2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
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
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.
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.
33 * Copyright (C) 2005 Csaba Henk.
34 * All rights reserved.
36 * Copyright (c) 2019 The FreeBSD Foundation
38 * Portions of this software were developed by BFF Storage Systems, LLC under
39 * sponsorship from the FreeBSD Foundation.
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
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.
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
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
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>
74 #include <sys/malloc.h>
75 #include <sys/queue.h>
77 #include <sys/mutex.h>
81 #include <sys/mount.h>
82 #include <sys/vnode.h>
83 #include <sys/namei.h>
85 #include <sys/unistd.h>
86 #include <sys/filedesc.h>
88 #include <sys/fcntl.h>
89 #include <sys/dirent.h>
92 #include <sys/sysctl.h>
96 #include "fuse_file.h"
97 #include "fuse_internal.h"
100 #include "fuse_node.h"
101 #include "fuse_file.h"
103 SDT_PROVIDER_DECLARE(fusefs);
106 * arg0: verbosity. Higher numbers give more verbose messages
107 * arg1: Textual message
109 SDT_PROBE_DEFINE2(fusefs, , internal, trace, "int", "char*");
111 #ifdef ZERO_PAD_INCOMPLETE_BUFS
112 static int isbzero(void *buf, size_t len);
116 counter_u64_t fuse_lookup_cache_hits;
117 counter_u64_t fuse_lookup_cache_misses;
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");
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");
126 fuse_internal_get_cached_vnode(struct mount* mp, ino_t ino, int flags,
130 struct thread *td = curthread;
131 uint64_t nodeid = ino;
136 error = vfs_hash_get(mp, fuse_vnode_hash(nodeid), flags, td, vpp,
137 fuse_vnode_cmp, &nodeid);
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
147 if (bintime_cmp(&(VTOFUD(*vpp)->entry_cache_timeout), &now, >)){
148 counter_u64_add(fuse_lookup_cache_hits, 1);
151 /* Entry cache timeout */
152 counter_u64_add(fuse_lookup_cache_misses, 1);
161 SDT_PROBE_DEFINE0(fusefs, , internal, access_vadmin);
162 /* Synchronously send a FUSE_ACCESS operation */
164 fuse_internal_access(struct vnode *vp,
170 uint32_t mask = F_OK;
174 struct fuse_dispatcher fdi;
175 struct fuse_access_in *fai;
176 struct fuse_data *data;
178 mp = vnode_mount(vp);
179 vtype = vnode_vtype(vp);
181 data = fuse_get_mpdata(mp);
182 dataflags = data->dataflags;
187 if (mode & VMODIFY_PERMS && vfs_isrdonly(mp)) {
188 switch (vp->v_type) {
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))
206 if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
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);
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.
219 SDT_PROBE0(fusefs, , internal, access_vadmin);
222 if (fsess_not_impl(mp, FUSE_ACCESS))
225 if ((mode & (VWRITE | VAPPEND)) != 0)
227 if ((mode & VREAD) != 0)
229 if ((mode & VEXEC) != 0)
232 fdisp_init(&fdi, sizeof(*fai));
233 fdisp_make_vp(&fdi, FUSE_ACCESS, vp, td, cred);
238 err = fdisp_wait_answ(&fdi);
242 fsess_set_notimpl(mp, FUSE_ACCESS);
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.
253 * If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do
254 * return the result to the caller).
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,
262 struct fuse_vnode_data *fvdat;
263 struct fuse_data *data;
264 struct vattr *vp_cache_at;
266 mp = vnode_mount(vp);
268 data = fuse_get_mpdata(mp);
270 ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs");
272 fuse_validity_2_bintime(attr_valid, attr_valid_nsec,
273 &fvdat->attr_cache_timeout);
275 if (vnode_isreg(vp) &&
276 fvdat->cached_attrs.va_size != VNOVAL &&
277 attr->size != fvdat->cached_attrs.va_size)
279 if ( data->cache_mode == FUSE_CACHE_WB &&
280 fvdat->flag & FN_SIZECHANGE)
285 * The server changed the file's size even though we're
286 * using writeback cacheing and and we have outstanding
287 * dirty writes! That's a server bug.
289 if (fuse_libabi_geq(data, 7, 23)) {
290 msg = "writeback cache incoherent!."
291 "To prevent data corruption, disable "
292 "the writeback cache according to your "
293 "FUSE server's documentation.";
295 msg = "writeback cache incoherent!."
296 "To prevent data corruption, disable "
297 "the writeback cache by setting "
298 "vfs.fusefs.data_cache_mode to 0 or 1.";
300 fuse_warn(data, FSESS_WARN_WB_CACHE_INCOHERENT, msg);
302 if (fuse_vnode_attr_cache_valid(vp) &&
303 data->cache_mode != FUSE_CACHE_UC)
306 * The server changed the file's size even though we
307 * have it cached and our cache has not yet expired.
310 fuse_warn(data, FSESS_WARN_CACHE_INCOHERENT,
313 "data corruption, disable the data cache "
314 "by mounting with -o direct_io, or as "
315 "directed otherwise by your FUSE server's "
320 /* Fix our buffers if the filesize changed without us knowing */
321 if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) {
322 (void)fuse_vnode_setsize(vp, attr->size, from_server);
323 fvdat->cached_attrs.va_size = attr->size;
326 if (attr_valid > 0 || attr_valid_nsec > 0)
327 vp_cache_at = &(fvdat->cached_attrs);
328 else if (vap != NULL)
333 vattr_null(vp_cache_at);
334 vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0];
335 vp_cache_at->va_fileid = attr->ino;
336 vp_cache_at->va_mode = attr->mode & ~S_IFMT;
337 vp_cache_at->va_nlink = attr->nlink;
338 vp_cache_at->va_uid = attr->uid;
339 vp_cache_at->va_gid = attr->gid;
340 vp_cache_at->va_rdev = attr->rdev;
341 vp_cache_at->va_size = attr->size;
342 /* XXX on i386, seconds are truncated to 32 bits */
343 vp_cache_at->va_atime.tv_sec = attr->atime;
344 vp_cache_at->va_atime.tv_nsec = attr->atimensec;
345 vp_cache_at->va_mtime.tv_sec = attr->mtime;
346 vp_cache_at->va_mtime.tv_nsec = attr->mtimensec;
347 vp_cache_at->va_ctime.tv_sec = attr->ctime;
348 vp_cache_at->va_ctime.tv_nsec = attr->ctimensec;
349 if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0)
350 vp_cache_at->va_blocksize = attr->blksize;
352 vp_cache_at->va_blocksize = PAGE_SIZE;
353 vp_cache_at->va_type = IFTOVT(attr->mode);
354 vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE;
355 vp_cache_at->va_flags = 0;
357 if (vap != vp_cache_at && vap != NULL)
358 memcpy(vap, vp_cache_at, sizeof(*vap));
364 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio)
366 if (tick->tk_aw_ohead.error == ENOSYS) {
367 fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick));
373 fuse_internal_fsync(struct vnode *vp,
378 struct fuse_fsync_in *ffsi = NULL;
379 struct fuse_dispatcher fdi;
380 struct fuse_filehandle *fufh;
381 struct fuse_vnode_data *fvdat = VTOFUD(vp);
382 struct mount *mp = vnode_mount(vp);
386 if (fsess_not_impl(vnode_mount(vp),
387 (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
393 if (fsess_not_impl(mp, op))
396 fdisp_init(&fdi, sizeof(*ffsi));
398 * fsync every open file handle for this file, because we can't be sure
399 * which file handle the caller is really referring to.
401 LIST_FOREACH(fufh, &fvdat->handles, next) {
402 fdi.iosize = sizeof(*ffsi);
404 fdisp_make_vp(&fdi, op, vp, td, NULL);
406 fdisp_refresh_vp(&fdi, op, vp, td, NULL);
408 ffsi->fh = fufh->fh_id;
409 ffsi->fsync_flags = 0;
412 ffsi->fsync_flags = FUSE_FSYNC_FDATASYNC;
414 if (waitfor == MNT_WAIT) {
415 err = fdisp_wait_answ(&fdi);
417 fuse_insert_callback(fdi.tick,
418 fuse_internal_fsync_callback);
419 fuse_insert_message(fdi.tick, false);
422 /* ENOSYS means "success, and don't call again" */
423 fsess_set_notimpl(mp, op);
433 /* Asynchronous invalidation */
434 SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry,
435 "struct vnode*", "struct fuse_notify_inval_entry_out*", "char*");
437 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio)
439 struct fuse_notify_inval_entry_out fnieo;
440 struct componentname cn;
441 struct vnode *dvp, *vp;
445 if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0)
448 if (fnieo.namelen >= sizeof(name))
451 if ((err = uiomove(name, fnieo.namelen, uio)) != 0)
453 name[fnieo.namelen] = '\0';
454 /* fusefs does not cache "." or ".." entries */
455 if (strncmp(name, ".", sizeof(".")) == 0 ||
456 strncmp(name, "..", sizeof("..")) == 0)
459 if (fnieo.parent == FUSE_ROOT_ID)
460 err = VFS_ROOT(mp, LK_SHARED, &dvp);
462 err = fuse_internal_get_cached_vnode( mp, fnieo.parent,
464 SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name);
466 * If dvp is not in the cache, then it must've been reclaimed. And
467 * since fuse_vnop_reclaim does a cache_purge, name's entry must've
468 * been invalidated already. So we can safely return if dvp == NULL
470 if (err != 0 || dvp == NULL)
473 * XXX we can't check dvp's generation because the FUSE invalidate
474 * entry message doesn't include it. Worse case is that we invalidate
475 * an entry that didn't need to be invalidated.
478 cn.cn_nameiop = LOOKUP;
479 cn.cn_flags = 0; /* !MAKEENTRY means free cached entry */
480 cn.cn_cred = curthread->td_ucred;
481 cn.cn_lkflags = LK_SHARED;
483 cn.cn_nameptr = name;
484 cn.cn_namelen = fnieo.namelen;
485 err = cache_lookup(dvp, &vp, &cn, NULL, NULL);
487 fuse_vnode_clear_attr_cache(dvp);
492 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode,
493 "struct vnode*", "struct fuse_notify_inval_inode_out *");
495 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio)
497 struct fuse_notify_inval_inode_out fniio;
501 if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0)
504 if (fniio.ino == FUSE_ROOT_ID)
505 err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp);
507 err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED,
509 SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio);
510 if (err != 0 || vp == NULL)
513 * XXX we can't check vp's generation because the FUSE invalidate
514 * entry message doesn't include it. Worse case is that we invalidate
515 * an inode that didn't need to be invalidated.
519 * Flush and invalidate buffers if off >= 0. Technically we only need
520 * to flush and invalidate the range of offsets [off, off + len), but
521 * for simplicity's sake we do everything.
524 fuse_io_invalbuf(vp, curthread);
525 fuse_vnode_clear_attr_cache(vp);
532 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp,
533 struct componentname *cnp, struct vattr *vap)
535 struct fuse_data *data;
536 struct fuse_mknod_in fmni;
539 data = fuse_get_mpdata(dvp->v_mount);
541 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
542 fmni.rdev = vap->va_rdev;
543 if (fuse_libabi_geq(data, 7, 12)) {
544 insize = sizeof(fmni);
545 fmni.umask = curthread->td_proc->p_pd->pd_cmask;
547 insize = FUSE_COMPAT_MKNOD_IN_SIZE;
549 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
550 insize, vap->va_type));
556 fuse_internal_readdir(struct vnode *vp,
559 struct fuse_filehandle *fufh,
560 struct fuse_iov *cookediov,
565 struct fuse_dispatcher fdi;
566 struct fuse_read_in *fri = NULL;
569 if (uio_resid(uio) == 0)
574 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p
579 * fnd_start is set non-zero once the offset in the directory gets
580 * to the startoff. This is done because directories must be read
581 * from the beginning (offset == 0) when fuse_vnop_readdir() needs
582 * to do an open of the directory.
583 * If it is not set non-zero here, it will be set non-zero in
584 * fuse_internal_readdir_processdata() when uio_offset == startoff.
587 if (uio->uio_offset == startoff)
589 while (uio_resid(uio) > 0) {
590 fdi.iosize = sizeof(*fri);
592 fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
594 fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
597 fri->fh = fufh->fh_id;
598 fri->offset = uio_offset(uio);
599 fri->size = MIN(uio->uio_resid,
600 fuse_get_mpdata(vp->v_mount)->max_read);
602 if ((err = fdisp_wait_answ(&fdi)))
604 if ((err = fuse_internal_readdir_processdata(uio, startoff,
605 &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov,
606 ncookies, &cookies)))
611 return ((err == -1) ? 0 : err);
615 * Return -1 to indicate that this readdir is finished, 0 if it copied
616 * all the directory data read in and it may be possible to read more
617 * and greater than 0 for a failure.
620 fuse_internal_readdir_processdata(struct uio *uio,
626 struct fuse_iov *cookediov,
635 struct fuse_dirent *fudge;
639 if (bufsize < FUSE_NAME_OFFSET)
642 if (bufsize < FUSE_NAME_OFFSET) {
646 fudge = (struct fuse_dirent *)buf;
647 freclen = FUSE_DIRENT_SIZE(fudge);
649 if (bufsize < freclen) {
651 * This indicates a partial directory entry at the
652 * end of the directory data.
657 #ifdef ZERO_PAD_INCOMPLETE_BUFS
658 if (isbzero(buf, FUSE_NAME_OFFSET)) {
664 if (!fudge->namelen || fudge->namelen > MAXNAMLEN) {
668 oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *)
671 if (oreclen > uio_resid(uio)) {
672 /* Out of space for the dir so we are done. */
677 * Don't start to copy the directory entries out until
678 * the requested offset in the directory is found.
680 if (*fnd_start != 0) {
681 fiov_adjust(cookediov, oreclen);
682 bzero(cookediov->base, oreclen);
684 de = (struct dirent *)cookediov->base;
685 de->d_fileno = fudge->ino;
686 de->d_off = fudge->off;
687 de->d_reclen = oreclen;
688 de->d_type = fudge->type;
689 de->d_namlen = fudge->namelen;
690 memcpy((char *)cookediov->base + sizeof(struct dirent) -
692 (char *)buf + FUSE_NAME_OFFSET, fudge->namelen);
693 dirent_terminate(de);
695 err = uiomove(cookediov->base, cookediov->len, uio);
698 if (cookies != NULL) {
699 if (*ncookies == 0) {
703 *cookies = fudge->off;
707 } else if (startoff == fudge->off)
709 buf = (char *)buf + freclen;
711 uio_setoffset(uio, fudge->off);
721 fuse_internal_remove(struct vnode *dvp,
723 struct componentname *cnp,
726 struct fuse_dispatcher fdi;
730 fdisp_init(&fdi, cnp->cn_namelen + 1);
731 fdisp_make_vp(&fdi, op, dvp, curthread, cnp->cn_cred);
733 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
734 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
736 err = fdisp_wait_answ(&fdi);
743 * Access the cached nlink even if the attr cached has expired. If
744 * it's inaccurate, the worst that will happen is:
745 * 1) We'll recycle the vnode even though the file has another link we
746 * don't know about, costing a bit of cpu time, or
747 * 2) We won't recycle the vnode even though all of its links are gone.
748 * It will linger around until vnlru reclaims it, costing a bit of
751 nlink = VTOFUD(vp)->cached_attrs.va_nlink--;
754 * Purge the parent's attribute cache because the daemon
755 * should've updated its mtime and ctime.
757 fuse_vnode_clear_attr_cache(dvp);
759 /* NB: nlink could be zero if it was never cached */
760 if (nlink <= 1 || vnode_vtype(vp) == VDIR) {
761 fuse_internal_vnode_disappear(vp);
764 fuse_vnode_update(vp, FN_CTIMECHANGE);
773 fuse_internal_rename(struct vnode *fdvp,
774 struct componentname *fcnp,
776 struct componentname *tcnp)
778 struct fuse_dispatcher fdi;
779 struct fuse_rename_in *fri;
782 fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2);
783 fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, curthread, tcnp->cn_cred);
786 fri->newdir = VTOI(tdvp);
787 memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr,
789 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0';
790 memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1,
791 tcnp->cn_nameptr, tcnp->cn_namelen);
792 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen +
793 tcnp->cn_namelen + 1] = '\0';
795 err = fdisp_wait_answ(&fdi);
802 /* entity creation */
805 fuse_internal_newentry_makerequest(struct mount *mp,
807 struct componentname *cnp,
811 struct fuse_dispatcher *fdip)
813 fdip->iosize = bufsize + cnp->cn_namelen + 1;
815 fdisp_make(fdip, op, mp, dnid, curthread, cnp->cn_cred);
816 memcpy(fdip->indata, buf, bufsize);
817 memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen);
818 ((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0';
822 fuse_internal_newentry_core(struct vnode *dvp,
824 struct componentname *cnp,
826 struct fuse_dispatcher *fdip)
829 struct fuse_entry_out *feo;
830 struct mount *mp = vnode_mount(dvp);
832 if ((err = fdisp_wait_answ(fdip))) {
837 if ((err = fuse_internal_checkentry(feo, vtyp))) {
840 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp);
842 fuse_internal_forget_send(mp, curthread, cnp->cn_cred,
848 * Purge the parent's attribute cache because the daemon should've
849 * updated its mtime and ctime
851 fuse_vnode_clear_attr_cache(dvp);
853 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
854 feo->attr_valid_nsec, NULL, true);
860 fuse_internal_newentry(struct vnode *dvp,
862 struct componentname *cnp,
869 struct fuse_dispatcher fdi;
870 struct mount *mp = vnode_mount(dvp);
873 fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf,
875 err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi);
881 /* entity destruction */
884 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio)
886 fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL,
887 ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1);
893 fuse_internal_forget_send(struct mount *mp,
900 struct fuse_dispatcher fdi;
901 struct fuse_forget_in *ffi;
904 * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu",
905 * (long long unsigned) nodeid));
908 fdisp_init(&fdi, sizeof(*ffi));
909 fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred);
912 ffi->nlookup = nlookup;
914 fuse_insert_message(fdi.tick, false);
918 /* Fetch the vnode's attributes from the daemon*/
920 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap,
921 struct ucred *cred, struct thread *td)
923 struct fuse_dispatcher fdi;
924 struct fuse_vnode_data *fvdat = VTOFUD(vp);
925 struct fuse_getattr_in *fgai;
926 struct fuse_attr_out *fao;
927 off_t old_filesize = fvdat->cached_attrs.va_size;
928 struct timespec old_ctime = fvdat->cached_attrs.va_ctime;
929 struct timespec old_mtime = fvdat->cached_attrs.va_mtime;
933 fdisp_init(&fdi, sizeof(*fgai));
934 fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred);
937 * We could look up a file handle and set it in fgai->fh, but that
938 * involves extra runtime work and I'm unaware of any file systems that
941 fgai->getattr_flags = 0;
942 if ((err = fdisp_wait_answ(&fdi))) {
944 fuse_internal_vnode_disappear(vp);
948 fao = (struct fuse_attr_out *)fdi.answ;
949 vtyp = IFTOVT(fao->attr.mode);
950 if (fvdat->flag & FN_SIZECHANGE)
951 fao->attr.size = old_filesize;
952 if (fvdat->flag & FN_CTIMECHANGE) {
953 fao->attr.ctime = old_ctime.tv_sec;
954 fao->attr.ctimensec = old_ctime.tv_nsec;
956 if (fvdat->flag & FN_MTIMECHANGE) {
957 fao->attr.mtime = old_mtime.tv_sec;
958 fao->attr.mtimensec = old_mtime.tv_nsec;
960 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
961 fao->attr_valid_nsec, vap, true);
962 if (vtyp != vnode_vtype(vp)) {
963 fuse_internal_vnode_disappear(vp);
972 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */
974 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred,
979 if ((attrs = VTOVA(vp)) != NULL) {
980 *vap = *attrs; /* struct copy */
984 return fuse_internal_do_getattr(vp, vap, cred, td);
988 fuse_internal_vnode_disappear(struct vnode *vp)
990 struct fuse_vnode_data *fvdat = VTOFUD(vp);
992 ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
993 fvdat->flag |= FN_REVOKED;
997 /* fuse start/stop */
999 SDT_PROBE_DEFINE2(fusefs, , internal, init_done,
1000 "struct fuse_data*", "struct fuse_init_out*");
1002 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio)
1005 struct fuse_data *data = tick->tk_data;
1006 struct fuse_init_out *fiio;
1008 if ((err = tick->tk_aw_ohead.error)) {
1011 if ((err = fticket_pull(tick, uio))) {
1014 fiio = fticket_resp(tick)->base;
1016 data->fuse_libabi_major = fiio->major;
1017 data->fuse_libabi_minor = fiio->minor;
1018 if (!fuse_libabi_geq(data, 7, 4)) {
1020 * With a little work we could support servers as old as 7.1.
1021 * But there would be little payoff.
1023 SDT_PROBE2(fusefs, , internal, trace, 1,
1024 "userpace version too low");
1025 err = EPROTONOSUPPORT;
1029 if (fuse_libabi_geq(data, 7, 5)) {
1030 if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) ||
1031 fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) {
1032 data->max_write = fiio->max_write;
1033 if (fiio->flags & FUSE_ASYNC_READ)
1034 data->dataflags |= FSESS_ASYNC_READ;
1035 if (fiio->flags & FUSE_POSIX_LOCKS)
1036 data->dataflags |= FSESS_POSIX_LOCKS;
1037 if (fiio->flags & FUSE_EXPORT_SUPPORT)
1038 data->dataflags |= FSESS_EXPORT_SUPPORT;
1039 if (fiio->flags & FUSE_NO_OPEN_SUPPORT)
1040 data->dataflags |= FSESS_NO_OPEN_SUPPORT;
1041 if (fiio->flags & FUSE_NO_OPENDIR_SUPPORT)
1042 data->dataflags |= FSESS_NO_OPENDIR_SUPPORT;
1044 * Don't bother to check FUSE_BIG_WRITES, because it's
1045 * redundant with max_write
1048 * max_background and congestion_threshold are not
1055 /* Old fixed values */
1056 data->max_write = 4096;
1059 if (fuse_libabi_geq(data, 7, 6))
1060 data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf;
1062 if (!fuse_libabi_geq(data, 7, 7))
1063 fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
1065 if (!fuse_libabi_geq(data, 7, 8)) {
1066 fsess_set_notimpl(data->mp, FUSE_BMAP);
1067 fsess_set_notimpl(data->mp, FUSE_DESTROY);
1070 if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 &&
1071 fiio->time_gran <= 1000000000)
1072 data->time_gran = fiio->time_gran;
1074 data->time_gran = 1;
1076 if (!fuse_libabi_geq(data, 7, 23))
1077 data->cache_mode = fuse_data_cache_mode;
1078 else if (fiio->flags & FUSE_WRITEBACK_CACHE)
1079 data->cache_mode = FUSE_CACHE_WB;
1081 data->cache_mode = FUSE_CACHE_WT;
1083 if (!fuse_libabi_geq(data, 7, 24))
1084 fsess_set_notimpl(data->mp, FUSE_LSEEK);
1086 if (!fuse_libabi_geq(data, 7, 28))
1087 fsess_set_notimpl(data->mp, FUSE_COPY_FILE_RANGE);
1091 fdata_set_dead(data);
1094 data->dataflags |= FSESS_INITED;
1095 SDT_PROBE2(fusefs, , internal, init_done, data, fiio);
1096 wakeup(&data->ticketer);
1103 fuse_internal_send_init(struct fuse_data *data, struct thread *td)
1105 struct fuse_init_in *fiii;
1106 struct fuse_dispatcher fdi;
1108 fdisp_init(&fdi, sizeof(*fiii));
1109 fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL);
1111 fiii->major = FUSE_KERNEL_VERSION;
1112 fiii->minor = FUSE_KERNEL_MINOR_VERSION;
1114 * fusefs currently reads ahead no more than one cache block at a time.
1115 * See fuse_read_biobackend
1117 fiii->max_readahead = maxbcachebuf;
1119 * Unsupported features:
1120 * FUSE_FILE_OPS: No known FUSE server or client supports it
1121 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it
1122 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even
1123 * when default ACLs are in use.
1124 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD
1125 * doesn't have splice(2).
1126 * FUSE_FLOCK_LOCKS: not yet implemented
1127 * FUSE_HAS_IOCTL_DIR: not yet implemented
1128 * FUSE_AUTO_INVAL_DATA: not yet implemented
1129 * FUSE_DO_READDIRPLUS: not yet implemented
1130 * FUSE_READDIRPLUS_AUTO: not yet implemented
1131 * FUSE_ASYNC_DIO: not yet implemented
1132 * FUSE_PARALLEL_DIROPS: not yet implemented
1133 * FUSE_HANDLE_KILLPRIV: not yet implemented
1134 * FUSE_POSIX_ACL: not yet implemented
1135 * FUSE_ABORT_ERROR: not yet implemented
1136 * FUSE_CACHE_SYMLINKS: not yet implemented
1137 * FUSE_MAX_PAGES: not yet implemented
1139 fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT
1140 | FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE
1141 | FUSE_NO_OPEN_SUPPORT | FUSE_NO_OPENDIR_SUPPORT;
1143 fuse_insert_callback(fdi.tick, fuse_internal_init_callback);
1144 fuse_insert_message(fdi.tick, false);
1145 fdisp_destroy(&fdi);
1149 * Send a FUSE_SETATTR operation with no permissions checks. If cred is NULL,
1150 * send the request with root credentials
1152 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap,
1153 struct thread *td, struct ucred *cred)
1155 struct fuse_vnode_data *fvdat;
1156 struct fuse_dispatcher fdi;
1157 struct fuse_setattr_in *fsai;
1159 pid_t pid = td->td_proc->p_pid;
1160 struct fuse_data *data;
1164 int sizechanged = -1;
1165 uint64_t newsize = 0;
1167 mp = vnode_mount(vp);
1169 data = fuse_get_mpdata(mp);
1170 dataflags = data->dataflags;
1172 fdisp_init(&fdi, sizeof(*fsai));
1173 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1181 if (vap->va_uid != (uid_t)VNOVAL) {
1182 fsai->uid = vap->va_uid;
1183 fsai->valid |= FATTR_UID;
1185 if (vap->va_gid != (gid_t)VNOVAL) {
1186 fsai->gid = vap->va_gid;
1187 fsai->valid |= FATTR_GID;
1189 if (vap->va_size != VNOVAL) {
1190 struct fuse_filehandle *fufh = NULL;
1192 /*Truncate to a new value. */
1193 fsai->size = vap->va_size;
1195 newsize = vap->va_size;
1196 fsai->valid |= FATTR_SIZE;
1198 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1200 fsai->fh = fufh->fh_id;
1201 fsai->valid |= FATTR_FH;
1203 VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1205 if (vap->va_atime.tv_sec != VNOVAL) {
1206 fsai->atime = vap->va_atime.tv_sec;
1207 fsai->atimensec = vap->va_atime.tv_nsec;
1208 fsai->valid |= FATTR_ATIME;
1209 if (vap->va_vaflags & VA_UTIMES_NULL)
1210 fsai->valid |= FATTR_ATIME_NOW;
1212 if (vap->va_mtime.tv_sec != VNOVAL) {
1213 fsai->mtime = vap->va_mtime.tv_sec;
1214 fsai->mtimensec = vap->va_mtime.tv_nsec;
1215 fsai->valid |= FATTR_MTIME;
1216 if (vap->va_vaflags & VA_UTIMES_NULL)
1217 fsai->valid |= FATTR_MTIME_NOW;
1218 } else if (fvdat->flag & FN_MTIMECHANGE) {
1219 fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec;
1220 fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
1221 fsai->valid |= FATTR_MTIME;
1223 if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) {
1224 fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec;
1225 fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
1226 fsai->valid |= FATTR_CTIME;
1228 if (vap->va_mode != (mode_t)VNOVAL) {
1229 fsai->mode = vap->va_mode & ALLPERMS;
1230 fsai->valid |= FATTR_MODE;
1236 if ((err = fdisp_wait_answ(&fdi)))
1238 vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1240 if (vnode_vtype(vp) != vtyp) {
1241 if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1242 SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! "
1243 "vnode_vtype is VNON and vtype isn't.");
1246 * STALE vnode, ditch
1248 * The vnode has changed its type "behind our back".
1249 * There's nothing really we can do, so let us just
1250 * force an internal revocation and tell the caller to
1251 * try again, if interested.
1253 fuse_internal_vnode_disappear(vp);
1258 struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1259 fuse_vnode_undirty_cached_timestamps(vp);
1260 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1261 fao->attr_valid_nsec, NULL, false);
1265 fdisp_destroy(&fdi);
1270 * FreeBSD clears the SUID and SGID bits on any write by a non-root user.
1273 fuse_internal_clear_suid_on_write(struct vnode *vp, struct ucred *cred,
1276 struct fuse_data *data;
1281 mp = vnode_mount(vp);
1282 data = fuse_get_mpdata(mp);
1283 dataflags = data->dataflags;
1285 ASSERT_VOP_LOCKED(vp, __func__);
1287 if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
1288 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1289 fuse_internal_getattr(vp, &va, cred, td);
1290 if (va.va_mode & (S_ISUID | S_ISGID)) {
1291 mode_t mode = va.va_mode & ~(S_ISUID | S_ISGID);
1292 /* Clear all vattr fields except mode */
1297 * Ignore fuse_internal_setattr's return value,
1298 * because at this point the write operation has
1299 * already succeeded and we don't want to return
1300 * failing status for that.
1302 (void)fuse_internal_setattr(vp, &va, td, NULL);
1308 #ifdef ZERO_PAD_INCOMPLETE_BUFS
1310 isbzero(void *buf, size_t len)
1314 for (i = 0; i < len; i++) {
1315 if (((char *)buf)[i])
1325 fuse_internal_init(void)
1327 fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK);
1328 fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK);
1332 fuse_internal_destroy(void)
1334 counter_u64_free(fuse_lookup_cache_hits);
1335 counter_u64_free(fuse_lookup_cache_misses);