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 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
45 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
61 #include <sys/param.h>
62 #include <sys/module.h>
63 #include <sys/systm.h>
64 #include <sys/errno.h>
65 #include <sys/kernel.h>
68 #include <sys/malloc.h>
69 #include <sys/queue.h>
71 #include <sys/rwlock.h>
74 #include <sys/mount.h>
75 #include <sys/vnode.h>
76 #include <sys/namei.h>
77 #include <sys/extattr.h>
79 #include <sys/unistd.h>
80 #include <sys/filedesc.h>
82 #include <sys/fcntl.h>
83 #include <sys/dirent.h>
86 #include <sys/sysctl.h>
87 #include <sys/vmmeter.h>
90 #include <vm/vm_extern.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_param.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vnode_pager.h>
98 #include <vm/vm_object.h>
101 #include "fuse_file.h"
102 #include "fuse_internal.h"
103 #include "fuse_ipc.h"
104 #include "fuse_node.h"
107 #include <sys/priv.h>
109 /* Maximum number of hardlinks to a single FUSE file */
110 #define FUSE_LINK_MAX UINT32_MAX
112 SDT_PROVIDER_DECLARE(fusefs);
115 * arg0: verbosity. Higher numbers give more verbose messages
116 * arg1: Textual message
118 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
121 static vop_access_t fuse_vnop_access;
122 static vop_advlock_t fuse_vnop_advlock;
123 static vop_bmap_t fuse_vnop_bmap;
124 static vop_close_t fuse_fifo_close;
125 static vop_close_t fuse_vnop_close;
126 static vop_create_t fuse_vnop_create;
127 static vop_deleteextattr_t fuse_vnop_deleteextattr;
128 static vop_fdatasync_t fuse_vnop_fdatasync;
129 static vop_fsync_t fuse_vnop_fsync;
130 static vop_getattr_t fuse_vnop_getattr;
131 static vop_getextattr_t fuse_vnop_getextattr;
132 static vop_inactive_t fuse_vnop_inactive;
133 static vop_link_t fuse_vnop_link;
134 static vop_listextattr_t fuse_vnop_listextattr;
135 static vop_lookup_t fuse_vnop_lookup;
136 static vop_mkdir_t fuse_vnop_mkdir;
137 static vop_mknod_t fuse_vnop_mknod;
138 static vop_open_t fuse_vnop_open;
139 static vop_pathconf_t fuse_vnop_pathconf;
140 static vop_read_t fuse_vnop_read;
141 static vop_readdir_t fuse_vnop_readdir;
142 static vop_readlink_t fuse_vnop_readlink;
143 static vop_reclaim_t fuse_vnop_reclaim;
144 static vop_remove_t fuse_vnop_remove;
145 static vop_rename_t fuse_vnop_rename;
146 static vop_rmdir_t fuse_vnop_rmdir;
147 static vop_setattr_t fuse_vnop_setattr;
148 static vop_setextattr_t fuse_vnop_setextattr;
149 static vop_strategy_t fuse_vnop_strategy;
150 static vop_symlink_t fuse_vnop_symlink;
151 static vop_write_t fuse_vnop_write;
152 static vop_getpages_t fuse_vnop_getpages;
153 static vop_print_t fuse_vnop_print;
154 static vop_vptofh_t fuse_vnop_vptofh;
156 struct vop_vector fuse_fifoops = {
157 .vop_default = &fifo_specops,
158 .vop_access = fuse_vnop_access,
159 .vop_close = fuse_fifo_close,
160 .vop_fsync = fuse_vnop_fsync,
161 .vop_getattr = fuse_vnop_getattr,
162 .vop_inactive = fuse_vnop_inactive,
163 .vop_pathconf = fuse_vnop_pathconf,
164 .vop_print = fuse_vnop_print,
165 .vop_read = VOP_PANIC,
166 .vop_reclaim = fuse_vnop_reclaim,
167 .vop_setattr = fuse_vnop_setattr,
168 .vop_write = VOP_PANIC,
169 .vop_vptofh = fuse_vnop_vptofh,
172 struct vop_vector fuse_vnops = {
173 .vop_allocate = VOP_EINVAL,
174 .vop_default = &default_vnodeops,
175 .vop_access = fuse_vnop_access,
176 .vop_advlock = fuse_vnop_advlock,
177 .vop_bmap = fuse_vnop_bmap,
178 .vop_close = fuse_vnop_close,
179 .vop_create = fuse_vnop_create,
180 .vop_deleteextattr = fuse_vnop_deleteextattr,
181 .vop_fsync = fuse_vnop_fsync,
182 .vop_fdatasync = fuse_vnop_fdatasync,
183 .vop_getattr = fuse_vnop_getattr,
184 .vop_getextattr = fuse_vnop_getextattr,
185 .vop_inactive = fuse_vnop_inactive,
187 * TODO: implement vop_ioctl after upgrading to protocol 7.16.
188 * FUSE_IOCTL was added in 7.11, but 32-bit compat is broken until
191 .vop_link = fuse_vnop_link,
192 .vop_listextattr = fuse_vnop_listextattr,
193 .vop_lookup = fuse_vnop_lookup,
194 .vop_mkdir = fuse_vnop_mkdir,
195 .vop_mknod = fuse_vnop_mknod,
196 .vop_open = fuse_vnop_open,
197 .vop_pathconf = fuse_vnop_pathconf,
199 * TODO: implement vop_poll after upgrading to protocol 7.21.
200 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
201 * 7.21, which adds the ability for the client to choose which poll
202 * events it wants, and for a client to deregister a file handle
204 .vop_read = fuse_vnop_read,
205 .vop_readdir = fuse_vnop_readdir,
206 .vop_readlink = fuse_vnop_readlink,
207 .vop_reclaim = fuse_vnop_reclaim,
208 .vop_remove = fuse_vnop_remove,
209 .vop_rename = fuse_vnop_rename,
210 .vop_rmdir = fuse_vnop_rmdir,
211 .vop_setattr = fuse_vnop_setattr,
212 .vop_setextattr = fuse_vnop_setextattr,
213 .vop_strategy = fuse_vnop_strategy,
214 .vop_symlink = fuse_vnop_symlink,
215 .vop_write = fuse_vnop_write,
216 .vop_getpages = fuse_vnop_getpages,
217 .vop_print = fuse_vnop_print,
218 .vop_vptofh = fuse_vnop_vptofh,
221 uma_zone_t fuse_pbuf_zone;
223 #define fuse_vm_page_lock(m) vm_page_lock((m));
224 #define fuse_vm_page_unlock(m) vm_page_unlock((m));
225 #define fuse_vm_page_lock_queues() ((void)0)
226 #define fuse_vm_page_unlock_queues() ((void)0)
228 /* Check permission for extattr operations, much like extattr_check_cred */
230 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
231 struct thread *td, accmode_t accmode)
233 struct mount *mp = vnode_mount(vp);
234 struct fuse_data *data = fuse_get_mpdata(mp);
237 * Kernel-invoked always succeeds.
243 * Do not allow privileged processes in jail to directly manipulate
247 case EXTATTR_NAMESPACE_SYSTEM:
248 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
249 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
252 case EXTATTR_NAMESPACE_USER:
253 return (fuse_internal_access(vp, accmode, td, cred));
259 /* Get a filehandle for a directory */
261 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
262 struct ucred *cred, pid_t pid)
264 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
266 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
269 /* Send FUSE_FLUSH for this vnode */
271 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
273 struct fuse_flush_in *ffi;
274 struct fuse_filehandle *fufh;
275 struct fuse_dispatcher fdi;
276 struct thread *td = curthread;
277 struct mount *mp = vnode_mount(vp);
280 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
283 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
287 fdisp_init(&fdi, sizeof(*ffi));
288 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
290 ffi->fh = fufh->fh_id;
292 * If the file has a POSIX lock then we're supposed to set lock_owner.
293 * If not, then lock_owner is undefined. So we may as well always set
296 ffi->lock_owner = td->td_proc->p_pid;
298 err = fdisp_wait_answ(&fdi);
300 fsess_set_notimpl(mp, FUSE_FLUSH);
307 /* Close wrapper for fifos. */
309 fuse_fifo_close(struct vop_close_args *ap)
311 return (fifo_specops.vop_close(ap));
315 struct vnop_access_args {
317 #if VOP_ACCESS_TAKES_ACCMODE_T
322 struct ucred *a_cred;
327 fuse_vnop_access(struct vop_access_args *ap)
329 struct vnode *vp = ap->a_vp;
330 int accmode = ap->a_accmode;
331 struct ucred *cred = ap->a_cred;
333 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
337 if (fuse_isdeadfs(vp)) {
338 if (vnode_isvroot(vp)) {
343 if (!(data->dataflags & FSESS_INITED)) {
344 if (vnode_isvroot(vp)) {
345 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
346 (fuse_match_cred(data->daemoncred, cred) == 0)) {
352 if (vnode_islnk(vp)) {
356 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
361 * struct vop_advlock_args {
362 * struct vop_generic_args a_gen;
363 * struct vnode *a_vp;
366 * struct flock *a_fl;
371 fuse_vnop_advlock(struct vop_advlock_args *ap)
373 struct vnode *vp = ap->a_vp;
374 struct flock *fl = ap->a_fl;
375 struct thread *td = curthread;
376 struct ucred *cred = td->td_ucred;
377 pid_t pid = td->td_proc->p_pid;
378 struct fuse_filehandle *fufh;
379 struct fuse_dispatcher fdi;
380 struct fuse_lk_in *fli;
381 struct fuse_lk_out *flo;
384 int flags = ap->a_flags;
386 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
388 if (fuse_isdeadfs(vp)) {
392 if (!(dataflags & FSESS_POSIX_LOCKS))
393 return vop_stdadvlock(ap);
394 /* FUSE doesn't properly support flock until protocol 7.17 */
396 return vop_stdadvlock(ap);
398 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
402 fdisp_init(&fdi, sizeof(*fli));
418 fdisp_make_vp(&fdi, op, vp, td, cred);
420 fli->fh = fufh->fh_id;
421 fli->owner = fl->l_pid;
422 fli->lk.start = fl->l_start;
424 fli->lk.end = fl->l_start + fl->l_len - 1;
426 fli->lk.end = INT64_MAX;
427 fli->lk.type = fl->l_type;
428 fli->lk.pid = fl->l_pid;
430 err = fdisp_wait_answ(&fdi);
433 if (err == 0 && op == FUSE_GETLK) {
435 fl->l_type = flo->lk.type;
436 fl->l_pid = flo->lk.pid;
437 if (flo->lk.type != F_UNLCK) {
438 fl->l_start = flo->lk.start;
439 if (flo->lk.end == INT64_MAX)
442 fl->l_len = flo->lk.end - flo->lk.start + 1;
443 fl->l_start = flo->lk.start;
453 struct bufobj **a_bop;
459 fuse_vnop_bmap(struct vop_bmap_args *ap)
461 struct vnode *vp = ap->a_vp;
462 struct bufobj **bo = ap->a_bop;
463 struct thread *td = curthread;
465 struct fuse_dispatcher fdi;
466 struct fuse_bmap_in *fbi;
467 struct fuse_bmap_out *fbo;
468 struct fuse_data *data;
471 daddr_t lbn = ap->a_bn;
472 daddr_t *pbn = ap->a_bnp;
473 int *runp = ap->a_runp;
474 int *runb = ap->a_runb;
478 if (fuse_isdeadfs(vp)) {
482 mp = vnode_mount(vp);
483 data = fuse_get_mpdata(mp);
484 biosize = fuse_iosize(vp);
485 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
486 data->max_readahead_blocks);
492 * The FUSE_BMAP operation does not include the runp and runb
493 * variables, so we must guess. Report nonzero contiguous runs so
494 * cluster_read will combine adjacent reads. It's worthwhile to reduce
495 * upcalls even if we don't know the true physical layout of the file.
497 * FUSE file systems may opt out of read clustering in two ways:
498 * * mounting with -onoclusterr
499 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
502 *runb = MIN(lbn, maxrun);
504 error = fuse_vnode_size(vp, &filesize, td->td_ucred, td);
506 *runp = MIN(MAX(0, filesize / biosize - lbn - 1),
512 if (fsess_isimpl(mp, FUSE_BMAP)) {
513 fdisp_init(&fdi, sizeof(*fbi));
514 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
517 fbi->blocksize = biosize;
518 error = fdisp_wait_answ(&fdi);
519 if (error == ENOSYS) {
521 fsess_set_notimpl(mp, FUSE_BMAP);
525 if (error == 0 && pbn != NULL)
532 /* If the daemon doesn't support BMAP, make up a sensible default */
534 *pbn = lbn * btodb(biosize);
539 struct vop_close_args {
542 struct ucred *a_cred;
547 fuse_vnop_close(struct vop_close_args *ap)
549 struct vnode *vp = ap->a_vp;
550 struct ucred *cred = ap->a_cred;
551 int fflag = ap->a_fflag;
552 struct thread *td = ap->a_td;
553 pid_t pid = td->td_proc->p_pid;
556 if (fuse_isdeadfs(vp))
560 if (fflag & IO_NDELAY)
563 err = fuse_flush(vp, cred, pid, fflag);
564 /* TODO: close the file handle, if we're sure it's no longer used */
565 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
566 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
572 fdisp_make_mknod_for_fallback(
573 struct fuse_dispatcher *fdip,
574 struct componentname *cnp,
580 enum fuse_opcode *op)
582 struct fuse_mknod_in *fmni;
584 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
586 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
590 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
592 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
595 struct vnop_create_args {
597 struct vnode **a_vpp;
598 struct componentname *a_cnp;
603 fuse_vnop_create(struct vop_create_args *ap)
605 struct vnode *dvp = ap->a_dvp;
606 struct vnode **vpp = ap->a_vpp;
607 struct componentname *cnp = ap->a_cnp;
608 struct vattr *vap = ap->a_vap;
609 struct thread *td = cnp->cn_thread;
610 struct ucred *cred = cnp->cn_cred;
612 struct fuse_data *data;
613 struct fuse_create_in *fci;
614 struct fuse_entry_out *feo;
615 struct fuse_open_out *foo;
616 struct fuse_dispatcher fdi, fdi2;
617 struct fuse_dispatcher *fdip = &fdi;
618 struct fuse_dispatcher *fdip2 = NULL;
622 struct mount *mp = vnode_mount(dvp);
623 data = fuse_get_mpdata(mp);
624 uint64_t parentnid = VTOFUD(dvp)->nid;
625 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
629 if (fuse_isdeadfs(dvp))
632 /* FUSE expects sockets to be created with FUSE_MKNOD */
633 if (vap->va_type == VSOCK)
634 return fuse_internal_mknod(dvp, vpp, cnp, vap);
637 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
638 * writable mode makes sense, and we might as well include readability
643 bzero(&fdi, sizeof(fdi));
645 if (vap->va_type != VREG)
648 if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
649 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
650 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
653 /* Use FUSE_CREATE */
657 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
658 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
661 fci->flags = O_CREAT | flags;
662 if (fuse_libabi_geq(data, 7, 12)) {
663 insize = sizeof(*fci);
664 fci->umask = td->td_proc->p_fd->fd_cmask;
666 insize = sizeof(struct fuse_open_in);
669 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
671 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
674 err = fdisp_wait_answ(fdip);
677 if (err == ENOSYS && op == FUSE_CREATE) {
678 fsess_set_notimpl(mp, FUSE_CREATE);
680 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
681 parentnid, td, cred, mode, &op);
682 err = fdisp_wait_answ(fdip);
690 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
694 if (op == FUSE_CREATE) {
695 foo = (struct fuse_open_out*)(feo + 1);
697 /* Issue a separate FUSE_OPEN */
698 struct fuse_open_in *foi;
701 fdisp_init(fdip2, sizeof(*foi));
702 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
706 err = fdisp_wait_answ(fdip2);
711 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
713 struct fuse_release_in *fri;
714 uint64_t nodeid = feo->nodeid;
715 uint64_t fh_id = foo->fh;
717 fdisp_init(fdip, sizeof(*fri));
718 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
722 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
723 fuse_insert_message(fdip->tick, false);
726 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
727 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
728 feo->attr_valid_nsec, NULL);
730 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
731 fuse_vnode_open(*vpp, foo->open_flags, td);
733 * Purge the parent's attribute cache because the daemon should've
734 * updated its mtime and ctime
736 fuse_vnode_clear_attr_cache(dvp);
737 cache_purge_negative(dvp);
741 fdisp_destroy(fdip2);
747 struct vnop_fdatasync_args {
748 struct vop_generic_args a_gen;
750 struct thread * a_td;
754 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
756 struct vnode *vp = ap->a_vp;
757 struct thread *td = ap->a_td;
758 int waitfor = MNT_WAIT;
762 if (fuse_isdeadfs(vp)) {
765 if ((err = vop_stdfdatasync_buf(ap)))
768 return fuse_internal_fsync(vp, td, waitfor, true);
772 struct vnop_fsync_args {
773 struct vop_generic_args a_gen;
776 struct thread * a_td;
780 fuse_vnop_fsync(struct vop_fsync_args *ap)
782 struct vnode *vp = ap->a_vp;
783 struct thread *td = ap->a_td;
784 int waitfor = ap->a_waitfor;
787 if (fuse_isdeadfs(vp)) {
790 if ((err = vop_stdfsync(ap)))
793 return fuse_internal_fsync(vp, td, waitfor, false);
797 struct vnop_getattr_args {
800 struct ucred *a_cred;
805 fuse_vnop_getattr(struct vop_getattr_args *ap)
807 struct vnode *vp = ap->a_vp;
808 struct vattr *vap = ap->a_vap;
809 struct ucred *cred = ap->a_cred;
810 struct thread *td = curthread;
815 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
817 /* Note that we are not bailing out on a dead file system just yet. */
819 if (!(dataflags & FSESS_INITED)) {
820 if (!vnode_isvroot(vp)) {
821 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
828 err = fuse_internal_getattr(vp, vap, cred, td);
829 if (err == ENOTCONN && vnode_isvroot(vp)) {
830 /* see comment in fuse_vfsop_statfs() */
837 bzero(vap, sizeof(*vap));
838 vap->va_type = vnode_vtype(vp);
844 struct vnop_inactive_args {
850 fuse_vnop_inactive(struct vop_inactive_args *ap)
852 struct vnode *vp = ap->a_vp;
853 struct thread *td = ap->a_td;
855 struct fuse_vnode_data *fvdat = VTOFUD(vp);
856 struct fuse_filehandle *fufh, *fufh_tmp;
860 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
861 if (need_flush && vp->v_type == VREG) {
862 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
863 fuse_vnode_savesize(vp, NULL, 0);
865 if ((fvdat->flag & FN_REVOKED) != 0)
866 fuse_io_invalbuf(vp, td);
868 fuse_io_flushbuf(vp, MNT_WAIT, td);
871 fuse_filehandle_close(vp, fufh, td, NULL);
874 if ((fvdat->flag & FN_REVOKED) != 0)
881 struct vnop_link_args {
882 struct vnode *a_tdvp;
884 struct componentname *a_cnp;
888 fuse_vnop_link(struct vop_link_args *ap)
890 struct vnode *vp = ap->a_vp;
891 struct vnode *tdvp = ap->a_tdvp;
892 struct componentname *cnp = ap->a_cnp;
894 struct vattr *vap = VTOVA(vp);
896 struct fuse_dispatcher fdi;
897 struct fuse_entry_out *feo;
898 struct fuse_link_in fli;
902 if (fuse_isdeadfs(vp)) {
905 if (vnode_mount(tdvp) != vnode_mount(vp)) {
910 * This is a seatbelt check to protect naive userspace filesystems from
911 * themselves and the limitations of the FUSE IPC protocol. If a
912 * filesystem does not allow attribute caching, assume it is capable of
913 * validating that nlink does not overflow.
915 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
917 fli.oldnodeid = VTOI(vp);
920 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
921 FUSE_LINK, &fli, sizeof(fli), &fdi);
922 if ((err = fdisp_wait_answ(&fdi))) {
927 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
930 * Purge the parent's attribute cache because the daemon
931 * should've updated its mtime and ctime
933 fuse_vnode_clear_attr_cache(tdvp);
934 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
935 feo->attr_valid_nsec, NULL);
942 struct fuse_lookup_alloc_arg {
943 struct fuse_entry_out *feo;
944 struct componentname *cnp;
949 /* Callback for vn_get_ino */
951 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
953 struct fuse_lookup_alloc_arg *flaa = arg;
955 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
959 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
960 "int", "struct timespec*", "struct timespec*");
962 struct vnop_lookup_args {
963 struct vnodeop_desc *a_desc;
965 struct vnode **a_vpp;
966 struct componentname *a_cnp;
970 fuse_vnop_lookup(struct vop_lookup_args *ap)
972 struct vnode *dvp = ap->a_dvp;
973 struct vnode **vpp = ap->a_vpp;
974 struct componentname *cnp = ap->a_cnp;
975 struct thread *td = cnp->cn_thread;
976 struct ucred *cred = cnp->cn_cred;
978 int nameiop = cnp->cn_nameiop;
979 int flags = cnp->cn_flags;
980 int wantparent = flags & (LOCKPARENT | WANTPARENT);
981 int islastcn = flags & ISLASTCN;
982 struct mount *mp = vnode_mount(dvp);
986 struct vnode *vp = NULL;
988 struct fuse_dispatcher fdi;
989 bool did_lookup = false;
990 struct fuse_entry_out *feo = NULL;
991 enum vtype vtyp; /* vnode type of target */
992 off_t filesize; /* filesize of target */
996 if (fuse_isdeadfs(dvp)) {
1000 if (!vnode_isdir(dvp))
1003 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1006 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1009 if (flags & ISDOTDOT) {
1010 KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1011 ("Looking up .. is TODO"));
1012 nid = VTOFUD(dvp)->parent_nid;
1015 /* .. is obviously a directory */
1018 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1020 /* . is obviously a directory */
1024 struct timespec now, timeout;
1026 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
1027 getnanouptime(&now);
1028 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1030 case -1: /* positive match */
1031 if (timespeccmp(&timeout, &now, >)) {
1032 counter_u64_add(fuse_lookup_cache_hits, 1);
1035 counter_u64_add(fuse_lookup_cache_misses, 1);
1037 &VTOFUD(*vpp)->entry_cache_timeout);
1048 case 0: /* no match in cache */
1049 counter_u64_add(fuse_lookup_cache_misses, 1);
1052 case ENOENT: /* negative match */
1053 getnanouptime(&now);
1054 if (timespeccmp(&timeout, &now, <=)) {
1056 cache_purge_negative(dvp);
1065 fdisp_init(&fdi, cnp->cn_namelen + 1);
1066 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1068 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1069 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1070 lookup_err = fdisp_wait_answ(&fdi);
1074 /* lookup call succeeded */
1075 feo = (struct fuse_entry_out *)fdi.answ;
1078 /* zero nodeid means ENOENT and cache it */
1079 struct timespec timeout;
1081 fdi.answ_stat = ENOENT;
1082 lookup_err = ENOENT;
1083 if (cnp->cn_flags & MAKEENTRY) {
1084 fuse_validity_2_timespec(feo, &timeout);
1085 cache_enter_time(dvp, *vpp, cnp,
1088 } else if (nid == FUSE_ROOT_ID) {
1089 lookup_err = EINVAL;
1091 vtyp = IFTOVT(feo->attr.mode);
1092 filesize = feo->attr.size;
1094 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1095 fdisp_destroy(&fdi);
1099 /* lookup_err, if non-zero, must be ENOENT at this point */
1102 /* Entry not found */
1103 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1104 err = fuse_internal_access(dvp, VWRITE, td, cred);
1107 * Set the SAVENAME flag to hold onto the
1108 * pathname for use later in VOP_CREATE or
1111 cnp->cn_flags |= SAVENAME;
1119 /* Entry was found */
1120 if (flags & ISDOTDOT) {
1121 struct fuse_lookup_alloc_arg flaa;
1127 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1130 } else if (nid == VTOI(dvp)) {
1134 struct fuse_vnode_data *fvdat;
1136 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1143 * In the case where we are looking up a FUSE node
1144 * represented by an existing cached vnode, and the
1145 * true size reported by FUSE_LOOKUP doesn't match
1146 * the vnode's cached size, then any cached writes
1147 * beyond the file's current size are lost.
1150 * * following attribute cache expiration, or
1151 * * due a bug in the daemon, or
1154 if (vnode_isreg(vp) &&
1155 filesize != fvdat->cached_attrs.va_size &&
1156 fvdat->flag & FN_SIZECHANGE) {
1158 * The FN_SIZECHANGE flag reflects a dirty
1159 * append. If userspace lets us know our cache
1160 * is invalid, that write was lost. (Dirty
1161 * writes that do not cause append are also
1162 * lost, but we don't detect them here.)
1164 * XXX: Maybe disable WB caching on this mount.
1166 printf("%s: WB cache incoherent on %s!\n",
1168 vnode_mount(vp)->mnt_stat.f_mntonname);
1170 fvdat->flag &= ~FN_SIZECHANGE;
1174 fuse_internal_cache_attrs(*vpp, &feo->attr,
1175 feo->attr_valid, feo->attr_valid_nsec, NULL);
1176 fuse_validity_2_bintime(feo->entry_valid,
1177 feo->entry_valid_nsec,
1178 &fvdat->entry_cache_timeout);
1180 if ((nameiop == DELETE || nameiop == RENAME) &&
1183 struct vattr dvattr;
1185 err = fuse_internal_access(dvp, VWRITE, td,
1190 * if the parent's sticky bit is set, check
1191 * whether we're allowed to remove the file.
1192 * Need to figure out the vnode locking to make
1195 fuse_internal_getattr(dvp, &dvattr, cred, td);
1196 if ((dvattr.va_mode & S_ISTXT) &&
1197 fuse_internal_access(dvp, VADMIN, td,
1199 fuse_internal_access(*vpp, VADMIN, td,
1207 (nameiop == DELETE) ||
1208 (nameiop == RENAME && wantparent))) {
1209 cnp->cn_flags |= SAVENAME;
1216 if (vp != NULL && dvp != vp)
1218 else if (vp != NULL)
1223 fdisp_destroy(&fdi);
1229 struct vnop_mkdir_args {
1230 struct vnode *a_dvp;
1231 struct vnode **a_vpp;
1232 struct componentname *a_cnp;
1233 struct vattr *a_vap;
1237 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1239 struct vnode *dvp = ap->a_dvp;
1240 struct vnode **vpp = ap->a_vpp;
1241 struct componentname *cnp = ap->a_cnp;
1242 struct vattr *vap = ap->a_vap;
1244 struct fuse_mkdir_in fmdi;
1246 if (fuse_isdeadfs(dvp)) {
1249 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1250 fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1252 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1253 sizeof(fmdi), VDIR));
1257 struct vnop_mknod_args {
1258 struct vnode *a_dvp;
1259 struct vnode **a_vpp;
1260 struct componentname *a_cnp;
1261 struct vattr *a_vap;
1265 fuse_vnop_mknod(struct vop_mknod_args *ap)
1268 struct vnode *dvp = ap->a_dvp;
1269 struct vnode **vpp = ap->a_vpp;
1270 struct componentname *cnp = ap->a_cnp;
1271 struct vattr *vap = ap->a_vap;
1273 if (fuse_isdeadfs(dvp))
1276 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1280 struct vop_open_args {
1283 struct ucred *a_cred;
1284 struct thread *a_td;
1285 int a_fdidx; / struct file *a_fp;
1289 fuse_vnop_open(struct vop_open_args *ap)
1291 struct vnode *vp = ap->a_vp;
1292 int a_mode = ap->a_mode;
1293 struct thread *td = ap->a_td;
1294 struct ucred *cred = ap->a_cred;
1295 pid_t pid = td->td_proc->p_pid;
1296 struct fuse_vnode_data *fvdat;
1298 if (fuse_isdeadfs(vp))
1300 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1301 return (EOPNOTSUPP);
1302 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1307 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1308 fuse_vnode_open(vp, 0, td);
1312 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1316 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1319 switch (ap->a_name) {
1320 case _PC_FILESIZEBITS:
1324 *ap->a_retval = NAME_MAX;
1327 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1329 case _PC_SYMLINK_MAX:
1330 *ap->a_retval = MAXPATHLEN;
1336 return (vop_stdpathconf(ap));
1341 struct vnop_read_args {
1345 struct ucred *a_cred;
1349 fuse_vnop_read(struct vop_read_args *ap)
1351 struct vnode *vp = ap->a_vp;
1352 struct uio *uio = ap->a_uio;
1353 int ioflag = ap->a_ioflag;
1354 struct ucred *cred = ap->a_cred;
1355 pid_t pid = curthread->td_proc->p_pid;
1357 if (fuse_isdeadfs(vp)) {
1361 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1362 ioflag |= IO_DIRECT;
1365 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1369 struct vnop_readdir_args {
1372 struct ucred *a_cred;
1379 fuse_vnop_readdir(struct vop_readdir_args *ap)
1381 struct vnode *vp = ap->a_vp;
1382 struct uio *uio = ap->a_uio;
1383 struct ucred *cred = ap->a_cred;
1384 struct fuse_filehandle *fufh = NULL;
1385 struct fuse_iov cookediov;
1391 bool closefufh = false;
1392 pid_t pid = curthread->td_proc->p_pid;
1396 if (fuse_isdeadfs(vp)) {
1399 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1400 (uio_resid(uio) < sizeof(struct dirent))) {
1404 tresid = uio->uio_resid;
1405 startoff = uio->uio_offset;
1406 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1407 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1409 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1410 * must implicitly open the directory here
1412 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1415 * When a directory is opened, it must be read from
1416 * the beginning. Hopefully, the "startoff" still
1417 * exists as an offset cookie for the directory.
1418 * If not, it will read the entire directory without
1419 * returning any entries and just return eof.
1421 uio->uio_offset = 0;
1427 if (ap->a_ncookies != NULL) {
1428 ncookies = uio->uio_resid /
1429 (offsetof(struct dirent, d_name) + 4) + 1;
1430 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1431 *ap->a_ncookies = ncookies;
1432 *ap->a_cookies = cookies;
1437 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1438 fiov_init(&cookediov, DIRCOOKEDSIZE);
1440 err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1441 &ncookies, cookies);
1443 fiov_teardown(&cookediov);
1445 fuse_filehandle_close(vp, fufh, curthread, cred);
1447 if (ap->a_ncookies != NULL) {
1449 *ap->a_ncookies -= ncookies;
1451 free(*ap->a_cookies, M_TEMP);
1452 *ap->a_ncookies = 0;
1453 *ap->a_cookies = NULL;
1456 if (err == 0 && tresid == uio->uio_resid)
1463 struct vnop_readlink_args {
1466 struct ucred *a_cred;
1470 fuse_vnop_readlink(struct vop_readlink_args *ap)
1472 struct vnode *vp = ap->a_vp;
1473 struct uio *uio = ap->a_uio;
1474 struct ucred *cred = ap->a_cred;
1476 struct fuse_dispatcher fdi;
1479 if (fuse_isdeadfs(vp)) {
1482 if (!vnode_islnk(vp)) {
1485 fdisp_init(&fdi, 0);
1486 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1490 if (((char *)fdi.answ)[0] == '/' &&
1491 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1492 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1494 err = uiomove(mpth, strlen(mpth), uio);
1497 err = uiomove(fdi.answ, fdi.iosize, uio);
1500 fdisp_destroy(&fdi);
1505 struct vnop_reclaim_args {
1507 struct thread *a_td;
1511 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1513 struct vnode *vp = ap->a_vp;
1514 struct thread *td = ap->a_td;
1515 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1516 struct fuse_filehandle *fufh, *fufh_tmp;
1519 panic("FUSE: no vnode data during recycling");
1521 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1522 printf("FUSE: vnode being reclaimed with open fufh "
1523 "(type=%#x)", fufh->fufh_type);
1524 fuse_filehandle_close(vp, fufh, td, NULL);
1527 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1528 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1531 fuse_vnode_setparent(vp, NULL);
1533 vfs_hash_remove(vp);
1534 vnode_destroy_vobject(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 struct vop_listextattr_args {
2223 struct vop_generic_args a_gen;
2225 int a_attrnamespace;
2228 struct ucred *a_cred;
2229 struct thread *a_td;
2233 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2235 struct vnode *vp = ap->a_vp;
2236 struct uio *uio = ap->a_uio;
2237 struct fuse_dispatcher fdi;
2238 struct fuse_listxattr_in *list_xattr_in;
2239 struct fuse_listxattr_out *list_xattr_out;
2240 struct mount *mp = vnode_mount(vp);
2241 struct thread *td = ap->a_td;
2242 struct ucred *cred = ap->a_cred;
2246 char *bsd_list = NULL;
2252 if (fuse_isdeadfs(vp))
2255 if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2258 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2263 * Add space for a NUL and the period separator if enabled.
2264 * Default to looking for user attributes.
2266 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2267 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2269 prefix = EXTATTR_NAMESPACE_USER_STRING;
2271 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2273 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2274 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2277 * Retrieve Linux / FUSE compatible list size.
2279 list_xattr_in = fdi.indata;
2280 list_xattr_in->size = 0;
2281 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2282 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2284 err = fdisp_wait_answ(&fdi);
2286 if (err == ENOSYS) {
2287 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2293 list_xattr_out = fdi.answ;
2294 linux_list_len = list_xattr_out->size;
2295 if (linux_list_len == 0) {
2296 if (ap->a_size != NULL)
2297 *ap->a_size = linux_list_len;
2302 * Retrieve Linux / FUSE compatible list values.
2304 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2305 list_xattr_in = fdi.indata;
2306 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2307 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2308 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2310 err = fdisp_wait_answ(&fdi);
2314 linux_list = fdi.answ;
2315 linux_list_len = fdi.iosize;
2318 * Retrieve the BSD compatible list values.
2319 * The Linux / FUSE attribute list format isn't the same
2320 * as FreeBSD's format. So we need to transform it into
2321 * FreeBSD's format before giving it to the user.
2323 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2324 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2325 bsd_list, &bsd_list_len);
2329 if (ap->a_size != NULL)
2330 *ap->a_size = bsd_list_len;
2333 err = uiomove(bsd_list, bsd_list_len, uio);
2336 free(bsd_list, M_TEMP);
2337 fdisp_destroy(&fdi);
2342 struct vop_deleteextattr_args {
2343 struct vop_generic_args a_gen;
2345 int a_attrnamespace;
2347 struct ucred *a_cred;
2348 struct thread *a_td;
2352 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2354 struct vnode *vp = ap->a_vp;
2355 struct fuse_dispatcher fdi;
2356 struct mount *mp = vnode_mount(vp);
2357 struct thread *td = ap->a_td;
2358 struct ucred *cred = ap->a_cred;
2364 if (fuse_isdeadfs(vp))
2367 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2370 if (vfs_isrdonly(mp))
2373 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2378 /* Default to looking for user attributes. */
2379 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2380 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2382 prefix = EXTATTR_NAMESPACE_USER_STRING;
2384 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2385 strlen(ap->a_name) + 1;
2387 fdisp_init(&fdi, len);
2388 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2390 attr_str = fdi.indata;
2391 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2394 err = fdisp_wait_answ(&fdi);
2395 if (err == ENOSYS) {
2396 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2400 fdisp_destroy(&fdi);
2405 struct vnop_print_args {
2410 fuse_vnop_print(struct vop_print_args *ap)
2412 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2414 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2415 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2416 (uintmax_t)fvdat->nlookup,
2423 * Get an NFS filehandle for a FUSE file.
2425 * This will only work for FUSE file systems that guarantee the uniqueness of
2426 * nodeid:generation, which most don't.
2430 IN struct vnode *a_vp;
2431 IN struct fid *a_fhp;
2435 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2437 struct vnode *vp = ap->a_vp;
2438 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2439 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2440 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2441 "FUSE fid type is too big");
2442 struct mount *mp = vnode_mount(vp);
2443 struct fuse_data *data = fuse_get_mpdata(mp);
2447 if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2450 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2454 /*ip = VTOI(ap->a_vp);*/
2455 /*ufhp = (struct ufid *)ap->a_fhp;*/
2456 fhp->len = sizeof(struct fuse_fid);
2457 fhp->nid = fvdat->nid;
2458 if (fvdat->generation <= UINT32_MAX)
2459 fhp->gen = fvdat->generation;