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)
261 struct fuse_vnode_data *fvdat;
262 struct fuse_data *data;
263 struct vattr *vp_cache_at;
265 mp = vnode_mount(vp);
267 data = fuse_get_mpdata(mp);
269 ASSERT_VOP_ELOCKED(vp, "fuse_internal_cache_attrs");
271 fuse_validity_2_bintime(attr_valid, attr_valid_nsec,
272 &fvdat->attr_cache_timeout);
274 /* Fix our buffers if the filesize changed without us knowing */
275 if (vnode_isreg(vp) && attr->size != fvdat->cached_attrs.va_size) {
276 (void)fuse_vnode_setsize(vp, attr->size);
277 fvdat->cached_attrs.va_size = attr->size;
280 if (attr_valid > 0 || attr_valid_nsec > 0)
281 vp_cache_at = &(fvdat->cached_attrs);
282 else if (vap != NULL)
287 vattr_null(vp_cache_at);
288 vp_cache_at->va_fsid = mp->mnt_stat.f_fsid.val[0];
289 vp_cache_at->va_fileid = attr->ino;
290 vp_cache_at->va_mode = attr->mode & ~S_IFMT;
291 vp_cache_at->va_nlink = attr->nlink;
292 vp_cache_at->va_uid = attr->uid;
293 vp_cache_at->va_gid = attr->gid;
294 vp_cache_at->va_rdev = attr->rdev;
295 vp_cache_at->va_size = attr->size;
296 /* XXX on i386, seconds are truncated to 32 bits */
297 vp_cache_at->va_atime.tv_sec = attr->atime;
298 vp_cache_at->va_atime.tv_nsec = attr->atimensec;
299 vp_cache_at->va_mtime.tv_sec = attr->mtime;
300 vp_cache_at->va_mtime.tv_nsec = attr->mtimensec;
301 vp_cache_at->va_ctime.tv_sec = attr->ctime;
302 vp_cache_at->va_ctime.tv_nsec = attr->ctimensec;
303 if (fuse_libabi_geq(data, 7, 9) && attr->blksize > 0)
304 vp_cache_at->va_blocksize = attr->blksize;
306 vp_cache_at->va_blocksize = PAGE_SIZE;
307 vp_cache_at->va_type = IFTOVT(attr->mode);
308 vp_cache_at->va_bytes = attr->blocks * S_BLKSIZE;
309 vp_cache_at->va_flags = 0;
311 if (vap != vp_cache_at && vap != NULL)
312 memcpy(vap, vp_cache_at, sizeof(*vap));
318 fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio)
320 if (tick->tk_aw_ohead.error == ENOSYS) {
321 fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick));
327 fuse_internal_fsync(struct vnode *vp,
332 struct fuse_fsync_in *ffsi = NULL;
333 struct fuse_dispatcher fdi;
334 struct fuse_filehandle *fufh;
335 struct fuse_vnode_data *fvdat = VTOFUD(vp);
336 struct mount *mp = vnode_mount(vp);
340 if (fsess_not_impl(vnode_mount(vp),
341 (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
347 if (fsess_not_impl(mp, op))
350 fdisp_init(&fdi, sizeof(*ffsi));
352 * fsync every open file handle for this file, because we can't be sure
353 * which file handle the caller is really referring to.
355 LIST_FOREACH(fufh, &fvdat->handles, next) {
356 fdi.iosize = sizeof(*ffsi);
358 fdisp_make_vp(&fdi, op, vp, td, NULL);
360 fdisp_refresh_vp(&fdi, op, vp, td, NULL);
362 ffsi->fh = fufh->fh_id;
363 ffsi->fsync_flags = 0;
366 ffsi->fsync_flags = 1;
368 if (waitfor == MNT_WAIT) {
369 err = fdisp_wait_answ(&fdi);
371 fuse_insert_callback(fdi.tick,
372 fuse_internal_fsync_callback);
373 fuse_insert_message(fdi.tick, false);
376 /* ENOSYS means "success, and don't call again" */
377 fsess_set_notimpl(mp, op);
387 /* Asynchronous invalidation */
388 SDT_PROBE_DEFINE3(fusefs, , internal, invalidate_entry,
389 "struct vnode*", "struct fuse_notify_inval_entry_out*", "char*");
391 fuse_internal_invalidate_entry(struct mount *mp, struct uio *uio)
393 struct fuse_notify_inval_entry_out fnieo;
394 struct componentname cn;
395 struct vnode *dvp, *vp;
399 if ((err = uiomove(&fnieo, sizeof(fnieo), uio)) != 0)
402 if (fnieo.namelen >= sizeof(name))
405 if ((err = uiomove(name, fnieo.namelen, uio)) != 0)
407 name[fnieo.namelen] = '\0';
408 /* fusefs does not cache "." or ".." entries */
409 if (strncmp(name, ".", sizeof(".")) == 0 ||
410 strncmp(name, "..", sizeof("..")) == 0)
413 if (fnieo.parent == FUSE_ROOT_ID)
414 err = VFS_ROOT(mp, LK_SHARED, &dvp);
416 err = fuse_internal_get_cached_vnode( mp, fnieo.parent,
418 SDT_PROBE3(fusefs, , internal, invalidate_entry, dvp, &fnieo, name);
420 * If dvp is not in the cache, then it must've been reclaimed. And
421 * since fuse_vnop_reclaim does a cache_purge, name's entry must've
422 * been invalidated already. So we can safely return if dvp == NULL
424 if (err != 0 || dvp == NULL)
427 * XXX we can't check dvp's generation because the FUSE invalidate
428 * entry message doesn't include it. Worse case is that we invalidate
429 * an entry that didn't need to be invalidated.
432 cn.cn_nameiop = LOOKUP;
433 cn.cn_flags = 0; /* !MAKEENTRY means free cached entry */
434 cn.cn_thread = curthread;
435 cn.cn_cred = curthread->td_ucred;
436 cn.cn_lkflags = LK_SHARED;
438 cn.cn_nameptr = name;
439 cn.cn_namelen = fnieo.namelen;
440 err = cache_lookup(dvp, &vp, &cn, NULL, NULL);
442 fuse_vnode_clear_attr_cache(dvp);
447 SDT_PROBE_DEFINE2(fusefs, , internal, invalidate_inode,
448 "struct vnode*", "struct fuse_notify_inval_inode_out *");
450 fuse_internal_invalidate_inode(struct mount *mp, struct uio *uio)
452 struct fuse_notify_inval_inode_out fniio;
456 if ((err = uiomove(&fniio, sizeof(fniio), uio)) != 0)
459 if (fniio.ino == FUSE_ROOT_ID)
460 err = VFS_ROOT(mp, LK_EXCLUSIVE, &vp);
462 err = fuse_internal_get_cached_vnode(mp, fniio.ino, LK_SHARED,
464 SDT_PROBE2(fusefs, , internal, invalidate_inode, vp, &fniio);
465 if (err != 0 || vp == NULL)
468 * XXX we can't check vp's generation because the FUSE invalidate
469 * entry message doesn't include it. Worse case is that we invalidate
470 * an inode that didn't need to be invalidated.
474 * Flush and invalidate buffers if off >= 0. Technically we only need
475 * to flush and invalidate the range of offsets [off, off + len), but
476 * for simplicity's sake we do everything.
479 fuse_io_invalbuf(vp, curthread);
480 fuse_vnode_clear_attr_cache(vp);
487 fuse_internal_mknod(struct vnode *dvp, struct vnode **vpp,
488 struct componentname *cnp, struct vattr *vap)
490 struct fuse_data *data;
491 struct fuse_mknod_in fmni;
494 data = fuse_get_mpdata(dvp->v_mount);
496 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
497 fmni.rdev = vap->va_rdev;
498 if (fuse_libabi_geq(data, 7, 12)) {
499 insize = sizeof(fmni);
500 fmni.umask = curthread->td_proc->p_pd->pd_cmask;
502 insize = FUSE_COMPAT_MKNOD_IN_SIZE;
504 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
505 insize, vap->va_type));
511 fuse_internal_readdir(struct vnode *vp,
514 struct fuse_filehandle *fufh,
515 struct fuse_iov *cookediov,
520 struct fuse_dispatcher fdi;
521 struct fuse_read_in *fri = NULL;
524 if (uio_resid(uio) == 0)
529 * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p
534 * fnd_start is set non-zero once the offset in the directory gets
535 * to the startoff. This is done because directories must be read
536 * from the beginning (offset == 0) when fuse_vnop_readdir() needs
537 * to do an open of the directory.
538 * If it is not set non-zero here, it will be set non-zero in
539 * fuse_internal_readdir_processdata() when uio_offset == startoff.
542 if (uio->uio_offset == startoff)
544 while (uio_resid(uio) > 0) {
545 fdi.iosize = sizeof(*fri);
547 fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
549 fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL);
552 fri->fh = fufh->fh_id;
553 fri->offset = uio_offset(uio);
554 fri->size = MIN(uio->uio_resid,
555 fuse_get_mpdata(vp->v_mount)->max_read);
557 if ((err = fdisp_wait_answ(&fdi)))
559 if ((err = fuse_internal_readdir_processdata(uio, startoff,
560 &fnd_start, fri->size, fdi.answ, fdi.iosize, cookediov,
561 ncookies, &cookies)))
566 return ((err == -1) ? 0 : err);
570 * Return -1 to indicate that this readdir is finished, 0 if it copied
571 * all the directory data read in and it may be possible to read more
572 * and greater than 0 for a failure.
575 fuse_internal_readdir_processdata(struct uio *uio,
581 struct fuse_iov *cookediov,
590 struct fuse_dirent *fudge;
594 if (bufsize < FUSE_NAME_OFFSET)
597 if (bufsize < FUSE_NAME_OFFSET) {
601 fudge = (struct fuse_dirent *)buf;
602 freclen = FUSE_DIRENT_SIZE(fudge);
604 if (bufsize < freclen) {
606 * This indicates a partial directory entry at the
607 * end of the directory data.
612 #ifdef ZERO_PAD_INCOMPLETE_BUFS
613 if (isbzero(buf, FUSE_NAME_OFFSET)) {
619 if (!fudge->namelen || fudge->namelen > MAXNAMLEN) {
623 oreclen = GENERIC_DIRSIZ((struct pseudo_dirent *)
626 if (oreclen > uio_resid(uio)) {
627 /* Out of space for the dir so we are done. */
632 * Don't start to copy the directory entries out until
633 * the requested offset in the directory is found.
635 if (*fnd_start != 0) {
636 fiov_adjust(cookediov, oreclen);
637 bzero(cookediov->base, oreclen);
639 de = (struct dirent *)cookediov->base;
640 de->d_fileno = fudge->ino;
641 de->d_off = fudge->off;
642 de->d_reclen = oreclen;
643 de->d_type = fudge->type;
644 de->d_namlen = fudge->namelen;
645 memcpy((char *)cookediov->base + sizeof(struct dirent) -
647 (char *)buf + FUSE_NAME_OFFSET, fudge->namelen);
648 dirent_terminate(de);
650 err = uiomove(cookediov->base, cookediov->len, uio);
653 if (cookies != NULL) {
654 if (*ncookies == 0) {
658 *cookies = fudge->off;
662 } else if (startoff == fudge->off)
664 buf = (char *)buf + freclen;
666 uio_setoffset(uio, fudge->off);
676 fuse_internal_remove(struct vnode *dvp,
678 struct componentname *cnp,
681 struct fuse_dispatcher fdi;
685 fdisp_init(&fdi, cnp->cn_namelen + 1);
686 fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred);
688 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
689 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
691 err = fdisp_wait_answ(&fdi);
698 * Access the cached nlink even if the attr cached has expired. If
699 * it's inaccurate, the worst that will happen is:
700 * 1) We'll recycle the vnode even though the file has another link we
701 * don't know about, costing a bit of cpu time, or
702 * 2) We won't recycle the vnode even though all of its links are gone.
703 * It will linger around until vnlru reclaims it, costing a bit of
706 nlink = VTOFUD(vp)->cached_attrs.va_nlink--;
709 * Purge the parent's attribute cache because the daemon
710 * should've updated its mtime and ctime.
712 fuse_vnode_clear_attr_cache(dvp);
714 /* NB: nlink could be zero if it was never cached */
715 if (nlink <= 1 || vnode_vtype(vp) == VDIR) {
716 fuse_internal_vnode_disappear(vp);
719 fuse_vnode_update(vp, FN_CTIMECHANGE);
728 fuse_internal_rename(struct vnode *fdvp,
729 struct componentname *fcnp,
731 struct componentname *tcnp)
733 struct fuse_dispatcher fdi;
734 struct fuse_rename_in *fri;
737 fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2);
738 fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred);
741 fri->newdir = VTOI(tdvp);
742 memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr,
744 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0';
745 memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1,
746 tcnp->cn_nameptr, tcnp->cn_namelen);
747 ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen +
748 tcnp->cn_namelen + 1] = '\0';
750 err = fdisp_wait_answ(&fdi);
757 /* entity creation */
760 fuse_internal_newentry_makerequest(struct mount *mp,
762 struct componentname *cnp,
766 struct fuse_dispatcher *fdip)
768 fdip->iosize = bufsize + cnp->cn_namelen + 1;
770 fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred);
771 memcpy(fdip->indata, buf, bufsize);
772 memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen);
773 ((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0';
777 fuse_internal_newentry_core(struct vnode *dvp,
779 struct componentname *cnp,
781 struct fuse_dispatcher *fdip)
784 struct fuse_entry_out *feo;
785 struct mount *mp = vnode_mount(dvp);
787 if ((err = fdisp_wait_answ(fdip))) {
792 if ((err = fuse_internal_checkentry(feo, vtyp))) {
795 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp);
797 fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred,
803 * Purge the parent's attribute cache because the daemon should've
804 * updated its mtime and ctime
806 fuse_vnode_clear_attr_cache(dvp);
808 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
809 feo->attr_valid_nsec, NULL);
815 fuse_internal_newentry(struct vnode *dvp,
817 struct componentname *cnp,
824 struct fuse_dispatcher fdi;
825 struct mount *mp = vnode_mount(dvp);
828 fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf,
830 err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi);
836 /* entity destruction */
839 fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio)
841 fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL,
842 ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1);
848 fuse_internal_forget_send(struct mount *mp,
855 struct fuse_dispatcher fdi;
856 struct fuse_forget_in *ffi;
859 * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu",
860 * (long long unsigned) nodeid));
863 fdisp_init(&fdi, sizeof(*ffi));
864 fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred);
867 ffi->nlookup = nlookup;
869 fuse_insert_message(fdi.tick, false);
873 /* Fetch the vnode's attributes from the daemon*/
875 fuse_internal_do_getattr(struct vnode *vp, struct vattr *vap,
876 struct ucred *cred, struct thread *td)
878 struct fuse_dispatcher fdi;
879 struct fuse_vnode_data *fvdat = VTOFUD(vp);
880 struct fuse_getattr_in *fgai;
881 struct fuse_attr_out *fao;
882 off_t old_filesize = fvdat->cached_attrs.va_size;
883 struct timespec old_ctime = fvdat->cached_attrs.va_ctime;
884 struct timespec old_mtime = fvdat->cached_attrs.va_mtime;
888 fdisp_init(&fdi, sizeof(*fgai));
889 fdisp_make_vp(&fdi, FUSE_GETATTR, vp, td, cred);
892 * We could look up a file handle and set it in fgai->fh, but that
893 * involves extra runtime work and I'm unaware of any file systems that
896 fgai->getattr_flags = 0;
897 if ((err = fdisp_wait_answ(&fdi))) {
899 fuse_internal_vnode_disappear(vp);
903 fao = (struct fuse_attr_out *)fdi.answ;
904 vtyp = IFTOVT(fao->attr.mode);
905 if (fvdat->flag & FN_SIZECHANGE)
906 fao->attr.size = old_filesize;
907 if (fvdat->flag & FN_CTIMECHANGE) {
908 fao->attr.ctime = old_ctime.tv_sec;
909 fao->attr.ctimensec = old_ctime.tv_nsec;
911 if (fvdat->flag & FN_MTIMECHANGE) {
912 fao->attr.mtime = old_mtime.tv_sec;
913 fao->attr.mtimensec = old_mtime.tv_nsec;
915 if (vnode_isreg(vp) &&
916 fvdat->cached_attrs.va_size != VNOVAL &&
917 fao->attr.size != fvdat->cached_attrs.va_size) {
919 * The server changed the file's size even though we had it
920 * cached! That's a server bug.
922 struct mount *mp = vnode_mount(vp);
923 struct fuse_data *data = fuse_get_mpdata(mp);
925 fuse_warn(data, FSESS_WARN_CACHE_INCOHERENT,
927 "To prevent data corruption, disable the data cache "
928 "by mounting with -o direct_io, or as directed "
929 "otherwise by your FUSE server's documentation.");
930 int iosize = fuse_iosize(vp);
931 v_inval_buf_range(vp, 0, INT64_MAX, iosize);
933 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
934 fao->attr_valid_nsec, vap);
935 if (vtyp != vnode_vtype(vp)) {
936 fuse_internal_vnode_disappear(vp);
945 /* Read a vnode's attributes from cache or fetch them from the fuse daemon */
947 fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred,
952 if ((attrs = VTOVA(vp)) != NULL) {
953 *vap = *attrs; /* struct copy */
957 return fuse_internal_do_getattr(vp, vap, cred, td);
961 fuse_internal_vnode_disappear(struct vnode *vp)
963 struct fuse_vnode_data *fvdat = VTOFUD(vp);
965 ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear");
966 fvdat->flag |= FN_REVOKED;
970 /* fuse start/stop */
972 SDT_PROBE_DEFINE2(fusefs, , internal, init_done,
973 "struct fuse_data*", "struct fuse_init_out*");
975 fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio)
978 struct fuse_data *data = tick->tk_data;
979 struct fuse_init_out *fiio;
981 if ((err = tick->tk_aw_ohead.error)) {
984 if ((err = fticket_pull(tick, uio))) {
987 fiio = fticket_resp(tick)->base;
989 data->fuse_libabi_major = fiio->major;
990 data->fuse_libabi_minor = fiio->minor;
991 if (!fuse_libabi_geq(data, 7, 4)) {
993 * With a little work we could support servers as old as 7.1.
994 * But there would be little payoff.
996 SDT_PROBE2(fusefs, , internal, trace, 1,
997 "userpace version too low");
998 err = EPROTONOSUPPORT;
1002 if (fuse_libabi_geq(data, 7, 5)) {
1003 if (fticket_resp(tick)->len == sizeof(struct fuse_init_out) ||
1004 fticket_resp(tick)->len == FUSE_COMPAT_22_INIT_OUT_SIZE) {
1005 data->max_write = fiio->max_write;
1006 if (fiio->flags & FUSE_ASYNC_READ)
1007 data->dataflags |= FSESS_ASYNC_READ;
1008 if (fiio->flags & FUSE_POSIX_LOCKS)
1009 data->dataflags |= FSESS_POSIX_LOCKS;
1010 if (fiio->flags & FUSE_EXPORT_SUPPORT)
1011 data->dataflags |= FSESS_EXPORT_SUPPORT;
1013 * Don't bother to check FUSE_BIG_WRITES, because it's
1014 * redundant with max_write
1017 * max_background and congestion_threshold are not
1024 /* Old fixed values */
1025 data->max_write = 4096;
1028 if (fuse_libabi_geq(data, 7, 6))
1029 data->max_readahead_blocks = fiio->max_readahead / maxbcachebuf;
1031 if (!fuse_libabi_geq(data, 7, 7))
1032 fsess_set_notimpl(data->mp, FUSE_INTERRUPT);
1034 if (!fuse_libabi_geq(data, 7, 8)) {
1035 fsess_set_notimpl(data->mp, FUSE_BMAP);
1036 fsess_set_notimpl(data->mp, FUSE_DESTROY);
1039 if (fuse_libabi_geq(data, 7, 23) && fiio->time_gran >= 1 &&
1040 fiio->time_gran <= 1000000000)
1041 data->time_gran = fiio->time_gran;
1043 data->time_gran = 1;
1045 if (!fuse_libabi_geq(data, 7, 23))
1046 data->cache_mode = fuse_data_cache_mode;
1047 else if (fiio->flags & FUSE_WRITEBACK_CACHE)
1048 data->cache_mode = FUSE_CACHE_WB;
1050 data->cache_mode = FUSE_CACHE_WT;
1052 if (!fuse_libabi_geq(data, 7, 24))
1053 fsess_set_notimpl(data->mp, FUSE_LSEEK);
1055 if (!fuse_libabi_geq(data, 7, 28))
1056 fsess_set_notimpl(data->mp, FUSE_COPY_FILE_RANGE);
1060 fdata_set_dead(data);
1063 data->dataflags |= FSESS_INITED;
1064 SDT_PROBE2(fusefs, , internal, init_done, data, fiio);
1065 wakeup(&data->ticketer);
1072 fuse_internal_send_init(struct fuse_data *data, struct thread *td)
1074 struct fuse_init_in *fiii;
1075 struct fuse_dispatcher fdi;
1077 fdisp_init(&fdi, sizeof(*fiii));
1078 fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL);
1080 fiii->major = FUSE_KERNEL_VERSION;
1081 fiii->minor = FUSE_KERNEL_MINOR_VERSION;
1083 * fusefs currently reads ahead no more than one cache block at a time.
1084 * See fuse_read_biobackend
1086 fiii->max_readahead = maxbcachebuf;
1088 * Unsupported features:
1089 * FUSE_FILE_OPS: No known FUSE server or client supports it
1090 * FUSE_ATOMIC_O_TRUNC: our VFS cannot support it
1091 * FUSE_DONT_MASK: unlike Linux, FreeBSD always applies the umask, even
1092 * when default ACLs are in use.
1093 * FUSE_SPLICE_WRITE, FUSE_SPLICE_MOVE, FUSE_SPLICE_READ: FreeBSD
1094 * doesn't have splice(2).
1095 * FUSE_FLOCK_LOCKS: not yet implemented
1096 * FUSE_HAS_IOCTL_DIR: not yet implemented
1097 * FUSE_AUTO_INVAL_DATA: not yet implemented
1098 * FUSE_DO_READDIRPLUS: not yet implemented
1099 * FUSE_READDIRPLUS_AUTO: not yet implemented
1100 * FUSE_ASYNC_DIO: not yet implemented
1101 * FUSE_NO_OPEN_SUPPORT: not yet implemented
1102 * FUSE_PARALLEL_DIROPS: not yet implemented
1103 * FUSE_HANDLE_KILLPRIV: not yet implemented
1104 * FUSE_POSIX_ACL: not yet implemented
1105 * FUSE_ABORT_ERROR: not yet implemented
1106 * FUSE_CACHE_SYMLINKS: not yet implemented
1107 * FUSE_MAX_PAGES: not yet implemented
1109 fiii->flags = FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_EXPORT_SUPPORT
1110 | FUSE_BIG_WRITES | FUSE_WRITEBACK_CACHE;
1112 fuse_insert_callback(fdi.tick, fuse_internal_init_callback);
1113 fuse_insert_message(fdi.tick, false);
1114 fdisp_destroy(&fdi);
1118 * Send a FUSE_SETATTR operation with no permissions checks. If cred is NULL,
1119 * send the request with root credentials
1121 int fuse_internal_setattr(struct vnode *vp, struct vattr *vap,
1122 struct thread *td, struct ucred *cred)
1124 struct fuse_vnode_data *fvdat;
1125 struct fuse_dispatcher fdi;
1126 struct fuse_setattr_in *fsai;
1128 pid_t pid = td->td_proc->p_pid;
1129 struct fuse_data *data;
1133 int sizechanged = -1;
1134 uint64_t newsize = 0;
1136 mp = vnode_mount(vp);
1138 data = fuse_get_mpdata(mp);
1139 dataflags = data->dataflags;
1141 fdisp_init(&fdi, sizeof(*fsai));
1142 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1150 if (vap->va_uid != (uid_t)VNOVAL) {
1151 fsai->uid = vap->va_uid;
1152 fsai->valid |= FATTR_UID;
1154 if (vap->va_gid != (gid_t)VNOVAL) {
1155 fsai->gid = vap->va_gid;
1156 fsai->valid |= FATTR_GID;
1158 if (vap->va_size != VNOVAL) {
1159 struct fuse_filehandle *fufh = NULL;
1161 /*Truncate to a new value. */
1162 fsai->size = vap->va_size;
1164 newsize = vap->va_size;
1165 fsai->valid |= FATTR_SIZE;
1167 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1169 fsai->fh = fufh->fh_id;
1170 fsai->valid |= FATTR_FH;
1172 VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1174 if (vap->va_atime.tv_sec != VNOVAL) {
1175 fsai->atime = vap->va_atime.tv_sec;
1176 fsai->atimensec = vap->va_atime.tv_nsec;
1177 fsai->valid |= FATTR_ATIME;
1178 if (vap->va_vaflags & VA_UTIMES_NULL)
1179 fsai->valid |= FATTR_ATIME_NOW;
1181 if (vap->va_mtime.tv_sec != VNOVAL) {
1182 fsai->mtime = vap->va_mtime.tv_sec;
1183 fsai->mtimensec = vap->va_mtime.tv_nsec;
1184 fsai->valid |= FATTR_MTIME;
1185 if (vap->va_vaflags & VA_UTIMES_NULL)
1186 fsai->valid |= FATTR_MTIME_NOW;
1187 } else if (fvdat->flag & FN_MTIMECHANGE) {
1188 fsai->mtime = fvdat->cached_attrs.va_mtime.tv_sec;
1189 fsai->mtimensec = fvdat->cached_attrs.va_mtime.tv_nsec;
1190 fsai->valid |= FATTR_MTIME;
1192 if (fuse_libabi_geq(data, 7, 23) && fvdat->flag & FN_CTIMECHANGE) {
1193 fsai->ctime = fvdat->cached_attrs.va_ctime.tv_sec;
1194 fsai->ctimensec = fvdat->cached_attrs.va_ctime.tv_nsec;
1195 fsai->valid |= FATTR_CTIME;
1197 if (vap->va_mode != (mode_t)VNOVAL) {
1198 fsai->mode = vap->va_mode & ALLPERMS;
1199 fsai->valid |= FATTR_MODE;
1205 if ((err = fdisp_wait_answ(&fdi)))
1207 vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1209 if (vnode_vtype(vp) != vtyp) {
1210 if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1211 SDT_PROBE2(fusefs, , internal, trace, 1, "FUSE: Dang! "
1212 "vnode_vtype is VNON and vtype isn't.");
1215 * STALE vnode, ditch
1217 * The vnode has changed its type "behind our back".
1218 * There's nothing really we can do, so let us just
1219 * force an internal revocation and tell the caller to
1220 * try again, if interested.
1222 fuse_internal_vnode_disappear(vp);
1227 struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1228 fuse_vnode_undirty_cached_timestamps(vp);
1229 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1230 fao->attr_valid_nsec, NULL);
1234 fdisp_destroy(&fdi);
1239 * FreeBSD clears the SUID and SGID bits on any write by a non-root user.
1242 fuse_internal_clear_suid_on_write(struct vnode *vp, struct ucred *cred,
1245 struct fuse_data *data;
1250 mp = vnode_mount(vp);
1251 data = fuse_get_mpdata(mp);
1252 dataflags = data->dataflags;
1254 ASSERT_VOP_LOCKED(vp, __func__);
1256 if (dataflags & FSESS_DEFAULT_PERMISSIONS) {
1257 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1258 fuse_internal_getattr(vp, &va, cred, td);
1259 if (va.va_mode & (S_ISUID | S_ISGID)) {
1260 mode_t mode = va.va_mode & ~(S_ISUID | S_ISGID);
1261 /* Clear all vattr fields except mode */
1266 * Ignore fuse_internal_setattr's return value,
1267 * because at this point the write operation has
1268 * already succeeded and we don't want to return
1269 * failing status for that.
1271 (void)fuse_internal_setattr(vp, &va, td, NULL);
1277 #ifdef ZERO_PAD_INCOMPLETE_BUFS
1279 isbzero(void *buf, size_t len)
1283 for (i = 0; i < len; i++) {
1284 if (((char *)buf)[i])
1294 fuse_internal_init(void)
1296 fuse_lookup_cache_misses = counter_u64_alloc(M_WAITOK);
1297 fuse_lookup_cache_hits = counter_u64_alloc(M_WAITOK);
1301 fuse_internal_destroy(void)
1303 counter_u64_free(fuse_lookup_cache_hits);
1304 counter_u64_free(fuse_lookup_cache_misses);