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,
186 * TODO: implement vop_ioctl after upgrading to protocol 7.16.
187 * FUSE_IOCTL was added in 7.11, but 32-bit compat is broken until
190 .vop_link = fuse_vnop_link,
191 .vop_listextattr = fuse_vnop_listextattr,
192 .vop_lookup = fuse_vnop_lookup,
193 .vop_mkdir = fuse_vnop_mkdir,
194 .vop_mknod = fuse_vnop_mknod,
195 .vop_open = fuse_vnop_open,
196 .vop_pathconf = fuse_vnop_pathconf,
198 * TODO: implement vop_poll after upgrading to protocol 7.21.
199 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
200 * 7.21, which adds the ability for the client to choose which poll
201 * events it wants, and for a client to deregister a file handle
203 .vop_read = fuse_vnop_read,
204 .vop_readdir = fuse_vnop_readdir,
205 .vop_readlink = fuse_vnop_readlink,
206 .vop_reclaim = fuse_vnop_reclaim,
207 .vop_remove = fuse_vnop_remove,
208 .vop_rename = fuse_vnop_rename,
209 .vop_rmdir = fuse_vnop_rmdir,
210 .vop_setattr = fuse_vnop_setattr,
211 .vop_setextattr = fuse_vnop_setextattr,
212 .vop_strategy = fuse_vnop_strategy,
213 .vop_symlink = fuse_vnop_symlink,
214 .vop_write = fuse_vnop_write,
215 .vop_getpages = fuse_vnop_getpages,
216 .vop_putpages = fuse_vnop_putpages,
217 .vop_print = fuse_vnop_print,
218 .vop_vptofh = fuse_vnop_vptofh,
221 u_long fuse_lookup_cache_hits = 0;
223 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
224 &fuse_lookup_cache_hits, 0, "number of positive cache hits in lookup");
226 u_long fuse_lookup_cache_misses = 0;
228 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
229 &fuse_lookup_cache_misses, 0, "number of cache misses in lookup");
232 * XXX: This feature is highly experimental and can bring to instabilities,
233 * needs revisiting before to be enabled by default.
235 static int fuse_reclaim_revoked = 0;
237 SYSCTL_INT(_vfs_fusefs, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
238 &fuse_reclaim_revoked, 0, "");
240 uma_zone_t fuse_pbuf_zone;
242 #define fuse_vm_page_lock(m) vm_page_lock((m));
243 #define fuse_vm_page_unlock(m) vm_page_unlock((m));
244 #define fuse_vm_page_lock_queues() ((void)0)
245 #define fuse_vm_page_unlock_queues() ((void)0)
247 /* Check permission for extattr operations, much like extattr_check_cred */
249 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
250 struct thread *td, accmode_t accmode)
252 struct mount *mp = vnode_mount(vp);
253 struct fuse_data *data = fuse_get_mpdata(mp);
256 * Kernel-invoked always succeeds.
262 * Do not allow privileged processes in jail to directly manipulate
266 case EXTATTR_NAMESPACE_SYSTEM:
267 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
268 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
271 case EXTATTR_NAMESPACE_USER:
272 return (fuse_internal_access(vp, accmode, td, cred));
278 /* Get a filehandle for a directory */
280 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
281 struct ucred *cred, pid_t pid)
283 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
285 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
288 /* Send FUSE_FLUSH for this vnode */
290 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
292 struct fuse_flush_in *ffi;
293 struct fuse_filehandle *fufh;
294 struct fuse_dispatcher fdi;
295 struct thread *td = curthread;
296 struct mount *mp = vnode_mount(vp);
299 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
302 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
306 fdisp_init(&fdi, sizeof(*ffi));
307 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
309 ffi->fh = fufh->fh_id;
311 * If the file has a POSIX lock then we're supposed to set lock_owner.
312 * If not, then lock_owner is undefined. So we may as well always set
315 ffi->lock_owner = td->td_proc->p_pid;
317 err = fdisp_wait_answ(&fdi);
319 fsess_set_notimpl(mp, FUSE_FLUSH);
326 /* Close wrapper for fifos. */
328 fuse_fifo_close(struct vop_close_args *ap)
330 return (fifo_specops.vop_close(ap));
334 struct vnop_access_args {
336 #if VOP_ACCESS_TAKES_ACCMODE_T
341 struct ucred *a_cred;
346 fuse_vnop_access(struct vop_access_args *ap)
348 struct vnode *vp = ap->a_vp;
349 int accmode = ap->a_accmode;
350 struct ucred *cred = ap->a_cred;
352 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
356 if (fuse_isdeadfs(vp)) {
357 if (vnode_isvroot(vp)) {
362 if (!(data->dataflags & FSESS_INITED)) {
363 if (vnode_isvroot(vp)) {
364 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
365 (fuse_match_cred(data->daemoncred, cred) == 0)) {
371 if (vnode_islnk(vp)) {
375 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
380 * struct vop_advlock_args {
381 * struct vop_generic_args a_gen;
382 * struct vnode *a_vp;
385 * struct flock *a_fl;
390 fuse_vnop_advlock(struct vop_advlock_args *ap)
392 struct vnode *vp = ap->a_vp;
393 struct flock *fl = ap->a_fl;
394 struct thread *td = curthread;
395 struct ucred *cred = td->td_ucred;
396 pid_t pid = td->td_proc->p_pid;
397 struct fuse_filehandle *fufh;
398 struct fuse_dispatcher fdi;
399 struct fuse_lk_in *fli;
400 struct fuse_lk_out *flo;
403 int flags = ap->a_flags;
405 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
407 if (fuse_isdeadfs(vp)) {
411 if (!(dataflags & FSESS_POSIX_LOCKS))
412 return vop_stdadvlock(ap);
413 /* FUSE doesn't properly support flock until protocol 7.17 */
415 return vop_stdadvlock(ap);
417 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
421 fdisp_init(&fdi, sizeof(*fli));
437 fdisp_make_vp(&fdi, op, vp, td, cred);
439 fli->fh = fufh->fh_id;
440 fli->owner = fl->l_pid;
441 fli->lk.start = fl->l_start;
443 fli->lk.end = fl->l_start + fl->l_len - 1;
445 fli->lk.end = INT64_MAX;
446 fli->lk.type = fl->l_type;
447 fli->lk.pid = fl->l_pid;
449 err = fdisp_wait_answ(&fdi);
452 if (err == 0 && op == FUSE_GETLK) {
454 fl->l_type = flo->lk.type;
455 fl->l_pid = flo->lk.pid;
456 if (flo->lk.type != F_UNLCK) {
457 fl->l_start = flo->lk.start;
458 if (flo->lk.end == INT64_MAX)
461 fl->l_len = flo->lk.end - flo->lk.start + 1;
462 fl->l_start = flo->lk.start;
470 struct vop_close_args {
473 struct ucred *a_cred;
478 fuse_vnop_close(struct vop_close_args *ap)
480 struct vnode *vp = ap->a_vp;
481 struct ucred *cred = ap->a_cred;
482 int fflag = ap->a_fflag;
483 struct thread *td = ap->a_td;
484 pid_t pid = td->td_proc->p_pid;
487 if (fuse_isdeadfs(vp))
491 if (fflag & IO_NDELAY)
494 err = fuse_flush(vp, cred, pid, fflag);
495 /* TODO: close the file handle, if we're sure it's no longer used */
496 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
497 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
503 fdisp_make_mknod_for_fallback(
504 struct fuse_dispatcher *fdip,
505 struct componentname *cnp,
511 enum fuse_opcode *op)
513 struct fuse_mknod_in *fmni;
515 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
517 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
521 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
523 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
526 struct vnop_create_args {
528 struct vnode **a_vpp;
529 struct componentname *a_cnp;
534 fuse_vnop_create(struct vop_create_args *ap)
536 struct vnode *dvp = ap->a_dvp;
537 struct vnode **vpp = ap->a_vpp;
538 struct componentname *cnp = ap->a_cnp;
539 struct vattr *vap = ap->a_vap;
540 struct thread *td = cnp->cn_thread;
541 struct ucred *cred = cnp->cn_cred;
543 struct fuse_data *data;
544 struct fuse_create_in *fci;
545 struct fuse_entry_out *feo;
546 struct fuse_open_out *foo;
547 struct fuse_dispatcher fdi, fdi2;
548 struct fuse_dispatcher *fdip = &fdi;
549 struct fuse_dispatcher *fdip2 = NULL;
553 struct mount *mp = vnode_mount(dvp);
554 data = fuse_get_mpdata(mp);
555 uint64_t parentnid = VTOFUD(dvp)->nid;
556 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
560 if (fuse_isdeadfs(dvp))
563 /* FUSE expects sockets to be created with FUSE_MKNOD */
564 if (vap->va_type == VSOCK)
565 return fuse_internal_mknod(dvp, vpp, cnp, vap);
568 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
569 * writable mode makes sense, and we might as well include readability
574 bzero(&fdi, sizeof(fdi));
576 if (vap->va_type != VREG)
579 if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
580 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
581 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
584 /* Use FUSE_CREATE */
588 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
589 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
592 fci->flags = O_CREAT | flags;
593 if (fuse_libabi_geq(data, 7, 12)) {
594 insize = sizeof(*fci);
595 fci->umask = td->td_proc->p_fd->fd_cmask;
597 insize = sizeof(struct fuse_open_in);
600 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
602 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
605 err = fdisp_wait_answ(fdip);
608 if (err == ENOSYS && op == FUSE_CREATE) {
609 fsess_set_notimpl(mp, FUSE_CREATE);
610 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
611 parentnid, td, cred, mode, &op);
612 err = fdisp_wait_answ(fdip);
620 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
624 if (op == FUSE_CREATE) {
625 foo = (struct fuse_open_out*)(feo + 1);
627 /* Issue a separate FUSE_OPEN */
628 struct fuse_open_in *foi;
631 fdisp_init(fdip2, sizeof(*foi));
632 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
636 err = fdisp_wait_answ(fdip2);
641 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
643 struct fuse_release_in *fri;
644 uint64_t nodeid = feo->nodeid;
645 uint64_t fh_id = foo->fh;
647 fdisp_init(fdip, sizeof(*fri));
648 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
652 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
653 fuse_insert_message(fdip->tick, false);
656 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
657 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
658 feo->attr_valid_nsec, NULL);
660 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
661 fuse_vnode_open(*vpp, foo->open_flags, td);
663 * Purge the parent's attribute cache because the daemon should've
664 * updated its mtime and ctime
666 fuse_vnode_clear_attr_cache(dvp);
667 cache_purge_negative(dvp);
671 fdisp_destroy(fdip2);
677 struct vnop_fdatasync_args {
678 struct vop_generic_args a_gen;
680 struct thread * a_td;
684 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
686 struct vnode *vp = ap->a_vp;
687 struct thread *td = ap->a_td;
688 int waitfor = MNT_WAIT;
692 if (fuse_isdeadfs(vp)) {
695 if ((err = vop_stdfdatasync_buf(ap)))
698 return fuse_internal_fsync(vp, td, waitfor, true);
702 struct vnop_fsync_args {
703 struct vop_generic_args a_gen;
706 struct thread * a_td;
710 fuse_vnop_fsync(struct vop_fsync_args *ap)
712 struct vnode *vp = ap->a_vp;
713 struct thread *td = ap->a_td;
714 int waitfor = ap->a_waitfor;
717 if (fuse_isdeadfs(vp)) {
720 if ((err = vop_stdfsync(ap)))
723 return fuse_internal_fsync(vp, td, waitfor, false);
727 struct vnop_getattr_args {
730 struct ucred *a_cred;
735 fuse_vnop_getattr(struct vop_getattr_args *ap)
737 struct vnode *vp = ap->a_vp;
738 struct vattr *vap = ap->a_vap;
739 struct ucred *cred = ap->a_cred;
740 struct thread *td = curthread;
745 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
747 /* Note that we are not bailing out on a dead file system just yet. */
749 if (!(dataflags & FSESS_INITED)) {
750 if (!vnode_isvroot(vp)) {
751 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
758 err = fuse_internal_getattr(vp, vap, cred, td);
759 if (err == ENOTCONN && vnode_isvroot(vp)) {
760 /* see comment in fuse_vfsop_statfs() */
767 bzero(vap, sizeof(*vap));
768 vap->va_type = vnode_vtype(vp);
774 struct vnop_inactive_args {
780 fuse_vnop_inactive(struct vop_inactive_args *ap)
782 struct vnode *vp = ap->a_vp;
783 struct thread *td = ap->a_td;
785 struct fuse_vnode_data *fvdat = VTOFUD(vp);
786 struct fuse_filehandle *fufh, *fufh_tmp;
790 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
791 if (need_flush && vp->v_type == VREG) {
792 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
793 fuse_vnode_savesize(vp, NULL, 0);
795 if ((fvdat->flag & FN_REVOKED) != 0)
796 fuse_io_invalbuf(vp, td);
798 fuse_io_flushbuf(vp, MNT_WAIT, td);
801 fuse_filehandle_close(vp, fufh, td, NULL);
804 if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
811 struct vnop_link_args {
812 struct vnode *a_tdvp;
814 struct componentname *a_cnp;
818 fuse_vnop_link(struct vop_link_args *ap)
820 struct vnode *vp = ap->a_vp;
821 struct vnode *tdvp = ap->a_tdvp;
822 struct componentname *cnp = ap->a_cnp;
824 struct vattr *vap = VTOVA(vp);
826 struct fuse_dispatcher fdi;
827 struct fuse_entry_out *feo;
828 struct fuse_link_in fli;
832 if (fuse_isdeadfs(vp)) {
835 if (vnode_mount(tdvp) != vnode_mount(vp)) {
840 * This is a seatbelt check to protect naive userspace filesystems from
841 * themselves and the limitations of the FUSE IPC protocol. If a
842 * filesystem does not allow attribute caching, assume it is capable of
843 * validating that nlink does not overflow.
845 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
847 fli.oldnodeid = VTOI(vp);
850 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
851 FUSE_LINK, &fli, sizeof(fli), &fdi);
852 if ((err = fdisp_wait_answ(&fdi))) {
857 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
860 * Purge the parent's attribute cache because the daemon
861 * should've updated its mtime and ctime
863 fuse_vnode_clear_attr_cache(tdvp);
864 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
865 feo->attr_valid_nsec, NULL);
872 struct fuse_lookup_alloc_arg {
873 struct fuse_entry_out *feo;
874 struct componentname *cnp;
879 /* Callback for vn_get_ino */
881 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
883 struct fuse_lookup_alloc_arg *flaa = arg;
885 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
889 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
890 "int", "struct timespec*", "struct timespec*");
892 struct vnop_lookup_args {
893 struct vnodeop_desc *a_desc;
895 struct vnode **a_vpp;
896 struct componentname *a_cnp;
900 fuse_vnop_lookup(struct vop_lookup_args *ap)
902 struct vnode *dvp = ap->a_dvp;
903 struct vnode **vpp = ap->a_vpp;
904 struct componentname *cnp = ap->a_cnp;
905 struct thread *td = cnp->cn_thread;
906 struct ucred *cred = cnp->cn_cred;
908 int nameiop = cnp->cn_nameiop;
909 int flags = cnp->cn_flags;
910 int wantparent = flags & (LOCKPARENT | WANTPARENT);
911 int islastcn = flags & ISLASTCN;
912 struct mount *mp = vnode_mount(dvp);
916 struct vnode *vp = NULL;
918 struct fuse_dispatcher fdi;
919 bool did_lookup = false;
920 struct fuse_entry_out *feo = NULL;
921 enum vtype vtyp; /* vnode type of target */
922 off_t filesize; /* filesize of target */
926 if (fuse_isdeadfs(dvp)) {
930 if (!vnode_isdir(dvp))
933 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
936 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
939 if (flags & ISDOTDOT) {
940 KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
941 ("Looking up .. is TODO"));
942 nid = VTOFUD(dvp)->parent_nid;
945 /* .. is obviously a directory */
948 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
950 /* . is obviously a directory */
954 struct timespec now, timeout;
956 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
958 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
960 case -1: /* positive match */
961 if (timespeccmp(&timeout, &now, >)) {
962 atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
965 atomic_add_acq_long(&fuse_lookup_cache_misses,
968 &VTOFUD(*vpp)->entry_cache_timeout);
979 case 0: /* no match in cache */
980 atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
983 case ENOENT: /* negative match */
985 if (timespeccmp(&timeout, &now, <=)) {
987 cache_purge_negative(dvp);
996 fdisp_init(&fdi, cnp->cn_namelen + 1);
997 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
999 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1000 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1001 lookup_err = fdisp_wait_answ(&fdi);
1005 /* lookup call succeeded */
1006 feo = (struct fuse_entry_out *)fdi.answ;
1009 /* zero nodeid means ENOENT and cache it */
1010 struct timespec timeout;
1012 fdi.answ_stat = ENOENT;
1013 lookup_err = ENOENT;
1014 if (cnp->cn_flags & MAKEENTRY) {
1015 fuse_validity_2_timespec(feo, &timeout);
1016 cache_enter_time(dvp, *vpp, cnp,
1019 } else if (nid == FUSE_ROOT_ID) {
1020 lookup_err = EINVAL;
1022 vtyp = IFTOVT(feo->attr.mode);
1023 filesize = feo->attr.size;
1025 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1026 fdisp_destroy(&fdi);
1030 /* lookup_err, if non-zero, must be ENOENT at this point */
1033 /* Entry not found */
1034 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1035 err = fuse_internal_access(dvp, VWRITE, td, cred);
1038 * Set the SAVENAME flag to hold onto the
1039 * pathname for use later in VOP_CREATE or
1042 cnp->cn_flags |= SAVENAME;
1050 /* Entry was found */
1051 if (flags & ISDOTDOT) {
1052 struct fuse_lookup_alloc_arg flaa;
1058 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1061 } else if (nid == VTOI(dvp)) {
1065 struct fuse_vnode_data *fvdat;
1067 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1074 * In the case where we are looking up a FUSE node
1075 * represented by an existing cached vnode, and the
1076 * true size reported by FUSE_LOOKUP doesn't match
1077 * the vnode's cached size, then any cached writes
1078 * beyond the file's current size are lost.
1081 * * following attribute cache expiration, or
1082 * * due a bug in the daemon, or
1085 if (vnode_isreg(vp) &&
1086 filesize != fvdat->cached_attrs.va_size &&
1087 fvdat->flag & FN_SIZECHANGE) {
1089 * The FN_SIZECHANGE flag reflects a dirty
1090 * append. If userspace lets us know our cache
1091 * is invalid, that write was lost. (Dirty
1092 * writes that do not cause append are also
1093 * lost, but we don't detect them here.)
1095 * XXX: Maybe disable WB caching on this mount.
1097 printf("%s: WB cache incoherent on %s!\n",
1099 vnode_mount(vp)->mnt_stat.f_mntonname);
1101 fvdat->flag &= ~FN_SIZECHANGE;
1105 fuse_internal_cache_attrs(*vpp, &feo->attr,
1106 feo->attr_valid, feo->attr_valid_nsec, NULL);
1107 fuse_validity_2_bintime(feo->entry_valid,
1108 feo->entry_valid_nsec,
1109 &fvdat->entry_cache_timeout);
1111 if ((nameiop == DELETE || nameiop == RENAME) &&
1114 struct vattr dvattr;
1116 err = fuse_internal_access(dvp, VWRITE, td,
1121 * if the parent's sticky bit is set, check
1122 * whether we're allowed to remove the file.
1123 * Need to figure out the vnode locking to make
1126 fuse_internal_getattr(dvp, &dvattr, cred, td);
1127 if ((dvattr.va_mode & S_ISTXT) &&
1128 fuse_internal_access(dvp, VADMIN, td,
1130 fuse_internal_access(*vpp, VADMIN, td,
1138 (nameiop == DELETE) ||
1139 (nameiop == RENAME && wantparent))) {
1140 cnp->cn_flags |= SAVENAME;
1147 if (vp != NULL && dvp != vp)
1149 else if (vp != NULL)
1154 fdisp_destroy(&fdi);
1160 struct vnop_mkdir_args {
1161 struct vnode *a_dvp;
1162 struct vnode **a_vpp;
1163 struct componentname *a_cnp;
1164 struct vattr *a_vap;
1168 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1170 struct vnode *dvp = ap->a_dvp;
1171 struct vnode **vpp = ap->a_vpp;
1172 struct componentname *cnp = ap->a_cnp;
1173 struct vattr *vap = ap->a_vap;
1175 struct fuse_mkdir_in fmdi;
1177 if (fuse_isdeadfs(dvp)) {
1180 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1181 fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1183 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1184 sizeof(fmdi), VDIR));
1188 struct vnop_mknod_args {
1189 struct vnode *a_dvp;
1190 struct vnode **a_vpp;
1191 struct componentname *a_cnp;
1192 struct vattr *a_vap;
1196 fuse_vnop_mknod(struct vop_mknod_args *ap)
1199 struct vnode *dvp = ap->a_dvp;
1200 struct vnode **vpp = ap->a_vpp;
1201 struct componentname *cnp = ap->a_cnp;
1202 struct vattr *vap = ap->a_vap;
1204 if (fuse_isdeadfs(dvp))
1207 return fuse_internal_mknod(dvp, vpp, cnp, vap);
1211 struct vop_open_args {
1214 struct ucred *a_cred;
1215 struct thread *a_td;
1216 int a_fdidx; / struct file *a_fp;
1220 fuse_vnop_open(struct vop_open_args *ap)
1222 struct vnode *vp = ap->a_vp;
1223 int a_mode = ap->a_mode;
1224 struct thread *td = ap->a_td;
1225 struct ucred *cred = ap->a_cred;
1226 pid_t pid = td->td_proc->p_pid;
1227 struct fuse_vnode_data *fvdat;
1229 if (fuse_isdeadfs(vp))
1231 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1232 return (EOPNOTSUPP);
1233 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1238 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1239 fuse_vnode_open(vp, 0, td);
1243 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1247 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1250 switch (ap->a_name) {
1251 case _PC_FILESIZEBITS:
1255 *ap->a_retval = NAME_MAX;
1258 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1260 case _PC_SYMLINK_MAX:
1261 *ap->a_retval = MAXPATHLEN;
1267 return (vop_stdpathconf(ap));
1272 struct vnop_read_args {
1276 struct ucred *a_cred;
1280 fuse_vnop_read(struct vop_read_args *ap)
1282 struct vnode *vp = ap->a_vp;
1283 struct uio *uio = ap->a_uio;
1284 int ioflag = ap->a_ioflag;
1285 struct ucred *cred = ap->a_cred;
1286 pid_t pid = curthread->td_proc->p_pid;
1288 if (fuse_isdeadfs(vp)) {
1292 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1293 ioflag |= IO_DIRECT;
1296 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1300 struct vnop_readdir_args {
1303 struct ucred *a_cred;
1310 fuse_vnop_readdir(struct vop_readdir_args *ap)
1312 struct vnode *vp = ap->a_vp;
1313 struct uio *uio = ap->a_uio;
1314 struct ucred *cred = ap->a_cred;
1315 struct fuse_filehandle *fufh = NULL;
1316 struct fuse_iov cookediov;
1322 bool closefufh = false;
1323 pid_t pid = curthread->td_proc->p_pid;
1327 if (fuse_isdeadfs(vp)) {
1330 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1331 (uio_resid(uio) < sizeof(struct dirent))) {
1335 tresid = uio->uio_resid;
1336 startoff = uio->uio_offset;
1337 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1338 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1340 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We
1341 * must implicitly open the directory here
1343 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1346 * When a directory is opened, it must be read from
1347 * the beginning. Hopefully, the "startoff" still
1348 * exists as an offset cookie for the directory.
1349 * If not, it will read the entire directory without
1350 * returning any entries and just return eof.
1352 uio->uio_offset = 0;
1358 if (ap->a_ncookies != NULL) {
1359 ncookies = uio->uio_resid /
1360 (offsetof(struct dirent, d_name) + 4) + 1;
1361 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1362 *ap->a_ncookies = ncookies;
1363 *ap->a_cookies = cookies;
1368 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1369 fiov_init(&cookediov, DIRCOOKEDSIZE);
1371 err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1372 &ncookies, cookies);
1374 fiov_teardown(&cookediov);
1376 fuse_filehandle_close(vp, fufh, curthread, cred);
1378 if (ap->a_ncookies != NULL) {
1380 *ap->a_ncookies -= ncookies;
1382 free(*ap->a_cookies, M_TEMP);
1383 *ap->a_ncookies = 0;
1384 *ap->a_cookies = NULL;
1387 if (err == 0 && tresid == uio->uio_resid)
1394 struct vnop_readlink_args {
1397 struct ucred *a_cred;
1401 fuse_vnop_readlink(struct vop_readlink_args *ap)
1403 struct vnode *vp = ap->a_vp;
1404 struct uio *uio = ap->a_uio;
1405 struct ucred *cred = ap->a_cred;
1407 struct fuse_dispatcher fdi;
1410 if (fuse_isdeadfs(vp)) {
1413 if (!vnode_islnk(vp)) {
1416 fdisp_init(&fdi, 0);
1417 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1421 if (((char *)fdi.answ)[0] == '/' &&
1422 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1423 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1425 err = uiomove(mpth, strlen(mpth), uio);
1428 err = uiomove(fdi.answ, fdi.iosize, uio);
1431 fdisp_destroy(&fdi);
1436 struct vnop_reclaim_args {
1438 struct thread *a_td;
1442 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1444 struct vnode *vp = ap->a_vp;
1445 struct thread *td = ap->a_td;
1446 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1447 struct fuse_filehandle *fufh, *fufh_tmp;
1450 panic("FUSE: no vnode data during recycling");
1452 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1453 printf("FUSE: vnode being reclaimed with open fufh "
1454 "(type=%#x)", fufh->fufh_type);
1455 fuse_filehandle_close(vp, fufh, td, NULL);
1458 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1459 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1462 fuse_vnode_setparent(vp, NULL);
1464 vfs_hash_remove(vp);
1465 vnode_destroy_vobject(vp);
1466 fuse_vnode_destroy(vp);
1472 struct vnop_remove_args {
1473 struct vnode *a_dvp;
1475 struct componentname *a_cnp;
1479 fuse_vnop_remove(struct vop_remove_args *ap)
1481 struct vnode *dvp = ap->a_dvp;
1482 struct vnode *vp = ap->a_vp;
1483 struct componentname *cnp = ap->a_cnp;
1487 if (fuse_isdeadfs(vp)) {
1490 if (vnode_isdir(vp)) {
1495 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1498 fuse_internal_vnode_disappear(vp);
1500 * Purge the parent's attribute cache because the daemon
1501 * should've updated its mtime and ctime
1503 fuse_vnode_clear_attr_cache(dvp);
1509 struct vnop_rename_args {
1510 struct vnode *a_fdvp;
1511 struct vnode *a_fvp;
1512 struct componentname *a_fcnp;
1513 struct vnode *a_tdvp;
1514 struct vnode *a_tvp;
1515 struct componentname *a_tcnp;
1519 fuse_vnop_rename(struct vop_rename_args *ap)
1521 struct vnode *fdvp = ap->a_fdvp;
1522 struct vnode *fvp = ap->a_fvp;
1523 struct componentname *fcnp = ap->a_fcnp;
1524 struct vnode *tdvp = ap->a_tdvp;
1525 struct vnode *tvp = ap->a_tvp;
1526 struct componentname *tcnp = ap->a_tcnp;
1527 struct fuse_data *data;
1528 bool newparent = fdvp != tdvp;
1529 bool isdir = fvp->v_type == VDIR;
1532 if (fuse_isdeadfs(fdvp)) {
1535 if (fvp->v_mount != tdvp->v_mount ||
1536 (tvp && fvp->v_mount != tvp->v_mount)) {
1537 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1544 * FUSE library is expected to check if target directory is not
1545 * under the source directory in the file system tree.
1546 * Linux performs this check at VFS level.
1549 * If source is a directory, and it will get a new parent, user must
1550 * have write permission to it, so ".." can be modified.
1552 data = fuse_get_mpdata(vnode_mount(tdvp));
1553 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1554 err = fuse_internal_access(fvp, VWRITE,
1555 tcnp->cn_thread, tcnp->cn_cred);
1559 sx_xlock(&data->rename_lock);
1560 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1563 fuse_vnode_setparent(fvp, tdvp);
1565 fuse_vnode_setparent(tvp, NULL);
1567 sx_unlock(&data->rename_lock);
1569 if (tvp != NULL && tvp != fvp) {
1572 if (vnode_isdir(fvp)) {
1573 if ((tvp != NULL) && vnode_isdir(tvp)) {
1594 struct vnop_rmdir_args {
1595 struct vnode *a_dvp;
1597 struct componentname *a_cnp;
1601 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1603 struct vnode *dvp = ap->a_dvp;
1604 struct vnode *vp = ap->a_vp;
1608 if (fuse_isdeadfs(vp)) {
1611 if (VTOFUD(vp) == VTOFUD(dvp)) {
1614 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1617 fuse_internal_vnode_disappear(vp);
1619 * Purge the parent's attribute cache because the daemon
1620 * should've updated its mtime and ctime
1622 fuse_vnode_clear_attr_cache(dvp);
1628 struct vnop_setattr_args {
1630 struct vattr *a_vap;
1631 struct ucred *a_cred;
1632 struct thread *a_td;
1636 fuse_vnop_setattr(struct vop_setattr_args *ap)
1638 struct vnode *vp = ap->a_vp;
1639 struct vattr *vap = ap->a_vap;
1640 struct ucred *cred = ap->a_cred;
1641 struct thread *td = curthread;
1643 struct fuse_data *data;
1644 struct vattr old_va;
1647 accmode_t accmode = 0;
1649 bool drop_suid = false;
1652 mp = vnode_mount(vp);
1653 data = fuse_get_mpdata(mp);
1654 dataflags = data->dataflags;
1655 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1656 if (cred->cr_ngroups > 0)
1657 cr_gid = cred->cr_groups[0];
1661 if (fuse_isdeadfs(vp)) {
1665 if (vap->va_uid != (uid_t)VNOVAL) {
1667 /* Only root may change a file's owner */
1668 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1670 /* As a special case, allow the null chown */
1671 err2 = fuse_internal_getattr(vp, &old_va, cred,
1675 if (vap->va_uid != old_va.va_uid)
1685 if (vap->va_gid != (gid_t)VNOVAL) {
1686 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1688 if (checkperm && !groupmember(vap->va_gid, cred))
1691 * Non-root users may only chgrp to one of their own
1694 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1696 /* As a special case, allow the null chgrp */
1697 err2 = fuse_internal_getattr(vp, &old_va, cred,
1701 if (vap->va_gid != old_va.va_gid)
1709 if (vap->va_size != VNOVAL) {
1710 switch (vp->v_type) {
1715 if (vfs_isrdonly(mp))
1720 * According to POSIX, the result is unspecified
1721 * for file types other than regular files,
1722 * directories and shared memory objects. We
1723 * don't support shared memory objects in the file
1724 * system, and have dubious support for truncating
1725 * symlinks. Just ignore the request in other cases.
1729 /* Don't set accmode. Permission to trunc is checked upstack */
1731 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1732 if (vap->va_vaflags & VA_UTIMES_NULL)
1738 if (vap->va_mode != (mode_t)VNOVAL)
1739 vap->va_mode &= ~(S_ISUID | S_ISGID);
1741 err = fuse_internal_getattr(vp, &old_va, cred, td);
1744 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1747 if (vap->va_mode != (mode_t)VNOVAL) {
1748 /* Only root may set the sticky bit on non-directories */
1749 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1750 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1752 if (checkperm && (vap->va_mode & S_ISGID)) {
1753 err = fuse_internal_getattr(vp, &old_va, cred, td);
1756 if (!groupmember(old_va.va_gid, cred)) {
1757 err = priv_check_cred(cred, PRIV_VFS_SETGID);
1765 if (vfs_isrdonly(mp))
1768 err = fuse_internal_access(vp, accmode, td, cred);
1772 return fuse_internal_setattr(vp, vap, td, cred);
1776 struct vnop_strategy_args {
1782 fuse_vnop_strategy(struct vop_strategy_args *ap)
1784 struct vnode *vp = ap->a_vp;
1785 struct buf *bp = ap->a_bp;
1787 if (!vp || fuse_isdeadfs(vp)) {
1788 bp->b_ioflags |= BIO_ERROR;
1789 bp->b_error = ENXIO;
1795 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1796 * fuse_io_strategy sets bp's error fields
1798 (void)fuse_io_strategy(vp, bp);
1805 struct vnop_symlink_args {
1806 struct vnode *a_dvp;
1807 struct vnode **a_vpp;
1808 struct componentname *a_cnp;
1809 struct vattr *a_vap;
1814 fuse_vnop_symlink(struct vop_symlink_args *ap)
1816 struct vnode *dvp = ap->a_dvp;
1817 struct vnode **vpp = ap->a_vpp;
1818 struct componentname *cnp = ap->a_cnp;
1819 const char *target = ap->a_target;
1821 struct fuse_dispatcher fdi;
1826 if (fuse_isdeadfs(dvp)) {
1830 * Unlike the other creator type calls, here we have to create a message
1831 * where the name of the new entry comes first, and the data describing
1832 * the entry comes second.
1833 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1834 * but put together the message manually and just call the core part.
1837 len = strlen(target) + 1;
1838 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1839 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1841 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1842 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1843 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1845 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1846 fdisp_destroy(&fdi);
1851 struct vnop_write_args {
1855 struct ucred *a_cred;
1859 fuse_vnop_write(struct vop_write_args *ap)
1861 struct vnode *vp = ap->a_vp;
1862 struct uio *uio = ap->a_uio;
1863 int ioflag = ap->a_ioflag;
1864 struct ucred *cred = ap->a_cred;
1865 pid_t pid = curthread->td_proc->p_pid;
1867 if (fuse_isdeadfs(vp)) {
1871 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1872 ioflag |= IO_DIRECT;
1875 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1878 SDT_PROBE_DEFINE1(fusefs, , vnops, vnop_getpages_error, "int");
1880 struct vnop_getpages_args {
1888 fuse_vnop_getpages(struct vop_getpages_args *ap)
1890 int i, error, nextoff, size, toff, count, npages;
1899 pid_t pid = curthread->td_proc->p_pid;
1902 KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1903 td = curthread; /* XXX */
1904 cred = curthread->td_ucred; /* XXX */
1906 npages = ap->a_count;
1908 if (!fsess_opt_mmap(vnode_mount(vp))) {
1909 SDT_PROBE2(fusefs, , vnops, trace, 1,
1910 "called on non-cacheable vnode??\n");
1911 return (VM_PAGER_ERROR);
1915 * If the last page is partially valid, just return it and allow
1916 * the pager to zero-out the blanks. Partially valid pages can
1917 * only occur at the file EOF.
1919 * XXXGL: is that true for FUSE, which is a local filesystem,
1920 * but still somewhat disconnected from the kernel?
1922 VM_OBJECT_WLOCK(vp->v_object);
1923 if (pages[npages - 1]->valid != 0 && --npages == 0)
1925 VM_OBJECT_WUNLOCK(vp->v_object);
1928 * We use only the kva address for the buffer, but this is extremely
1929 * convenient and fast.
1931 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1933 kva = (vm_offset_t)bp->b_data;
1934 pmap_qenter(kva, pages, npages);
1935 VM_CNT_INC(v_vnodein);
1936 VM_CNT_ADD(v_vnodepgsin, npages);
1938 count = npages << PAGE_SHIFT;
1939 iov.iov_base = (caddr_t)kva;
1940 iov.iov_len = count;
1943 uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1944 uio.uio_resid = count;
1945 uio.uio_segflg = UIO_SYSSPACE;
1946 uio.uio_rw = UIO_READ;
1949 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
1950 pmap_qremove(kva, npages);
1952 uma_zfree(fuse_pbuf_zone, bp);
1954 if (error && (uio.uio_resid == count)) {
1955 SDT_PROBE1(fusefs, , vnops, vnop_getpages_error, error);
1956 return VM_PAGER_ERROR;
1959 * Calculate the number of bytes read and validate only that number
1960 * of bytes. Note that due to pending writes, size may be 0. This
1961 * does not mean that the remaining data is invalid!
1964 size = count - uio.uio_resid;
1965 VM_OBJECT_WLOCK(vp->v_object);
1966 fuse_vm_page_lock_queues();
1967 for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1970 nextoff = toff + PAGE_SIZE;
1973 if (nextoff <= size) {
1975 * Read operation filled an entire page
1977 m->valid = VM_PAGE_BITS_ALL;
1978 KASSERT(m->dirty == 0,
1979 ("fuse_getpages: page %p is dirty", m));
1980 } else if (size > toff) {
1982 * Read operation filled a partial page.
1985 vm_page_set_valid_range(m, 0, size - toff);
1986 KASSERT(m->dirty == 0,
1987 ("fuse_getpages: page %p is dirty", m));
1990 * Read operation was short. If no error occurred
1991 * we may have hit a zero-fill section. We simply
1992 * leave valid set to 0.
1997 fuse_vm_page_unlock_queues();
1999 VM_OBJECT_WUNLOCK(vp->v_object);
2004 return (VM_PAGER_OK);
2008 struct vnop_putpages_args {
2014 vm_ooffset_t a_offset;
2018 fuse_vnop_putpages(struct vop_putpages_args *ap)
2024 int i, error, npages, count;
2032 pid_t pid = curthread->td_proc->p_pid;
2035 KASSERT(vp->v_object, ("objectless vp passed to putpages"));
2036 fsize = vp->v_object->un_pager.vnp.vnp_size;
2037 td = curthread; /* XXX */
2038 cred = curthread->td_ucred; /* XXX */
2040 count = ap->a_count;
2041 rtvals = ap->a_rtvals;
2042 npages = btoc(count);
2043 offset = IDX_TO_OFF(pages[0]->pindex);
2045 if (!fsess_opt_mmap(vnode_mount(vp))) {
2046 SDT_PROBE2(fusefs, , vnops, trace, 1,
2047 "called on non-cacheable vnode??\n");
2049 for (i = 0; i < npages; i++)
2050 rtvals[i] = VM_PAGER_AGAIN;
2053 * When putting pages, do not extend file past EOF.
2056 if (offset + count > fsize) {
2057 count = fsize - offset;
2062 * We use only the kva address for the buffer, but this is extremely
2063 * convenient and fast.
2065 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
2067 kva = (vm_offset_t)bp->b_data;
2068 pmap_qenter(kva, pages, npages);
2069 VM_CNT_INC(v_vnodeout);
2070 VM_CNT_ADD(v_vnodepgsout, count);
2072 iov.iov_base = (caddr_t)kva;
2073 iov.iov_len = count;
2076 uio.uio_offset = offset;
2077 uio.uio_resid = count;
2078 uio.uio_segflg = UIO_SYSSPACE;
2079 uio.uio_rw = UIO_WRITE;
2082 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
2084 pmap_qremove(kva, npages);
2085 uma_zfree(fuse_pbuf_zone, bp);
2088 int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
2090 for (i = 0; i < nwritten; i++) {
2091 rtvals[i] = VM_PAGER_OK;
2092 VM_OBJECT_WLOCK(pages[i]->object);
2093 vm_page_undirty(pages[i]);
2094 VM_OBJECT_WUNLOCK(pages[i]->object);
2100 static const char extattr_namespace_separator = '.';
2103 struct vop_getextattr_args {
2104 struct vop_generic_args a_gen;
2106 int a_attrnamespace;
2110 struct ucred *a_cred;
2111 struct thread *a_td;
2115 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2117 struct vnode *vp = ap->a_vp;
2118 struct uio *uio = ap->a_uio;
2119 struct fuse_dispatcher fdi;
2120 struct fuse_getxattr_in *get_xattr_in;
2121 struct fuse_getxattr_out *get_xattr_out;
2122 struct mount *mp = vnode_mount(vp);
2123 struct thread *td = ap->a_td;
2124 struct ucred *cred = ap->a_cred;
2130 if (fuse_isdeadfs(vp))
2133 if (!fsess_isimpl(mp, FUSE_GETXATTR))
2136 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2140 /* Default to looking for user attributes. */
2141 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2142 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2144 prefix = EXTATTR_NAMESPACE_USER_STRING;
2146 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2147 strlen(ap->a_name) + 1;
2149 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2150 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2152 get_xattr_in = fdi.indata;
2154 * Check to see whether we're querying the available size or
2155 * issuing the actual request. If we pass in 0, we get back struct
2156 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2157 * that much data, without the struct fuse_getxattr_out header.
2160 get_xattr_in->size = 0;
2162 get_xattr_in->size = uio->uio_resid;
2164 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2165 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2168 err = fdisp_wait_answ(&fdi);
2170 if (err == ENOSYS) {
2171 fsess_set_notimpl(mp, FUSE_GETXATTR);
2177 get_xattr_out = fdi.answ;
2179 if (ap->a_size != NULL)
2180 *ap->a_size = get_xattr_out->size;
2183 err = uiomove(fdi.answ, fdi.iosize, uio);
2186 fdisp_destroy(&fdi);
2191 struct vop_setextattr_args {
2192 struct vop_generic_args a_gen;
2194 int a_attrnamespace;
2197 struct ucred *a_cred;
2198 struct thread *a_td;
2202 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2204 struct vnode *vp = ap->a_vp;
2205 struct uio *uio = ap->a_uio;
2206 struct fuse_dispatcher fdi;
2207 struct fuse_setxattr_in *set_xattr_in;
2208 struct mount *mp = vnode_mount(vp);
2209 struct thread *td = ap->a_td;
2210 struct ucred *cred = ap->a_cred;
2216 if (fuse_isdeadfs(vp))
2219 if (!fsess_isimpl(mp, FUSE_SETXATTR))
2222 if (vfs_isrdonly(mp))
2225 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2226 if (ap->a_uio == NULL) {
2228 * If we got here as fallback from VOP_DELETEEXTATTR, then
2229 * return EOPNOTSUPP.
2231 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2232 return (EOPNOTSUPP);
2237 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2242 /* Default to looking for user attributes. */
2243 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2244 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2246 prefix = EXTATTR_NAMESPACE_USER_STRING;
2248 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2249 strlen(ap->a_name) + 1;
2251 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2252 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2254 set_xattr_in = fdi.indata;
2255 set_xattr_in->size = uio->uio_resid;
2257 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2258 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2261 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2262 uio->uio_resid, uio);
2267 err = fdisp_wait_answ(&fdi);
2269 if (err == ENOSYS) {
2270 fsess_set_notimpl(mp, FUSE_SETXATTR);
2273 if (err == ERESTART) {
2274 /* Can't restart after calling uiomove */
2279 fdisp_destroy(&fdi);
2284 * The Linux / FUSE extended attribute list is simply a collection of
2285 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2286 * byte length followed by a non-NUL terminated string. So, this allows
2287 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2288 * Linux attribute names are reported with the namespace as a prefix (e.g.
2289 * "user.attribute_name"), but in FreeBSD they are reported without the
2290 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2292 * user.attr_name1\0user.attr_name2\0
2296 * <num>attr_name1<num>attr_name2
2298 * Where "<num>" is a single byte number of characters in the attribute name.
2301 * prefix - exattr namespace prefix string
2302 * list, list_len - input list with namespace prefixes
2303 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2306 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2307 char *bsd_list, int *bsd_list_len)
2309 int len, pos, dist_to_next, prefix_len;
2313 prefix_len = strlen(prefix);
2315 while (pos < list_len && list[pos] != '\0') {
2316 dist_to_next = strlen(&list[pos]) + 1;
2317 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2318 list[pos + prefix_len] == extattr_namespace_separator) {
2319 len = dist_to_next -
2320 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2321 if (len >= EXTATTR_MAXNAMELEN)
2322 return (ENAMETOOLONG);
2324 bsd_list[*bsd_list_len] = len;
2325 memcpy(&bsd_list[*bsd_list_len + 1],
2326 &list[pos + prefix_len +
2327 sizeof(extattr_namespace_separator)], len);
2329 *bsd_list_len += len + 1;
2332 pos += dist_to_next;
2339 struct vop_listextattr_args {
2340 struct vop_generic_args a_gen;
2342 int a_attrnamespace;
2345 struct ucred *a_cred;
2346 struct thread *a_td;
2350 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2352 struct vnode *vp = ap->a_vp;
2353 struct uio *uio = ap->a_uio;
2354 struct fuse_dispatcher fdi;
2355 struct fuse_listxattr_in *list_xattr_in;
2356 struct fuse_listxattr_out *list_xattr_out;
2357 struct mount *mp = vnode_mount(vp);
2358 struct thread *td = ap->a_td;
2359 struct ucred *cred = ap->a_cred;
2363 char *bsd_list = NULL;
2369 if (fuse_isdeadfs(vp))
2372 if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2375 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2380 * Add space for a NUL and the period separator if enabled.
2381 * Default to looking for user attributes.
2383 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2384 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2386 prefix = EXTATTR_NAMESPACE_USER_STRING;
2388 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2390 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2391 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2394 * Retrieve Linux / FUSE compatible list size.
2396 list_xattr_in = fdi.indata;
2397 list_xattr_in->size = 0;
2398 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2399 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2401 err = fdisp_wait_answ(&fdi);
2403 if (err == ENOSYS) {
2404 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2410 list_xattr_out = fdi.answ;
2411 linux_list_len = list_xattr_out->size;
2412 if (linux_list_len == 0) {
2413 if (ap->a_size != NULL)
2414 *ap->a_size = linux_list_len;
2419 * Retrieve Linux / FUSE compatible list values.
2421 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2422 list_xattr_in = fdi.indata;
2423 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2424 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2425 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2427 err = fdisp_wait_answ(&fdi);
2431 linux_list = fdi.answ;
2432 linux_list_len = fdi.iosize;
2435 * Retrieve the BSD compatible list values.
2436 * The Linux / FUSE attribute list format isn't the same
2437 * as FreeBSD's format. So we need to transform it into
2438 * FreeBSD's format before giving it to the user.
2440 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2441 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2442 bsd_list, &bsd_list_len);
2446 if (ap->a_size != NULL)
2447 *ap->a_size = bsd_list_len;
2450 err = uiomove(bsd_list, bsd_list_len, uio);
2453 free(bsd_list, M_TEMP);
2454 fdisp_destroy(&fdi);
2459 struct vop_deleteextattr_args {
2460 struct vop_generic_args a_gen;
2462 int a_attrnamespace;
2464 struct ucred *a_cred;
2465 struct thread *a_td;
2469 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2471 struct vnode *vp = ap->a_vp;
2472 struct fuse_dispatcher fdi;
2473 struct mount *mp = vnode_mount(vp);
2474 struct thread *td = ap->a_td;
2475 struct ucred *cred = ap->a_cred;
2481 if (fuse_isdeadfs(vp))
2484 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2487 if (vfs_isrdonly(mp))
2490 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2495 /* Default to looking for user attributes. */
2496 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2497 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2499 prefix = EXTATTR_NAMESPACE_USER_STRING;
2501 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2502 strlen(ap->a_name) + 1;
2504 fdisp_init(&fdi, len);
2505 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2507 attr_str = fdi.indata;
2508 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2511 err = fdisp_wait_answ(&fdi);
2512 if (err == ENOSYS) {
2513 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2517 fdisp_destroy(&fdi);
2522 struct vnop_print_args {
2527 fuse_vnop_print(struct vop_print_args *ap)
2529 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2531 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2532 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2533 (uintmax_t)fvdat->nlookup,
2540 * Get an NFS filehandle for a FUSE file.
2542 * This will only work for FUSE file systems that guarantee the uniqueness of
2543 * nodeid:generation, which most don't.
2547 IN struct vnode *a_vp;
2548 IN struct fid *a_fhp;
2552 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2554 struct vnode *vp = ap->a_vp;
2555 struct fuse_vnode_data *fvdat = VTOFUD(vp);
2556 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2557 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2558 "FUSE fid type is too big");
2559 struct mount *mp = vnode_mount(vp);
2560 struct fuse_data *data = fuse_get_mpdata(mp);
2564 if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2567 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2571 /*ip = VTOI(ap->a_vp);*/
2572 /*ufhp = (struct ufid *)ap->a_fhp;*/
2573 fhp->len = sizeof(struct fuse_fid);
2574 fhp->nid = fvdat->nid;
2575 if (fvdat->generation <= UINT32_MAX)
2576 fhp->gen = fvdat->generation;