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_close_t fuse_fifo_close;
124 static vop_close_t fuse_vnop_close;
125 static vop_create_t fuse_vnop_create;
126 static vop_deleteextattr_t fuse_vnop_deleteextattr;
127 static vop_fdatasync_t fuse_vnop_fdatasync;
128 static vop_fsync_t fuse_vnop_fsync;
129 static vop_getattr_t fuse_vnop_getattr;
130 static vop_getextattr_t fuse_vnop_getextattr;
131 static vop_inactive_t fuse_vnop_inactive;
132 static vop_link_t fuse_vnop_link;
133 static vop_listextattr_t fuse_vnop_listextattr;
134 static vop_lookup_t fuse_vnop_lookup;
135 static vop_mkdir_t fuse_vnop_mkdir;
136 static vop_mknod_t fuse_vnop_mknod;
137 static vop_open_t fuse_vnop_open;
138 static vop_pathconf_t fuse_vnop_pathconf;
139 static vop_read_t fuse_vnop_read;
140 static vop_readdir_t fuse_vnop_readdir;
141 static vop_readlink_t fuse_vnop_readlink;
142 static vop_reclaim_t fuse_vnop_reclaim;
143 static vop_remove_t fuse_vnop_remove;
144 static vop_rename_t fuse_vnop_rename;
145 static vop_rmdir_t fuse_vnop_rmdir;
146 static vop_setattr_t fuse_vnop_setattr;
147 static vop_setextattr_t fuse_vnop_setextattr;
148 static vop_strategy_t fuse_vnop_strategy;
149 static vop_symlink_t fuse_vnop_symlink;
150 static vop_write_t fuse_vnop_write;
151 static vop_getpages_t fuse_vnop_getpages;
152 static vop_putpages_t fuse_vnop_putpages;
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_close = fuse_vnop_close,
178 .vop_create = fuse_vnop_create,
179 .vop_deleteextattr = fuse_vnop_deleteextattr,
180 .vop_fsync = fuse_vnop_fsync,
181 .vop_fdatasync = fuse_vnop_fdatasync,
182 .vop_getattr = fuse_vnop_getattr,
183 .vop_getextattr = fuse_vnop_getextattr,
184 .vop_inactive = fuse_vnop_inactive,
185 .vop_link = fuse_vnop_link,
186 .vop_listextattr = fuse_vnop_listextattr,
187 .vop_lookup = fuse_vnop_lookup,
188 .vop_mkdir = fuse_vnop_mkdir,
189 .vop_mknod = fuse_vnop_mknod,
190 .vop_open = fuse_vnop_open,
191 .vop_pathconf = fuse_vnop_pathconf,
192 .vop_read = fuse_vnop_read,
193 .vop_readdir = fuse_vnop_readdir,
194 .vop_readlink = fuse_vnop_readlink,
195 .vop_reclaim = fuse_vnop_reclaim,
196 .vop_remove = fuse_vnop_remove,
197 .vop_rename = fuse_vnop_rename,
198 .vop_rmdir = fuse_vnop_rmdir,
199 .vop_setattr = fuse_vnop_setattr,
200 .vop_setextattr = fuse_vnop_setextattr,
201 .vop_strategy = fuse_vnop_strategy,
202 .vop_symlink = fuse_vnop_symlink,
203 .vop_write = fuse_vnop_write,
204 .vop_getpages = fuse_vnop_getpages,
205 .vop_putpages = fuse_vnop_putpages,
206 .vop_print = fuse_vnop_print,
207 .vop_vptofh = fuse_vnop_vptofh,
210 static u_long fuse_lookup_cache_hits = 0;
212 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
213 &fuse_lookup_cache_hits, 0, "number of positive cache hits in lookup");
215 static u_long fuse_lookup_cache_misses = 0;
217 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
218 &fuse_lookup_cache_misses, 0, "number of cache misses in lookup");
221 * XXX: This feature is highly experimental and can bring to instabilities,
222 * needs revisiting before to be enabled by default.
224 static int fuse_reclaim_revoked = 0;
226 SYSCTL_INT(_vfs_fusefs, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
227 &fuse_reclaim_revoked, 0, "");
229 uma_zone_t fuse_pbuf_zone;
231 #define fuse_vm_page_lock(m) vm_page_lock((m));
232 #define fuse_vm_page_unlock(m) vm_page_unlock((m));
233 #define fuse_vm_page_lock_queues() ((void)0)
234 #define fuse_vm_page_unlock_queues() ((void)0)
236 /* Check permission for extattr operations, much like extattr_check_cred */
238 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
239 struct thread *td, accmode_t accmode)
241 struct mount *mp = vnode_mount(vp);
242 struct fuse_data *data = fuse_get_mpdata(mp);
245 * Kernel-invoked always succeeds.
251 * Do not allow privileged processes in jail to directly manipulate
255 case EXTATTR_NAMESPACE_SYSTEM:
256 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
257 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
260 case EXTATTR_NAMESPACE_USER:
261 return (fuse_internal_access(vp, accmode, td, cred));
267 /* Get a filehandle for a directory */
269 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
270 struct ucred *cred, pid_t pid)
272 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
274 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
277 /* Send FUSE_FLUSH for this vnode */
279 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
281 struct fuse_flush_in *ffi;
282 struct fuse_filehandle *fufh;
283 struct fuse_dispatcher fdi;
284 struct thread *td = curthread;
285 struct mount *mp = vnode_mount(vp);
288 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
291 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
295 fdisp_init(&fdi, sizeof(*ffi));
296 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
298 ffi->fh = fufh->fh_id;
300 * If the file has a POSIX lock then we're supposed to set lock_owner.
301 * If not, then lock_owner is undefined. So we may as well always set
304 ffi->lock_owner = td->td_proc->p_pid;
306 err = fdisp_wait_answ(&fdi);
308 fsess_set_notimpl(mp, FUSE_FLUSH);
315 /* Close wrapper for fifos. */
317 fuse_fifo_close(struct vop_close_args *ap)
319 return (fifo_specops.vop_close(ap));
323 struct vnop_access_args {
325 #if VOP_ACCESS_TAKES_ACCMODE_T
330 struct ucred *a_cred;
335 fuse_vnop_access(struct vop_access_args *ap)
337 struct vnode *vp = ap->a_vp;
338 int accmode = ap->a_accmode;
339 struct ucred *cred = ap->a_cred;
341 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
345 if (fuse_isdeadfs(vp)) {
346 if (vnode_isvroot(vp)) {
351 if (!(data->dataflags & FSESS_INITED)) {
352 if (vnode_isvroot(vp)) {
353 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
354 (fuse_match_cred(data->daemoncred, cred) == 0)) {
360 if (vnode_islnk(vp)) {
364 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
369 * struct vop_advlock_args {
370 * struct vop_generic_args a_gen;
371 * struct vnode *a_vp;
374 * struct flock *a_fl;
379 fuse_vnop_advlock(struct vop_advlock_args *ap)
381 struct vnode *vp = ap->a_vp;
382 struct flock *fl = ap->a_fl;
383 struct thread *td = curthread;
384 struct ucred *cred = td->td_ucred;
385 pid_t pid = td->td_proc->p_pid;
386 struct fuse_filehandle *fufh;
387 struct fuse_dispatcher fdi;
388 struct fuse_lk_in *fli;
389 struct fuse_lk_out *flo;
392 int flags = ap->a_flags;
394 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
396 if (fuse_isdeadfs(vp)) {
400 if (!(dataflags & FSESS_POSIX_LOCKS))
401 return vop_stdadvlock(ap);
402 /* FUSE doesn't properly support flock until protocol 7.17 */
404 return vop_stdadvlock(ap);
406 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
410 fdisp_init(&fdi, sizeof(*fli));
426 fdisp_make_vp(&fdi, op, vp, td, cred);
428 fli->fh = fufh->fh_id;
429 fli->owner = fl->l_pid;
430 fli->lk.start = fl->l_start;
432 fli->lk.end = fl->l_start + fl->l_len - 1;
434 fli->lk.end = INT64_MAX;
435 fli->lk.type = fl->l_type;
436 fli->lk.pid = fl->l_pid;
438 err = fdisp_wait_answ(&fdi);
441 if (err == 0 && op == FUSE_GETLK) {
443 fl->l_type = flo->lk.type;
444 fl->l_pid = flo->lk.pid;
445 if (flo->lk.type != F_UNLCK) {
446 fl->l_start = flo->lk.start;
447 if (flo->lk.end == INT64_MAX)
450 fl->l_len = flo->lk.end - flo->lk.start + 1;
451 fl->l_start = flo->lk.start;
459 struct vop_close_args {
462 struct ucred *a_cred;
467 fuse_vnop_close(struct vop_close_args *ap)
469 struct vnode *vp = ap->a_vp;
470 struct ucred *cred = ap->a_cred;
471 int fflag = ap->a_fflag;
472 struct thread *td = ap->a_td;
473 pid_t pid = td->td_proc->p_pid;
476 if (fuse_isdeadfs(vp))
480 if (fflag & IO_NDELAY)
483 err = fuse_flush(vp, cred, pid, fflag);
484 /* TODO: close the file handle, if we're sure it's no longer used */
485 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
486 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
492 fdisp_make_mknod_for_fallback(
493 struct fuse_dispatcher *fdip,
494 struct componentname *cnp,
500 enum fuse_opcode *op)
502 struct fuse_mknod_in *fmni;
504 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
506 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
510 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
512 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
515 struct vnop_create_args {
517 struct vnode **a_vpp;
518 struct componentname *a_cnp;
523 fuse_vnop_create(struct vop_create_args *ap)
525 struct vnode *dvp = ap->a_dvp;
526 struct vnode **vpp = ap->a_vpp;
527 struct componentname *cnp = ap->a_cnp;
528 struct vattr *vap = ap->a_vap;
529 struct thread *td = cnp->cn_thread;
530 struct ucred *cred = cnp->cn_cred;
532 struct fuse_open_in *foi;
533 struct fuse_entry_out *feo;
534 struct fuse_open_out *foo;
535 struct fuse_dispatcher fdi, fdi2;
536 struct fuse_dispatcher *fdip = &fdi;
537 struct fuse_dispatcher *fdip2 = NULL;
541 struct mount *mp = vnode_mount(dvp);
542 uint64_t parentnid = VTOFUD(dvp)->nid;
543 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
547 if (fuse_isdeadfs(dvp))
550 /* FUSE expects sockets to be created with FUSE_MKNOD */
551 if (vap->va_type == VSOCK)
552 return fuse_internal_mknod(dvp, vpp, cnp, vap);
555 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
556 * writable mode makes sense, and we might as well include readability
561 bzero(&fdi, sizeof(fdi));
563 if (vap->va_type != VREG)
566 if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
567 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
568 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
571 /* Use FUSE_CREATE */
573 fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
574 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
577 foi->flags = O_CREAT | flags;
578 memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
580 ((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
583 err = fdisp_wait_answ(fdip);
586 if (err == ENOSYS && op == FUSE_CREATE) {
587 fsess_set_notimpl(mp, FUSE_CREATE);
588 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
589 parentnid, td, cred, mode, &op);
590 err = fdisp_wait_answ(fdip);
598 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
602 if (op == FUSE_CREATE) {
603 foo = (struct fuse_open_out*)(feo + 1);
605 /* Issue a separate FUSE_OPEN */
607 fdisp_init(fdip2, sizeof(*foi));
608 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
613 err = fdisp_wait_answ(fdip2);
618 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
620 struct fuse_release_in *fri;
621 uint64_t nodeid = feo->nodeid;
622 uint64_t fh_id = foo->fh;
624 fdisp_init(fdip, sizeof(*fri));
625 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
629 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
630 fuse_insert_message(fdip->tick, false);
633 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
634 fuse_internal_cache_attrs(*vpp, cred, &feo->attr, feo->attr_valid,
635 feo->attr_valid_nsec, NULL);
637 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
638 fuse_vnode_open(*vpp, foo->open_flags, td);
640 * Purge the parent's attribute cache because the daemon should've
641 * updated its mtime and ctime
643 fuse_vnode_clear_attr_cache(dvp);
644 cache_purge_negative(dvp);
648 fdisp_destroy(fdip2);
654 struct vnop_fdatasync_args {
655 struct vop_generic_args a_gen;
657 struct thread * a_td;
661 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
663 struct vnode *vp = ap->a_vp;
664 struct thread *td = ap->a_td;
665 int waitfor = MNT_WAIT;
669 if (fuse_isdeadfs(vp)) {
672 if ((err = vop_stdfdatasync_buf(ap)))
675 return fuse_internal_fsync(vp, td, waitfor, true);
679 struct vnop_fsync_args {
680 struct vop_generic_args a_gen;
683 struct thread * a_td;
687 fuse_vnop_fsync(struct vop_fsync_args *ap)
689 struct vnode *vp = ap->a_vp;
690 struct thread *td = ap->a_td;
691 int waitfor = ap->a_waitfor;
694 if (fuse_isdeadfs(vp)) {
697 if ((err = vop_stdfsync(ap)))
700 return fuse_internal_fsync(vp, td, waitfor, false);
704 struct vnop_getattr_args {
707 struct ucred *a_cred;
712 fuse_vnop_getattr(struct vop_getattr_args *ap)
714 struct vnode *vp = ap->a_vp;
715 struct vattr *vap = ap->a_vap;
716 struct ucred *cred = ap->a_cred;
717 struct thread *td = curthread;
722 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
724 /* Note that we are not bailing out on a dead file system just yet. */
726 if (!(dataflags & FSESS_INITED)) {
727 if (!vnode_isvroot(vp)) {
728 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
735 err = fuse_internal_getattr(vp, vap, cred, td);
736 if (err == ENOTCONN && vnode_isvroot(vp)) {
737 /* see comment in fuse_vfsop_statfs() */
744 bzero(vap, sizeof(*vap));
745 vap->va_type = vnode_vtype(vp);
751 struct vnop_inactive_args {
757 fuse_vnop_inactive(struct vop_inactive_args *ap)
759 struct vnode *vp = ap->a_vp;
760 struct thread *td = ap->a_td;
762 struct fuse_vnode_data *fvdat = VTOFUD(vp);
763 struct fuse_filehandle *fufh, *fufh_tmp;
767 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
768 if (need_flush && vp->v_type == VREG) {
769 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
770 fuse_vnode_savesize(vp, NULL, 0);
772 if ((fvdat->flag & FN_REVOKED) != 0)
773 fuse_io_invalbuf(vp, td);
775 fuse_io_flushbuf(vp, MNT_WAIT, td);
778 fuse_filehandle_close(vp, fufh, td, NULL);
781 if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
788 struct vnop_link_args {
789 struct vnode *a_tdvp;
791 struct componentname *a_cnp;
795 fuse_vnop_link(struct vop_link_args *ap)
797 struct vnode *vp = ap->a_vp;
798 struct vnode *tdvp = ap->a_tdvp;
799 struct componentname *cnp = ap->a_cnp;
800 struct ucred *cred = cnp->cn_cred;
802 struct vattr *vap = VTOVA(vp);
804 struct fuse_dispatcher fdi;
805 struct fuse_entry_out *feo;
806 struct fuse_link_in fli;
810 if (fuse_isdeadfs(vp)) {
813 if (vnode_mount(tdvp) != vnode_mount(vp)) {
818 * This is a seatbelt check to protect naive userspace filesystems from
819 * themselves and the limitations of the FUSE IPC protocol. If a
820 * filesystem does not allow attribute caching, assume it is capable of
821 * validating that nlink does not overflow.
823 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
825 fli.oldnodeid = VTOI(vp);
828 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
829 FUSE_LINK, &fli, sizeof(fli), &fdi);
830 if ((err = fdisp_wait_answ(&fdi))) {
835 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
838 * Purge the parent's attribute cache because the daemon
839 * should've updated its mtime and ctime
841 fuse_vnode_clear_attr_cache(tdvp);
842 fuse_internal_cache_attrs(vp, cred, &feo->attr, feo->attr_valid,
843 feo->attr_valid_nsec, NULL);
850 struct fuse_lookup_alloc_arg {
851 struct fuse_entry_out *feo;
852 struct componentname *cnp;
857 /* Callback for vn_get_ino */
859 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
861 struct fuse_lookup_alloc_arg *flaa = arg;
863 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
867 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
868 "int", "struct timespec*", "struct timespec*");
870 struct vnop_lookup_args {
871 struct vnodeop_desc *a_desc;
873 struct vnode **a_vpp;
874 struct componentname *a_cnp;
878 fuse_vnop_lookup(struct vop_lookup_args *ap)
880 struct vnode *dvp = ap->a_dvp;
881 struct vnode **vpp = ap->a_vpp;
882 struct componentname *cnp = ap->a_cnp;
883 struct thread *td = cnp->cn_thread;
884 struct ucred *cred = cnp->cn_cred;
886 int nameiop = cnp->cn_nameiop;
887 int flags = cnp->cn_flags;
888 int wantparent = flags & (LOCKPARENT | WANTPARENT);
889 int islastcn = flags & ISLASTCN;
890 struct mount *mp = vnode_mount(dvp);
894 struct vnode *vp = NULL;
896 struct fuse_dispatcher fdi;
897 bool did_lookup = false;
898 struct fuse_entry_out *feo = NULL;
899 enum vtype vtyp; /* vnode type of target */
900 off_t filesize; /* filesize of target */
904 if (fuse_isdeadfs(dvp)) {
908 if (!vnode_isdir(dvp))
911 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
914 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
917 if (flags & ISDOTDOT) {
918 KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
919 ("Looking up .. is TODO"));
920 nid = VTOFUD(dvp)->parent_nid;
923 /* .. is obviously a directory */
926 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
928 /* . is obviously a directory */
932 struct timespec now, timeout;
934 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
936 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
938 case -1: /* positive match */
939 if (timespeccmp(&timeout, &now, >)) {
940 atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
943 atomic_add_acq_long(&fuse_lookup_cache_misses,
955 case 0: /* no match in cache */
956 atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
959 case ENOENT: /* negative match */
961 if (timespeccmp(&timeout, &now, <=)) {
963 cache_purge_negative(dvp);
972 fdisp_init(&fdi, cnp->cn_namelen + 1);
973 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
975 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
976 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
977 lookup_err = fdisp_wait_answ(&fdi);
981 /* lookup call succeeded */
982 feo = (struct fuse_entry_out *)fdi.answ;
985 /* zero nodeid means ENOENT and cache it */
986 struct timespec timeout;
988 fdi.answ_stat = ENOENT;
990 if (cnp->cn_flags & MAKEENTRY) {
991 fuse_validity_2_timespec(feo, &timeout);
992 cache_enter_time(dvp, *vpp, cnp,
995 } else if (nid == FUSE_ROOT_ID) {
998 vtyp = IFTOVT(feo->attr.mode);
999 filesize = feo->attr.size;
1001 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1002 fdisp_destroy(&fdi);
1006 /* lookup_err, if non-zero, must be ENOENT at this point */
1009 /* Entry not found */
1010 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1011 err = fuse_internal_access(dvp, VWRITE, td, cred);
1014 * Set the SAVENAME flag to hold onto the
1015 * pathname for use later in VOP_CREATE or
1018 cnp->cn_flags |= SAVENAME;
1026 /* Entry was found */
1027 if (flags & ISDOTDOT) {
1028 struct fuse_lookup_alloc_arg flaa;
1034 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1037 } else if (nid == VTOI(dvp)) {
1041 struct fuse_vnode_data *fvdat;
1043 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1050 * In the case where we are looking up a FUSE node
1051 * represented by an existing cached vnode, and the
1052 * true size reported by FUSE_LOOKUP doesn't match
1053 * the vnode's cached size, then any cached writes
1054 * beyond the file's current size are lost.
1057 * * following attribute cache expiration, or
1058 * * due a bug in the daemon, or
1061 if (vnode_isreg(vp) &&
1062 filesize != fvdat->cached_attrs.va_size &&
1063 fvdat->flag & FN_SIZECHANGE) {
1065 * The FN_SIZECHANGE flag reflects a dirty
1066 * append. If userspace lets us know our cache
1067 * is invalid, that write was lost. (Dirty
1068 * writes that do not cause append are also
1069 * lost, but we don't detect them here.)
1071 * XXX: Maybe disable WB caching on this mount.
1073 printf("%s: WB cache incoherent on %s!\n",
1075 vnode_mount(vp)->mnt_stat.f_mntonname);
1077 fvdat->flag &= ~FN_SIZECHANGE;
1081 fuse_internal_cache_attrs(*vpp, cred, &feo->attr,
1082 feo->attr_valid, feo->attr_valid_nsec, NULL);
1084 if ((nameiop == DELETE || nameiop == RENAME) &&
1087 struct vattr dvattr;
1089 err = fuse_internal_access(dvp, VWRITE, td,
1094 * if the parent's sticky bit is set, check
1095 * whether we're allowed to remove the file.
1096 * Need to figure out the vnode locking to make
1099 fuse_internal_getattr(dvp, &dvattr, cred, td);
1100 if ((dvattr.va_mode & S_ISTXT) &&
1101 fuse_internal_access(dvp, VADMIN, td,
1103 fuse_internal_access(*vpp, VADMIN, td,
1111 (nameiop == DELETE) ||
1112 (nameiop == RENAME && wantparent))) {
1113 cnp->cn_flags |= SAVENAME;
1120 if (vp != NULL && dvp != vp)
1122 else if (vp != NULL)
1127 fdisp_destroy(&fdi);
1133 struct vnop_mkdir_args {
1134 struct vnode *a_dvp;
1135 struct vnode **a_vpp;
1136 struct componentname *a_cnp;
1137 struct vattr *a_vap;
1141 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1143 struct vnode *dvp = ap->a_dvp;
1144 struct vnode **vpp = ap->a_vpp;
1145 struct componentname *cnp = ap->a_cnp;
1146 struct vattr *vap = ap->a_vap;
1148 struct fuse_mkdir_in fmdi;
1150 if (fuse_isdeadfs(dvp)) {
1153 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1155 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1156 sizeof(fmdi), VDIR));
1160 struct vnop_mknod_args {
1161 struct vnode *a_dvp;
1162 struct vnode **a_vpp;
1163 struct componentname *a_cnp;
1164 struct vattr *a_vap;
1168 fuse_vnop_mknod(struct vop_mknod_args *ap)
1171 struct vnode *dvp = ap->a_dvp;
1172 struct vnode **vpp = ap->a_vpp;
1173 struct componentname *cnp = ap->a_cnp;
1174 struct vattr *vap = ap->a_vap;
1176 if (fuse_isdeadfs(dvp))
1179 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1183 struct vop_open_args {
1186 struct ucred *a_cred;
1187 struct thread *a_td;
1188 int a_fdidx; / struct file *a_fp;
1192 fuse_vnop_open(struct vop_open_args *ap)
1194 struct vnode *vp = ap->a_vp;
1195 int a_mode = ap->a_mode;
1196 struct thread *td = ap->a_td;
1197 struct ucred *cred = ap->a_cred;
1198 pid_t pid = td->td_proc->p_pid;
1199 struct fuse_vnode_data *fvdat;
1201 if (fuse_isdeadfs(vp))
1203 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1204 return (EOPNOTSUPP);
1205 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1210 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1211 fuse_vnode_open(vp, 0, td);
1215 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1219 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1222 switch (ap->a_name) {
1223 case _PC_FILESIZEBITS:
1227 *ap->a_retval = NAME_MAX;
1230 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1232 case _PC_SYMLINK_MAX:
1233 *ap->a_retval = MAXPATHLEN;
1239 return (vop_stdpathconf(ap));
1244 struct vnop_read_args {
1248 struct ucred *a_cred;
1252 fuse_vnop_read(struct vop_read_args *ap)
1254 struct vnode *vp = ap->a_vp;
1255 struct uio *uio = ap->a_uio;
1256 int ioflag = ap->a_ioflag;
1257 struct ucred *cred = ap->a_cred;
1258 pid_t pid = curthread->td_proc->p_pid;
1260 if (fuse_isdeadfs(vp)) {
1264 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1265 ioflag |= IO_DIRECT;
1268 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1272 struct vnop_readdir_args {
1275 struct ucred *a_cred;
1282 fuse_vnop_readdir(struct vop_readdir_args *ap)
1284 struct vnode *vp = ap->a_vp;
1285 struct uio *uio = ap->a_uio;
1286 struct ucred *cred = ap->a_cred;
1287 struct fuse_filehandle *fufh = NULL;
1288 struct fuse_iov cookediov;
1294 bool closefufh = false;
1295 pid_t pid = curthread->td_proc->p_pid;
1299 if (fuse_isdeadfs(vp)) {
1302 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1303 (uio_resid(uio) < sizeof(struct dirent))) {
1307 tresid = uio->uio_resid;
1308 startoff = uio->uio_offset;
1309 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1310 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1312 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1313 * must implicitly open the directory here
1315 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1318 * When a directory is opened, it must be read from
1319 * the beginning. Hopefully, the "startoff" still
1320 * exists as an offset cookie for the directory.
1321 * If not, it will read the entire directory without
1322 * returning any entries and just return eof.
1324 uio->uio_offset = 0;
1330 if (ap->a_ncookies != NULL) {
1331 ncookies = uio->uio_resid /
1332 (offsetof(struct dirent, d_name) + 4) + 1;
1333 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1334 *ap->a_ncookies = ncookies;
1335 *ap->a_cookies = cookies;
1340 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1341 fiov_init(&cookediov, DIRCOOKEDSIZE);
1343 err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1344 &ncookies, cookies);
1346 fiov_teardown(&cookediov);
1348 fuse_filehandle_close(vp, fufh, curthread, cred);
1350 if (ap->a_ncookies != NULL) {
1352 *ap->a_ncookies -= ncookies;
1354 free(*ap->a_cookies, M_TEMP);
1355 *ap->a_ncookies = 0;
1356 *ap->a_cookies = NULL;
1359 if (err == 0 && tresid == uio->uio_resid)
1366 struct vnop_readlink_args {
1369 struct ucred *a_cred;
1373 fuse_vnop_readlink(struct vop_readlink_args *ap)
1375 struct vnode *vp = ap->a_vp;
1376 struct uio *uio = ap->a_uio;
1377 struct ucred *cred = ap->a_cred;
1379 struct fuse_dispatcher fdi;
1382 if (fuse_isdeadfs(vp)) {
1385 if (!vnode_islnk(vp)) {
1388 fdisp_init(&fdi, 0);
1389 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1393 if (((char *)fdi.answ)[0] == '/' &&
1394 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1395 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1397 err = uiomove(mpth, strlen(mpth), uio);
1400 err = uiomove(fdi.answ, fdi.iosize, uio);
1403 fdisp_destroy(&fdi);
1408 struct vnop_reclaim_args {
1410 struct thread *a_td;
1414 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1416 struct vnode *vp = ap->a_vp;
1417 struct thread *td = ap->a_td;
1418 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1419 struct fuse_filehandle *fufh, *fufh_tmp;
1422 panic("FUSE: no vnode data during recycling");
1424 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1425 printf("FUSE: vnode being reclaimed with open fufh "
1426 "(type=%#x)", fufh->fufh_type);
1427 fuse_filehandle_close(vp, fufh, td, NULL);
1430 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1431 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1434 fuse_vnode_setparent(vp, NULL);
1436 vfs_hash_remove(vp);
1437 vnode_destroy_vobject(vp);
1438 fuse_vnode_destroy(vp);
1444 struct vnop_remove_args {
1445 struct vnode *a_dvp;
1447 struct componentname *a_cnp;
1451 fuse_vnop_remove(struct vop_remove_args *ap)
1453 struct vnode *dvp = ap->a_dvp;
1454 struct vnode *vp = ap->a_vp;
1455 struct componentname *cnp = ap->a_cnp;
1459 if (fuse_isdeadfs(vp)) {
1462 if (vnode_isdir(vp)) {
1467 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1470 fuse_internal_vnode_disappear(vp);
1472 * Purge the parent's attribute cache because the daemon
1473 * should've updated its mtime and ctime
1475 fuse_vnode_clear_attr_cache(dvp);
1481 struct vnop_rename_args {
1482 struct vnode *a_fdvp;
1483 struct vnode *a_fvp;
1484 struct componentname *a_fcnp;
1485 struct vnode *a_tdvp;
1486 struct vnode *a_tvp;
1487 struct componentname *a_tcnp;
1491 fuse_vnop_rename(struct vop_rename_args *ap)
1493 struct vnode *fdvp = ap->a_fdvp;
1494 struct vnode *fvp = ap->a_fvp;
1495 struct componentname *fcnp = ap->a_fcnp;
1496 struct vnode *tdvp = ap->a_tdvp;
1497 struct vnode *tvp = ap->a_tvp;
1498 struct componentname *tcnp = ap->a_tcnp;
1499 struct fuse_data *data;
1500 bool newparent = fdvp != tdvp;
1501 bool isdir = fvp->v_type == VDIR;
1504 if (fuse_isdeadfs(fdvp)) {
1507 if (fvp->v_mount != tdvp->v_mount ||
1508 (tvp && fvp->v_mount != tvp->v_mount)) {
1509 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1516 * FUSE library is expected to check if target directory is not
1517 * under the source directory in the file system tree.
1518 * Linux performs this check at VFS level.
1521 * If source is a directory, and it will get a new parent, user must
1522 * have write permission to it, so ".." can be modified.
1524 data = fuse_get_mpdata(vnode_mount(tdvp));
1525 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1526 err = fuse_internal_access(fvp, VWRITE,
1527 tcnp->cn_thread, tcnp->cn_cred);
1531 sx_xlock(&data->rename_lock);
1532 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1535 fuse_vnode_setparent(fvp, tdvp);
1537 fuse_vnode_setparent(tvp, NULL);
1539 sx_unlock(&data->rename_lock);
1541 if (tvp != NULL && tvp != fvp) {
1544 if (vnode_isdir(fvp)) {
1545 if ((tvp != NULL) && vnode_isdir(tvp)) {
1566 struct vnop_rmdir_args {
1567 struct vnode *a_dvp;
1569 struct componentname *a_cnp;
1573 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1575 struct vnode *dvp = ap->a_dvp;
1576 struct vnode *vp = ap->a_vp;
1580 if (fuse_isdeadfs(vp)) {
1583 if (VTOFUD(vp) == VTOFUD(dvp)) {
1586 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1589 fuse_internal_vnode_disappear(vp);
1591 * Purge the parent's attribute cache because the daemon
1592 * should've updated its mtime and ctime
1594 fuse_vnode_clear_attr_cache(dvp);
1600 struct vnop_setattr_args {
1602 struct vattr *a_vap;
1603 struct ucred *a_cred;
1604 struct thread *a_td;
1608 fuse_vnop_setattr(struct vop_setattr_args *ap)
1610 struct vnode *vp = ap->a_vp;
1611 struct vattr *vap = ap->a_vap;
1612 struct ucred *cred = ap->a_cred;
1613 struct thread *td = curthread;
1615 struct fuse_data *data;
1616 struct vattr old_va;
1619 accmode_t accmode = 0;
1621 bool drop_suid = false;
1624 mp = vnode_mount(vp);
1625 data = fuse_get_mpdata(mp);
1626 dataflags = data->dataflags;
1627 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1628 if (cred->cr_ngroups > 0)
1629 cr_gid = cred->cr_groups[0];
1633 if (fuse_isdeadfs(vp)) {
1637 if (vap->va_uid != (uid_t)VNOVAL) {
1639 /* Only root may change a file's owner */
1640 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1642 /* As a special case, allow the null chown */
1643 err2 = fuse_internal_getattr(vp, &old_va, cred,
1647 if (vap->va_uid != old_va.va_uid)
1657 if (vap->va_gid != (gid_t)VNOVAL) {
1658 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1660 if (checkperm && !groupmember(vap->va_gid, cred))
1663 * Non-root users may only chgrp to one of their own
1666 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1668 /* As a special case, allow the null chgrp */
1669 err2 = fuse_internal_getattr(vp, &old_va, cred,
1673 if (vap->va_gid != old_va.va_gid)
1681 if (vap->va_size != VNOVAL) {
1682 switch (vp->v_type) {
1687 if (vfs_isrdonly(mp))
1692 * According to POSIX, the result is unspecified
1693 * for file types other than regular files,
1694 * directories and shared memory objects. We
1695 * don't support shared memory objects in the file
1696 * system, and have dubious support for truncating
1697 * symlinks. Just ignore the request in other cases.
1701 /* Don't set accmode. Permission to trunc is checked upstack */
1703 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1704 if (vap->va_vaflags & VA_UTIMES_NULL)
1710 if (vap->va_mode != (mode_t)VNOVAL)
1711 vap->va_mode &= ~(S_ISUID | S_ISGID);
1713 err = fuse_internal_getattr(vp, &old_va, cred, td);
1716 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1719 if (vap->va_mode != (mode_t)VNOVAL) {
1720 /* Only root may set the sticky bit on non-directories */
1721 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1722 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1724 if (checkperm && (vap->va_mode & S_ISGID)) {
1725 err = fuse_internal_getattr(vp, &old_va, cred, td);
1728 if (!groupmember(old_va.va_gid, cred)) {
1729 err = priv_check_cred(cred, PRIV_VFS_SETGID);
1737 if (vfs_isrdonly(mp))
1740 err = fuse_internal_access(vp, accmode, td, cred);
1744 return fuse_internal_setattr(vp, vap, td, cred);
1748 struct vnop_strategy_args {
1754 fuse_vnop_strategy(struct vop_strategy_args *ap)
1756 struct vnode *vp = ap->a_vp;
1757 struct buf *bp = ap->a_bp;
1759 if (!vp || fuse_isdeadfs(vp)) {
1760 bp->b_ioflags |= BIO_ERROR;
1761 bp->b_error = ENXIO;
1767 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1768 * fuse_io_strategy sets bp's error fields
1770 (void)fuse_io_strategy(vp, bp);
1777 struct vnop_symlink_args {
1778 struct vnode *a_dvp;
1779 struct vnode **a_vpp;
1780 struct componentname *a_cnp;
1781 struct vattr *a_vap;
1786 fuse_vnop_symlink(struct vop_symlink_args *ap)
1788 struct vnode *dvp = ap->a_dvp;
1789 struct vnode **vpp = ap->a_vpp;
1790 struct componentname *cnp = ap->a_cnp;
1791 const char *target = ap->a_target;
1793 struct fuse_dispatcher fdi;
1798 if (fuse_isdeadfs(dvp)) {
1802 * Unlike the other creator type calls, here we have to create a message
1803 * where the name of the new entry comes first, and the data describing
1804 * the entry comes second.
1805 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1806 * but put together the message manually and just call the core part.
1809 len = strlen(target) + 1;
1810 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1811 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1813 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1814 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1815 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1817 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1818 fdisp_destroy(&fdi);
1823 struct vnop_write_args {
1827 struct ucred *a_cred;
1831 fuse_vnop_write(struct vop_write_args *ap)
1833 struct vnode *vp = ap->a_vp;
1834 struct uio *uio = ap->a_uio;
1835 int ioflag = ap->a_ioflag;
1836 struct ucred *cred = ap->a_cred;
1837 pid_t pid = curthread->td_proc->p_pid;
1839 if (fuse_isdeadfs(vp)) {
1843 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1844 ioflag |= IO_DIRECT;
1847 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1850 SDT_PROBE_DEFINE1(fusefs, , vnops, vnop_getpages_error, "int");
1852 struct vnop_getpages_args {
1860 fuse_vnop_getpages(struct vop_getpages_args *ap)
1862 int i, error, nextoff, size, toff, count, npages;
1871 pid_t pid = curthread->td_proc->p_pid;
1874 KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1875 td = curthread; /* XXX */
1876 cred = curthread->td_ucred; /* XXX */
1878 npages = ap->a_count;
1880 if (!fsess_opt_mmap(vnode_mount(vp))) {
1881 SDT_PROBE2(fusefs, , vnops, trace, 1,
1882 "called on non-cacheable vnode??\n");
1883 return (VM_PAGER_ERROR);
1887 * If the last page is partially valid, just return it and allow
1888 * the pager to zero-out the blanks. Partially valid pages can
1889 * only occur at the file EOF.
1891 * XXXGL: is that true for FUSE, which is a local filesystem,
1892 * but still somewhat disconnected from the kernel?
1894 VM_OBJECT_WLOCK(vp->v_object);
1895 if (pages[npages - 1]->valid != 0 && --npages == 0)
1897 VM_OBJECT_WUNLOCK(vp->v_object);
1900 * We use only the kva address for the buffer, but this is extremely
1901 * convenient and fast.
1903 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1905 kva = (vm_offset_t)bp->b_data;
1906 pmap_qenter(kva, pages, npages);
1907 VM_CNT_INC(v_vnodein);
1908 VM_CNT_ADD(v_vnodepgsin, npages);
1910 count = npages << PAGE_SHIFT;
1911 iov.iov_base = (caddr_t)kva;
1912 iov.iov_len = count;
1915 uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1916 uio.uio_resid = count;
1917 uio.uio_segflg = UIO_SYSSPACE;
1918 uio.uio_rw = UIO_READ;
1921 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
1922 pmap_qremove(kva, npages);
1924 uma_zfree(fuse_pbuf_zone, bp);
1926 if (error && (uio.uio_resid == count)) {
1927 SDT_PROBE1(fusefs, , vnops, vnop_getpages_error, error);
1928 return VM_PAGER_ERROR;
1931 * Calculate the number of bytes read and validate only that number
1932 * of bytes. Note that due to pending writes, size may be 0. This
1933 * does not mean that the remaining data is invalid!
1936 size = count - uio.uio_resid;
1937 VM_OBJECT_WLOCK(vp->v_object);
1938 fuse_vm_page_lock_queues();
1939 for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1942 nextoff = toff + PAGE_SIZE;
1945 if (nextoff <= size) {
1947 * Read operation filled an entire page
1949 m->valid = VM_PAGE_BITS_ALL;
1950 KASSERT(m->dirty == 0,
1951 ("fuse_getpages: page %p is dirty", m));
1952 } else if (size > toff) {
1954 * Read operation filled a partial page.
1957 vm_page_set_valid_range(m, 0, size - toff);
1958 KASSERT(m->dirty == 0,
1959 ("fuse_getpages: page %p is dirty", m));
1962 * Read operation was short. If no error occurred
1963 * we may have hit a zero-fill section. We simply
1964 * leave valid set to 0.
1969 fuse_vm_page_unlock_queues();
1971 VM_OBJECT_WUNLOCK(vp->v_object);
1976 return (VM_PAGER_OK);
1980 struct vnop_putpages_args {
1986 vm_ooffset_t a_offset;
1990 fuse_vnop_putpages(struct vop_putpages_args *ap)
1996 int i, error, npages, count;
2004 pid_t pid = curthread->td_proc->p_pid;
2007 KASSERT(vp->v_object, ("objectless vp passed to putpages"));
2008 fsize = vp->v_object->un_pager.vnp.vnp_size;
2009 td = curthread; /* XXX */
2010 cred = curthread->td_ucred; /* XXX */
2012 count = ap->a_count;
2013 rtvals = ap->a_rtvals;
2014 npages = btoc(count);
2015 offset = IDX_TO_OFF(pages[0]->pindex);
2017 if (!fsess_opt_mmap(vnode_mount(vp))) {
2018 SDT_PROBE2(fusefs, , vnops, trace, 1,
2019 "called on non-cacheable vnode??\n");
2021 for (i = 0; i < npages; i++)
2022 rtvals[i] = VM_PAGER_AGAIN;
2025 * When putting pages, do not extend file past EOF.
2028 if (offset + count > fsize) {
2029 count = fsize - offset;
2034 * We use only the kva address for the buffer, but this is extremely
2035 * convenient and fast.
2037 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
2039 kva = (vm_offset_t)bp->b_data;
2040 pmap_qenter(kva, pages, npages);
2041 VM_CNT_INC(v_vnodeout);
2042 VM_CNT_ADD(v_vnodepgsout, count);
2044 iov.iov_base = (caddr_t)kva;
2045 iov.iov_len = count;
2048 uio.uio_offset = offset;
2049 uio.uio_resid = count;
2050 uio.uio_segflg = UIO_SYSSPACE;
2051 uio.uio_rw = UIO_WRITE;
2054 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
2056 pmap_qremove(kva, npages);
2057 uma_zfree(fuse_pbuf_zone, bp);
2060 int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
2062 for (i = 0; i < nwritten; i++) {
2063 rtvals[i] = VM_PAGER_OK;
2064 VM_OBJECT_WLOCK(pages[i]->object);
2065 vm_page_undirty(pages[i]);
2066 VM_OBJECT_WUNLOCK(pages[i]->object);
2072 static const char extattr_namespace_separator = '.';
2075 struct vop_getextattr_args {
2076 struct vop_generic_args a_gen;
2078 int a_attrnamespace;
2082 struct ucred *a_cred;
2083 struct thread *a_td;
2087 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2089 struct vnode *vp = ap->a_vp;
2090 struct uio *uio = ap->a_uio;
2091 struct fuse_dispatcher fdi;
2092 struct fuse_getxattr_in *get_xattr_in;
2093 struct fuse_getxattr_out *get_xattr_out;
2094 struct mount *mp = vnode_mount(vp);
2095 struct thread *td = ap->a_td;
2096 struct ucred *cred = ap->a_cred;
2102 if (fuse_isdeadfs(vp))
2105 if (!fsess_isimpl(mp, FUSE_GETXATTR))
2108 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2112 /* Default to looking for user attributes. */
2113 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2114 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2116 prefix = EXTATTR_NAMESPACE_USER_STRING;
2118 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2119 strlen(ap->a_name) + 1;
2121 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2122 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2124 get_xattr_in = fdi.indata;
2126 * Check to see whether we're querying the available size or
2127 * issuing the actual request. If we pass in 0, we get back struct
2128 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2129 * that much data, without the struct fuse_getxattr_out header.
2132 get_xattr_in->size = 0;
2134 get_xattr_in->size = uio->uio_resid;
2136 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2137 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2140 err = fdisp_wait_answ(&fdi);
2142 if (err == ENOSYS) {
2143 fsess_set_notimpl(mp, FUSE_GETXATTR);
2149 get_xattr_out = fdi.answ;
2151 if (ap->a_size != NULL)
2152 *ap->a_size = get_xattr_out->size;
2155 err = uiomove(fdi.answ, fdi.iosize, uio);
2158 fdisp_destroy(&fdi);
2163 struct vop_setextattr_args {
2164 struct vop_generic_args a_gen;
2166 int a_attrnamespace;
2169 struct ucred *a_cred;
2170 struct thread *a_td;
2174 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2176 struct vnode *vp = ap->a_vp;
2177 struct uio *uio = ap->a_uio;
2178 struct fuse_dispatcher fdi;
2179 struct fuse_setxattr_in *set_xattr_in;
2180 struct mount *mp = vnode_mount(vp);
2181 struct thread *td = ap->a_td;
2182 struct ucred *cred = ap->a_cred;
2188 if (fuse_isdeadfs(vp))
2191 if (!fsess_isimpl(mp, FUSE_SETXATTR))
2194 if (vfs_isrdonly(mp))
2197 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2198 if (ap->a_uio == NULL) {
2200 * If we got here as fallback from VOP_DELETEEXTATTR, then
2201 * return EOPNOTSUPP.
2203 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2204 return (EOPNOTSUPP);
2209 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2214 /* Default to looking for user attributes. */
2215 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2216 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2218 prefix = EXTATTR_NAMESPACE_USER_STRING;
2220 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2221 strlen(ap->a_name) + 1;
2223 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2224 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2226 set_xattr_in = fdi.indata;
2227 set_xattr_in->size = uio->uio_resid;
2229 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2230 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2233 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2234 uio->uio_resid, uio);
2239 err = fdisp_wait_answ(&fdi);
2241 if (err == ENOSYS) {
2242 fsess_set_notimpl(mp, FUSE_SETXATTR);
2245 if (err == ERESTART) {
2246 /* Can't restart after calling uiomove */
2251 fdisp_destroy(&fdi);
2256 * The Linux / FUSE extended attribute list is simply a collection of
2257 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2258 * byte length followed by a non-NUL terminated string. So, this allows
2259 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2260 * Linux attribute names are reported with the namespace as a prefix (e.g.
2261 * "user.attribute_name"), but in FreeBSD they are reported without the
2262 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2264 * user.attr_name1\0user.attr_name2\0
2268 * <num>attr_name1<num>attr_name2
2270 * Where "<num>" is a single byte number of characters in the attribute name.
2273 * prefix - exattr namespace prefix string
2274 * list, list_len - input list with namespace prefixes
2275 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2278 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2279 char *bsd_list, int *bsd_list_len)
2281 int len, pos, dist_to_next, prefix_len;
2285 prefix_len = strlen(prefix);
2287 while (pos < list_len && list[pos] != '\0') {
2288 dist_to_next = strlen(&list[pos]) + 1;
2289 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2290 list[pos + prefix_len] == extattr_namespace_separator) {
2291 len = dist_to_next -
2292 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2293 if (len >= EXTATTR_MAXNAMELEN)
2294 return (ENAMETOOLONG);
2296 bsd_list[*bsd_list_len] = len;
2297 memcpy(&bsd_list[*bsd_list_len + 1],
2298 &list[pos + prefix_len +
2299 sizeof(extattr_namespace_separator)], len);
2301 *bsd_list_len += len + 1;
2304 pos += dist_to_next;
2311 struct vop_listextattr_args {
2312 struct vop_generic_args a_gen;
2314 int a_attrnamespace;
2317 struct ucred *a_cred;
2318 struct thread *a_td;
2322 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2324 struct vnode *vp = ap->a_vp;
2325 struct uio *uio = ap->a_uio;
2326 struct fuse_dispatcher fdi;
2327 struct fuse_listxattr_in *list_xattr_in;
2328 struct fuse_listxattr_out *list_xattr_out;
2329 struct mount *mp = vnode_mount(vp);
2330 struct thread *td = ap->a_td;
2331 struct ucred *cred = ap->a_cred;
2335 char *bsd_list = NULL;
2341 if (fuse_isdeadfs(vp))
2344 if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2347 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2352 * Add space for a NUL and the period separator if enabled.
2353 * Default to looking for user attributes.
2355 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2356 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2358 prefix = EXTATTR_NAMESPACE_USER_STRING;
2360 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2362 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2363 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2366 * Retrieve Linux / FUSE compatible list size.
2368 list_xattr_in = fdi.indata;
2369 list_xattr_in->size = 0;
2370 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2371 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2373 err = fdisp_wait_answ(&fdi);
2375 if (err == ENOSYS) {
2376 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2382 list_xattr_out = fdi.answ;
2383 linux_list_len = list_xattr_out->size;
2384 if (linux_list_len == 0) {
2385 if (ap->a_size != NULL)
2386 *ap->a_size = linux_list_len;
2391 * Retrieve Linux / FUSE compatible list values.
2393 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2394 list_xattr_in = fdi.indata;
2395 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2396 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2397 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2399 err = fdisp_wait_answ(&fdi);
2403 linux_list = fdi.answ;
2404 linux_list_len = fdi.iosize;
2407 * Retrieve the BSD compatible list values.
2408 * The Linux / FUSE attribute list format isn't the same
2409 * as FreeBSD's format. So we need to transform it into
2410 * FreeBSD's format before giving it to the user.
2412 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2413 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2414 bsd_list, &bsd_list_len);
2418 if (ap->a_size != NULL)
2419 *ap->a_size = bsd_list_len;
2422 err = uiomove(bsd_list, bsd_list_len, uio);
2425 free(bsd_list, M_TEMP);
2426 fdisp_destroy(&fdi);
2431 struct vop_deleteextattr_args {
2432 struct vop_generic_args a_gen;
2434 int a_attrnamespace;
2436 struct ucred *a_cred;
2437 struct thread *a_td;
2441 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2443 struct vnode *vp = ap->a_vp;
2444 struct fuse_dispatcher fdi;
2445 struct mount *mp = vnode_mount(vp);
2446 struct thread *td = ap->a_td;
2447 struct ucred *cred = ap->a_cred;
2453 if (fuse_isdeadfs(vp))
2456 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2459 if (vfs_isrdonly(mp))
2462 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2467 /* Default to looking for user attributes. */
2468 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2469 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2471 prefix = EXTATTR_NAMESPACE_USER_STRING;
2473 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2474 strlen(ap->a_name) + 1;
2476 fdisp_init(&fdi, len);
2477 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2479 attr_str = fdi.indata;
2480 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2483 err = fdisp_wait_answ(&fdi);
2484 if (err == ENOSYS) {
2485 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2489 fdisp_destroy(&fdi);
2494 struct vnop_print_args {
2499 fuse_vnop_print(struct vop_print_args *ap)
2501 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2503 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2504 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2505 (uintmax_t)fvdat->nlookup,
2512 * Get an NFS filehandle for a FUSE file.
2514 * This will only work for FUSE file systems that guarantee the uniqueness of
2515 * nodeid:generation, which most don't
2519 IN struct vnode *a_vp;
2520 IN struct fid *a_fhp;
2524 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2526 struct vnode *vp = ap->a_vp;
2527 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2528 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2529 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2530 "FUSE fid type is too big");
2531 struct mount *mp = vnode_mount(vp);
2532 struct fuse_data *data = fuse_get_mpdata(mp);
2536 if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2539 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2543 /*ip = VTOI(ap->a_vp);*/
2544 /*ufhp = (struct ufid *)ap->a_fhp;*/
2545 fhp->len = sizeof(struct fuse_fid);
2546 fhp->nid = fvdat->nid;
2547 if (fvdat->generation <= UINT32_MAX)
2548 fhp->gen = fvdat->generation;