fnieo.namelen >= sizeof(name))
403 		return (EINVAL);
404 
405 	if ((err = uiomove(name, fnieo.namelen, uio)) != 0)
406 		return (err);
407 	name[fnieo.namelen] = '\0';
408 	/* fusefs does not cache "." or ".." entries */
409 	if (strncmp(name, ".", sizeof(".")) == 0 ||
410 	    strncmp(name, "..", sizeof("..")) == 0)
411 		return (0);
412 
413 	if (fnieo.parent == FUSE_ROOT_ID)
414 		err = VFS_ROOT(mp, LK_SHARED, &dvp);
415 	else
416 		err = fuse_internal_get_cached_vnode( mp, fnieo.parent,
417 			LK_SHARED, &dvp);
418 	SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name);
419 	/*
420 	 * If dvp is not in the cache, then it must've been reclaimed.  And
421 	 * since fuse_vnop_reclaim does a cache_purge, name's entry must've
422 	 * been invalidated already.  So we can safely return if dvp == NULL
423 	 */
424 	if (err != 0 || dvp == NULL)
425 		return (err);
426 	/*
427 	 * XXX we can't check dvp's generation because the FUSE invalidate
428 	 * entry message doesn't include it.  Worse case is that we invalidate
429 	 * an entry that didn't need to be invalidated.
430 	 */
431 
432 	cn.cn_nameiop = LOOKUP;
433 	cn.cn_flags = 0;	/* !MAKEENTRY means free cached entry */
434 	cn.cn_thread = curthread;
435 	cn.cn_cred = curthread->td_ucred;
436 	cn.cn_lkflags = LK_SHARED;
437 	cn.cn_pnbuf = NULL;
438 	cn.cn_nameptr = name;
439 	cn.cn_namelen = fnieo.namelen;
440 	err = cache_lookup(dvp, &vp, &cn, NULL, NULL);
441 	MPASS(err == 0);
442 	fuse_vnode_clear_attr_cache(dvp);
443 	vput(dvp);
444 	return (0);
445 }
446 
447 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode,
448 	"struct vnode*", "struct fuse_notify_inval_inode_out *");
449 int
450 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio)
451 {
452 	struct fuse_notify_inval_inode_out fniio;
453 	struct vnode *vp;
454 	int err;
455 
456 	if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0)
457 		return (err);
458 
459 	if (fniio.ino == FUSE_ROOT_ID)
460 		err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp);
461 	else
462 		err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED,
463 			&vp);
464 	SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio);
465 	if (err != 0 || vp == NULL)
466 		return (err);
467 	/*
468 	 * XXX we can't check vp's generation because the FUSE invalidate
469 	 * entry message doesn't include it.  Worse case is that we invalidate
470 	 * an inode that didn't need to be invalidated.
471 	 */
472 
473 	/*
474 	 * Flush and invalidate buffers if off >= 0.  Technically we only need
475 	 * to flush and invalidate the range of offsets [off, off + len), but
476 	 * for simplicity's sake we do everything.
477 	 */
478 	if (fniio.off >= 0)
479 		fuse_io_invalbuf(vp, curthread);
480 	fuse_vnode_clear_attr_cache(vp);
481 	vput(vp);
482 	return (0);
483 }
484 
485 /* mknod */
486 int
487 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp,
488 	struct componentname *cnp, struct vattr *vap)
489 {
490 	struct fuse_data *data;
491 	struct fuse_mknod_in fmni;
492 	size_t insize;
493 
494 	data = fuse_get_mpdata(dvp->v_mount);
495 
496 	fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
497 	fmni.rdev = vap->va_rdev;
498 	if (fuse_libabi_geq(data, 7, 12)) {
499 		insize = sizeof(fmni);
500 		fmni.umask = curthread->td_proc->p_pd->pd_cmask;
501 	} else {
502 		insize = FUSE_COMPAT_MKNOD_IN_SIZE;
503 	}
504 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
505 	    insize, vap->va_type));
506 }
507 
508 /* readdir */
509 
510 int
511 fuse_internal_readdir(struct vnode *vp,
512     struct uio *uio,
513     off_t startoff,
514     struct fuse_filehandle *fufh,
515     struct fuse_iov *cookediov,
516     int *ncookies,
517     u_long *cookies)
518 {
519 	int err = 0;
520 	struct fuse_dispatcher fdi;
521 	struct fuse_read_in *fri = NULL;
522 	int fnd_start;
523 
524 	if (uio_resid(uio) == 0)
525 		return 0;
526 	fdisp_init(&fdi, 0);
527 
528 	/*
529 	 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p
530 	 * I/O).
531 	 */
532 
533 	/*
534 	 * fnd_start is set non-zero once the offset in the directory gets
535 	 * to the startoff.  This is done because directories must be read
536 	 * from the beginning (offset == 0) when fuse_vnop_readdir() needs
537 	 * to do an open of the directory.
538 	 * If it is not set non-zero here, it will be set non-zero in
539 	 * fuse_internal_readdir_processdata() when uio_offset == startoff.
540 	 */
541 	fnd_start = 0;
542 	if (uio->uio_offset == startoff)
543 		fnd_start = 1;
544 	while (uio_resid(uio) > 0) {
545 		fdi.iosize = sizeof(*fri);
546 		if (fri == NULL)
547 			fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
548 		else
549 			fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
550 
551 		fri = fdi.indata;
552 		fri->fh = fufh->fh_id;
553 		fri->offset = uio_offset(uio);
554 		fri->size = MIN(uio->uio_resid,
555 		    fuse_get_mpdata(vp->v_mount)->max_read);
556 
557 		if ((err = fdisp_wait_answ(&fdi)))
558 			break;
559 		if ((err = fuse_internal_readdir_processdata(uio, startoff,
560 		    &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov,
561 		    ncookies, &cookies)))
562 			break;
563 	}
564 
565 	fdisp_destroy(&fdi);
566 	return ((err == -1) ? 0 : err);
567 }
568 
569 /*
570  * Return -1 to indicate that this readdir is finished, 0 if it copied
571  * all the directory data read in and it may be possible to read more
572  * and greater than 0 for a failure.
573  */
574 int
575 fuse_internal_readdir_processdata(struct uio *uio,
576     off_t startoff,
577     int *fnd_start,
578     size_t reqsize,
579     void *buf,
580     size_t bufsize,
581     struct fuse_iov *cookediov,
582     int *ncookies,
583     u_long **cookiesp)
584 {
585 	int err = 0;
586 	int oreclen;
587 	size_t freclen;
588 
589 	struct dirent *de;
590 	struct fuse_dirent *fudge;
591 	u_long *cookies;
592 
593 	cookies = *cookiesp;
594 	if (bufsize < FUSE_NAME_OFFSET)
595 		return -1;
596 	for (;;) {
597 		if (bufsize < FUSE_NAME_OFFSET) {
598 			err = -1;
599 			break;
600 		}
601 		fudge = (struct fuse_dirent *)buf;
602 		freclen = FUSE_DIRENT_SIZE(fudge);
603 
604 		if (bufsize < freclen) {
605 			/*
606 			 * This indicates a partial directory entry at the
607 			 * end of the directory data.
608 			 */
609 			err = -1;
610 			break;
611 		}
612 #ifdef ZERO_PAD_INCOMPLETE_BUFS
613 		if (isbzero(buf, FUSE_NAME_OFFSET)) {
614 			err = -1;
615 			break;
616 		}
617 #endif
618 
619 		if (!fudge->namelen || fudge->namelen > MAXNAMLEN) {
620 			err = EINVAL;
621 			break;
622 		}
623 		oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *)
624 					    &fudge->namelen);
625 
626 		if (oreclen > uio_resid(uio)) {
627 			/* Out of space for the dir so we are done. */
628 			err = -1;
629 			break;
630 		}
631 		/*
632 		 * Don't start to copy the directory entries out until
633 		 * the requested offset in the directory is found.
634 		 */
635 		if (*fnd_start != 0) {
636 			fiov_adjust(cookediov, oreclen);
637 			bzero(cookediov->base, oreclen);
638 
639 			de = (struct dirent *)cookediov->base;
640 			de->d_fileno = fudge->ino;
641 			de->d_off = fudge->off;
642 			de->d_reclen = oreclen;
643 			de->d_type = fudge->type;
644 			de->d_namlen = fudge->namelen;
645 			memcpy((char *)cookediov->base + sizeof(struct dirent) -
646 			       MAXNAMLEN - 1,
647 			       (char *)buf + FUSE_NAME_OFFSET, fudge->namelen);
648 			dirent_terminate(de);
649 
650 			err = uiomove(cookediov->base, cookediov->len, uio);
651 			if (err)
652 				break;
653 			if (cookies != NULL) {
654 				if (*ncookies == 0) {
655 					err = -1;
656 					break;
657 				}
658 				*cookies = fudge->off;
659 				cookies++;
660 				(*ncookies)--;
661 			}
662 		} else if (startoff == fudge->off)
663 			*fnd_start = 1;
664 		buf = (char *)buf + freclen;
665 		bufsize -= freclen;
666 		uio_setoffset(uio, fudge->off);
667 	}
668 	*cookiesp = cookies;
669 
670 	return err;
671 }
672 
673 /* remove */
674 
675 int
676 fuse_internal_remove(struct vnode *dvp,
677     struct vnode *vp,
678     struct componentname *cnp,
679     enum fuse_opcode op)
680 {
681 	struct fuse_dispatcher fdi;
682 	nlink_t nlink;
683 	int err = 0;
684 
685 	fdisp_init(&fdi, cnp->cn_namelen + 1);
686 	fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred);
687 
688 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
689 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
690 
691 	err = fdisp_wait_answ(&fdi);
692 	fdisp_destroy(&fdi);
693 
694 	if (err)
695 		return (err);
696 
697 	/*
698 	 * Access the cached nlink even if the attr cached has expired.  If
699 	 * it's inaccurate, the worst that will happen is:
700 	 * 1) We'll recycle the vnode even though the file has another link we
701 	 *    don't know about, costing a bit of cpu time, or
702 	 * 2) We won't recycle the vnode even though all of its links are gone.
703 	 *    It will linger around until vnlru reclaims it, costing a bit of
704 	 *    temporary memory.
705 	 */
706 	nlink = VTOFUD(vp)->cached_attrs.va_nlink--;
707 
708 	/*
709 	 * Purge the parent's attribute cache because the daemon
710 	 * should've updated its mtime and ctime.
711 	 */
712 	fuse_vnode_clear_attr_cache(dvp);
713 
714 	/* NB: nlink could be zero if it was never cached */
715 	if (nlink <= 1 || vnode_vtype(vp) == VDIR) {
716 		fuse_internal_vnode_disappear(vp);
717 	} else {
718 		cache_purge(vp);
719 		fuse_vnode_update(vp, FN_CTIMECHANGE);
720 	}
721 
722 	return err;
723 }
724 
725 /* rename */
726 
727 int
728 fuse_internal_rename(struct vnode *fdvp,
729     struct componentname *fcnp,
730     struct vnode *tdvp,
731     struct componentname *tcnp)
732 {
733 	struct fuse_dispatcher fdi;
734 	struct fuse_rename_in *fri;
735 	int err = 0;
736 
737 	fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2);
738 	fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred);
739 
740 	fri = fdi.indata;
741 	fri->newdir = VTOI(tdvp);
742 	memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr,
743 	    fcnp->cn_namelen);
744 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0';
745 	memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1,
746 	    tcnp->cn_nameptr, tcnp->cn_namelen);
747 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen +
748 	    tcnp->cn_namelen + 1] = '\0';
749 
750 	err = fdisp_wait_answ(&fdi);
751 	fdisp_destroy(&fdi);
752 	return err;
753 }
754 
755 /* strategy */
756 
757 /* entity creation */
758 
759 void
760 fuse_internal_newentry_makerequest(struct mount *mp,
761     uint64_t dnid,
762     struct componentname *cnp,
763     enum fuse_opcode op,
764     void *buf,
765     size_t bufsize,
766     struct fuse_dispatcher *fdip)
767 {
768 	fdip->iosize = bufsize + cnp->cn_namelen + 1;
769 
770 	fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred);
771 	memcpy(fdip->indata, buf, bufsize);
772 	memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen);
773 	((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0';
774 }
775 
776 int
777 fuse_internal_newentry_core(struct vnode *dvp,
778     struct vnode **vpp,
779     struct componentname *cnp,
780     enum vtype vtyp,
781     struct fuse_dispatcher *fdip)
782 {
783 	int err = 0;
784 	struct fuse_entry_out *feo;
785 	struct mount *mp = vnode_mount(dvp);
786 
787 	if ((err = fdisp_wait_answ(fdip))) {
788 		return err;
789 	}
790 	feo = fdip->answ;
791 
792 	if ((err = fuse_internal_checkentry(feo, vtyp))) {
793 		return err;
794 	}
795 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp);
796 	if (err) {
797 		fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred,
798 		    feo->nodeid, 1);
799 		return err;
800 	}
801 
802 	/*
803 	 * Purge the parent's attribute cache because the daemon should've
804 	 * updated its mtime and ctime
805 	 */
806 	fuse_vnode_clear_attr_cache(dvp);
807 
808 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
809 		feo->attr_valid_nsec, NULL);
810 
811 	return err;
812 }
813 
814 int
815 fuse_internal_newentry(struct vnode *dvp,
816     struct vnode **vpp,
817     struct componentname *cnp,
818     enum fuse_opcode op,
819     void *buf,
820     size_t bufsize,
821     enum vtype vtype)
822 {
823 	int err;
824 	struct fuse_dispatcher fdi;
825 	struct mount *mp = vnode_mount(dvp);
826 
827 	fdisp_init(&fdi, 0);
828 	fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf,
829 	    bufsize, &fdi);
830 	err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi);
831 	fdisp_destroy(&fdi);
832 
833 	return err;
834 }
835 
836 /* entity destruction */
837 
838 int
839 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio)
840 {
841 	fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL,
842 	    ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1);
843 
844 	return 0;
845 }
846 
847 void
848 fuse_internal_forget_send(struct mount *mp,
849     struct thread *td,
850     struct ucred *cred,
851     uint64_t nodeid,
852     uint64_t nlookup)
853 {
854 
855 	struct fuse_dispatcher fdi;
856 	struct fuse_forget_in *ffi;
857 
858 	/*
859          * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu",
860          *         (long long unsigned) nodeid));
861          */
862 
863 	fdisp_init(&fdi, sizeof(*ffi));
864 	fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred);
865 
866 	ffi = fdi.indata;
867 	ffi->nlookup = nlookup;
868 
869 	fuse_insert_message(fdi.tick, false);
870 	fdisp_destroy(&fdi);
871 }
872 
873 SDT_PROBE_DEFINE2(fusefs, , internal, getattr_cache_incoherent,
874 	"struct vnode*", "struct fuse_attr_out*");
875 
876 /* Fetch the vnode's attributes from the daemon*/
877 int
878 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap,
879 	struct ucred *cred, struct thread *td)
880 {
881 	struct fuse_dispatcher fdi;
882 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
883 	struct fuse_getattr_in *fgai;
884 	struct fuse_attr_out *fao;
885 	off_t old_filesize = fvdat->cached_attrs.va_size;
886 	struct timespec old_ctime = fvdat->cached_attrs.va_ctime;
887 	struct timespec old_mtime = fvdat->cached_attrs.va_mtime;
888 	enum vtype vtyp;
889 	int err;
890 
891 	fdisp_init(&fdi, sizeof(*fgai));
892 	fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred);
893 	fgai = fdi.indata;
894 	/*
895 	 * We could look up a file handle and set it in fgai->fh, but that
896 	 * involves extra runtime work and I'm unaware of any file systems that
897 	 * care.
898 	 */
899 	fgai->getattr_flags = 0;
900 	if ((err = fdisp_wait_answ(&fdi))) {
901 		if (err == ENOENT)
902 			fuse_internal_vnode_disappear(vp);
903 		goto out;
904 	}
905 
906 	fao = (struct fuse_attr_out *)fdi.answ;
907 	vtyp = IFTOVT(fao->attr.mode);
908 	if (fvdat->flag & FN_SIZECHANGE)
909 		fao->attr.size = old_filesize;
910 	if (fvdat->flag & FN_CTIMECHANGE) {
911 		fao->attr.ctime = old_ctime.tv_sec;
912 		fao->attr.ctimensec = old_ctime.tv_nsec;
913 	}
914 	if (fvdat->flag & FN_MTIMECHANGE) {
915 		fao->attr.mtime = old_mtime.tv_sec;
916 		fao->attr.mtimensec = old_mtime.tv_nsec;
917 	}
918 	if (vnode_isreg(vp) &&
919 	    fvdat->cached_attrs.va_size != VNOVAL &&
920 	    fao->attr.size != fvdat->cached_attrs.va_size) {
921 		/*
922 		 * The server changed the file's size even though we had it
923 		 * cached!  That's a server bug.
924 		 */
925 		SDT_PROBE2(fusefs, , internal, getattr_cache_incoherent, vp,
926 		    fao);
927 		printf("%s: cache incoherent on %s!  "
928 		    "Buggy FUSE server detected.  To prevent data corruption, "
929 		    "disable the data cache by mounting with -o direct_io, or "
930 		    "as directed otherwise by your FUSE server's "
931 		    "documentation\n", __func__,
932 		    vnode_mount(vp)->mnt_stat.f_mntonname);
933 		int iosize = fuse_iosize(vp);
934 		v_inval_buf_range(vp, 0, INT64_MAX, iosize);
935 	}
936 	fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
937 		fao->attr_valid_nsec, vap);
938 	if (vtyp != vnode_vtype(vp)) {
939 		fuse_internal_vnode_disappear(vp);
940 		err = ENOENT;
941 	}
942 
943 out:
944 	fdisp_destroy(&fdi);
945 	return err;
946 }
947 
948 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */
949 int
950 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred,
951 	struct thread *td)
952 {
953 	struct vattr *attrs;
954 
955 	if ((attrs = VTOVA(vp)) != NULL) {
956 		*vap = *attrs;	/* struct copy */
957 		return 0;
958 	}
959 
960 	return fuse_internal_do_getattr(vp, vap, cred, td);
961 }
962 
963 void
964 fuse_internal_vnode_disappear(struct vnode *vp)
965 {
966 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
967 
968 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
969 	fvdat->flag |= FN_REVOKED;
970 	cache_purge(vp);
971 }
972 
973 /* fuse start/stop */
974 
975 SDT_PROBE_DEFINE2(fusefs, , internal, init_done,
976 	"struct fuse_data*", "struct fuse_init_out*");
977 int
978 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio)
979 {
980 	int err = 0;
981 	struct fuse_data *data = tick->tk_data;
982 	struct fuse_init_out *fiio;
983 
984 	if ((err = tick->tk_aw_ohead.error)) {
985 		goto out;
986 	}
987 	if ((err = fticket_pull(tick, uio))) {
988 		goto out;
989 	}
990 	fiio = fticket_resp(tick)->base;
991 
992 	data->fuse_libabi_major = fiio->major;
993 	data->fuse_libabi_minor = fiio->minor;
994 	if (!fuse_libabi_geq(data, 7, 4)) {
995 		/*
996 		 * With a little work we could support servers as old as 7.1.
997 		 * But there would be little payoff.
998 		 */
999 		SDT_PROBE2(fusefs, , internal, trace, 1,
1000 			"userpace version too low");
1001 		err = EPROTONOSUPPORT;
1002 		goto out;
1003 	}
1004 
1005 	if (fuse_libabi_geq(data, 7, 5)) {
1006 		if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) ||
1007 		    fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) {
1008 			data->max_write = fiio->max_write;
1009 			if (fiio->flags & FUSE_ASYNC_READ)
1010 				data->dataflags |= FSESS_ASYNC_READ;
1011 			if (fiio->flags & FUSE_POSIX_LOCKS)
1012 				data->dataflags |= FSESS_POSIX_LOCKS;
1013 			if (fiio->flags & FUSE_EXPORT_SUPPORT)
1014 				data->dataflags |= FSESS_EXPORT_SUPPORT;
1015 			/*
1016 			 * Don't bother to check FUSE_BIG_WRITES, because it's
1017 			 * redundant with max_write
1018 			 */
1019 			/*
1020 			 * max_background and congestion_threshold are not
1021 			 * implemented
1022 			 */
1023 		} else {
1024 			err = EINVAL;
1025 		}
1026 	} else {
1027 		/* Old fixed values */
1028 		data->max_write = 4096;
1029 	}
1030 
1031 	if (fuse_libabi_geq(data, 7, 6))
1032 		data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf;
1033 
1034 	if (!fuse_libabi_geq(data, 7, 7))
1035 		fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
1036 
1037 	if (!fuse_libabi_geq(data, 7, 8)) {
1038 		fsess_set_notimpl(data->mp, FUSE_BMAP);
1039 		fsess_set_notimpl(data->mp, FUSE_DESTROY);
1040 	}
1041 
1042 	if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 &&
1043 	    fiio->time_gran <= 1000000000)
1044 		data->time_gran = fiio->time_gran;
1045 	else
1046 		data->time_gran = 1;
1047 
1048 	if (!fuse_libabi_geq(data, 7, 23))
1049 		data->cache_mode = fuse_data_cache_mode;
1050 	else if (fiio->flags & FUSE_WRITEBACK_CACHE)
1051 		data->cache_mode = FUSE_CACHE_WB;
1052 	else
1053 		data->cache_mode = FUSE_CACHE_WT;
1054 
1055 	if (!fuse_libabi_geq(data, 7, 24))
1056 		fsess_set_notimpl(data->mp, FUSE_LSEEK);
1057 
1058 	if (!fuse_libabi_geq(data, 7, 28))
1059 		fsess_set_notimpl(data->mp, FUSE_COPY_FILE_RANGE);
1060 
1061 out:
1062 	if (err) {
1063 		fdata_set_dead(data);
1064 	}
1065 	FUSE_LOCK();
1066 	data->dataflags |= FSESS_INITED;
1067 	SDT_PROBE2(fusefs, , internal, init_done, data, fiio);
1068 	wakeup(&data->ticketer);
1069 	FUSE_UNLOCK();
1070 
1071 	return 0;
1072 }
1073 
1074 void
1075 fuse_internal_send_init(struct fuse_data *data, struct thread *td)
1076 {
1077 	struct fuse_init_in *fiii;
1078 	struct fuse_dispatcher fdi;
1079 
1080 	fdisp_init(&fdi, sizeof(*fiii));
1081 	fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL);
1082 	fiii = fdi.indata;
1083 	fiii->major = FUSE_KERNEL_VERSION;
1084 	fiii->minor = FUSE_KERNEL_MINOR_VERSION;
1085 	/*
1086 	 * fusefs currently reads ahead no more than one cache block at a time.
1087 	 * See fuse_read_biobackend
1088 	 */
1089 	fiii->max_readahead = maxbcachebuf;
1090 	/*
1091 	 * Unsupported features:
1092 	 * FUSE_FILE_OPS: No known FUSE server or client supports it
1093 	 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it
1094 	 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even
1095 	 *	when default ACLs are in use.
1096 	 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD
1097 	 *	doesn't have splice(2).
1098 	 * FUSE_FLOCK_LOCKS: not yet implemented
1099 	 * FUSE_HAS_IOCTL_DIR: not yet implemented
1100 	 * FUSE_AUTO_INVAL_DATA: not yet implemented
1101 	 * FUSE_DO_READDIRPLUS: not yet implemented
1102 	 * FUSE_READDIRPLUS_AUTO: not yet implemented
1103 	 * FUSE_ASYNC_DIO: not yet implemented
1104 	 * FUSE_NO_OPEN_SUPPORT: not yet implemented
1105 	 * FUSE_PARALLEL_DIROPS: not yet implemented
1106 	 * FUSE_HANDLE_KILLPRIV: not yet implemented
1107 	 * FUSE_POSIX_ACL: not yet implemented
1108 	 * FUSE_ABORT_ERROR: not yet implemented
1109 	 * FUSE_CACHE_SYMLINKS: not yet implemented
1110 	 * FUSE_MAX_PAGES: not yet implemented
1111 	 */
1112 	fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT
1113 		| FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE;
1114 
1115 	fuse_insert_callback(fdi.tick, fuse_internal_init_callback);
1116 	fuse_insert_message(fdi.tick, false);
1117 	fdisp_destroy(&fdi);
1118 }
1119 
1120 /*
1121  * Send a FUSE_SETATTR operation with no permissions checks.  If cred is NULL,
1122  * send the request with root credentials
1123  */
1124 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap,
1125 	struct thread *td, struct ucred *cred)
1126 {
1127 	struct fuse_vnode_data *fvdat;
1128 	struct fuse_dispatcher fdi;
1129 	struct fuse_setattr_in *fsai;
1130 	struct mount *mp;
1131 	pid_t pid = td->td_proc->p_pid;
1132 	struct fuse_data *data;
1133 	int dataflags;
1134 	int err = 0;
1135 	enum vtype vtyp;
1136 	int sizechanged = -1;
1137 	uint64_t newsize = 0;
1138 
1139 	mp = vnode_mount(vp);
1140 	fvdat = VTOFUD(vp);
1141 	data = fuse_get_mpdata(mp);
1142 	dataflags = data->dataflags;
1143 
1144 	fdisp_init(&fdi, sizeof(*fsai));
1145 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1146 	if (!cred) {
1147 		fdi.finh->uid = 0;
1148 		fdi.finh->gid = 0;
1149 	}
1150 	fsai = fdi.indata;
1151 	fsai->valid = 0;
1152 
1153 	if (vap->va_uid != (uid_t)VNOVAL) {
1154 		fsai->uid = vap->va_uid;
1155 		fsai->valid |= FATTR_UID;
1156 	}
1157 	if (vap->va_gid != (gid_t)VNOVAL) {
1158 		fsai->gid = vap->va_gid;
1159 		fsai->valid |= FATTR_GID;
1160 	}
1161 	if (vap->va_size != VNOVAL) {
1162 		struct fuse_filehandle *fufh = NULL;
1163 
1164 		/*Truncate to a new value. */
1165 		fsai->size = vap->va_size;
1166 		sizechanged = 1;
1167 		newsize = vap->va_size;
1168 		fsai->valid |= FATTR_SIZE;
1169 
1170 		fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1171 		if (fufh) {
1172 			fsai->fh = fufh->fh_id;
1173 			fsai->valid |= FATTR_FH;
1174 		}
1175 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1176 	}
1177 	if (vap->va_atime.tv_sec != VNOVAL) {
1178 		fsai->atime = vap->va_atime.tv_sec;
1179 		fsai->atimensec = vap->va_atime.tv_nsec;
1180 		fsai->valid |= FATTR_ATIME;
1181 		if (vap->va_vaflags & VA_UTIMES_NULL)
1182 			fsai->valid |= FATTR_ATIME_NOW;
1183 	}
1184 	if (vap->va_mtime.tv_sec != VNOVAL) {
1185 		fsai->mtime = vap->va_mtime.tv_sec;
1186 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1187 		fsai->valid |= FATTR_MTIME;
1188 		if (vap->va_vaflags & VA_UTIMES_NULL)
1189 			fsai->valid |= FATTR_MTIME_NOW;
1190 	} else if (fvdat->flag & FN_MTIMECHANGE) {
1191 		fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec;
1192 		fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
1193 		fsai->valid |= FATTR_MTIME;
1194 	}
1195 	if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) {
1196 		fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec;
1197 		fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
1198 		fsai->valid |= FATTR_CTIME;
1199 	}
1200 	if (vap->va_mode != (mode_t)VNOVAL) {
1201 		fsai->mode = vap->va_mode & ALLPERMS;
1202 		fsai->valid |= FATTR_MODE;
1203 	}
1204 	if (!fsai->valid) {
1205 		goto out;
1206 	}
1207 
1208 	if ((err = fdisp_wait_answ(&fdi)))
1209 		goto out;
1210 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1211 
1212 	if (vnode_vtype(vp) != vtyp) {
1213 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1214 			SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! "
1215 				"vnode_vtype is VNON and vtype isn't.");
1216 		} else {
1217 			/*
1218 	                 * STALE vnode, ditch
1219 	                 *
1220 			 * The vnode has changed its type "behind our back".
1221 			 * There's nothing really we can do, so let us just
1222 			 * force an internal revocation and tell the caller to
1223 			 * try again, if interested.
1224 	                 */
1225 			fuse_internal_vnode_disappear(vp);
1226 			err = EAGAIN;
1227 		}
1228 	}
1229 	if (err == 0) {
1230 		struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1231 		fuse_vnode_undirty_cached_timestamps(vp);
1232 		fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1233 			fao->attr_valid_nsec, NULL);
1234 	}
1235 
1236 out:
1237 	fdisp_destroy(&fdi);
1238 	return err;
1239 }
1240 
1241 /*
1242  * FreeBSD clears the SUID and SGID bits on any write by a non-root user.
1243  */
1244 void
1245 fuse_internal_clear_suid_on_write(struct vnode *vp, struct ucred *cred,
1246 	struct thread *td)
1247 {
1248 	struct fuse_data *data;
1249 	struct mount *mp;
1250 	struct vattr va;
1251 	int dataflags;
1252 
1253 	mp = vnode_mount(vp);
1254 	data = fuse_get_mpdata(mp);
1255 	dataflags = data->dataflags;
1256 
1257 	ASSERT_VOP_LOCKED(vp, __func__);
1258 
1259 	if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
1260 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1261 			fuse_internal_getattr(vp, &va, cred, td);
1262 			if (va.va_mode & (S_ISUID | S_ISGID)) {
1263 				mode_t mode = va.va_mode & ~(S_ISUID | S_ISGID);
1264 				/* Clear all vattr fields except mode */
1265 				vattr_null(&va);
1266 				va.va_mode = mode;
1267 
1268 				/*
1269 				 * Ignore fuse_internal_setattr's return value,
1270 				 * because at this point the write operation has
1271 				 * already succeeded and we don't want to return
1272 				 * failing status for that.
1273 				 */
1274 				(void)fuse_internal_setattr(vp, &va, td, NULL);
1275 			}
1276 		}
1277 	}
1278 }
1279 
1280 #ifdef ZERO_PAD_INCOMPLETE_BUFS
1281 static int
1282 isbzero(void *buf, size_t len)
1283 {
1284 	int i;
1285 
1286 	for (i = 0; i < len; i++) {
1287 		if (((char *)buf)[i])
1288 			return (0);
1289 	}
1290 
1291 	return (1);
1292 }
1293 
1294 #endif
1295 
1296 void
1297 fuse_internal_init(void)
1298 {
1299 	fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK);
1300 	fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK);
1301 }
1302 
1303 void
1304 fuse_internal_destroy(void)
1305 {
1306 	counter_u64_free(fuse_lookup_cache_hits);
1307 	counter_u64_free(fuse_lookup_cache_misses);
1308 }
1309