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/module.h>
68 #include <sys/systm.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
73 #include <sys/malloc.h>
74 #include <sys/queue.h>
75 #include <sys/limits.h>
77 #include <sys/rwlock.h>
80 #include <sys/mount.h>
81 #include <sys/vnode.h>
82 #include <sys/namei.h>
83 #include <sys/extattr.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>
93 #include <sys/vmmeter.h>
96 #include <vm/vm_extern.h>
98 #include <vm/vm_map.h>
99 #include <vm/vm_page.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_pager.h>
103 #include <vm/vnode_pager.h>
104 #include <vm/vm_object.h>
107 #include "fuse_file.h"
108 #include "fuse_internal.h"
109 #include "fuse_ipc.h"
110 #include "fuse_node.h"
113 #include <sys/priv.h>
115 /* Maximum number of hardlinks to a single FUSE file */
116 #define FUSE_LINK_MAX UINT32_MAX
118 SDT_PROVIDER_DECLARE(fusefs);
121 * arg0: verbosity. Higher numbers give more verbose messages
122 * arg1: Textual message
124 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
127 static vop_access_t fuse_vnop_access;
128 static vop_advlock_t fuse_vnop_advlock;
129 static vop_bmap_t fuse_vnop_bmap;
130 static vop_close_t fuse_fifo_close;
131 static vop_close_t fuse_vnop_close;
132 static vop_create_t fuse_vnop_create;
133 static vop_deleteextattr_t fuse_vnop_deleteextattr;
134 static vop_fdatasync_t fuse_vnop_fdatasync;
135 static vop_fsync_t fuse_vnop_fsync;
136 static vop_getattr_t fuse_vnop_getattr;
137 static vop_getextattr_t fuse_vnop_getextattr;
138 static vop_inactive_t fuse_vnop_inactive;
139 static vop_link_t fuse_vnop_link;
140 static vop_listextattr_t fuse_vnop_listextattr;
141 static vop_lookup_t fuse_vnop_lookup;
142 static vop_mkdir_t fuse_vnop_mkdir;
143 static vop_mknod_t fuse_vnop_mknod;
144 static vop_open_t fuse_vnop_open;
145 static vop_pathconf_t fuse_vnop_pathconf;
146 static vop_read_t fuse_vnop_read;
147 static vop_readdir_t fuse_vnop_readdir;
148 static vop_readlink_t fuse_vnop_readlink;
149 static vop_reclaim_t fuse_vnop_reclaim;
150 static vop_remove_t fuse_vnop_remove;
151 static vop_rename_t fuse_vnop_rename;
152 static vop_rmdir_t fuse_vnop_rmdir;
153 static vop_setattr_t fuse_vnop_setattr;
154 static vop_setextattr_t fuse_vnop_setextattr;
155 static vop_strategy_t fuse_vnop_strategy;
156 static vop_symlink_t fuse_vnop_symlink;
157 static vop_write_t fuse_vnop_write;
158 static vop_getpages_t fuse_vnop_getpages;
159 static vop_print_t fuse_vnop_print;
160 static vop_vptofh_t fuse_vnop_vptofh;
162 struct vop_vector fuse_fifoops = {
163 .vop_default = &fifo_specops,
164 .vop_access = fuse_vnop_access,
165 .vop_close = fuse_fifo_close,
166 .vop_fsync = fuse_vnop_fsync,
167 .vop_getattr = fuse_vnop_getattr,
168 .vop_inactive = fuse_vnop_inactive,
169 .vop_pathconf = fuse_vnop_pathconf,
170 .vop_print = fuse_vnop_print,
171 .vop_read = VOP_PANIC,
172 .vop_reclaim = fuse_vnop_reclaim,
173 .vop_setattr = fuse_vnop_setattr,
174 .vop_write = VOP_PANIC,
175 .vop_vptofh = fuse_vnop_vptofh,
178 struct vop_vector fuse_vnops = {
179 .vop_allocate = VOP_EINVAL,
180 .vop_default = &default_vnodeops,
181 .vop_access = fuse_vnop_access,
182 .vop_advlock = fuse_vnop_advlock,
183 .vop_bmap = fuse_vnop_bmap,
184 .vop_close = fuse_vnop_close,
185 .vop_create = fuse_vnop_create,
186 .vop_deleteextattr = fuse_vnop_deleteextattr,
187 .vop_fsync = fuse_vnop_fsync,
188 .vop_fdatasync = fuse_vnop_fdatasync,
189 .vop_getattr = fuse_vnop_getattr,
190 .vop_getextattr = fuse_vnop_getextattr,
191 .vop_inactive = fuse_vnop_inactive,
193 * TODO: implement vop_ioctl after upgrading to protocol 7.16.
194 * FUSE_IOCTL was added in 7.11, but 32-bit compat is broken until
197 .vop_link = fuse_vnop_link,
198 .vop_listextattr = fuse_vnop_listextattr,
199 .vop_lookup = fuse_vnop_lookup,
200 .vop_mkdir = fuse_vnop_mkdir,
201 .vop_mknod = fuse_vnop_mknod,
202 .vop_open = fuse_vnop_open,
203 .vop_pathconf = fuse_vnop_pathconf,
205 * TODO: implement vop_poll after upgrading to protocol 7.21.
206 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
207 * 7.21, which adds the ability for the client to choose which poll
208 * events it wants, and for a client to deregister a file handle
210 .vop_read = fuse_vnop_read,
211 .vop_readdir = fuse_vnop_readdir,
212 .vop_readlink = fuse_vnop_readlink,
213 .vop_reclaim = fuse_vnop_reclaim,
214 .vop_remove = fuse_vnop_remove,
215 .vop_rename = fuse_vnop_rename,
216 .vop_rmdir = fuse_vnop_rmdir,
217 .vop_setattr = fuse_vnop_setattr,
218 .vop_setextattr = fuse_vnop_setextattr,
219 .vop_strategy = fuse_vnop_strategy,
220 .vop_symlink = fuse_vnop_symlink,
221 .vop_write = fuse_vnop_write,
222 .vop_getpages = fuse_vnop_getpages,
223 .vop_print = fuse_vnop_print,
224 .vop_vptofh = fuse_vnop_vptofh,
227 uma_zone_t fuse_pbuf_zone;
229 /* Check permission for extattr operations, much like extattr_check_cred */
231 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
232 struct thread *td, accmode_t accmode)
234 struct mount *mp = vnode_mount(vp);
235 struct fuse_data *data = fuse_get_mpdata(mp);
238 * Kernel-invoked always succeeds.
244 * Do not allow privileged processes in jail to directly manipulate
248 case EXTATTR_NAMESPACE_SYSTEM:
249 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
250 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
253 case EXTATTR_NAMESPACE_USER:
254 return (fuse_internal_access(vp, accmode, td, cred));
260 /* Get a filehandle for a directory */
262 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
263 struct ucred *cred, pid_t pid)
265 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
267 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
270 /* Send FUSE_FLUSH for this vnode */
272 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
274 struct fuse_flush_in *ffi;
275 struct fuse_filehandle *fufh;
276 struct fuse_dispatcher fdi;
277 struct thread *td = curthread;
278 struct mount *mp = vnode_mount(vp);
281 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
284 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
288 fdisp_init(&fdi, sizeof(*ffi));
289 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
291 ffi->fh = fufh->fh_id;
293 * If the file has a POSIX lock then we're supposed to set lock_owner.
294 * If not, then lock_owner is undefined. So we may as well always set
297 ffi->lock_owner = td->td_proc->p_pid;
299 err = fdisp_wait_answ(&fdi);
301 fsess_set_notimpl(mp, FUSE_FLUSH);
308 /* Close wrapper for fifos. */
310 fuse_fifo_close(struct vop_close_args *ap)
312 return (fifo_specops.vop_close(ap));
316 struct vnop_access_args {
318 #if VOP_ACCESS_TAKES_ACCMODE_T
323 struct ucred *a_cred;
328 fuse_vnop_access(struct vop_access_args *ap)
330 struct vnode *vp = ap->a_vp;
331 int accmode = ap->a_accmode;
332 struct ucred *cred = ap->a_cred;
334 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
338 if (fuse_isdeadfs(vp)) {
339 if (vnode_isvroot(vp)) {
344 if (!(data->dataflags & FSESS_INITED)) {
345 if (vnode_isvroot(vp)) {
346 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
347 (fuse_match_cred(data->daemoncred, cred) == 0)) {
353 if (vnode_islnk(vp)) {
357 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
362 * struct vop_advlock_args {
363 * struct vop_generic_args a_gen;
364 * struct vnode *a_vp;
367 * struct flock *a_fl;
372 fuse_vnop_advlock(struct vop_advlock_args *ap)
374 struct vnode *vp = ap->a_vp;
375 struct flock *fl = ap->a_fl;
376 struct thread *td = curthread;
377 struct ucred *cred = td->td_ucred;
378 pid_t pid = td->td_proc->p_pid;
379 struct fuse_filehandle *fufh;
380 struct fuse_dispatcher fdi;
381 struct fuse_lk_in *fli;
382 struct fuse_lk_out *flo;
385 int flags = ap->a_flags;
387 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
389 if (fuse_isdeadfs(vp)) {
393 if (!(dataflags & FSESS_POSIX_LOCKS))
394 return vop_stdadvlock(ap);
395 /* FUSE doesn't properly support flock until protocol 7.17 */
397 return vop_stdadvlock(ap);
399 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
403 fdisp_init(&fdi, sizeof(*fli));
419 fdisp_make_vp(&fdi, op, vp, td, cred);
421 fli->fh = fufh->fh_id;
422 fli->owner = fl->l_pid;
423 fli->lk.start = fl->l_start;
425 fli->lk.end = fl->l_start + fl->l_len - 1;
427 fli->lk.end = INT64_MAX;
428 fli->lk.type = fl->l_type;
429 fli->lk.pid = fl->l_pid;
431 err = fdisp_wait_answ(&fdi);
434 if (err == 0 && op == FUSE_GETLK) {
436 fl->l_type = flo->lk.type;
437 fl->l_pid = flo->lk.pid;
438 if (flo->lk.type != F_UNLCK) {
439 fl->l_start = flo->lk.start;
440 if (flo->lk.end == INT64_MAX)
443 fl->l_len = flo->lk.end - flo->lk.start + 1;
444 fl->l_start = flo->lk.start;
454 struct bufobj **a_bop;
460 fuse_vnop_bmap(struct vop_bmap_args *ap)
462 struct vnode *vp = ap->a_vp;
463 struct bufobj **bo = ap->a_bop;
464 struct thread *td = curthread;
466 struct fuse_dispatcher fdi;
467 struct fuse_bmap_in *fbi;
468 struct fuse_bmap_out *fbo;
469 struct fuse_data *data;
472 daddr_t lbn = ap->a_bn;
473 daddr_t *pbn = ap->a_bnp;
474 int *runp = ap->a_runp;
475 int *runb = ap->a_runb;
479 if (fuse_isdeadfs(vp)) {
483 mp = vnode_mount(vp);
484 data = fuse_get_mpdata(mp);
485 biosize = fuse_iosize(vp);
486 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
487 data->max_readahead_blocks);
493 * The FUSE_BMAP operation does not include the runp and runb
494 * variables, so we must guess. Report nonzero contiguous runs so
495 * cluster_read will combine adjacent reads. It's worthwhile to reduce
496 * upcalls even if we don't know the true physical layout of the file.
498 * FUSE file systems may opt out of read clustering in two ways:
499 * * mounting with -onoclusterr
500 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
503 *runb = MIN(lbn, maxrun);
505 error = fuse_vnode_size(vp, &filesize, td->td_ucred, td);
507 *runp = MIN(MAX(0, filesize / (off_t)biosize - lbn - 1),
513 if (fsess_isimpl(mp, FUSE_BMAP)) {
514 fdisp_init(&fdi, sizeof(*fbi));
515 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
518 fbi->blocksize = biosize;
519 error = fdisp_wait_answ(&fdi);
520 if (error == ENOSYS) {
522 fsess_set_notimpl(mp, FUSE_BMAP);
526 if (error == 0 && pbn != NULL)
533 /* If the daemon doesn't support BMAP, make up a sensible default */
535 *pbn = lbn * btodb(biosize);
540 struct vop_close_args {
543 struct ucred *a_cred;
548 fuse_vnop_close(struct vop_close_args *ap)
550 struct vnode *vp = ap->a_vp;
551 struct ucred *cred = ap->a_cred;
552 int fflag = ap->a_fflag;
553 struct thread *td = ap->a_td;
554 pid_t pid = td->td_proc->p_pid;
557 if (fuse_isdeadfs(vp))
561 if (fflag & IO_NDELAY)
564 err = fuse_flush(vp, cred, pid, fflag);
565 /* TODO: close the file handle, if we're sure it's no longer used */
566 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
567 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
573 fdisp_make_mknod_for_fallback(
574 struct fuse_dispatcher *fdip,
575 struct componentname *cnp,
581 enum fuse_opcode *op)
583 struct fuse_mknod_in *fmni;
585 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
587 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
591 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
593 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
596 struct vnop_create_args {
598 struct vnode **a_vpp;
599 struct componentname *a_cnp;
604 fuse_vnop_create(struct vop_create_args *ap)
606 struct vnode *dvp = ap->a_dvp;
607 struct vnode **vpp = ap->a_vpp;
608 struct componentname *cnp = ap->a_cnp;
609 struct vattr *vap = ap->a_vap;
610 struct thread *td = cnp->cn_thread;
611 struct ucred *cred = cnp->cn_cred;
613 struct fuse_data *data;
614 struct fuse_create_in *fci;
615 struct fuse_entry_out *feo;
616 struct fuse_open_out *foo;
617 struct fuse_dispatcher fdi, fdi2;
618 struct fuse_dispatcher *fdip = &fdi;
619 struct fuse_dispatcher *fdip2 = NULL;
623 struct mount *mp = vnode_mount(dvp);
624 data = fuse_get_mpdata(mp);
625 uint64_t parentnid = VTOFUD(dvp)->nid;
626 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
630 if (fuse_isdeadfs(dvp))
633 /* FUSE expects sockets to be created with FUSE_MKNOD */
634 if (vap->va_type == VSOCK)
635 return fuse_internal_mknod(dvp, vpp, cnp, vap);
638 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
639 * writable mode makes sense, and we might as well include readability
644 bzero(&fdi, sizeof(fdi));
646 if (vap->va_type != VREG)
649 if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
650 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
651 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
654 /* Use FUSE_CREATE */
658 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
659 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
662 fci->flags = O_CREAT | flags;
663 if (fuse_libabi_geq(data, 7, 12)) {
664 insize = sizeof(*fci);
665 fci->umask = td->td_proc->p_fd->fd_cmask;
667 insize = sizeof(struct fuse_open_in);
670 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
672 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
675 err = fdisp_wait_answ(fdip);
678 if (err == ENOSYS && op == FUSE_CREATE) {
679 fsess_set_notimpl(mp, FUSE_CREATE);
681 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
682 parentnid, td, cred, mode, &op);
683 err = fdisp_wait_answ(fdip);
691 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
695 if (op == FUSE_CREATE) {
696 foo = (struct fuse_open_out*)(feo + 1);
698 /* Issue a separate FUSE_OPEN */
699 struct fuse_open_in *foi;
702 fdisp_init(fdip2, sizeof(*foi));
703 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
707 err = fdisp_wait_answ(fdip2);
712 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
714 struct fuse_release_in *fri;
715 uint64_t nodeid = feo->nodeid;
716 uint64_t fh_id = foo->fh;
718 fdisp_init(fdip, sizeof(*fri));
719 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
723 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
724 fuse_insert_message(fdip->tick, false);
727 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
728 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
729 feo->attr_valid_nsec, NULL);
731 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
732 fuse_vnode_open(*vpp, foo->open_flags, td);
734 * Purge the parent's attribute cache because the daemon should've
735 * updated its mtime and ctime
737 fuse_vnode_clear_attr_cache(dvp);
738 cache_purge_negative(dvp);
742 fdisp_destroy(fdip2);
748 struct vnop_fdatasync_args {
749 struct vop_generic_args a_gen;
751 struct thread * a_td;
755 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
757 struct vnode *vp = ap->a_vp;
758 struct thread *td = ap->a_td;
759 int waitfor = MNT_WAIT;
763 if (fuse_isdeadfs(vp)) {
766 if ((err = vop_stdfdatasync_buf(ap)))
769 return fuse_internal_fsync(vp, td, waitfor, true);
773 struct vnop_fsync_args {
774 struct vop_generic_args a_gen;
777 struct thread * a_td;
781 fuse_vnop_fsync(struct vop_fsync_args *ap)
783 struct vnode *vp = ap->a_vp;
784 struct thread *td = ap->a_td;
785 int waitfor = ap->a_waitfor;
788 if (fuse_isdeadfs(vp)) {
791 if ((err = vop_stdfsync(ap)))
794 return fuse_internal_fsync(vp, td, waitfor, false);
798 struct vnop_getattr_args {
801 struct ucred *a_cred;
806 fuse_vnop_getattr(struct vop_getattr_args *ap)
808 struct vnode *vp = ap->a_vp;
809 struct vattr *vap = ap->a_vap;
810 struct ucred *cred = ap->a_cred;
811 struct thread *td = curthread;
816 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
818 /* Note that we are not bailing out on a dead file system just yet. */
820 if (!(dataflags & FSESS_INITED)) {
821 if (!vnode_isvroot(vp)) {
822 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
829 err = fuse_internal_getattr(vp, vap, cred, td);
830 if (err == ENOTCONN && vnode_isvroot(vp)) {
831 /* see comment in fuse_vfsop_statfs() */
838 bzero(vap, sizeof(*vap));
839 vap->va_type = vnode_vtype(vp);
845 struct vnop_inactive_args {
851 fuse_vnop_inactive(struct vop_inactive_args *ap)
853 struct vnode *vp = ap->a_vp;
854 struct thread *td = ap->a_td;
856 struct fuse_vnode_data *fvdat = VTOFUD(vp);
857 struct fuse_filehandle *fufh, *fufh_tmp;
861 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
862 if (need_flush && vp->v_type == VREG) {
863 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
864 fuse_vnode_savesize(vp, NULL, 0);
866 if ((fvdat->flag & FN_REVOKED) != 0)
867 fuse_io_invalbuf(vp, td);
869 fuse_io_flushbuf(vp, MNT_WAIT, td);
872 fuse_filehandle_close(vp, fufh, td, NULL);
875 if ((fvdat->flag & FN_REVOKED) != 0)
882 struct vnop_link_args {
883 struct vnode *a_tdvp;
885 struct componentname *a_cnp;
889 fuse_vnop_link(struct vop_link_args *ap)
891 struct vnode *vp = ap->a_vp;
892 struct vnode *tdvp = ap->a_tdvp;
893 struct componentname *cnp = ap->a_cnp;
895 struct vattr *vap = VTOVA(vp);
897 struct fuse_dispatcher fdi;
898 struct fuse_entry_out *feo;
899 struct fuse_link_in fli;
903 if (fuse_isdeadfs(vp)) {
906 if (vnode_mount(tdvp) != vnode_mount(vp)) {
911 * This is a seatbelt check to protect naive userspace filesystems from
912 * themselves and the limitations of the FUSE IPC protocol. If a
913 * filesystem does not allow attribute caching, assume it is capable of
914 * validating that nlink does not overflow.
916 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
918 fli.oldnodeid = VTOI(vp);
921 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
922 FUSE_LINK, &fli, sizeof(fli), &fdi);
923 if ((err = fdisp_wait_answ(&fdi))) {
928 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
931 * Purge the parent's attribute cache because the daemon
932 * should've updated its mtime and ctime
934 fuse_vnode_clear_attr_cache(tdvp);
935 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
936 feo->attr_valid_nsec, NULL);
943 struct fuse_lookup_alloc_arg {
944 struct fuse_entry_out *feo;
945 struct componentname *cnp;
950 /* Callback for vn_get_ino */
952 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
954 struct fuse_lookup_alloc_arg *flaa = arg;
956 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
960 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
961 "int", "struct timespec*", "struct timespec*");
963 struct vnop_lookup_args {
964 struct vnodeop_desc *a_desc;
966 struct vnode **a_vpp;
967 struct componentname *a_cnp;
971 fuse_vnop_lookup(struct vop_lookup_args *ap)
973 struct vnode *dvp = ap->a_dvp;
974 struct vnode **vpp = ap->a_vpp;
975 struct componentname *cnp = ap->a_cnp;
976 struct thread *td = cnp->cn_thread;
977 struct ucred *cred = cnp->cn_cred;
979 int nameiop = cnp->cn_nameiop;
980 int flags = cnp->cn_flags;
981 int wantparent = flags & (LOCKPARENT | WANTPARENT);
982 int islastcn = flags & ISLASTCN;
983 struct mount *mp = vnode_mount(dvp);
987 struct vnode *vp = NULL;
989 struct fuse_dispatcher fdi;
990 bool did_lookup = false;
991 struct fuse_entry_out *feo = NULL;
992 enum vtype vtyp; /* vnode type of target */
993 off_t filesize; /* filesize of target */
997 if (fuse_isdeadfs(dvp)) {
1001 if (!vnode_isdir(dvp))
1004 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1007 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1010 if (flags & ISDOTDOT) {
1011 KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1012 ("Looking up .. is TODO"));
1013 nid = VTOFUD(dvp)->parent_nid;
1016 /* .. is obviously a directory */
1019 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1021 /* . is obviously a directory */
1025 struct timespec now, timeout;
1027 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
1028 getnanouptime(&now);
1029 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1031 case -1: /* positive match */
1032 if (timespeccmp(&timeout, &now, >)) {
1033 counter_u64_add(fuse_lookup_cache_hits, 1);
1036 counter_u64_add(fuse_lookup_cache_misses, 1);
1038 &VTOFUD(*vpp)->entry_cache_timeout);
1049 case 0: /* no match in cache */
1050 counter_u64_add(fuse_lookup_cache_misses, 1);
1053 case ENOENT: /* negative match */
1054 getnanouptime(&now);
1055 if (timespeccmp(&timeout, &now, <=)) {
1057 cache_purge_negative(dvp);
1066 fdisp_init(&fdi, cnp->cn_namelen + 1);
1067 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1069 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1070 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1071 lookup_err = fdisp_wait_answ(&fdi);
1075 /* lookup call succeeded */
1076 feo = (struct fuse_entry_out *)fdi.answ;
1079 /* zero nodeid means ENOENT and cache it */
1080 struct timespec timeout;
1082 fdi.answ_stat = ENOENT;
1083 lookup_err = ENOENT;
1084 if (cnp->cn_flags & MAKEENTRY) {
1085 fuse_validity_2_timespec(feo, &timeout);
1086 cache_enter_time(dvp, *vpp, cnp,
1089 } else if (nid == FUSE_ROOT_ID) {
1090 lookup_err = EINVAL;
1092 vtyp = IFTOVT(feo->attr.mode);
1093 filesize = feo->attr.size;
1095 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1096 fdisp_destroy(&fdi);
1100 /* lookup_err, if non-zero, must be ENOENT at this point */
1103 /* Entry not found */
1104 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1105 err = fuse_internal_access(dvp, VWRITE, td, cred);
1108 * Set the SAVENAME flag to hold onto the
1109 * pathname for use later in VOP_CREATE or
1112 cnp->cn_flags |= SAVENAME;
1120 /* Entry was found */
1121 if (flags & ISDOTDOT) {
1122 struct fuse_lookup_alloc_arg flaa;
1128 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1131 } else if (nid == VTOI(dvp)) {
1135 struct fuse_vnode_data *fvdat;
1137 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1144 * In the case where we are looking up a FUSE node
1145 * represented by an existing cached vnode, and the
1146 * true size reported by FUSE_LOOKUP doesn't match
1147 * the vnode's cached size, then any cached writes
1148 * beyond the file's current size are lost.
1151 * * following attribute cache expiration, or
1152 * * due a bug in the daemon, or
1155 if (vnode_isreg(vp) &&
1156 filesize != fvdat->cached_attrs.va_size &&
1157 fvdat->flag & FN_SIZECHANGE) {
1159 * The FN_SIZECHANGE flag reflects a dirty
1160 * append. If userspace lets us know our cache
1161 * is invalid, that write was lost. (Dirty
1162 * writes that do not cause append are also
1163 * lost, but we don't detect them here.)
1165 * XXX: Maybe disable WB caching on this mount.
1167 printf("%s: WB cache incoherent on %s!\n",
1169 vnode_mount(vp)->mnt_stat.f_mntonname);
1171 fvdat->flag &= ~FN_SIZECHANGE;
1175 fuse_internal_cache_attrs(*vpp, &feo->attr,
1176 feo->attr_valid, feo->attr_valid_nsec, NULL);
1177 fuse_validity_2_bintime(feo->entry_valid,
1178 feo->entry_valid_nsec,
1179 &fvdat->entry_cache_timeout);
1181 if ((nameiop == DELETE || nameiop == RENAME) &&
1184 struct vattr dvattr;
1186 err = fuse_internal_access(dvp, VWRITE, td,
1191 * if the parent's sticky bit is set, check
1192 * whether we're allowed to remove the file.
1193 * Need to figure out the vnode locking to make
1196 fuse_internal_getattr(dvp, &dvattr, cred, td);
1197 if ((dvattr.va_mode & S_ISTXT) &&
1198 fuse_internal_access(dvp, VADMIN, td,
1200 fuse_internal_access(*vpp, VADMIN, td,
1208 (nameiop == DELETE) ||
1209 (nameiop == RENAME && wantparent))) {
1210 cnp->cn_flags |= SAVENAME;
1217 if (vp != NULL && dvp != vp)
1219 else if (vp != NULL)
1224 fdisp_destroy(&fdi);
1230 struct vnop_mkdir_args {
1231 struct vnode *a_dvp;
1232 struct vnode **a_vpp;
1233 struct componentname *a_cnp;
1234 struct vattr *a_vap;
1238 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1240 struct vnode *dvp = ap->a_dvp;
1241 struct vnode **vpp = ap->a_vpp;
1242 struct componentname *cnp = ap->a_cnp;
1243 struct vattr *vap = ap->a_vap;
1245 struct fuse_mkdir_in fmdi;
1247 if (fuse_isdeadfs(dvp)) {
1250 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1251 fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1253 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1254 sizeof(fmdi), VDIR));
1258 struct vnop_mknod_args {
1259 struct vnode *a_dvp;
1260 struct vnode **a_vpp;
1261 struct componentname *a_cnp;
1262 struct vattr *a_vap;
1266 fuse_vnop_mknod(struct vop_mknod_args *ap)
1269 struct vnode *dvp = ap->a_dvp;
1270 struct vnode **vpp = ap->a_vpp;
1271 struct componentname *cnp = ap->a_cnp;
1272 struct vattr *vap = ap->a_vap;
1274 if (fuse_isdeadfs(dvp))
1277 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1281 struct vop_open_args {
1284 struct ucred *a_cred;
1285 struct thread *a_td;
1286 int a_fdidx; / struct file *a_fp;
1290 fuse_vnop_open(struct vop_open_args *ap)
1292 struct vnode *vp = ap->a_vp;
1293 int a_mode = ap->a_mode;
1294 struct thread *td = ap->a_td;
1295 struct ucred *cred = ap->a_cred;
1296 pid_t pid = td->td_proc->p_pid;
1297 struct fuse_vnode_data *fvdat;
1299 if (fuse_isdeadfs(vp))
1301 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1302 return (EOPNOTSUPP);
1303 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1308 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1309 fuse_vnode_open(vp, 0, td);
1313 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1317 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1320 switch (ap->a_name) {
1321 case _PC_FILESIZEBITS:
1325 *ap->a_retval = NAME_MAX;
1328 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1330 case _PC_SYMLINK_MAX:
1331 *ap->a_retval = MAXPATHLEN;
1337 return (vop_stdpathconf(ap));
1342 struct vnop_read_args {
1346 struct ucred *a_cred;
1350 fuse_vnop_read(struct vop_read_args *ap)
1352 struct vnode *vp = ap->a_vp;
1353 struct uio *uio = ap->a_uio;
1354 int ioflag = ap->a_ioflag;
1355 struct ucred *cred = ap->a_cred;
1356 pid_t pid = curthread->td_proc->p_pid;
1358 if (fuse_isdeadfs(vp)) {
1362 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1363 ioflag |= IO_DIRECT;
1366 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1370 struct vnop_readdir_args {
1373 struct ucred *a_cred;
1380 fuse_vnop_readdir(struct vop_readdir_args *ap)
1382 struct vnode *vp = ap->a_vp;
1383 struct uio *uio = ap->a_uio;
1384 struct ucred *cred = ap->a_cred;
1385 struct fuse_filehandle *fufh = NULL;
1386 struct fuse_iov cookediov;
1392 bool closefufh = false;
1393 pid_t pid = curthread->td_proc->p_pid;
1397 if (fuse_isdeadfs(vp)) {
1400 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1401 (uio_resid(uio) < sizeof(struct dirent))) {
1405 tresid = uio->uio_resid;
1406 startoff = uio->uio_offset;
1407 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1408 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1410 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1411 * must implicitly open the directory here
1413 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1416 * When a directory is opened, it must be read from
1417 * the beginning. Hopefully, the "startoff" still
1418 * exists as an offset cookie for the directory.
1419 * If not, it will read the entire directory without
1420 * returning any entries and just return eof.
1422 uio->uio_offset = 0;
1428 if (ap->a_ncookies != NULL) {
1429 ncookies = uio->uio_resid /
1430 (offsetof(struct dirent, d_name) + 4) + 1;
1431 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1432 *ap->a_ncookies = ncookies;
1433 *ap->a_cookies = cookies;
1438 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1439 fiov_init(&cookediov, DIRCOOKEDSIZE);
1441 err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1442 &ncookies, cookies);
1444 fiov_teardown(&cookediov);
1446 fuse_filehandle_close(vp, fufh, curthread, cred);
1448 if (ap->a_ncookies != NULL) {
1450 *ap->a_ncookies -= ncookies;
1452 free(*ap->a_cookies, M_TEMP);
1453 *ap->a_ncookies = 0;
1454 *ap->a_cookies = NULL;
1457 if (err == 0 && tresid == uio->uio_resid)
1464 struct vnop_readlink_args {
1467 struct ucred *a_cred;
1471 fuse_vnop_readlink(struct vop_readlink_args *ap)
1473 struct vnode *vp = ap->a_vp;
1474 struct uio *uio = ap->a_uio;
1475 struct ucred *cred = ap->a_cred;
1477 struct fuse_dispatcher fdi;
1480 if (fuse_isdeadfs(vp)) {
1483 if (!vnode_islnk(vp)) {
1486 fdisp_init(&fdi, 0);
1487 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1491 if (((char *)fdi.answ)[0] == '/' &&
1492 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1493 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1495 err = uiomove(mpth, strlen(mpth), uio);
1498 err = uiomove(fdi.answ, fdi.iosize, uio);
1501 fdisp_destroy(&fdi);
1506 struct vnop_reclaim_args {
1508 struct thread *a_td;
1512 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1514 struct vnode *vp = ap->a_vp;
1515 struct thread *td = ap->a_td;
1516 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1517 struct fuse_filehandle *fufh, *fufh_tmp;
1520 panic("FUSE: no vnode data during recycling");
1522 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1523 printf("FUSE: vnode being reclaimed with open fufh "
1524 "(type=%#x)", fufh->fufh_type);
1525 fuse_filehandle_close(vp, fufh, td, NULL);
1528 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1529 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1532 fuse_vnode_setparent(vp, NULL);
1534 vfs_hash_remove(vp);
1535 fuse_vnode_destroy(vp);
1541 struct vnop_remove_args {
1542 struct vnode *a_dvp;
1544 struct componentname *a_cnp;
1548 fuse_vnop_remove(struct vop_remove_args *ap)
1550 struct vnode *dvp = ap->a_dvp;
1551 struct vnode *vp = ap->a_vp;
1552 struct componentname *cnp = ap->a_cnp;
1556 if (fuse_isdeadfs(vp)) {
1559 if (vnode_isdir(vp)) {
1563 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1569 struct vnop_rename_args {
1570 struct vnode *a_fdvp;
1571 struct vnode *a_fvp;
1572 struct componentname *a_fcnp;
1573 struct vnode *a_tdvp;
1574 struct vnode *a_tvp;
1575 struct componentname *a_tcnp;
1579 fuse_vnop_rename(struct vop_rename_args *ap)
1581 struct vnode *fdvp = ap->a_fdvp;
1582 struct vnode *fvp = ap->a_fvp;
1583 struct componentname *fcnp = ap->a_fcnp;
1584 struct vnode *tdvp = ap->a_tdvp;
1585 struct vnode *tvp = ap->a_tvp;
1586 struct componentname *tcnp = ap->a_tcnp;
1587 struct fuse_data *data;
1588 bool newparent = fdvp != tdvp;
1589 bool isdir = fvp->v_type == VDIR;
1592 if (fuse_isdeadfs(fdvp)) {
1595 if (fvp->v_mount != tdvp->v_mount ||
1596 (tvp && fvp->v_mount != tvp->v_mount)) {
1597 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1604 * FUSE library is expected to check if target directory is not
1605 * under the source directory in the file system tree.
1606 * Linux performs this check at VFS level.
1609 * If source is a directory, and it will get a new parent, user must
1610 * have write permission to it, so ".." can be modified.
1612 data = fuse_get_mpdata(vnode_mount(tdvp));
1613 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1614 err = fuse_internal_access(fvp, VWRITE,
1615 tcnp->cn_thread, tcnp->cn_cred);
1619 sx_xlock(&data->rename_lock);
1620 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1623 fuse_vnode_setparent(fvp, tdvp);
1625 fuse_vnode_setparent(tvp, NULL);
1627 sx_unlock(&data->rename_lock);
1629 if (tvp != NULL && tvp != fvp) {
1632 if (vnode_isdir(fvp)) {
1633 if ((tvp != NULL) && vnode_isdir(tvp)) {
1654 struct vnop_rmdir_args {
1655 struct vnode *a_dvp;
1657 struct componentname *a_cnp;
1661 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1663 struct vnode *dvp = ap->a_dvp;
1664 struct vnode *vp = ap->a_vp;
1668 if (fuse_isdeadfs(vp)) {
1671 if (VTOFUD(vp) == VTOFUD(dvp)) {
1674 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1680 struct vnop_setattr_args {
1682 struct vattr *a_vap;
1683 struct ucred *a_cred;
1684 struct thread *a_td;
1688 fuse_vnop_setattr(struct vop_setattr_args *ap)
1690 struct vnode *vp = ap->a_vp;
1691 struct vattr *vap = ap->a_vap;
1692 struct ucred *cred = ap->a_cred;
1693 struct thread *td = curthread;
1695 struct fuse_data *data;
1696 struct vattr old_va;
1699 accmode_t accmode = 0;
1701 bool drop_suid = false;
1704 mp = vnode_mount(vp);
1705 data = fuse_get_mpdata(mp);
1706 dataflags = data->dataflags;
1707 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1708 if (cred->cr_ngroups > 0)
1709 cr_gid = cred->cr_groups[0];
1713 if (fuse_isdeadfs(vp)) {
1717 if (vap->va_uid != (uid_t)VNOVAL) {
1719 /* Only root may change a file's owner */
1720 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1722 /* As a special case, allow the null chown */
1723 err2 = fuse_internal_getattr(vp, &old_va, cred,
1727 if (vap->va_uid != old_va.va_uid)
1737 if (vap->va_gid != (gid_t)VNOVAL) {
1738 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1740 if (checkperm && !groupmember(vap->va_gid, cred))
1743 * Non-root users may only chgrp to one of their own
1746 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1748 /* As a special case, allow the null chgrp */
1749 err2 = fuse_internal_getattr(vp, &old_va, cred,
1753 if (vap->va_gid != old_va.va_gid)
1761 if (vap->va_size != VNOVAL) {
1762 switch (vp->v_type) {
1767 if (vfs_isrdonly(mp))
1772 * According to POSIX, the result is unspecified
1773 * for file types other than regular files,
1774 * directories and shared memory objects. We
1775 * don't support shared memory objects in the file
1776 * system, and have dubious support for truncating
1777 * symlinks. Just ignore the request in other cases.
1781 /* Don't set accmode. Permission to trunc is checked upstack */
1783 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1784 if (vap->va_vaflags & VA_UTIMES_NULL)
1790 if (vap->va_mode != (mode_t)VNOVAL)
1791 vap->va_mode &= ~(S_ISUID | S_ISGID);
1793 err = fuse_internal_getattr(vp, &old_va, cred, td);
1796 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1799 if (vap->va_mode != (mode_t)VNOVAL) {
1800 /* Only root may set the sticky bit on non-directories */
1801 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1802 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1804 if (checkperm && (vap->va_mode & S_ISGID)) {
1805 err = fuse_internal_getattr(vp, &old_va, cred, td);
1808 if (!groupmember(old_va.va_gid, cred)) {
1809 err = priv_check_cred(cred, PRIV_VFS_SETGID);
1817 if (vfs_isrdonly(mp))
1820 err = fuse_internal_access(vp, accmode, td, cred);
1824 return fuse_internal_setattr(vp, vap, td, cred);
1828 struct vnop_strategy_args {
1834 fuse_vnop_strategy(struct vop_strategy_args *ap)
1836 struct vnode *vp = ap->a_vp;
1837 struct buf *bp = ap->a_bp;
1839 if (!vp || fuse_isdeadfs(vp)) {
1840 bp->b_ioflags |= BIO_ERROR;
1841 bp->b_error = ENXIO;
1847 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1848 * fuse_io_strategy sets bp's error fields
1850 (void)fuse_io_strategy(vp, bp);
1857 struct vnop_symlink_args {
1858 struct vnode *a_dvp;
1859 struct vnode **a_vpp;
1860 struct componentname *a_cnp;
1861 struct vattr *a_vap;
1866 fuse_vnop_symlink(struct vop_symlink_args *ap)
1868 struct vnode *dvp = ap->a_dvp;
1869 struct vnode **vpp = ap->a_vpp;
1870 struct componentname *cnp = ap->a_cnp;
1871 const char *target = ap->a_target;
1873 struct fuse_dispatcher fdi;
1878 if (fuse_isdeadfs(dvp)) {
1882 * Unlike the other creator type calls, here we have to create a message
1883 * where the name of the new entry comes first, and the data describing
1884 * the entry comes second.
1885 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1886 * but put together the message manually and just call the core part.
1889 len = strlen(target) + 1;
1890 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1891 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1893 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1894 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1895 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1897 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1898 fdisp_destroy(&fdi);
1903 struct vnop_write_args {
1907 struct ucred *a_cred;
1911 fuse_vnop_write(struct vop_write_args *ap)
1913 struct vnode *vp = ap->a_vp;
1914 struct uio *uio = ap->a_uio;
1915 int ioflag = ap->a_ioflag;
1916 struct ucred *cred = ap->a_cred;
1917 pid_t pid = curthread->td_proc->p_pid;
1919 if (fuse_isdeadfs(vp)) {
1923 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1924 ioflag |= IO_DIRECT;
1927 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1931 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1933 const int biosize = fuse_iosize(vp);
1935 return (off / biosize);
1939 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1943 const int biosize = fuse_iosize(vp);
1945 err = fuse_vnode_size(vp, &filesize, NULL, NULL);
1946 KASSERT(err == 0, ("vfs_bio_getpages can't handle errors here"));
1950 if ((off_t)lbn * biosize >= filesize) {
1952 } else if ((off_t)(lbn + 1) * biosize > filesize) {
1953 blksz = filesize - (off_t)lbn *biosize;
1961 struct vnop_getpages_args {
1969 fuse_vnop_getpages(struct vop_getpages_args *ap)
1971 struct vnode *vp = ap->a_vp;
1973 if (!fsess_opt_mmap(vnode_mount(vp))) {
1974 SDT_PROBE2(fusefs, , vnops, trace, 1,
1975 "called on non-cacheable vnode??\n");
1976 return (VM_PAGER_ERROR);
1979 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1980 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
1983 static const char extattr_namespace_separator = '.';
1986 struct vop_getextattr_args {
1987 struct vop_generic_args a_gen;
1989 int a_attrnamespace;
1993 struct ucred *a_cred;
1994 struct thread *a_td;
1998 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2000 struct vnode *vp = ap->a_vp;
2001 struct uio *uio = ap->a_uio;
2002 struct fuse_dispatcher fdi;
2003 struct fuse_getxattr_in *get_xattr_in;
2004 struct fuse_getxattr_out *get_xattr_out;
2005 struct mount *mp = vnode_mount(vp);
2006 struct thread *td = ap->a_td;
2007 struct ucred *cred = ap->a_cred;
2013 if (fuse_isdeadfs(vp))
2016 if (!fsess_isimpl(mp, FUSE_GETXATTR))
2019 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2023 /* Default to looking for user attributes. */
2024 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2025 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2027 prefix = EXTATTR_NAMESPACE_USER_STRING;
2029 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2030 strlen(ap->a_name) + 1;
2032 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2033 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2035 get_xattr_in = fdi.indata;
2037 * Check to see whether we're querying the available size or
2038 * issuing the actual request. If we pass in 0, we get back struct
2039 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2040 * that much data, without the struct fuse_getxattr_out header.
2043 get_xattr_in->size = 0;
2045 get_xattr_in->size = uio->uio_resid;
2047 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2048 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2051 err = fdisp_wait_answ(&fdi);
2053 if (err == ENOSYS) {
2054 fsess_set_notimpl(mp, FUSE_GETXATTR);
2060 get_xattr_out = fdi.answ;
2062 if (ap->a_size != NULL)
2063 *ap->a_size = get_xattr_out->size;
2066 err = uiomove(fdi.answ, fdi.iosize, uio);
2069 fdisp_destroy(&fdi);
2074 struct vop_setextattr_args {
2075 struct vop_generic_args a_gen;
2077 int a_attrnamespace;
2080 struct ucred *a_cred;
2081 struct thread *a_td;
2085 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2087 struct vnode *vp = ap->a_vp;
2088 struct uio *uio = ap->a_uio;
2089 struct fuse_dispatcher fdi;
2090 struct fuse_setxattr_in *set_xattr_in;
2091 struct mount *mp = vnode_mount(vp);
2092 struct thread *td = ap->a_td;
2093 struct ucred *cred = ap->a_cred;
2099 if (fuse_isdeadfs(vp))
2102 if (!fsess_isimpl(mp, FUSE_SETXATTR))
2105 if (vfs_isrdonly(mp))
2108 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2109 if (ap->a_uio == NULL) {
2111 * If we got here as fallback from VOP_DELETEEXTATTR, then
2112 * return EOPNOTSUPP.
2114 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2115 return (EOPNOTSUPP);
2120 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2125 /* Default to looking for user attributes. */
2126 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2127 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2129 prefix = EXTATTR_NAMESPACE_USER_STRING;
2131 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2132 strlen(ap->a_name) + 1;
2134 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2135 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2137 set_xattr_in = fdi.indata;
2138 set_xattr_in->size = uio->uio_resid;
2140 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2141 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2144 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2145 uio->uio_resid, uio);
2150 err = fdisp_wait_answ(&fdi);
2152 if (err == ENOSYS) {
2153 fsess_set_notimpl(mp, FUSE_SETXATTR);
2156 if (err == ERESTART) {
2157 /* Can't restart after calling uiomove */
2162 fdisp_destroy(&fdi);
2167 * The Linux / FUSE extended attribute list is simply a collection of
2168 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2169 * byte length followed by a non-NUL terminated string. So, this allows
2170 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2171 * Linux attribute names are reported with the namespace as a prefix (e.g.
2172 * "user.attribute_name"), but in FreeBSD they are reported without the
2173 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2175 * user.attr_name1\0user.attr_name2\0
2179 * <num>attr_name1<num>attr_name2
2181 * Where "<num>" is a single byte number of characters in the attribute name.
2184 * prefix - exattr namespace prefix string
2185 * list, list_len - input list with namespace prefixes
2186 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2189 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2190 char *bsd_list, int *bsd_list_len)
2192 int len, pos, dist_to_next, prefix_len;
2196 prefix_len = strlen(prefix);
2198 while (pos < list_len && list[pos] != '\0') {
2199 dist_to_next = strlen(&list[pos]) + 1;
2200 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2201 list[pos + prefix_len] == extattr_namespace_separator) {
2202 len = dist_to_next -
2203 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2204 if (len >= EXTATTR_MAXNAMELEN)
2205 return (ENAMETOOLONG);
2207 bsd_list[*bsd_list_len] = len;
2208 memcpy(&bsd_list[*bsd_list_len + 1],
2209 &list[pos + prefix_len +
2210 sizeof(extattr_namespace_separator)], len);
2212 *bsd_list_len += len + 1;
2215 pos += dist_to_next;
2222 * List extended attributes
2224 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2225 * has a number of differences compared to its FreeBSD equivalent,
2226 * extattr_list_file:
2228 * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2229 * whereas listxattr(2) only returns attributes for a single namespace
2230 * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2231 * - If the provided buffer is not large enough to hold the result,
2232 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2233 * return as many results as will fit.
2236 struct vop_listextattr_args {
2237 struct vop_generic_args a_gen;
2239 int a_attrnamespace;
2242 struct ucred *a_cred;
2243 struct thread *a_td;
2247 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2249 struct vnode *vp = ap->a_vp;
2250 struct uio *uio = ap->a_uio;
2251 struct fuse_dispatcher fdi;
2252 struct fuse_listxattr_in *list_xattr_in;
2253 struct fuse_listxattr_out *list_xattr_out;
2254 struct mount *mp = vnode_mount(vp);
2255 struct thread *td = ap->a_td;
2256 struct ucred *cred = ap->a_cred;
2258 char *bsd_list = NULL;
2264 if (fuse_isdeadfs(vp))
2267 if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2270 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2275 * Add space for a NUL and the period separator if enabled.
2276 * Default to looking for user attributes.
2278 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2279 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2281 prefix = EXTATTR_NAMESPACE_USER_STRING;
2283 fdisp_init(&fdi, sizeof(*list_xattr_in));
2284 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2287 * Retrieve Linux / FUSE compatible list size.
2289 list_xattr_in = fdi.indata;
2290 list_xattr_in->size = 0;
2292 err = fdisp_wait_answ(&fdi);
2294 if (err == ENOSYS) {
2295 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2301 list_xattr_out = fdi.answ;
2302 linux_list_len = list_xattr_out->size;
2303 if (linux_list_len == 0) {
2304 if (ap->a_size != NULL)
2305 *ap->a_size = linux_list_len;
2310 * Retrieve Linux / FUSE compatible list values.
2312 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2313 list_xattr_in = fdi.indata;
2314 list_xattr_in->size = linux_list_len;
2316 err = fdisp_wait_answ(&fdi);
2317 if (err == ERANGE) {
2319 * Race detected. The attribute list must've grown since the
2320 * first FUSE_LISTXATTR call. Start over. Go all the way back
2321 * to userland so we can process signals, if necessary, before
2326 } else if (err != 0)
2329 linux_list = fdi.answ;
2330 /* FUSE doesn't allow the server to return more data than requested */
2331 if (fdi.iosize > linux_list_len) {
2332 printf("WARNING: FUSE protocol violation. Server returned "
2333 "more extended attribute data than requested; "
2334 "should've returned ERANGE instead");
2336 /* But returning less data is fine */
2337 linux_list_len = fdi.iosize;
2341 * Retrieve the BSD compatible list values.
2342 * The Linux / FUSE attribute list format isn't the same
2343 * as FreeBSD's format. So we need to transform it into
2344 * FreeBSD's format before giving it to the user.
2346 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2347 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2348 bsd_list, &bsd_list_len);
2352 if (ap->a_size != NULL)
2353 *ap->a_size = bsd_list_len;
2356 err = uiomove(bsd_list, bsd_list_len, uio);
2359 free(bsd_list, M_TEMP);
2360 fdisp_destroy(&fdi);
2365 struct vop_deleteextattr_args {
2366 struct vop_generic_args a_gen;
2368 int a_attrnamespace;
2370 struct ucred *a_cred;
2371 struct thread *a_td;
2375 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2377 struct vnode *vp = ap->a_vp;
2378 struct fuse_dispatcher fdi;
2379 struct mount *mp = vnode_mount(vp);
2380 struct thread *td = ap->a_td;
2381 struct ucred *cred = ap->a_cred;
2387 if (fuse_isdeadfs(vp))
2390 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2393 if (vfs_isrdonly(mp))
2396 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2401 /* Default to looking for user attributes. */
2402 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2403 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2405 prefix = EXTATTR_NAMESPACE_USER_STRING;
2407 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2408 strlen(ap->a_name) + 1;
2410 fdisp_init(&fdi, len);
2411 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2413 attr_str = fdi.indata;
2414 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2417 err = fdisp_wait_answ(&fdi);
2418 if (err == ENOSYS) {
2419 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2423 fdisp_destroy(&fdi);
2428 struct vnop_print_args {
2433 fuse_vnop_print(struct vop_print_args *ap)
2435 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2437 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2438 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2439 (uintmax_t)fvdat->nlookup,
2446 * Get an NFS filehandle for a FUSE file.
2448 * This will only work for FUSE file systems that guarantee the uniqueness of
2449 * nodeid:generation, which most don't.
2453 IN struct vnode *a_vp;
2454 IN struct fid *a_fhp;
2458 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2460 struct vnode *vp = ap->a_vp;
2461 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2462 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2463 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2464 "FUSE fid type is too big");
2465 struct mount *mp = vnode_mount(vp);
2466 struct fuse_data *data = fuse_get_mpdata(mp);
2470 if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2473 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2477 /*ip = VTOI(ap->a_vp);*/
2478 /*ufhp = (struct ufid *)ap->a_fhp;*/
2479 fhp->len = sizeof(struct fuse_fid);
2480 fhp->nid = fvdat->nid;
2481 if (fvdat->generation <= UINT32_MAX)
2482 fhp->gen = fvdat->generation;