xref: /src/sys/fs/fuse/fuse_internal.c (revision 7e68af7ce2c1b892954df415774fe59fd2f1b62f)
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
fuse_internal_get_cached_vnode(struct mount * mp,ino_t ino,int flags,struct vnode ** vpp)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
fuse_internal_access(struct vnode * vp,accmode_t mode,struct thread * td,struct ucred * cred)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
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)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_CACHED_ATTRS_LOCKED(vp);
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 	    fvdat->flag & FN_SIZECHANGE &&
273 	    attr->size != fvdat->cached_attrs.va_size)
274 	{
275 		if (data->cache_mode == FUSE_CACHE_WB)
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 	vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0];
329 	vp_cache_at->va_fileid = attr->ino;
330 	vp_cache_at->va_mode = attr->mode & ~S_IFMT;
331 	vp_cache_at->va_nlink     = attr->nlink;
332 	vp_cache_at->va_uid       = attr->uid;
333 	vp_cache_at->va_gid       = attr->gid;
334 	vp_cache_at->va_rdev      = attr->rdev;
335 	vp_cache_at->va_size      = attr->size;
336 	/* XXX on i386, seconds are truncated to 32 bits */
337 	vp_cache_at->va_atime.tv_sec  = attr->atime;
338 	vp_cache_at->va_atime.tv_nsec = attr->atimensec;
339 	vp_cache_at->va_mtime.tv_sec  = attr->mtime;
340 	vp_cache_at->va_mtime.tv_nsec = attr->mtimensec;
341 	vp_cache_at->va_ctime.tv_sec  = attr->ctime;
342 	vp_cache_at->va_ctime.tv_nsec = attr->ctimensec;
343 	if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0)
344 		vp_cache_at->va_blocksize = attr->blksize;
345 	else
346 		vp_cache_at->va_blocksize = PAGE_SIZE;
347 	vp_cache_at->va_type = IFTOVT(attr->mode);
348 	vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE;
349 	vp_cache_at->va_flags = 0;
350 
351 	if (vap != vp_cache_at && vap != NULL)
352 		memcpy(vap, vp_cache_at, sizeof(*vap));
353 }
354 
355 /* fsync */
356 
357 int
fuse_internal_fsync_callback(struct fuse_ticket * tick,struct uio * uio)358 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio)
359 {
360 	if (tick->tk_aw_ohead.error == ENOSYS) {
361 		fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick));
362 	}
363 	return 0;
364 }
365 
366 int
fuse_internal_fsync(struct vnode * vp,struct thread * td,int waitfor,bool datasync)367 fuse_internal_fsync(struct vnode *vp,
368     struct thread *td,
369     int waitfor,
370     bool datasync)
371 {
372 	struct fuse_fsync_in *ffsi = NULL;
373 	struct fuse_dispatcher fdi;
374 	struct fuse_filehandle *fufh;
375 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
376 	struct mount *mp = vnode_mount(vp);
377 	int op = FUSE_FSYNC;
378 	int err = 0;
379 
380 	if (fsess_not_impl(vnode_mount(vp),
381 	    (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
382 		return 0;
383 	}
384 	if (vnode_isdir(vp))
385 		op = FUSE_FSYNCDIR;
386 
387 	if (fsess_not_impl(mp, op))
388 		return 0;
389 
390 	fdisp_init(&fdi, sizeof(*ffsi));
391 	/*
392 	 * fsync every open file handle for this file, because we can't be sure
393 	 * which file handle the caller is really referring to.
394 	 */
395 	LIST_FOREACH(fufh, &fvdat->handles, next) {
396 		fdi.iosize = sizeof(*ffsi);
397 		if (ffsi == NULL)
398 			fdisp_make_vp(&fdi, op, vp, td, NULL);
399 		else
400 			fdisp_refresh_vp(&fdi, op, vp, td, NULL);
401 		ffsi = fdi.indata;
402 		ffsi->fh = fufh->fh_id;
403 		ffsi->fsync_flags = 0;
404 
405 		if (datasync)
406 			ffsi->fsync_flags = FUSE_FSYNC_FDATASYNC;
407 
408 		if (waitfor == MNT_WAIT) {
409 			err = fdisp_wait_answ(&fdi);
410 		} else {
411 			fuse_insert_callback(fdi.tick,
412 				fuse_internal_fsync_callback);
413 			fuse_insert_message(fdi.tick, false);
414 		}
415 		if (err == ENOSYS) {
416 			/* ENOSYS means "success, and don't call again" */
417 			fsess_set_notimpl(mp, op);
418 			err = 0;
419 			break;
420 		}
421 	}
422 	fdisp_destroy(&fdi);
423 
424 	return err;
425 }
426 
427 /* Asynchronous invalidation */
428 SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry,
429 	"struct vnode*", "struct fuse_notify_inval_entry_out*", "char*");
430 int
fuse_internal_invalidate_entry(struct mount * mp,struct uio * uio)431 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio)
432 {
433 	struct fuse_notify_inval_entry_out fnieo;
434 	struct componentname cn;
435 	struct vnode *dvp, *vp;
436 	char name[PATH_MAX];
437 	int err;
438 
439 	if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0)
440 		return (err);
441 
442 	if (fnieo.namelen >= sizeof(name))
443 		return (EINVAL);
444 
445 	if ((err = uiomove(name, fnieo.namelen, uio)) != 0)
446 		return (err);
447 	name[fnieo.namelen] = '\0';
448 	/* fusefs does not cache "." or ".." entries */
449 	if (strncmp(name, ".", sizeof(".")) == 0 ||
450 	    strncmp(name, "..", sizeof("..")) == 0)
451 		return (0);
452 
453 	if (fnieo.parent == FUSE_ROOT_ID)
454 		err = VFS_ROOT(mp, LK_SHARED, &dvp);
455 	else
456 		err = fuse_internal_get_cached_vnode( mp, fnieo.parent,
457 			LK_SHARED, &dvp);
458 	SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name);
459 	/*
460 	 * If dvp is not in the cache, then it must've been reclaimed.  And
461 	 * since fuse_vnop_reclaim does a cache_purge, name's entry must've
462 	 * been invalidated already.  So we can safely return if dvp == NULL
463 	 */
464 	if (err != 0 || dvp == NULL)
465 		return (err);
466 	/*
467 	 * XXX we can't check dvp's generation because the FUSE invalidate
468 	 * entry message doesn't include it.  Worse case is that we invalidate
469 	 * an entry that didn't need to be invalidated.
470 	 */
471 
472 	cn.cn_nameiop = LOOKUP;
473 	cn.cn_flags = 0;	/* !MAKEENTRY means free cached entry */
474 	cn.cn_cred = curthread->td_ucred;
475 	cn.cn_lkflags = LK_SHARED;
476 	cn.cn_pnbuf = NULL;
477 	cn.cn_nameptr = name;
478 	cn.cn_namelen = fnieo.namelen;
479 	err = cache_lookup(dvp, &vp, &cn, NULL, NULL);
480 	MPASS(err == 0);
481 	CACHED_ATTR_LOCK(dvp);
482 	fuse_vnode_clear_attr_cache(dvp);
483 	CACHED_ATTR_UNLOCK(dvp);
484 	vput(dvp);
485 	return (0);
486 }
487 
488 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode,
489 	"struct vnode*", "struct fuse_notify_inval_inode_out *");
490 int
fuse_internal_invalidate_inode(struct mount * mp,struct uio * uio)491 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio)
492 {
493 	struct fuse_notify_inval_inode_out fniio;
494 	struct vnode *vp;
495 	int err;
496 
497 	if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0)
498 		return (err);
499 
500 	if (fniio.ino == FUSE_ROOT_ID)
501 		err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp);
502 	else
503 		err = fuse_internal_get_cached_vnode(mp, fniio.ino,
504 		    LK_EXCLUSIVE, &vp);
505 	SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio);
506 	if (err != 0 || vp == NULL)
507 		return (err);
508 	/*
509 	 * XXX we can't check vp's generation because the FUSE invalidate
510 	 * entry message doesn't include it.  Worse case is that we invalidate
511 	 * an inode that didn't need to be invalidated.
512 	 */
513 
514 	/*
515 	 * Flush and invalidate buffers if off >= 0.  Technically we only need
516 	 * to flush and invalidate the range of offsets [off, off + len), but
517 	 * for simplicity's sake we do everything.
518 	 */
519 	if (fniio.off >= 0)
520 		fuse_io_invalbuf(vp, curthread);
521 	fuse_vnode_clear_attr_cache(vp);
522 	vput(vp);
523 	return (0);
524 }
525 
526 /* mknod */
527 int
fuse_internal_mknod(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct vattr * vap)528 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp,
529 	struct componentname *cnp, struct vattr *vap)
530 {
531 	struct fuse_data *data;
532 	struct fuse_mknod_in fmni;
533 	size_t insize;
534 
535 	data = fuse_get_mpdata(dvp->v_mount);
536 
537 	fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
538 	fmni.rdev = vap->va_rdev;
539 	if (fuse_libabi_geq(data, 7, 12)) {
540 		insize = sizeof(fmni);
541 		fmni.umask = curthread->td_proc->p_pd->pd_cmask;
542 		fmni.padding = 0;
543 	} else {
544 		insize = FUSE_COMPAT_MKNOD_IN_SIZE;
545 	}
546 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
547 	    insize, vap->va_type));
548 }
549 
550 /* readdir */
551 
552 int
fuse_internal_readdir(struct vnode * vp,struct uio * uio,struct fuse_filehandle * fufh,struct fuse_iov * cookediov,int * ncookies,uint64_t * cookies)553 fuse_internal_readdir(struct vnode *vp,
554     struct uio *uio,
555     struct fuse_filehandle *fufh,
556     struct fuse_iov *cookediov,
557     int *ncookies,
558     uint64_t *cookies)
559 {
560 	int err = 0;
561 	struct fuse_dispatcher fdi;
562 	struct fuse_read_in *fri = NULL;
563 
564 	if (uio_resid(uio) == 0)
565 		return 0;
566 	fdisp_init(&fdi, 0);
567 
568 	/*
569 	 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p
570 	 * I/O).
571 	 */
572 	while (uio_resid(uio) > 0) {
573 		fdi.iosize = sizeof(*fri);
574 		fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
575 		fri = fdi.indata;
576 		fri->fh = fufh->fh_id;
577 		fri->offset = uio_offset(uio);
578 		fri->size = MIN(uio->uio_resid,
579 		    fuse_get_mpdata(vp->v_mount)->max_read);
580 
581 		if ((err = fdisp_wait_answ(&fdi)))
582 			break;
583 		if ((err = fuse_internal_readdir_processdata(uio, fri->size,
584 			fdi.answ, fdi.iosize, cookediov, ncookies, &cookies)))
585 			break;
586 	}
587 
588 	fdisp_destroy(&fdi);
589 	return ((err == -1) ? 0 : err);
590 }
591 
592 /*
593  * Return -1 to indicate that this readdir is finished, 0 if it copied
594  * all the directory data read in and it may be possible to read more
595  * and greater than 0 for a failure.
596  */
597 int
fuse_internal_readdir_processdata(struct uio * uio,size_t reqsize,void * buf,size_t bufsize,struct fuse_iov * cookediov,int * ncookies,uint64_t ** cookiesp)598 fuse_internal_readdir_processdata(struct uio *uio,
599     size_t reqsize,
600     void *buf,
601     size_t bufsize,
602     struct fuse_iov *cookediov,
603     int *ncookies,
604     uint64_t **cookiesp)
605 {
606 	int err = 0;
607 	int oreclen;
608 	size_t freclen;
609 
610 	struct dirent *de;
611 	struct fuse_dirent *fudge;
612 	uint64_t *cookies;
613 
614 	cookies = *cookiesp;
615 	if (bufsize < FUSE_NAME_OFFSET)
616 		return -1;
617 	for (;;) {
618 		if (bufsize < FUSE_NAME_OFFSET) {
619 			err = -1;
620 			break;
621 		}
622 		fudge = (struct fuse_dirent *)buf;
623 		freclen = FUSE_DIRENT_SIZE(fudge);
624 
625 		if (bufsize < freclen) {
626 			/*
627 			 * This indicates a partial directory entry at the
628 			 * end of the directory data.
629 			 */
630 			err = -1;
631 			break;
632 		}
633 #ifdef ZERO_PAD_INCOMPLETE_BUFS
634 		if (isbzero(buf, FUSE_NAME_OFFSET)) {
635 			err = -1;
636 			break;
637 		}
638 #endif
639 
640 		if (!fudge->namelen || fudge->namelen > MAXNAMLEN) {
641 			err = EINVAL;
642 			break;
643 		}
644 		oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *)
645 					    &fudge->namelen);
646 
647 		if (oreclen > uio_resid(uio)) {
648 			/* Out of space for the dir so we are done. */
649 			err = -1;
650 			break;
651 		}
652 		fiov_adjust(cookediov, oreclen);
653 		bzero(cookediov->base, oreclen);
654 
655 		de = (struct dirent *)cookediov->base;
656 		de->d_fileno = fudge->ino;
657 		de->d_off = fudge->off;
658 		de->d_reclen = oreclen;
659 		de->d_type = fudge->type;
660 		de->d_namlen = fudge->namelen;
661 		memcpy((char *)cookediov->base + sizeof(struct dirent) -
662 		       MAXNAMLEN - 1,
663 		       (char *)buf + FUSE_NAME_OFFSET, fudge->namelen);
664 		dirent_terminate(de);
665 
666 		err = uiomove(cookediov->base, cookediov->len, uio);
667 		if (err)
668 			break;
669 		if (cookies != NULL) {
670 			if (*ncookies == 0) {
671 				err = -1;
672 				break;
673 			}
674 			*cookies = fudge->off;
675 			cookies++;
676 			(*ncookies)--;
677 		}
678 		buf = (char *)buf + freclen;
679 		bufsize -= freclen;
680 		uio_setoffset(uio, fudge->off);
681 	}
682 	*cookiesp = cookies;
683 
684 	return err;
685 }
686 
687 /* remove */
688 
689 int
fuse_internal_remove(struct vnode * dvp,struct vnode * vp,struct componentname * cnp,enum fuse_opcode op)690 fuse_internal_remove(struct vnode *dvp,
691     struct vnode *vp,
692     struct componentname *cnp,
693     enum fuse_opcode op)
694 {
695 	struct fuse_dispatcher fdi;
696 	nlink_t nlink;
697 	int err = 0;
698 
699 	ASSERT_CACHED_ATTRS_LOCKED(vp);
700 
701 	fdisp_init(&fdi, cnp->cn_namelen + 1);
702 	fdisp_make_vp(&fdi, op, dvp, curthread, cnp->cn_cred);
703 
704 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
705 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
706 
707 	err = fdisp_wait_answ(&fdi);
708 	fdisp_destroy(&fdi);
709 
710 	if (err)
711 		return (err);
712 
713 	/*
714 	 * Access the cached nlink even if the attr cached has expired.  If
715 	 * it's inaccurate, the worst that will happen is:
716 	 * 1) We'll recycle the vnode even though the file has another link we
717 	 *    don't know about, costing a bit of cpu time, or
718 	 * 2) We won't recycle the vnode even though all of its links are gone.
719 	 *    It will linger around until vnlru reclaims it, costing a bit of
720 	 *    temporary memory.
721 	 */
722 	nlink = VTOFUD(vp)->cached_attrs.va_nlink--;
723 
724 	/*
725 	 * Purge the parent's attribute cache because the daemon
726 	 * should've updated its mtime and ctime.
727 	 */
728 	fuse_vnode_clear_attr_cache(dvp);
729 
730 	/* NB: nlink could be zero if it was never cached */
731 	if (nlink <= 1 || vnode_vtype(vp) == VDIR) {
732 		fuse_internal_vnode_disappear(vp);
733 	} else {
734 		cache_purge(vp);
735 		fuse_vnode_update(vp, FN_CTIMECHANGE);
736 	}
737 
738 	return err;
739 }
740 
741 /* rename */
742 
743 int
fuse_internal_rename(struct vnode * fdvp,struct componentname * fcnp,struct vnode * tdvp,struct componentname * tcnp)744 fuse_internal_rename(struct vnode *fdvp,
745     struct componentname *fcnp,
746     struct vnode *tdvp,
747     struct componentname *tcnp)
748 {
749 	struct fuse_dispatcher fdi;
750 	struct fuse_rename_in *fri;
751 	int err = 0;
752 
753 	fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2);
754 	fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, curthread, tcnp->cn_cred);
755 
756 	fri = fdi.indata;
757 	fri->newdir = VTOI(tdvp);
758 	memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr,
759 	    fcnp->cn_namelen);
760 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0';
761 	memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1,
762 	    tcnp->cn_nameptr, tcnp->cn_namelen);
763 	((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen +
764 	    tcnp->cn_namelen + 1] = '\0';
765 
766 	err = fdisp_wait_answ(&fdi);
767 	fdisp_destroy(&fdi);
768 	return err;
769 }
770 
771 /* strategy */
772 
773 /* entity creation */
774 
775 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)776 fuse_internal_newentry_makerequest(struct mount *mp,
777     uint64_t dnid,
778     struct componentname *cnp,
779     enum fuse_opcode op,
780     void *buf,
781     size_t bufsize,
782     struct fuse_dispatcher *fdip)
783 {
784 	fdip->iosize = bufsize + cnp->cn_namelen + 1;
785 
786 	fdisp_make(fdip, op, mp, dnid, curthread, cnp->cn_cred);
787 	memcpy(fdip->indata, buf, bufsize);
788 	memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen);
789 	((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0';
790 }
791 
792 int
fuse_internal_newentry_core(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,__enum_uint8 (vtype)vtyp,struct fuse_dispatcher * fdip)793 fuse_internal_newentry_core(struct vnode *dvp,
794     struct vnode **vpp,
795     struct componentname *cnp,
796     __enum_uint8(vtype) vtyp,
797     struct fuse_dispatcher *fdip)
798 {
799 	int err = 0;
800 	struct fuse_entry_out *feo;
801 	struct mount *mp = vnode_mount(dvp);
802 
803 	if ((err = fdisp_wait_answ(fdip))) {
804 		return err;
805 	}
806 	feo = fdip->answ;
807 
808 	if ((err = fuse_internal_checkentry(feo, vtyp))) {
809 		return err;
810 	}
811 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp);
812 	if (err) {
813 		fuse_internal_forget_send(mp, curthread, cnp->cn_cred,
814 		    feo->nodeid, 1);
815 		return err;
816 	}
817 
818 	/*
819 	 * Purge the parent's attribute cache because the daemon should've
820 	 * updated its mtime and ctime
821 	 */
822 	fuse_vnode_clear_attr_cache(dvp);
823 
824 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
825 		feo->attr_valid_nsec, NULL, true);
826 
827 	return err;
828 }
829 
830 int
fuse_internal_newentry(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,enum fuse_opcode op,void * buf,size_t bufsize,__enum_uint8 (vtype)vtype)831 fuse_internal_newentry(struct vnode *dvp,
832     struct vnode **vpp,
833     struct componentname *cnp,
834     enum fuse_opcode op,
835     void *buf,
836     size_t bufsize,
837     __enum_uint8(vtype) vtype)
838 {
839 	int err;
840 	struct fuse_dispatcher fdi;
841 	struct mount *mp = vnode_mount(dvp);
842 
843 	fdisp_init(&fdi, 0);
844 	fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf,
845 	    bufsize, &fdi);
846 	err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi);
847 	fdisp_destroy(&fdi);
848 
849 	return err;
850 }
851 
852 /* entity destruction */
853 
854 int
fuse_internal_forget_callback(struct fuse_ticket * ftick,struct uio * uio)855 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio)
856 {
857 	fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL,
858 	    ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1);
859 
860 	return 0;
861 }
862 
863 void
fuse_internal_forget_send(struct mount * mp,struct thread * td,struct ucred * cred,uint64_t nodeid,uint64_t nlookup)864 fuse_internal_forget_send(struct mount *mp,
865     struct thread *td,
866     struct ucred *cred,
867     uint64_t nodeid,
868     uint64_t nlookup)
869 {
870 
871 	struct fuse_dispatcher fdi;
872 	struct fuse_forget_in *ffi;
873 
874 	/*
875          * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu",
876          *         (long long unsigned) nodeid));
877          */
878 
879 	fdisp_init(&fdi, sizeof(*ffi));
880 	fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred);
881 
882 	ffi = fdi.indata;
883 	ffi->nlookup = nlookup;
884 
885 	fuse_insert_message(fdi.tick, false);
886 	fdisp_destroy(&fdi);
887 }
888 
889 /* Fetch the vnode's attributes from the daemon*/
890 int
fuse_internal_do_getattr(struct vnode * vp,struct vattr * vap,struct ucred * cred,struct thread * td)891 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap,
892 	struct ucred *cred, struct thread *td)
893 {
894 	struct fuse_dispatcher fdi;
895 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
896 	struct fuse_getattr_in *fgai;
897 	struct fuse_attr_out *fao;
898 	__enum_uint8(vtype) vtyp;
899 	int err;
900 
901 	fdisp_init(&fdi, sizeof(*fgai));
902 	fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred);
903 	fgai = fdi.indata;
904 	/*
905 	 * We could look up a file handle and set it in fgai->fh, but that
906 	 * involves extra runtime work and I'm unaware of any file systems that
907 	 * care.
908 	 */
909 	fgai->getattr_flags = 0;
910 	if ((err = fdisp_wait_answ(&fdi))) {
911 		if (err == ENOENT)
912 			fuse_internal_vnode_disappear(vp);
913 		goto out;
914 	}
915 
916 	fao = (struct fuse_attr_out *)fdi.answ;
917 	vtyp = IFTOVT(fao->attr.mode);
918 
919 	CACHED_ATTR_LOCK(vp);
920 	if (fvdat->flag & FN_SIZECHANGE)
921 		fao->attr.size = fvdat->cached_attrs.va_size;
922 	if (fvdat->flag & FN_ATIMECHANGE) {
923 		fao->attr.atime = fvdat->cached_attrs.va_atime.tv_sec;
924 		fao->attr.atimensec = fvdat->cached_attrs.va_atime.tv_nsec;
925 	}
926 	if (fvdat->flag & FN_CTIMECHANGE) {
927 		fao->attr.ctime = fvdat->cached_attrs.va_ctime.tv_sec;
928 		fao->attr.ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
929 	}
930 	if (fvdat->flag & FN_MTIMECHANGE) {
931 		fao->attr.mtime = fvdat->cached_attrs.va_mtime.tv_sec;
932 		fao->attr.mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
933 	}
934 
935 	fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
936 		fao->attr_valid_nsec, vap, true);
937 
938 	CACHED_ATTR_UNLOCK(vp);
939 	if (vtyp != vnode_vtype(vp)) {
940 		fuse_internal_vnode_disappear(vp);
941 		err = ENOENT;
942 	}
943 
944 out:
945 	fdisp_destroy(&fdi);
946 	return err;
947 }
948 
949 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */
950 int
fuse_internal_getattr(struct vnode * vp,struct vattr * vap,struct ucred * cred,struct thread * td)951 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred,
952 	struct thread *td)
953 {
954 	struct vattr *attrs;
955 
956 	CACHED_ATTR_LOCK(vp);
957 	if ((attrs = VTOVA(vp)) != NULL) {
958 		*vap = *attrs;	/* struct copy */
959 		CACHED_ATTR_UNLOCK(vp);
960 		return 0;
961 	} else
962 		CACHED_ATTR_UNLOCK(vp);
963 
964 	return fuse_internal_do_getattr(vp, vap, cred, td);
965 }
966 
967 void
fuse_internal_vnode_disappear(struct vnode * vp)968 fuse_internal_vnode_disappear(struct vnode *vp)
969 {
970 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
971 
972 	ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
973 	fvdat->flag |= FN_REVOKED;
974 	cache_purge(vp);
975 }
976 
977 /* fuse start/stop */
978 
979 SDT_PROBE_DEFINE2(fusefs, , internal, init_done,
980 	"struct fuse_data*", "struct fuse_init_out*");
981 int
fuse_internal_init_callback(struct fuse_ticket * tick,struct uio * uio)982 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio)
983 {
984 	int err = 0;
985 	struct fuse_data *data = tick->tk_data;
986 	struct fuse_init_out *fiio = NULL;
987 
988 	if (fdata_get_dead(data))
989 		goto out;
990 
991 	if ((err = tick->tk_aw_ohead.error)) {
992 		goto out;
993 	}
994 	if ((err = fticket_pull(tick, uio))) {
995 		goto out;
996 	}
997 	fiio = fticket_resp(tick)->base;
998 
999 	data->fuse_libabi_major = fiio->major;
1000 	data->fuse_libabi_minor = fiio->minor;
1001 	if (!fuse_libabi_geq(data, 7, 4)) {
1002 		/*
1003 		 * With a little work we could support servers as old as 7.1.
1004 		 * But there would be little payoff.
1005 		 */
1006 		SDT_PROBE2(fusefs, , internal, trace, 1,
1007 			"userspace version too low");
1008 		err = EPROTONOSUPPORT;
1009 		goto out;
1010 	}
1011 
1012 	if (fuse_libabi_geq(data, 7, 5)) {
1013 		if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) ||
1014 		    fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) {
1015 			data->max_write = fiio->max_write;
1016 			if (fiio->flags & FUSE_ASYNC_READ)
1017 				data->dataflags |= FSESS_ASYNC_READ;
1018 			if (fiio->flags & FUSE_POSIX_LOCKS)
1019 				data->dataflags |= FSESS_POSIX_LOCKS;
1020 			if (fiio->flags & FUSE_EXPORT_SUPPORT)
1021 				data->dataflags |= FSESS_EXPORT_SUPPORT;
1022 			/*
1023 			 * Don't bother to check FUSE_BIG_WRITES, because it's
1024 			 * redundant with max_write
1025 			 */
1026 			/*
1027 			 * max_background and congestion_threshold are not
1028 			 * implemented
1029 			 */
1030 		} else {
1031 			err = EINVAL;
1032 		}
1033 	} else {
1034 		/* Old fixed values */
1035 		data->max_write = 4096;
1036 	}
1037 
1038 	if (fuse_libabi_geq(data, 7, 6))
1039 		data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf;
1040 
1041 	if (!fuse_libabi_geq(data, 7, 7))
1042 		fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
1043 
1044 	if (!fuse_libabi_geq(data, 7, 8)) {
1045 		fsess_set_notimpl(data->mp, FUSE_BMAP);
1046 		fsess_set_notimpl(data->mp, FUSE_DESTROY);
1047 	}
1048 
1049 	if (!fuse_libabi_geq(data, 7, 19)) {
1050 		fsess_set_notimpl(data->mp, FUSE_FALLOCATE);
1051 	}
1052 
1053 	if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 &&
1054 	    fiio->time_gran <= 1000000000)
1055 		data->time_gran = fiio->time_gran;
1056 	else
1057 		data->time_gran = 1;
1058 
1059 	if (!fuse_libabi_geq(data, 7, 23))
1060 		data->cache_mode = fuse_data_cache_mode;
1061 	else if (fiio->flags & FUSE_WRITEBACK_CACHE)
1062 		data->cache_mode = FUSE_CACHE_WB;
1063 	else
1064 		data->cache_mode = FUSE_CACHE_WT;
1065 
1066 	if (!fuse_libabi_geq(data, 7, 24))
1067 		fsess_set_notimpl(data->mp, FUSE_LSEEK);
1068 
1069 	if (!fuse_libabi_geq(data, 7, 28))
1070 		fsess_set_notimpl(data->mp, FUSE_COPY_FILE_RANGE);
1071 
1072 	if (fuse_libabi_geq(data, 7, 33) && (fiio->flags & FUSE_SETXATTR_EXT))
1073 		data->dataflags |= FSESS_SETXATTR_EXT;
1074 out:
1075 	if (err) {
1076 		fdata_set_dead(data);
1077 	}
1078 	FUSE_LOCK();
1079 	data->dataflags |= FSESS_INITED;
1080 	SDT_PROBE2(fusefs, , internal, init_done, data, fiio);
1081 	wakeup(&data->ticketer);
1082 	FUSE_UNLOCK();
1083 
1084 	return 0;
1085 }
1086 
1087 void
fuse_internal_send_init(struct fuse_data * data,struct thread * td)1088 fuse_internal_send_init(struct fuse_data *data, struct thread *td)
1089 {
1090 	struct fuse_init_in *fiii;
1091 	struct fuse_dispatcher fdi;
1092 
1093 	fdisp_init(&fdi, sizeof(*fiii));
1094 	fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL);
1095 	fiii = fdi.indata;
1096 	fiii->major = FUSE_KERNEL_VERSION;
1097 	fiii->minor = FUSE_KERNEL_MINOR_VERSION;
1098 	/*
1099 	 * fusefs currently reads ahead no more than one cache block at a time.
1100 	 * See fuse_read_biobackend
1101 	 */
1102 	fiii->max_readahead = maxbcachebuf;
1103 	/*
1104 	 * Unsupported features:
1105 	 * FUSE_FILE_OPS: No known FUSE server or client supports it
1106 	 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it
1107 	 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even
1108 	 *	when default ACLs are in use.
1109 	 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD
1110 	 *	doesn't have splice(2).
1111 	 * FUSE_FLOCK_LOCKS: not yet implemented
1112 	 * FUSE_AUTO_INVAL_DATA: not yet implemented
1113 	 * FUSE_DO_READDIRPLUS: not yet implemented
1114 	 * FUSE_READDIRPLUS_AUTO: not yet implemented
1115 	 * FUSE_ASYNC_DIO: not yet implemented
1116 	 * FUSE_PARALLEL_DIROPS: not yet implemented
1117 	 * FUSE_HANDLE_KILLPRIV: not yet implemented
1118 	 * FUSE_POSIX_ACL: not yet implemented
1119 	 * FUSE_ABORT_ERROR: not yet implemented
1120 	 * FUSE_CACHE_SYMLINKS: not yet implemented
1121 	 * FUSE_MAX_PAGES: not yet implemented
1122 	 */
1123 	fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT
1124 		| FUSE_BIG_WRITES | FUSE_HAS_IOCTL_DIR | FUSE_WRITEBACK_CACHE
1125 		| FUSE_NO_OPEN_SUPPORT | FUSE_NO_OPENDIR_SUPPORT
1126 		| FUSE_SETXATTR_EXT;
1127 
1128 	fuse_insert_callback(fdi.tick, fuse_internal_init_callback);
1129 	fuse_insert_message(fdi.tick, false);
1130 	fdisp_destroy(&fdi);
1131 }
1132 
1133 /*
1134  * Send a FUSE_SETATTR operation with no permissions checks.  If cred is NULL,
1135  * send the request with root credentials
1136  */
fuse_internal_setattr(struct vnode * vp,struct vattr * vap,struct thread * td,struct ucred * cred)1137 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap,
1138 	struct thread *td, struct ucred *cred)
1139 {
1140 	struct fuse_vnode_data *fvdat;
1141 	struct fuse_dispatcher fdi;
1142 	struct fuse_setattr_in *fsai;
1143 	struct mount *mp;
1144 	pid_t pid = td->td_proc->p_pid;
1145 	struct fuse_data *data;
1146 	int err = 0;
1147 	__enum_uint8(vtype) vtyp;
1148 
1149 	ASSERT_CACHED_ATTRS_LOCKED(vp);
1150 
1151 	mp = vnode_mount(vp);
1152 	fvdat = VTOFUD(vp);
1153 	data = fuse_get_mpdata(mp);
1154 
1155 	fdisp_init(&fdi, sizeof(*fsai));
1156 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1157 	if (!cred) {
1158 		fdi.finh->uid = 0;
1159 		fdi.finh->gid = 0;
1160 	}
1161 	fsai = fdi.indata;
1162 	fsai->valid = 0;
1163 
1164 	if (vap->va_uid != (uid_t)VNOVAL) {
1165 		fsai->uid = vap->va_uid;
1166 		fsai->valid |= FATTR_UID;
1167 	}
1168 	if (vap->va_gid != (gid_t)VNOVAL) {
1169 		fsai->gid = vap->va_gid;
1170 		fsai->valid |= FATTR_GID;
1171 	}
1172 	if (vap->va_size != VNOVAL) {
1173 		struct fuse_filehandle *fufh = NULL;
1174 
1175 		/*Truncate to a new value. */
1176 		fsai->size = vap->va_size;
1177 		fsai->valid |= FATTR_SIZE;
1178 
1179 		fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1180 		if (fufh) {
1181 			fsai->fh = fufh->fh_id;
1182 			fsai->valid |= FATTR_FH;
1183 		}
1184 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1185 	}
1186 	if (vap->va_atime.tv_sec != VNOVAL) {
1187 		fsai->atime = vap->va_atime.tv_sec;
1188 		fsai->atimensec = vap->va_atime.tv_nsec;
1189 		fsai->valid |= FATTR_ATIME;
1190 		if (vap->va_vaflags & VA_UTIMES_NULL)
1191 			fsai->valid |= FATTR_ATIME_NOW;
1192 	} else if (fvdat->flag & FN_ATIMECHANGE) {
1193 		fsai->atime = fvdat->cached_attrs.va_atime.tv_sec;
1194 		fsai->atimensec = fvdat->cached_attrs.va_atime.tv_nsec;
1195 		fsai->valid |= FATTR_ATIME;
1196 	}
1197 	if (vap->va_mtime.tv_sec != VNOVAL) {
1198 		fsai->mtime = vap->va_mtime.tv_sec;
1199 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1200 		fsai->valid |= FATTR_MTIME;
1201 		if (vap->va_vaflags & VA_UTIMES_NULL)
1202 			fsai->valid |= FATTR_MTIME_NOW;
1203 	} else if (fvdat->flag & FN_MTIMECHANGE) {
1204 		fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec;
1205 		fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
1206 		fsai->valid |= FATTR_MTIME;
1207 	}
1208 	if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) {
1209 		fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec;
1210 		fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
1211 		fsai->valid |= FATTR_CTIME;
1212 	}
1213 	if (vap->va_mode != (mode_t)VNOVAL) {
1214 		fsai->mode = vap->va_mode & ALLPERMS;
1215 		fsai->valid |= FATTR_MODE;
1216 	}
1217 	if (!fsai->valid) {
1218 		goto out;
1219 	}
1220 
1221 	if ((err = fdisp_wait_answ(&fdi)))
1222 		goto out;
1223 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1224 
1225 	if (vnode_vtype(vp) != vtyp) {
1226 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1227 			SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! "
1228 				"vnode_vtype is VNON and vtype isn't.");
1229 		} else {
1230 			/*
1231 	                 * STALE vnode, ditch
1232 	                 *
1233 			 * The vnode has changed its type "behind our back".
1234 			 * This probably means that the file got deleted and
1235 			 * recreated on the server, with the same inode.
1236 			 * There's nothing really we can do, so let us just
1237 			 * return ENOENT.  After all, the entry must not have
1238 			 * existed in the recent past.  If the user tries
1239 			 * again, it will work.
1240 	                 */
1241 			fuse_internal_vnode_disappear(vp);
1242 			err = ENOENT;
1243 		}
1244 	}
1245 	if (err == 0) {
1246 		struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1247 		fuse_vnode_undirty_cached_timestamps(vp, true);
1248 		fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1249 			fao->attr_valid_nsec, NULL, false);
1250 		getnanouptime(&fvdat->last_local_modify);
1251 	}
1252 
1253 out:
1254 	fdisp_destroy(&fdi);
1255 	return err;
1256 }
1257 
1258 /*
1259  * FreeBSD clears the SUID and SGID bits on any write by a non-root user.
1260  */
1261 void
fuse_internal_clear_suid_on_write(struct vnode * vp,struct ucred * cred,struct thread * td)1262 fuse_internal_clear_suid_on_write(struct vnode *vp, struct ucred *cred,
1263 	struct thread *td)
1264 {
1265 	struct fuse_data *data;
1266 	struct mount *mp;
1267 	struct vattr va;
1268 	int dataflags;
1269 
1270 	mp = vnode_mount(vp);
1271 	data = fuse_get_mpdata(mp);
1272 	dataflags = data->dataflags;
1273 
1274 	ASSERT_VOP_LOCKED(vp, __func__);
1275 
1276 	if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
1277 		if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1278 			fuse_internal_getattr(vp, &va, cred, td);
1279 			if (va.va_mode & (S_ISUID | S_ISGID)) {
1280 				mode_t mode = va.va_mode & ~(S_ISUID | S_ISGID);
1281 				/* Clear all vattr fields except mode */
1282 				vattr_null(&va);
1283 				va.va_mode = mode;
1284 
1285 				/*
1286 				 * Ignore fuse_internal_setattr's return value,
1287 				 * because at this point the write operation has
1288 				 * already succeeded and we don't want to return
1289 				 * failing status for that.
1290 				 */
1291 				(void)fuse_internal_setattr(vp, &va, td, NULL);
1292 			}
1293 		}
1294 	}
1295 }
1296 
1297 #ifdef ZERO_PAD_INCOMPLETE_BUFS
1298 static int
isbzero(void * buf,size_t len)1299 isbzero(void *buf, size_t len)
1300 {
1301 	int i;
1302 
1303 	for (i = 0; i < len; i++) {
1304 		if (((char *)buf)[i])
1305 			return (0);
1306 	}
1307 
1308 	return (1);
1309 }
1310 
1311 #endif
1312 
1313 void
fuse_internal_init(void)1314 fuse_internal_init(void)
1315 {
1316 	fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK);
1317 	fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK);
1318 }
1319 
1320 void
fuse_internal_destroy(void)1321 fuse_internal_destroy(void)
1322 {
1323 	counter_u64_free(fuse_lookup_cache_hits);
1324 	counter_u64_free(fuse_lookup_cache_misses);
1325 }
1326