xref: /freebsd/sys/fs/fuse/fuse_internal.c (revision fedf5b965fd623957a9b851418ba3301ac122e27)
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/param.h>
64 #include <sys/systm.h>
65 #include <sys/counter.h>
66 #include <sys/module.h>
67 #include <sys/errno.h>
68 #include <sys/kernel.h>
69 #include <sys/conf.h>
70 #include <sys/uio.h>
71 #include <sys/malloc.h>
72 #include <sys/queue.h>
73 #include <sys/lock.h>
74 #include <sys/mutex.h>
75 #include <sys/sdt.h>
76 #include <sys/sx.h>
77 #include <sys/proc.h>
78 #include <sys/mount.h>
79 #include <sys/vnode.h>
80 #include <sys/namei.h>
81 #include <sys/stat.h>
82 #include <sys/unistd.h>
83 #include <sys/filedesc.h>
84 #include <sys/file.h>
85 #include <sys/fcntl.h>
86 #include <sys/dirent.h>
87 #include <sys/bio.h>
88 #include <sys/buf.h>
89 #include <sys/sysctl.h>
90 #include <sys/priv.h>
91 
92 #include "fuse.h"
93 #include "fuse_file.h"
94 #include "fuse_internal.h"
95 #include "fuse_io.h"
96 #include "fuse_ipc.h"
97 #include "fuse_node.h"
98 #include "fuse_file.h"
99 
100 SDT_PROVIDER_DECLARE(fusefs);
101 /*
102  * Fuse trace probe:
103  * arg0: verbosity.  Higher numbers give more verbose messages
104  * arg1: Textual message
105  */
106 SDT_PROBE_DEFINE2(fusefs, , internal, trace, "int", "char*");
107 
108 #ifdef ZERO_PAD_INCOMPLETE_BUFS
109 static int isbzero(void *buf, size_t len);
110 
111 #endif
112 
113 counter_u64_t fuse_lookup_cache_hits;
114 counter_u64_t fuse_lookup_cache_misses;
115 
116 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
117     &fuse_lookup_cache_hits, "number of positive cache hits in lookup");
118 
119 SYSCTL_COUNTER_U64(_vfs_fusefs_stats, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
120     &fuse_lookup_cache_misses, "number of cache misses in lookup");
121 
122 int
123 fuse_internal_get_cached_vnode(struct mount* mp, ino_t ino, int flags,
124 	struct vnode **vpp)
125 {
126 	struct bintime now;
127 	struct thread *td = curthread;
128 	uint64_t nodeid = ino;
129 	int error;
130 
131 	*vpp = NULL;
132 
133 	error = vfs_hash_get(mp, fuse_vnode_hash(nodeid), flags, td, vpp,
134 	    fuse_vnode_cmp, &nodeid);
135 	if (error)
136 		return error;
137 	/*
138 	 * Check the entry cache timeout.  We have to do this within fusefs
139 	 * instead of by using cache_enter_time/cache_lookup because those
140 	 * routines are only intended to work with pathnames, not inodes
141 	 */
142 	if (*vpp != NULL) {
143 		getbinuptime(&now);
144 		if (bintime_cmp(&(VTOFUD(*vpp)->entry_cache_timeout), &now, >)){
145 			counter_u64_add(fuse_lookup_cache_hits, 1);
146 			return 0;
147 		} else {
148 			/* Entry cache timeout */
149 			counter_u64_add(fuse_lookup_cache_misses, 1);
150 			cache_purge(*vpp);
151 			vput(*vpp);
152 			*vpp = NULL;
153 		}
154 	}
155 	return 0;
156 }
157 
158 SDT_PROBE_DEFINE0(fusefs, , internal, access_vadmin);
159 /* Synchronously send a FUSE_ACCESS operation */
160 int
161 fuse_internal_access(struct vnode *vp,
162     accmode_t mode,
163     struct thread *td,
164     struct ucred *cred)
165 {
166 	int err = 0;
167 	uint32_t mask = F_OK;
168 	int dataflags;
169 	struct mount *mp;
170 	struct fuse_dispatcher fdi;
171 	struct fuse_access_in *fai;
172 	struct fuse_data *data;
173 
174 	mp = vnode_mount(vp);
175 
176 	data = fuse_get_mpdata(mp);
177 	dataflags = data->dataflags;
178 
179 	if (mode == 0)
180 		return 0;
181 
182 	if (mode & VMODIFY_PERMS && vfs_isrdonly(mp)) {
183 		switch (vp->v_type) {
184 		case VDIR:
185 			/* FALLTHROUGH */
186 		case VLNK:
187 			/* FALLTHROUGH */
188 		case VREG:
189 			return EROFS;
190 		default:
191 			break;
192 		}
193 	}
194 
195 	/* Unless explicitly permitted, deny everyone except the fs owner. */
196 	if (!(dataflags & FSESS_DAEMON_CAN_SPY)) {
197 		if (fuse_match_cred(data->daemoncred, cred))
198 			return EPERM;
199 	}
200 
201 	if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
202 		struct vattr va;
203 
204 		fuse_internal_getattr(vp, &va, cred, td);
205 		return vaccess(vp->v_type, va.va_mode, va.va_uid,
206 		    va.va_gid, mode, cred);
207 	}
208 
209 	if (mode & VADMIN) {
210 		/*
211 		 * The FUSE protocol doesn't have an equivalent of VADMIN, so
212 		 * it's a bug if we ever reach this point with that bit set.
213 		 */
214 		SDT_PROBE0(fusefs, , internal, access_vadmin);
215 	}
216 
217 	if (fsess_not_impl(mp, FUSE_ACCESS))
218 		return 0;
219 
220 	if ((mode & (VWRITE | VAPPEND)) != 0)
221 		mask |= W_OK;
222 	if ((mode & VREAD) != 0)
223 		mask |= R_OK;
224 	if ((mode & VEXEC) != 0)
225 		mask |= X_OK;
226 
227 	fdisp_init(&fdi, sizeof(*fai));
228 	fdisp_make_vp(&fdi, FUSE_ACCESS, vp, td, cred);
229 
230 	fai = fdi.indata;
231 	fai->mask = mask;
232 
233 	err = fdisp_wait_answ(&fdi);
234 	fdisp_destroy(&fdi);
235 
236 	if (err == ENOSYS) {
237 		fsess_set_notimpl(mp, FUSE_ACCESS);
238 		err = 0;
239 	}
240 	return err;
241 }
242 
243 /*
244  * Cache FUSE attributes from attr, in attribute cache associated with vnode
245  * 'vp'.  Optionally, if argument 'vap' is not NULL, store a copy of the
246  * converted attributes there as well.
247  *
248  * If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do
249  * return the result to the caller).
250  */
251 void
252 fuse_internal_cache_attrs(struct vnode *vp, struct fuse_attr *attr,
253 	uint64_t attr_valid, uint32_t attr_valid_nsec, struct vattr *vap,
254 	bool from_server)
255 {
256 	struct mount *mp;
257 	struct fuse_vnode_data *fvdat;
258 	struct fuse_data *data;
259 	struct vattr *vp_cache_at;
260 
261 	mp = vnode_mount(vp);
262 	fvdat = VTOFUD(vp);
263 	data = fuse_get_mpdata(mp);
264 
265 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs");
266 
267 	fuse_validity_2_bintime(attr_valid, attr_valid_nsec,
268 		&fvdat->attr_cache_timeout);
269 
270 	if (vnode_isreg(vp) &&
271 	    fvdat->cached_attrs.va_size != VNOVAL &&
272 	    attr->size != fvdat->cached_attrs.va_size)
273 	{
274 		if ( data->cache_mode == FUSE_CACHE_WB &&
275 		    fvdat->flag & FN_SIZECHANGE)
276 		{
277 			const char *msg;
278 
279 			/*
280 			 * The server changed the file's size even though we're
281 			 * using writeback cacheing and and we have outstanding
282 			 * dirty writes!  That's a server bug.
283 			 */
284 			if (fuse_libabi_geq(data, 7, 23)) {
285 				msg = "writeback cache incoherent!."
286 				    "To prevent data corruption, disable "
287 				    "the writeback cache according to your "
288 				    "FUSE server's documentation.";
289 			} else {
290 				msg = "writeback cache incoherent!."
291 				    "To prevent data corruption, disable "
292 				    "the writeback cache by setting "
293 				    "vfs.fusefs.data_cache_mode to 0 or 1.";
294 			}
295 			fuse_warn(data, FSESS_WARN_WB_CACHE_INCOHERENT, msg);
296 		}
297 		if (fuse_vnode_attr_cache_valid(vp) &&
298 		    data->cache_mode != FUSE_CACHE_UC)
299 		{
300 			/*
301 			 * The server changed the file's size even though we
302 			 * have it cached and our cache has not yet expired.
303 			 * That's a bug.
304 			 */
305 			fuse_warn(data, FSESS_WARN_CACHE_INCOHERENT,
306 			    "cache incoherent!  "
307 			    "To prevent "
308 			    "data corruption, disable the data cache "
309 			    "by mounting with -o direct_io, or as "
310 			    "directed otherwise by your FUSE server's "
311 			    "documentation.");
312 		}
313 	}
314 
315 	/* Fix our buffers if the filesize changed without us knowing */
316 	if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) {
317 		(void)fuse_vnode_setsize(vp, attr->size, from_server);
318 		fvdat->cached_attrs.va_size = attr->size;
319 	}
320 
321 	if (attr_valid > 0 || attr_valid_nsec > 0)
322 		vp_cache_at = &(fvdat->cached_attrs);
323 	else if (vap != NULL)
324 		vp_cache_at = vap;
325 	else
326 		return;
327 
328 	vattr_null(vp_cache_at);
329 	vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0];
330 	vp_cache_at->va_fileid = attr->ino;
331 	vp_cache_at->va_mode = attr->mode & ~S_IFMT;
332 	vp_cache_at->va_nlink     = attr->nlink;
333 	vp_cache_at->va_uid       = attr->uid;
334 	vp_cache_at->va_gid       = attr->gid;
335 	vp_cache_at->va_rdev      = attr->rdev;
336 	vp_cache_at->va_size      = attr->size;
337 	/* XXX on i386, seconds are truncated to 32 bits */
338 	vp_cache_at->va_atime.tv_sec  = attr->atime;
339 	vp_cache_at->va_atime.tv_nsec = attr->atimensec;
340 	vp_cache_at->va_mtime.tv_sec  = attr->mtime;
341 	vp_cache_at->va_mtime.tv_nsec = attr->mtimensec;
342 	vp_cache_at->va_ctime.tv_sec  = attr->ctime;
343 	vp_cache_at->va_ctime.tv_nsec = attr->ctimensec;
344 	if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0)
345 		vp_cache_at->va_blocksize = attr->blksize;
346 	else
347 		vp_cache_at->va_blocksize = PAGE_SIZE;
348 	vp_cache_at->va_type = IFTOVT(attr->mode);
349 	vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE;
350 	vp_cache_at->va_flags = 0;
351 
352 	if (vap != vp_cache_at && vap != NULL)
353 		memcpy(vap, vp_cache_at, sizeof(*vap));
354 }
355 
356 /* fsync */
357 
358 int
359 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio)
360 {
361 	if (tick->tk_aw_ohead.error == ENOSYS) {
362 		fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick));
363 	}
364 	return 0;
365 }
366 
367 int
368 fuse_internal_fsync(struct vnode *vp,
369     struct thread *td,
370     int waitfor,
371     bool datasync)
372 {
373 	struct fuse_fsync_in *ffsi = NULL;
374 	struct fuse_dispatcher fdi;
375 	struct fuse_filehandle *fufh;
376 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
377 	struct mount *mp = vnode_mount(vp);
378 	int op = FUSE_FSYNC;
379 	int err = 0;
380 
381 	if (fsess_not_impl(vnode_mount(vp),
382 	    (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
383 		return 0;
384 	}
385 	if (vnode_isdir(vp))
386 		op = FUSE_FSYNCDIR;
387 
388 	if (fsess_not_impl(mp, op))
389 		return 0;
390 
391 	fdisp_init(&fdi, sizeof(*ffsi));
392 	/*
393 	 * fsync every open file handle for this file, because we can't be sure
394 	 * which file handle the caller is really referring to.
395 	 */
396 	LIST_FOREACH(fufh, &fvdat->handles, next) {
397 		fdi.iosize = sizeof(*ffsi);
398 		if (ffsi == NULL)
399 			fdisp_make_vp(&fdi, op, vp, td, NULL);
400 		else
401 			fdisp_refresh_vp(&fdi, op, vp, td, NULL);
402 		ffsi = fdi.indata;
403 		ffsi->fh = fufh->fh_id;
404 		ffsi->fsync_flags = 0;
405 
406 		if (datasync)
407 			ffsi->fsync_flags = FUSE_FSYNC_FDATASYNC;
408 
409 		if (waitfor == MNT_WAIT) {
410 			err = fdisp_wait_answ(&fdi);
411 		} else {
412 			fuse_insert_callback(fdi.tick,
413 				fuse_internal_fsync_callback);
414 			fuse_insert_message(fdi.tick, false);
415 		}
416 		if (err == ENOSYS) {
417 			/* ENOSYS means "success, and don't call again" */
418 			fsess_set_notimpl(mp, op);
419 			err = 0;
420 			break;
421 		}
422 	}
423 	fdisp_destroy(&fdi);
424 
425 	return err;
426 }
427 
428 /* Asynchronous invalidation */
429 SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry,
430 	"struct vnode*", "struct fuse_notify_inval_entry_out*", "char*");
431 int
432 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio)
433 {
434 	struct fuse_notify_inval_entry_out fnieo;
435 	struct componentname cn;
436 	struct vnode *dvp, *vp;
437 	char name[PATH_MAX];
438 	int err;
439 
440 	if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0)
441 		return (err);
442 
443 	if (fnieo.namelen >= sizeof(name))
444 		return (EINVAL);
445 
446 	if ((err = uiomove(name, fnieo.namelen, uio)) != 0)
447 		return (err);
448 	name[fnieo.namelen] = '\0';
449 	/* fusefs does not cache "." or ".." entries */
450 	if (strncmp(name, ".", sizeof(".")) == 0 ||
451 	    strncmp(name, "..", sizeof("..")) == 0)
452 		return (0);
453 
454 	if (fnieo.parent == FUSE_ROOT_ID)
455 		err = VFS_ROOT(mp, LK_SHARED, &dvp);
456 	else
457 		err = fuse_internal_get_cached_vnode( mp, fnieo.parent,
458 			LK_SHARED, &dvp);
459 	SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name);
460 	/*
461 	 * If dvp is not in the cache, then it must've been reclaimed.  And
462 	 * since fuse_vnop_reclaim does a cache_purge, name's entry must've
463 	 * been invalidated already.  So we can safely return if dvp == NULL
464 	 */
465 	if (err != 0 || dvp == NULL)
466 		return (err);
467 	/*
468 	 * XXX we can't check dvp's generation because the FUSE invalidate
469 	 * entry message doesn't include it.  Worse case is that we invalidate
470 	 * an entry that didn't need to be invalidated.
471 	 */
472 
473 	cn.cn_nameiop = LOOKUP;
474 	cn.cn_flags = 0;	/* !MAKEENTRY means free cached entry */
475 	cn.cn_cred = curthread->td_ucred;
476 	cn.cn_lkflags = LK_SHARED;
477 	cn.cn_pnbuf = NULL;
478 	cn.cn_nameptr = name;
479 	cn.cn_namelen = fnieo.namelen;
480 	err = cache_lookup(dvp, &vp, &cn, NULL, NULL);
481 	MPASS(err == 0);
482 	fuse_vnode_clear_attr_cache(dvp);
483 	vput(dvp);
484 	return (0);
485 }
486 
487 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode,
488 	"struct vnode*", "struct fuse_notify_inval_inode_out *");
489 int
490 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio)
491 {
492 	struct fuse_notify_inval_inode_out fniio;
493 	struct vnode *vp;
494 	int err;
495 
496 	if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0)
497 		return (err);
498 
499 	if (fniio.ino == FUSE_ROOT_ID)
500 		err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp);
501 	else
502 		err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED,
503 			&vp);
504 	SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio);
505 	if (err != 0 || vp == NULL)
506 		return (err);
507 	/*
508 	 * XXX we can't check vp's generation because the FUSE invalidate
509 	 * entry message doesn't include it.  Worse case is that we invalidate
510 	 * an inode that didn't need to be invalidated.
511 	 */
512 
513 	/*
514 	 * Flush and invalidate buffers if off >= 0.  Technically we only need
515 	 * to flush and invalidate the range of offsets [off, off + len), but
516 	 * for simplicity's sake we do everything.
517 	 */
518 	if (fniio.off >= 0)
519 		fuse_io_invalbuf(vp, curthread);
520 	fuse_vnode_clear_attr_cache(vp);
521 	vput(vp);
522 	return (0);
523 }
524 
525 /* mknod */
526 int
527 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp,
528 	struct componentname *cnp, struct vattr *vap)
529 {
530 	struct fuse_data *data;
531 	struct fuse_mknod_in fmni;
532 	size_t insize;
533 
534 	data = fuse_get_mpdata(dvp->v_mount);
535 
536 	fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
537 	fmni.rdev = vap->va_rdev;
538 	if (fuse_libabi_geq(data, 7, 12)) {
539 		insize = sizeof(fmni);
540 		fmni.umask = curthread->td_proc->p_pd->pd_cmask;
541 		fmni.padding = 0;
542 	} else {
543 		insize = FUSE_COMPAT_MKNOD_IN_SIZE;
544 	}
545 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
546 	    insize, vap->va_type));
547 }
548 
549 /* readdir */
550 
551 int
552 fuse_internal_readdir(struct vnode *vp,
553     struct uio *uio,
554     struct fuse_filehandle *fufh,
555     struct fuse_iov *cookediov,
556     int *ncookies,
557     uint64_t *cookies)
558 {
559 	int err = 0;
560 	struct fuse_dispatcher fdi;
561 	struct fuse_read_in *fri = NULL;
562 
563 	if (uio_resid(uio) == 0)
564 		return 0;
565 	fdisp_init(&fdi, 0);
566 
567 	/*
568 	 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p
569 	 * I/O).
570 	 */
571 	while (uio_resid(uio) > 0) {
572 		fdi.iosize = sizeof(*fri);
573 		fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
574 		fri = fdi.indata;
575 		fri->fh = fufh->fh_id;
576 		fri->offset = uio_offset(uio);
577 		fri->size = MIN(uio->uio_resid,
578 		    fuse_get_mpdata(vp->v_mount)->max_read);
579 
580 		if ((err = fdisp_wait_answ(&fdi)))
581 			break;
582 		if ((err = fuse_internal_readdir_processdata(uio, fri->size,
583 			fdi.answ, fdi.iosize, cookediov, ncookies, &cookies)))
584 			break;
585 	}
586 
587 	fdisp_destroy(&fdi);
588 	return ((err == -1) ? 0 : err);
589 }
590 
591 /*
592  * Return -1 to indicate that this readdir is finished, 0 if it copied
593  * all the directory data read in and it may be possible to read more
594  * and greater than 0 for a failure.
595  */
596 int
597 fuse_internal_readdir_processdata(struct uio *uio,
598     size_t reqsize,
599     void *buf,
600     size_t bufsize,
601     struct fuse_iov *cookediov,
602     int *ncookies,
603     uint64_t **cookiesp)
604 {
605 	int err = 0;
606 	int oreclen;
607 	size_t freclen;
608 
609 	struct dirent *de;
610 	struct fuse_dirent *fudge;
611 	uint64_t *cookies;
612 
613 	cookies = *cookiesp;
614 	if (bufsize < FUSE_NAME_OFFSET)
615 		return -1;
616 	for (;;) {
617 		if (bufsize < FUSE_NAME_OFFSET) {
618 			err = -1;
619 			break;
620 		}
621 		fudge = (struct fuse_dirent *)buf;
622 		freclen = FUSE_DIRENT_SIZE(fudge);
623 
624 		if (bufsize < freclen) {
625 			/*
626 			 * This indicates a partial directory entry at the
627 			 * end of the directory data.
628 			 */
629 			err = -1;
630 			break;
631 		}
632 #ifdef ZERO_PAD_INCOMPLETE_BUFS
633 		if (isbzero(buf, FUSE_NAME_OFFSET)) {
634 			err = -1;
635 			break;
636 		}
637 #endif
638 
639 		if (!fudge->namelen || fudge->namelen > MAXNAMLEN) {
640 			err = EINVAL;
641 			break;
642 		}
643 		oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *)
644 					    &fudge->namelen);
645 
646 		if (oreclen > uio_resid(uio)) {
647 			/* Out of space for the dir so we are done. */
648 			err = -1;
649 			break;
650 		}
651 		fiov_adjust(cookediov, oreclen);
652 		bzero(cookediov->base, oreclen);
653 
654 		de = (struct dirent *)cookediov->base;
655 		de->d_fileno = fudge->ino;
656 		de->d_off = fudge->off;
657 		de->d_reclen = oreclen;
658 		de->d_type = fudge->type;
659 		de->d_namlen = fudge->namelen;
660 		memcpy((char *)cookediov->base + sizeof(struct dirent) -
661 		       MAXNAMLEN - 1,
662 		       (char *)buf + FUSE_NAME_OFFSET, fudge->namelen);
663 		dirent_terminate(de);
664 
665 		err = uiomove(cookediov->base, cookediov->len, uio);
666 		if (err)
667 			break;
668 		if (cookies != NULL) {
669 			if (*ncookies == 0) {
670 				err = -1;
671 				break;
672 			}
673 			*cookies = fudge->off;
674 			cookies++;
675 			(*ncookies)--;
676 		}
677 		buf = (char *)buf + freclen;
678 		bufsize -= freclen;
679 		uio_setoffset(uio, fudge->off);
680 	}
681 	*cookiesp = cookies;
682 
683 	return err;
684 }
685 
686 /* remove */
687 
688 int
689 fuse_internal_remove(struct vnode *dvp,
690     struct vnode *vp,
691     struct componentname *cnp,
692     enum fuse_opcode op)
693 {
694 	struct fuse_dispatcher fdi;
695 	nlink_t nlink;
696 	int err = 0;
697 
698 	fdisp_init(&fdi, cnp->cn_namelen + 1);
699 	fdisp_make_vp(&fdi, op, dvp, curthread, cnp->cn_cred);
700 
701 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
702 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
703 
704 	err = fdisp_wait_answ(&fdi);
705 	fdisp_destroy(&fdi);
706 
707 	if (err)
708 		return (err);
709 
710 	/*
711 	 * Access the cached nlink even if the attr cached has expired.  If
712 	 * it's inaccurate, the worst that will happen is:
713 	 * 1) We'll recycle the vnode even though the file has another link we
714 	 *    don't know about, costing a bit of cpu time, or
715 	 * 2) We won't recycle the vnode even though all of its links are gone.
716 	 *    It will linger around until vnlru reclaims it, costing a bit of
717 	 *    temporary memory.
718 	 */
719 	nlink = VTOFUD(vp)->cached_attrs.va_nlink--;
720 
721 	/*
722 	 * Purge the parent's attribute cache because the daemon
723 	 * should've updated its mtime and ctime.
724 	 */
725 	fuse_vnode_clear_attr_cache(dvp);
726 
727 	/* NB: nlink could be zero if it was never cached */
728 	if (nlink <= 1 || vnode_vtype(vp) == VDIR) {
729 		fuse_internal_vnode_disappear(vp);
730 	} else {
731 		cache_purge(vp);
732 		fuse_vnode_update(vp, FN_CTIMECHANGE);
733 	}
734 
735 	return err;
736 }
737 
738 /* rename */
739 
740 int
741 fuse_internal_rename(struct vnode *fdvp,
742     struct componentname *fcnp,
743     struct vnode *tdvp,
744     struct componentname *tcnp)
745 {
746 	struct fuse_dispatcher fdi;
747 	struct fuse_rename_in *fri;
748 	int err = 0;
749 
750 	fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2);
751 	fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, curthread, tcnp->cn_cred);
752 
753 	fri = fdi.indata;
754 	fri->newdir = VTOI(tdvp);
755 	memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr,
756 	    fcnp->cn_namelen);
757 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0';
758 	memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1,
759 	    tcnp->cn_nameptr, tcnp->cn_namelen);
760 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen +
761 	    tcnp->cn_namelen + 1] = '\0';
762 
763 	err = fdisp_wait_answ(&fdi);
764 	fdisp_destroy(&fdi);
765 	return err;
766 }
767 
768 /* strategy */
769 
770 /* entity creation */
771 
772 void
773 fuse_internal_newentry_makerequest(struct mount *mp,
774     uint64_t dnid,
775     struct componentname *cnp,
776     enum fuse_opcode op,
777     void *buf,
778     size_t bufsize,
779     struct fuse_dispatcher *fdip)
780 {
781 	fdip->iosize = bufsize + cnp->cn_namelen + 1;
782 
783 	fdisp_make(fdip, op, mp, dnid, curthread, cnp->cn_cred);
784 	memcpy(fdip->indata, buf, bufsize);
785 	memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen);
786 	((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0';
787 }
788 
789 int
790 fuse_internal_newentry_core(struct vnode *dvp,
791     struct vnode **vpp,
792     struct componentname *cnp,
793     __enum_uint8(vtype) vtyp,
794     struct fuse_dispatcher *fdip)
795 {
796 	int err = 0;
797 	struct fuse_entry_out *feo;
798 	struct mount *mp = vnode_mount(dvp);
799 
800 	if ((err = fdisp_wait_answ(fdip))) {
801 		return err;
802 	}
803 	feo = fdip->answ;
804 
805 	if ((err = fuse_internal_checkentry(feo, vtyp))) {
806 		return err;
807 	}
808 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp);
809 	if (err) {
810 		fuse_internal_forget_send(mp, curthread, cnp->cn_cred,
811 		    feo->nodeid, 1);
812 		return err;
813 	}
814 
815 	/*
816 	 * Purge the parent's attribute cache because the daemon should've
817 	 * updated its mtime and ctime
818 	 */
819 	fuse_vnode_clear_attr_cache(dvp);
820 
821 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
822 		feo->attr_valid_nsec, NULL, true);
823 
824 	return err;
825 }
826 
827 int
828 fuse_internal_newentry(struct vnode *dvp,
829     struct vnode **vpp,
830     struct componentname *cnp,
831     enum fuse_opcode op,
832     void *buf,
833     size_t bufsize,
834     __enum_uint8(vtype) vtype)
835 {
836 	int err;
837 	struct fuse_dispatcher fdi;
838 	struct mount *mp = vnode_mount(dvp);
839 
840 	fdisp_init(&fdi, 0);
841 	fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf,
842 	    bufsize, &fdi);
843 	err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi);
844 	fdisp_destroy(&fdi);
845 
846 	return err;
847 }
848 
849 /* entity destruction */
850 
851 int
852 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio)
853 {
854 	fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL,
855 	    ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1);
856 
857 	return 0;
858 }
859 
860 void
861 fuse_internal_forget_send(struct mount *mp,
862     struct thread *td,
863     struct ucred *cred,
864     uint64_t nodeid,
865     uint64_t nlookup)
866 {
867 
868 	struct fuse_dispatcher fdi;
869 	struct fuse_forget_in *ffi;
870 
871 	/*
872          * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu",
873          *         (long long unsigned) nodeid));
874          */
875 
876 	fdisp_init(&fdi, sizeof(*ffi));
877 	fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred);
878 
879 	ffi = fdi.indata;
880 	ffi->nlookup = nlookup;
881 
882 	fuse_insert_message(fdi.tick, false);
883 	fdisp_destroy(&fdi);
884 }
885 
886 /* Fetch the vnode's attributes from the daemon*/
887 int
888 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap,
889 	struct ucred *cred, struct thread *td)
890 {
891 	struct fuse_dispatcher fdi;
892 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
893 	struct fuse_getattr_in *fgai;
894 	struct fuse_attr_out *fao;
895 	off_t old_filesize = fvdat->cached_attrs.va_size;
896 	struct timespec old_atime = fvdat->cached_attrs.va_atime;
897 	struct timespec old_ctime = fvdat->cached_attrs.va_ctime;
898 	struct timespec old_mtime = fvdat->cached_attrs.va_mtime;
899 	__enum_uint8(vtype) vtyp;
900 	int err;
901 
902 	ASSERT_VOP_LOCKED(vp, __func__);
903 
904 	fdisp_init(&fdi, sizeof(*fgai));
905 	fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred);
906 	fgai = fdi.indata;
907 	/*
908 	 * We could look up a file handle and set it in fgai->fh, but that
909 	 * involves extra runtime work and I'm unaware of any file systems that
910 	 * care.
911 	 */
912 	fgai->getattr_flags = 0;
913 	if ((err = fdisp_wait_answ(&fdi))) {
914 		if (err == ENOENT)
915 			fuse_internal_vnode_disappear(vp);
916 		goto out;
917 	}
918 
919 	fao = (struct fuse_attr_out *)fdi.answ;
920 	vtyp = IFTOVT(fao->attr.mode);
921 	if (fvdat->flag & FN_SIZECHANGE)
922 		fao->attr.size = old_filesize;
923 	if (fvdat->flag & FN_ATIMECHANGE) {
924 		fao->attr.atime = old_atime.tv_sec;
925 		fao->attr.atimensec = old_atime.tv_nsec;
926 	}
927 	if (fvdat->flag & FN_CTIMECHANGE) {
928 		fao->attr.ctime = old_ctime.tv_sec;
929 		fao->attr.ctimensec = old_ctime.tv_nsec;
930 	}
931 	if (fvdat->flag & FN_MTIMECHANGE) {
932 		fao->attr.mtime = old_mtime.tv_sec;
933 		fao->attr.mtimensec = old_mtime.tv_nsec;
934 	}
935 	fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
936 		fao->attr_valid_nsec, vap, true);
937 	if (vtyp != vnode_vtype(vp)) {
938 		fuse_internal_vnode_disappear(vp);
939 		err = ENOENT;
940 	}
941 
942 out:
943 	fdisp_destroy(&fdi);
944 	return err;
945 }
946 
947 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */
948 int
949 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred,
950 	struct thread *td)
951 {
952 	struct vattr *attrs;
953 
954 	if ((attrs = VTOVA(vp)) != NULL) {
955 		*vap = *attrs;	/* struct copy */
956 		return 0;
957 	}
958 
959 	return fuse_internal_do_getattr(vp, vap, cred, td);
960 }
961 
962 void
963 fuse_internal_vnode_disappear(struct vnode *vp)
964 {
965 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
966 
967 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
968 	fvdat->flag |= FN_REVOKED;
969 	cache_purge(vp);
970 }
971 
972 /* fuse start/stop */
973 
974 SDT_PROBE_DEFINE2(fusefs, , internal, init_done,
975 	"struct fuse_data*", "struct fuse_init_out*");
976 int
977 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio)
978 {
979 	int err = 0;
980 	struct fuse_data *data = tick->tk_data;
981 	struct fuse_init_out *fiio = NULL;
982 
983 	if ((err = tick->tk_aw_ohead.error)) {
984 		goto out;
985 	}
986 	if ((err = fticket_pull(tick, uio))) {
987 		goto out;
988 	}
989 	fiio = fticket_resp(tick)->base;
990 
991 	data->fuse_libabi_major = fiio->major;
992 	data->fuse_libabi_minor = fiio->minor;
993 	if (!fuse_libabi_geq(data, 7, 4)) {
994 		/*
995 		 * With a little work we could support servers as old as 7.1.
996 		 * But there would be little payoff.
997 		 */
998 		SDT_PROBE2(fusefs, , internal, trace, 1,
999 			"userspace version too low");
1000 		err = EPROTONOSUPPORT;
1001 		goto out;
1002 	}
1003 
1004 	if (fuse_libabi_geq(data, 7, 5)) {
1005 		if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) ||
1006 		    fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) {
1007 			data->max_write = fiio->max_write;
1008 			if (fiio->flags & FUSE_ASYNC_READ)
1009 				data->dataflags |= FSESS_ASYNC_READ;
1010 			if (fiio->flags & FUSE_POSIX_LOCKS)
1011 				data->dataflags |= FSESS_POSIX_LOCKS;
1012 			if (fiio->flags & FUSE_EXPORT_SUPPORT)
1013 				data->dataflags |= FSESS_EXPORT_SUPPORT;
1014 			if (fiio->flags & FUSE_NO_OPEN_SUPPORT)
1015 				data->dataflags |= FSESS_NO_OPEN_SUPPORT;
1016 			if (fiio->flags & FUSE_NO_OPENDIR_SUPPORT)
1017 				data->dataflags |= FSESS_NO_OPENDIR_SUPPORT;
1018 			/*
1019 			 * Don't bother to check FUSE_BIG_WRITES, because it's
1020 			 * redundant with max_write
1021 			 */
1022 			/*
1023 			 * max_background and congestion_threshold are not
1024 			 * implemented
1025 			 */
1026 		} else {
1027 			err = EINVAL;
1028 		}
1029 	} else {
1030 		/* Old fixed values */
1031 		data->max_write = 4096;
1032 	}
1033 
1034 	if (fuse_libabi_geq(data, 7, 6))
1035 		data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf;
1036 
1037 	if (!fuse_libabi_geq(data, 7, 7))
1038 		fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
1039 
1040 	if (!fuse_libabi_geq(data, 7, 8)) {
1041 		fsess_set_notimpl(data->mp, FUSE_BMAP);
1042 		fsess_set_notimpl(data->mp, FUSE_DESTROY);
1043 	}
1044 
1045 	if (!fuse_libabi_geq(data, 7, 19)) {
1046 		fsess_set_notimpl(data->mp, FUSE_FALLOCATE);
1047 	}
1048 
1049 	if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 &&
1050 	    fiio->time_gran <= 1000000000)
1051 		data->time_gran = fiio->time_gran;
1052 	else
1053 		data->time_gran = 1;
1054 
1055 	if (!fuse_libabi_geq(data, 7, 23))
1056 		data->cache_mode = fuse_data_cache_mode;
1057 	else if (fiio->flags & FUSE_WRITEBACK_CACHE)
1058 		data->cache_mode = FUSE_CACHE_WB;
1059 	else
1060 		data->cache_mode = FUSE_CACHE_WT;
1061 
1062 	if (!fuse_libabi_geq(data, 7, 24))
1063 		fsess_set_notimpl(data->mp, FUSE_LSEEK);
1064 
1065 	if (!fuse_libabi_geq(data, 7, 28))
1066 		fsess_set_notimpl(data->mp, FUSE_COPY_FILE_RANGE);
1067 
1068 out:
1069 	if (err) {
1070 		fdata_set_dead(data);
1071 	}
1072 	FUSE_LOCK();
1073 	data->dataflags |= FSESS_INITED;
1074 	SDT_PROBE2(fusefs, , internal, init_done, data, fiio);
1075 	wakeup(&data->ticketer);
1076 	FUSE_UNLOCK();
1077 
1078 	return 0;
1079 }
1080 
1081 void
1082 fuse_internal_send_init(struct fuse_data *data, struct thread *td)
1083 {
1084 	struct fuse_init_in *fiii;
1085 	struct fuse_dispatcher fdi;
1086 
1087 	fdisp_init(&fdi, sizeof(*fiii));
1088 	fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL);
1089 	fiii = fdi.indata;
1090 	fiii->major = FUSE_KERNEL_VERSION;
1091 	fiii->minor = FUSE_KERNEL_MINOR_VERSION;
1092 	/*
1093 	 * fusefs currently reads ahead no more than one cache block at a time.
1094 	 * See fuse_read_biobackend
1095 	 */
1096 	fiii->max_readahead = maxbcachebuf;
1097 	/*
1098 	 * Unsupported features:
1099 	 * FUSE_FILE_OPS: No known FUSE server or client supports it
1100 	 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it
1101 	 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even
1102 	 *	when default ACLs are in use.
1103 	 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD
1104 	 *	doesn't have splice(2).
1105 	 * FUSE_FLOCK_LOCKS: not yet implemented
1106 	 * FUSE_HAS_IOCTL_DIR: not yet implemented
1107 	 * FUSE_AUTO_INVAL_DATA: not yet implemented
1108 	 * FUSE_DO_READDIRPLUS: not yet implemented
1109 	 * FUSE_READDIRPLUS_AUTO: not yet implemented
1110 	 * FUSE_ASYNC_DIO: not yet implemented
1111 	 * FUSE_PARALLEL_DIROPS: not yet implemented
1112 	 * FUSE_HANDLE_KILLPRIV: not yet implemented
1113 	 * FUSE_POSIX_ACL: not yet implemented
1114 	 * FUSE_ABORT_ERROR: not yet implemented
1115 	 * FUSE_CACHE_SYMLINKS: not yet implemented
1116 	 * FUSE_MAX_PAGES: not yet implemented
1117 	 */
1118 	fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT
1119 		| FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE
1120 		| FUSE_NO_OPEN_SUPPORT | FUSE_NO_OPENDIR_SUPPORT;
1121 
1122 	fuse_insert_callback(fdi.tick, fuse_internal_init_callback);
1123 	fuse_insert_message(fdi.tick, false);
1124 	fdisp_destroy(&fdi);
1125 }
1126 
1127 /*
1128  * Send a FUSE_SETATTR operation with no permissions checks.  If cred is NULL,
1129  * send the request with root credentials
1130  */
1131 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap,
1132 	struct thread *td, struct ucred *cred)
1133 {
1134 	struct fuse_vnode_data *fvdat;
1135 	struct fuse_dispatcher fdi;
1136 	struct fuse_setattr_in *fsai;
1137 	struct mount *mp;
1138 	pid_t pid = td->td_proc->p_pid;
1139 	struct fuse_data *data;
1140 	int err = 0;
1141 	__enum_uint8(vtype) vtyp;
1142 
1143 	ASSERT_VOP_ELOCKED(vp, __func__);
1144 
1145 	mp = vnode_mount(vp);
1146 	fvdat = VTOFUD(vp);
1147 	data = fuse_get_mpdata(mp);
1148 
1149 	fdisp_init(&fdi, sizeof(*fsai));
1150 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1151 	if (!cred) {
1152 		fdi.finh->uid = 0;
1153 		fdi.finh->gid = 0;
1154 	}
1155 	fsai = fdi.indata;
1156 	fsai->valid = 0;
1157 
1158 	if (vap->va_uid != (uid_t)VNOVAL) {
1159 		fsai->uid = vap->va_uid;
1160 		fsai->valid |= FATTR_UID;
1161 	}
1162 	if (vap->va_gid != (gid_t)VNOVAL) {
1163 		fsai->gid = vap->va_gid;
1164 		fsai->valid |= FATTR_GID;
1165 	}
1166 	if (vap->va_size != VNOVAL) {
1167 		struct fuse_filehandle *fufh = NULL;
1168 
1169 		/*Truncate to a new value. */
1170 		fsai->size = vap->va_size;
1171 		fsai->valid |= FATTR_SIZE;
1172 
1173 		fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1174 		if (fufh) {
1175 			fsai->fh = fufh->fh_id;
1176 			fsai->valid |= FATTR_FH;
1177 		}
1178 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1179 	}
1180 	if (vap->va_atime.tv_sec != VNOVAL) {
1181 		fsai->atime = vap->va_atime.tv_sec;
1182 		fsai->atimensec = vap->va_atime.tv_nsec;
1183 		fsai->valid |= FATTR_ATIME;
1184 		if (vap->va_vaflags & VA_UTIMES_NULL)
1185 			fsai->valid |= FATTR_ATIME_NOW;
1186 	} else if (fvdat->flag & FN_ATIMECHANGE) {
1187 		fsai->atime = fvdat->cached_attrs.va_atime.tv_sec;
1188 		fsai->atimensec = fvdat->cached_attrs.va_atime.tv_nsec;
1189 		fsai->valid |= FATTR_ATIME;
1190 	}
1191 	if (vap->va_mtime.tv_sec != VNOVAL) {
1192 		fsai->mtime = vap->va_mtime.tv_sec;
1193 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1194 		fsai->valid |= FATTR_MTIME;
1195 		if (vap->va_vaflags & VA_UTIMES_NULL)
1196 			fsai->valid |= FATTR_MTIME_NOW;
1197 	} else if (fvdat->flag & FN_MTIMECHANGE) {
1198 		fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec;
1199 		fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
1200 		fsai->valid |= FATTR_MTIME;
1201 	}
1202 	if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) {
1203 		fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec;
1204 		fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
1205 		fsai->valid |= FATTR_CTIME;
1206 	}
1207 	if (vap->va_mode != (mode_t)VNOVAL) {
1208 		fsai->mode = vap->va_mode & ALLPERMS;
1209 		fsai->valid |= FATTR_MODE;
1210 	}
1211 	if (!fsai->valid) {
1212 		goto out;
1213 	}
1214 
1215 	if ((err = fdisp_wait_answ(&fdi)))
1216 		goto out;
1217 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1218 
1219 	if (vnode_vtype(vp) != vtyp) {
1220 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1221 			SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! "
1222 				"vnode_vtype is VNON and vtype isn't.");
1223 		} else {
1224 			/*
1225 	                 * STALE vnode, ditch
1226 	                 *
1227 			 * The vnode has changed its type "behind our back".
1228 			 * This probably means that the file got deleted and
1229 			 * recreated on the server, with the same inode.
1230 			 * There's nothing really we can do, so let us just
1231 			 * return ENOENT.  After all, the entry must not have
1232 			 * existed in the recent past.  If the user tries
1233 			 * again, it will work.
1234 	                 */
1235 			fuse_internal_vnode_disappear(vp);
1236 			err = ENOENT;
1237 		}
1238 	}
1239 	if (err == 0) {
1240 		struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1241 		fuse_vnode_undirty_cached_timestamps(vp, true);
1242 		fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1243 			fao->attr_valid_nsec, NULL, false);
1244 		getnanouptime(&fvdat->last_local_modify);
1245 	}
1246 
1247 out:
1248 	fdisp_destroy(&fdi);
1249 	return err;
1250 }
1251 
1252 /*
1253  * FreeBSD clears the SUID and SGID bits on any write by a non-root user.
1254  */
1255 void
1256 fuse_internal_clear_suid_on_write(struct vnode *vp, struct ucred *cred,
1257 	struct thread *td)
1258 {
1259 	struct fuse_data *data;
1260 	struct mount *mp;
1261 	struct vattr va;
1262 	int dataflags;
1263 
1264 	mp = vnode_mount(vp);
1265 	data = fuse_get_mpdata(mp);
1266 	dataflags = data->dataflags;
1267 
1268 	ASSERT_VOP_LOCKED(vp, __func__);
1269 
1270 	if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
1271 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1272 			fuse_internal_getattr(vp, &va, cred, td);
1273 			if (va.va_mode & (S_ISUID | S_ISGID)) {
1274 				mode_t mode = va.va_mode & ~(S_ISUID | S_ISGID);
1275 				/* Clear all vattr fields except mode */
1276 				vattr_null(&va);
1277 				va.va_mode = mode;
1278 
1279 				/*
1280 				 * Ignore fuse_internal_setattr's return value,
1281 				 * because at this point the write operation has
1282 				 * already succeeded and we don't want to return
1283 				 * failing status for that.
1284 				 */
1285 				(void)fuse_internal_setattr(vp, &va, td, NULL);
1286 			}
1287 		}
1288 	}
1289 }
1290 
1291 #ifdef ZERO_PAD_INCOMPLETE_BUFS
1292 static int
1293 isbzero(void *buf, size_t len)
1294 {
1295 	int i;
1296 
1297 	for (i = 0; i < len; i++) {
1298 		if (((char *)buf)[i])
1299 			return (0);
1300 	}
1301 
1302 	return (1);
1303 }
1304 
1305 #endif
1306 
1307 void
1308 fuse_internal_init(void)
1309 {
1310 	fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK);
1311 	fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK);
1312 }
1313 
1314 void
1315 fuse_internal_destroy(void)
1316 {
1317 	counter_u64_free(fuse_lookup_cache_hits);
1318 	counter_u64_free(fuse_lookup_cache_misses);
1319 }
1320