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(fuse);
115 * arg0: verbosity. Higher numbers give more verbose messages
116 * arg1: Textual message
118 SDT_PROBE_DEFINE2(fuse, , vnops, trace, "int", "char*");
121 static vop_access_t fuse_vnop_access;
122 static vop_close_t fuse_vnop_close;
123 static vop_create_t fuse_vnop_create;
124 static vop_deleteextattr_t fuse_vnop_deleteextattr;
125 static vop_fdatasync_t fuse_vnop_fdatasync;
126 static vop_fsync_t fuse_vnop_fsync;
127 static vop_getattr_t fuse_vnop_getattr;
128 static vop_getextattr_t fuse_vnop_getextattr;
129 static vop_inactive_t fuse_vnop_inactive;
130 static vop_link_t fuse_vnop_link;
131 static vop_listextattr_t fuse_vnop_listextattr;
132 static vop_lookup_t fuse_vnop_lookup;
133 static vop_mkdir_t fuse_vnop_mkdir;
134 static vop_mknod_t fuse_vnop_mknod;
135 static vop_open_t fuse_vnop_open;
136 static vop_pathconf_t fuse_vnop_pathconf;
137 static vop_read_t fuse_vnop_read;
138 static vop_readdir_t fuse_vnop_readdir;
139 static vop_readlink_t fuse_vnop_readlink;
140 static vop_reclaim_t fuse_vnop_reclaim;
141 static vop_remove_t fuse_vnop_remove;
142 static vop_rename_t fuse_vnop_rename;
143 static vop_rmdir_t fuse_vnop_rmdir;
144 static vop_setattr_t fuse_vnop_setattr;
145 static vop_setextattr_t fuse_vnop_setextattr;
146 static vop_strategy_t fuse_vnop_strategy;
147 static vop_symlink_t fuse_vnop_symlink;
148 static vop_write_t fuse_vnop_write;
149 static vop_getpages_t fuse_vnop_getpages;
150 static vop_putpages_t fuse_vnop_putpages;
151 static vop_print_t fuse_vnop_print;
153 struct vop_vector fuse_vnops = {
154 .vop_default = &default_vnodeops,
155 .vop_access = fuse_vnop_access,
156 .vop_close = fuse_vnop_close,
157 .vop_create = fuse_vnop_create,
158 .vop_deleteextattr = fuse_vnop_deleteextattr,
159 .vop_fsync = fuse_vnop_fsync,
160 .vop_fdatasync = fuse_vnop_fdatasync,
161 .vop_getattr = fuse_vnop_getattr,
162 .vop_getextattr = fuse_vnop_getextattr,
163 .vop_inactive = fuse_vnop_inactive,
164 .vop_link = fuse_vnop_link,
165 .vop_listextattr = fuse_vnop_listextattr,
166 .vop_lookup = fuse_vnop_lookup,
167 .vop_mkdir = fuse_vnop_mkdir,
168 .vop_mknod = fuse_vnop_mknod,
169 .vop_open = fuse_vnop_open,
170 .vop_pathconf = fuse_vnop_pathconf,
171 .vop_read = fuse_vnop_read,
172 .vop_readdir = fuse_vnop_readdir,
173 .vop_readlink = fuse_vnop_readlink,
174 .vop_reclaim = fuse_vnop_reclaim,
175 .vop_remove = fuse_vnop_remove,
176 .vop_rename = fuse_vnop_rename,
177 .vop_rmdir = fuse_vnop_rmdir,
178 .vop_setattr = fuse_vnop_setattr,
179 .vop_setextattr = fuse_vnop_setextattr,
180 .vop_strategy = fuse_vnop_strategy,
181 .vop_symlink = fuse_vnop_symlink,
182 .vop_write = fuse_vnop_write,
183 .vop_getpages = fuse_vnop_getpages,
184 .vop_putpages = fuse_vnop_putpages,
185 .vop_print = fuse_vnop_print,
188 static u_long fuse_lookup_cache_hits = 0;
190 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
191 &fuse_lookup_cache_hits, 0, "number of positive cache hits in lookup");
193 static u_long fuse_lookup_cache_misses = 0;
195 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
196 &fuse_lookup_cache_misses, 0, "number of cache misses in lookup");
199 * XXX: This feature is highly experimental and can bring to instabilities,
200 * needs revisiting before to be enabled by default.
202 static int fuse_reclaim_revoked = 0;
204 SYSCTL_INT(_vfs_fusefs, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
205 &fuse_reclaim_revoked, 0, "");
207 uma_zone_t fuse_pbuf_zone;
209 #define fuse_vm_page_lock(m) vm_page_lock((m));
210 #define fuse_vm_page_unlock(m) vm_page_unlock((m));
211 #define fuse_vm_page_lock_queues() ((void)0)
212 #define fuse_vm_page_unlock_queues() ((void)0)
214 /* Check permission for extattr operations, much like extattr_check_cred */
216 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
217 struct thread *td, accmode_t accmode)
219 struct mount *mp = vnode_mount(vp);
220 struct fuse_data *data = fuse_get_mpdata(mp);
223 * Kernel-invoked always succeeds.
229 * Do not allow privileged processes in jail to directly manipulate
233 case EXTATTR_NAMESPACE_SYSTEM:
234 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
235 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
238 case EXTATTR_NAMESPACE_USER:
239 return (fuse_internal_access(vp, accmode, td, cred));
245 /* Get a filehandle for a directory */
247 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
248 struct ucred *cred, pid_t pid)
250 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
252 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
255 /* Send FUSE_FLUSH for this vnode */
257 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
259 struct fuse_flush_in *ffi;
260 struct fuse_filehandle *fufh;
261 struct fuse_dispatcher fdi;
262 struct thread *td = curthread;
263 struct mount *mp = vnode_mount(vp);
266 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
269 err = fuse_filehandle_get(vp, fflag, &fufh, cred, pid);
273 fdisp_init(&fdi, sizeof(*ffi));
274 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
276 ffi->fh = fufh->fh_id;
278 err = fdisp_wait_answ(&fdi);
280 fsess_set_notimpl(mp, FUSE_FLUSH);
288 struct vnop_access_args {
290 #if VOP_ACCESS_TAKES_ACCMODE_T
295 struct ucred *a_cred;
300 fuse_vnop_access(struct vop_access_args *ap)
302 struct vnode *vp = ap->a_vp;
303 int accmode = ap->a_accmode;
304 struct ucred *cred = ap->a_cred;
306 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
310 if (fuse_isdeadfs(vp)) {
311 if (vnode_isvroot(vp)) {
316 if (!(data->dataflags & FSESS_INITED)) {
317 if (vnode_isvroot(vp)) {
318 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
319 (fuse_match_cred(data->daemoncred, cred) == 0)) {
325 if (vnode_islnk(vp)) {
329 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
334 struct vop_close_args {
337 struct ucred *a_cred;
342 fuse_vnop_close(struct vop_close_args *ap)
344 struct vnode *vp = ap->a_vp;
345 struct ucred *cred = ap->a_cred;
346 int fflag = ap->a_fflag;
347 struct thread *td = ap->a_td;
348 pid_t pid = td->td_proc->p_pid;
351 if (fuse_isdeadfs(vp))
355 if (fflag & IO_NDELAY)
358 err = fuse_flush(vp, cred, pid, fflag);
359 /* TODO: close the file handle, if we're sure it's no longer used */
360 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
361 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
367 fdisp_make_mknod_for_fallback(
368 struct fuse_dispatcher *fdip,
369 struct componentname *cnp,
375 enum fuse_opcode *op)
377 struct fuse_mknod_in *fmni;
379 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
381 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
385 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
387 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
390 struct vnop_create_args {
392 struct vnode **a_vpp;
393 struct componentname *a_cnp;
398 fuse_vnop_create(struct vop_create_args *ap)
400 struct vnode *dvp = ap->a_dvp;
401 struct vnode **vpp = ap->a_vpp;
402 struct componentname *cnp = ap->a_cnp;
403 struct vattr *vap = ap->a_vap;
404 struct thread *td = cnp->cn_thread;
405 struct ucred *cred = cnp->cn_cred;
407 struct fuse_open_in *foi;
408 struct fuse_entry_out *feo;
409 struct fuse_open_out *foo;
410 struct fuse_dispatcher fdi, fdi2;
411 struct fuse_dispatcher *fdip = &fdi;
412 struct fuse_dispatcher *fdip2 = NULL;
416 struct mount *mp = vnode_mount(dvp);
417 uint64_t parentnid = VTOFUD(dvp)->nid;
418 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
423 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
424 * writable mode makes sense, and we might as well include readability
429 if (fuse_isdeadfs(dvp)) {
432 bzero(&fdi, sizeof(fdi));
434 if ((vap->va_type != VREG))
437 if (!fsess_isimpl(mp, FUSE_CREATE)) {
438 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
439 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
442 /* Use FUSE_CREATE */
444 fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
445 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
448 foi->flags = O_CREAT | flags;
449 memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
451 ((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
454 err = fdisp_wait_answ(fdip);
457 if (err == ENOSYS && op == FUSE_CREATE) {
458 fsess_set_notimpl(mp, FUSE_CREATE);
459 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
460 parentnid, td, cred, mode, &op);
461 err = fdisp_wait_answ(fdip);
469 if ((err = fuse_internal_checkentry(feo, VREG))) {
473 if (op == FUSE_CREATE) {
474 foo = (struct fuse_open_out*)(feo + 1);
476 /* Issue a separate FUSE_OPEN */
478 fdisp_init(fdip2, sizeof(*foi));
479 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
484 err = fdisp_wait_answ(fdip2);
489 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, VREG);
491 struct fuse_release_in *fri;
492 uint64_t nodeid = feo->nodeid;
493 uint64_t fh_id = foo->fh;
495 fdisp_init(fdip, sizeof(*fri));
496 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
500 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
501 fuse_insert_message(fdip->tick);
504 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
505 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
506 feo->attr_valid_nsec, NULL);
508 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
509 fuse_vnode_open(*vpp, foo->open_flags, td);
510 cache_purge_negative(dvp);
514 fdisp_destroy(fdip2);
520 struct vnop_fdatasync_args {
521 struct vop_generic_args a_gen;
523 struct thread * a_td;
527 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
529 struct vnode *vp = ap->a_vp;
530 struct thread *td = ap->a_td;
531 int waitfor = MNT_WAIT;
535 if (fuse_isdeadfs(vp)) {
538 if ((err = vop_stdfdatasync_buf(ap)))
541 return fuse_internal_fsync(vp, td, waitfor, true);
545 struct vnop_fsync_args {
546 struct vop_generic_args a_gen;
549 struct thread * a_td;
553 fuse_vnop_fsync(struct vop_fsync_args *ap)
555 struct vnode *vp = ap->a_vp;
556 struct thread *td = ap->a_td;
557 int waitfor = ap->a_waitfor;
560 if (fuse_isdeadfs(vp)) {
563 if ((err = vop_stdfsync(ap)))
566 return fuse_internal_fsync(vp, td, waitfor, false);
570 struct vnop_getattr_args {
573 struct ucred *a_cred;
578 fuse_vnop_getattr(struct vop_getattr_args *ap)
580 struct vnode *vp = ap->a_vp;
581 struct vattr *vap = ap->a_vap;
582 struct ucred *cred = ap->a_cred;
583 struct thread *td = curthread;
588 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
590 /* Note that we are not bailing out on a dead file system just yet. */
592 if (!(dataflags & FSESS_INITED)) {
593 if (!vnode_isvroot(vp)) {
594 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
601 err = fuse_internal_getattr(vp, vap, cred, td);
602 if (err == ENOTCONN && vnode_isvroot(vp)) {
603 /* see comment in fuse_vfsop_statfs() */
610 bzero(vap, sizeof(*vap));
611 vap->va_type = vnode_vtype(vp);
617 struct vnop_inactive_args {
623 fuse_vnop_inactive(struct vop_inactive_args *ap)
625 struct vnode *vp = ap->a_vp;
626 struct thread *td = ap->a_td;
628 struct fuse_vnode_data *fvdat = VTOFUD(vp);
629 struct fuse_filehandle *fufh, *fufh_tmp;
633 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
634 if (need_flush && vp->v_type == VREG) {
635 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
636 fuse_vnode_savesize(vp, NULL, 0);
638 if (fuse_data_cache_invalidate ||
639 (fvdat->flag & FN_REVOKED) != 0)
640 fuse_io_invalbuf(vp, td);
642 fuse_io_flushbuf(vp, MNT_WAIT, td);
645 fuse_filehandle_close(vp, fufh, td, NULL);
648 if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
655 struct vnop_link_args {
656 struct vnode *a_tdvp;
658 struct componentname *a_cnp;
662 fuse_vnop_link(struct vop_link_args *ap)
664 struct vnode *vp = ap->a_vp;
665 struct vnode *tdvp = ap->a_tdvp;
666 struct componentname *cnp = ap->a_cnp;
668 struct vattr *vap = VTOVA(vp);
670 struct fuse_dispatcher fdi;
671 struct fuse_entry_out *feo;
672 struct fuse_link_in fli;
676 if (fuse_isdeadfs(vp)) {
679 if (vnode_mount(tdvp) != vnode_mount(vp)) {
684 * This is a seatbelt check to protect naive userspace filesystems from
685 * themselves and the limitations of the FUSE IPC protocol. If a
686 * filesystem does not allow attribute caching, assume it is capable of
687 * validating that nlink does not overflow.
689 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
691 fli.oldnodeid = VTOI(vp);
694 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
695 FUSE_LINK, &fli, sizeof(fli), &fdi);
696 if ((err = fdisp_wait_answ(&fdi))) {
701 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
707 SDT_PROBE_DEFINE3(fuse, , vnops, cache_lookup,
708 "int", "struct timespec*", "struct timespec*");
710 struct vnop_lookup_args {
711 struct vnodeop_desc *a_desc;
713 struct vnode **a_vpp;
714 struct componentname *a_cnp;
718 fuse_vnop_lookup(struct vop_lookup_args *ap)
720 struct vnode *dvp = ap->a_dvp;
721 struct vnode **vpp = ap->a_vpp;
722 struct componentname *cnp = ap->a_cnp;
723 struct thread *td = cnp->cn_thread;
724 struct ucred *cred = cnp->cn_cred;
726 int nameiop = cnp->cn_nameiop;
727 int flags = cnp->cn_flags;
728 int wantparent = flags & (LOCKPARENT | WANTPARENT);
729 int islastcn = flags & ISLASTCN;
730 struct mount *mp = vnode_mount(dvp);
734 struct vnode *vp = NULL;
736 struct fuse_dispatcher fdi;
738 struct fuse_entry_out *feo = NULL;
739 struct fuse_attr_out *fao = NULL;
740 struct fuse_attr *fattr = NULL;
744 if (fuse_isdeadfs(dvp)) {
748 if (!vnode_isdir(dvp)) {
751 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) {
755 * We do access check prior to doing anything else only in the case
756 * when we are at fs root (we'd like to say, "we are at the first
757 * component", but that's not exactly the same... nevermind).
758 * See further comments at further access checks.
761 /* TODO: consider eliminating this. Is there any good reason for it? */
762 if (vnode_isvroot(dvp)) { /* early permission check hack */
763 if ((err = fuse_internal_access(dvp, VEXEC, td, cred))) {
767 if (flags & ISDOTDOT) {
768 nid = VTOFUD(dvp)->parent_nid;
775 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
781 struct timespec now, timeout;
783 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
785 SDT_PROBE3(fuse, , vnops, cache_lookup, err, &timeout, &now);
787 case -1: /* positive match */
788 if (timespeccmp(&timeout, &now, >)) {
789 atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
792 atomic_add_acq_long(&fuse_lookup_cache_misses,
794 fuse_internal_vnode_disappear(*vpp);
804 case 0: /* no match in cache */
805 atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
808 case ENOENT: /* negative match */
810 if (timespeccmp(&timeout, &now, <=)) {
812 cache_purge_negative(dvp);
821 fdisp_init(&fdi, cnp->cn_namelen + 1);
825 fdisp_make(&fdi, op, mp, nid, td, cred);
827 if (op == FUSE_LOOKUP) {
828 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
829 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
831 lookup_err = fdisp_wait_answ(&fdi);
833 if ((op == FUSE_LOOKUP) && !lookup_err) {
834 /* lookup call succeeded */
835 nid = ((struct fuse_entry_out *)fdi.answ)->nodeid;
837 /* zero nodeid means ENOENT and cache it */
838 struct timespec timeout;
840 fdi.answ_stat = ENOENT;
842 if (cnp->cn_flags & MAKEENTRY) {
843 feo = (struct fuse_entry_out *)fdi.answ;
844 fuse_validity_2_timespec(feo, &timeout);
845 cache_enter_time(dvp, *vpp, cnp, &timeout,
848 } else if (nid == FUSE_ROOT_ID) {
853 (!fdi.answ_stat || lookup_err != ENOENT || op != FUSE_LOOKUP)) {
857 /* lookup_err, if non-zero, must be ENOENT at this point */
860 /* Entry not found */
861 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
862 err = fuse_internal_access(dvp, VWRITE, td, cred);
867 * Possibly record the position of a slot in the
868 * directory large enough for the new component name.
869 * This can be recorded in the vnode private data for
870 * dvp. Set the SAVENAME flag to hold onto the
871 * pathname for use later in VOP_CREATE or VOP_RENAME.
873 cnp->cn_flags |= SAVENAME;
883 /* Entry was found */
884 if (op == FUSE_GETATTR) {
885 fattr = &((struct fuse_attr_out *)fdi.answ)->attr;
887 feo = (struct fuse_entry_out *)fdi.answ;
888 fattr = &(feo->attr);
891 if ((nameiop == DELETE || nameiop == RENAME) && islastcn) {
892 err = fuse_internal_access(dvp, VWRITE, td, cred);
896 * TODO: if the parent's sticky bit is set, check
897 * whether we're allowed to remove the file.
898 * Need to figure out the vnode locking to make this
903 fuse_internal_getattr(dvp, &dvattr, cred, td);
904 if ((dvattr.va_mode & S_ISTXT) &&
905 fuse_internal_access(dvp, VADMIN, td, cred) &&
906 fuse_internal_access(*vpp, VADMIN, td, cred)) {
914 * If deleting, and at end of pathname, return parameters
915 * which can be used to remove file. If the wantparent flag
916 * isn't set, we return only the directory, otherwise we go on
917 * and lock the inode, being careful with ".".
919 if (nameiop == DELETE && islastcn) {
920 if (nid == VTOI(dvp)) {
924 err = fuse_vnode_get(dvp->v_mount, feo, nid,
925 dvp, &vp, cnp, IFTOVT(fattr->mode));
932 * Save the name for use in VOP_RMDIR and VOP_REMOVE
935 cnp->cn_flags |= SAVENAME;
940 * If rewriting (RENAME), return the inode and the
941 * information required to rewrite the present directory
942 * Must get inode of directory entry to verify it's a
943 * regular file, or empty directory.
945 if (nameiop == RENAME && wantparent && islastcn) {
949 if (nid == VTOI(dvp)) {
953 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
954 &vp, cnp, IFTOVT(fattr->mode));
960 * Save the name for use in VOP_RENAME later.
962 cnp->cn_flags |= SAVENAME;
966 if (flags & ISDOTDOT) {
971 * Expanded copy of vn_vget_ino() so that
972 * fuse_vnode_get() can be used.
975 ltype = VOP_ISLOCKED(dvp);
976 err = vfs_busy(mp, MBF_NOWAIT);
980 err = vfs_busy(mp, 0);
981 vn_lock(dvp, ltype | LK_RETRY);
985 if ((dvp->v_iflag & VI_DOOMED) != 0) {
992 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, NULL,
993 &vp, cnp, IFTOVT(fattr->mode));
995 vn_lock(dvp, ltype | LK_RETRY);
996 if ((dvp->v_iflag & VI_DOOMED) != 0) {
1004 } else if (nid == VTOI(dvp)) {
1008 struct fuse_vnode_data *fvdat;
1010 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1011 &vp, cnp, IFTOVT(fattr->mode));
1015 fuse_vnode_setparent(vp, dvp);
1018 * In the case where we are looking up a FUSE node
1019 * represented by an existing cached vnode, and the
1020 * true size reported by FUSE_LOOKUP doesn't match
1021 * the vnode's cached size, fix the vnode cache to
1022 * match the real object size.
1024 * This can occur via FUSE distributed filesystems,
1025 * irregular files, etc.
1028 if (vnode_isreg(vp) &&
1029 fattr->size != fvdat->filesize) {
1031 * The FN_SIZECHANGE flag reflects a dirty
1032 * append. If userspace lets us know our cache
1033 * is invalid, that write was lost. (Dirty
1034 * writes that do not cause append are also
1035 * lost, but we don't detect them here.)
1037 * XXX: Maybe disable WB caching on this mount.
1039 if (fvdat->flag & FN_SIZECHANGE)
1040 printf("%s: WB cache incoherent on "
1042 vnode_mount(vp)->mnt_stat.f_mntonname);
1044 (void)fuse_vnode_setsize(vp, cred, fattr->size);
1045 fvdat->flag &= ~FN_SIZECHANGE;
1050 if (op == FUSE_GETATTR) {
1051 fao = (struct fuse_attr_out*)fdi.answ;
1052 fuse_internal_cache_attrs(*vpp,
1053 &fao->attr, fao->attr_valid,
1054 fao->attr_valid_nsec, NULL);
1056 fuse_internal_cache_attrs(*vpp,
1057 &feo->attr, feo->attr_valid,
1058 feo->attr_valid_nsec, NULL);
1064 /* No lookup error; need to clean up. */
1066 if (err) { /* Found inode; exit with no vnode. */
1067 if (op == FUSE_LOOKUP) {
1068 fuse_internal_forget_send(vnode_mount(dvp), td, cred,
1071 fdisp_destroy(&fdi);
1074 #ifndef NO_EARLY_PERM_CHECK_HACK
1077 * We have the attributes of the next item
1078 * *now*, and it's a fact, and we do not
1079 * have to do extra work for it (ie, beg the
1080 * daemon), and it neither depends on such
1081 * accidental things like attr caching. So
1082 * the big idea: check credentials *now*,
1083 * not at the beginning of the next call to
1086 * The first item of the lookup chain (fs root)
1087 * won't be checked then here, of course, as
1088 * its never "the next". But go and see that
1089 * the root is taken care about at the very
1090 * beginning of this function.
1092 * Now, given we want to do the access check
1093 * this way, one might ask: so then why not
1094 * do the access check just after fetching
1095 * the inode and its attributes from the
1096 * daemon? Why bother with producing the
1097 * corresponding vnode at all if something
1098 * is not OK? We know what's the deal as
1099 * soon as we get those attrs... There is
1100 * one bit of info though not given us by
1101 * the daemon: whether his response is
1102 * authoritative or not... His response should
1103 * be ignored if something is mounted over
1104 * the dir in question. But that can be
1105 * known only by having the vnode...
1107 int tmpvtype = vnode_vtype(*vpp);
1109 if ((tmpvtype != VDIR) && (tmpvtype != VLNK)) {
1112 if (!err && !vnode_mountedhere(*vpp)) {
1113 err = fuse_internal_access(*vpp, VEXEC,
1117 if (tmpvtype == VLNK)
1118 SDT_PROBE2(fuse, , vnops, trace,
1119 1, "weird, permission "
1120 "error with a symlink?");
1128 fdisp_destroy(&fdi);
1134 struct vnop_mkdir_args {
1135 struct vnode *a_dvp;
1136 struct vnode **a_vpp;
1137 struct componentname *a_cnp;
1138 struct vattr *a_vap;
1142 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1144 struct vnode *dvp = ap->a_dvp;
1145 struct vnode **vpp = ap->a_vpp;
1146 struct componentname *cnp = ap->a_cnp;
1147 struct vattr *vap = ap->a_vap;
1149 struct fuse_mkdir_in fmdi;
1151 if (fuse_isdeadfs(dvp)) {
1154 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1156 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1157 sizeof(fmdi), VDIR));
1161 struct vnop_mknod_args {
1162 struct vnode *a_dvp;
1163 struct vnode **a_vpp;
1164 struct componentname *a_cnp;
1165 struct vattr *a_vap;
1169 fuse_vnop_mknod(struct vop_mknod_args *ap)
1172 struct vnode *dvp = ap->a_dvp;
1173 struct vnode **vpp = ap->a_vpp;
1174 struct componentname *cnp = ap->a_cnp;
1175 struct vattr *vap = ap->a_vap;
1176 struct fuse_mknod_in fmni;
1178 if (fuse_isdeadfs(dvp))
1181 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1182 fmni.rdev = vap->va_rdev;
1183 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
1184 sizeof(fmni), vap->va_type));
1188 struct vnop_open_args {
1191 struct ucred *a_cred;
1192 struct thread *a_td;
1193 int a_fdidx; / struct file *a_fp;
1197 fuse_vnop_open(struct vop_open_args *ap)
1199 struct vnode *vp = ap->a_vp;
1200 int a_mode = ap->a_mode;
1201 struct thread *td = ap->a_td;
1202 struct ucred *cred = ap->a_cred;
1203 pid_t pid = td->td_proc->p_pid;
1204 struct fuse_vnode_data *fvdat;
1206 if (fuse_isdeadfs(vp))
1208 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1209 return (EOPNOTSUPP);
1210 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1215 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1216 fuse_vnode_open(vp, 0, td);
1220 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1224 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1227 switch (ap->a_name) {
1228 case _PC_FILESIZEBITS:
1232 *ap->a_retval = NAME_MAX;
1235 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1237 case _PC_SYMLINK_MAX:
1238 *ap->a_retval = MAXPATHLEN;
1244 return (vop_stdpathconf(ap));
1249 struct vnop_read_args {
1253 struct ucred *a_cred;
1257 fuse_vnop_read(struct vop_read_args *ap)
1259 struct vnode *vp = ap->a_vp;
1260 struct uio *uio = ap->a_uio;
1261 int ioflag = ap->a_ioflag;
1262 struct ucred *cred = ap->a_cred;
1263 pid_t pid = curthread->td_proc->p_pid;
1265 if (fuse_isdeadfs(vp)) {
1269 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1270 ioflag |= IO_DIRECT;
1273 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1277 struct vnop_readdir_args {
1280 struct ucred *a_cred;
1287 fuse_vnop_readdir(struct vop_readdir_args *ap)
1289 struct vnode *vp = ap->a_vp;
1290 struct uio *uio = ap->a_uio;
1291 struct ucred *cred = ap->a_cred;
1292 struct fuse_filehandle *fufh = NULL;
1293 struct fuse_iov cookediov;
1295 pid_t pid = curthread->td_proc->p_pid;
1297 if (fuse_isdeadfs(vp)) {
1300 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1301 (uio_resid(uio) < sizeof(struct dirent))) {
1305 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1308 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1309 fiov_init(&cookediov, DIRCOOKEDSIZE);
1311 err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1313 fiov_teardown(&cookediov);
1319 struct vnop_readlink_args {
1322 struct ucred *a_cred;
1326 fuse_vnop_readlink(struct vop_readlink_args *ap)
1328 struct vnode *vp = ap->a_vp;
1329 struct uio *uio = ap->a_uio;
1330 struct ucred *cred = ap->a_cred;
1332 struct fuse_dispatcher fdi;
1335 if (fuse_isdeadfs(vp)) {
1338 if (!vnode_islnk(vp)) {
1341 fdisp_init(&fdi, 0);
1342 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1346 if (((char *)fdi.answ)[0] == '/' &&
1347 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1348 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1350 err = uiomove(mpth, strlen(mpth), uio);
1353 err = uiomove(fdi.answ, fdi.iosize, uio);
1356 fdisp_destroy(&fdi);
1361 struct vnop_reclaim_args {
1363 struct thread *a_td;
1367 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1369 struct vnode *vp = ap->a_vp;
1370 struct thread *td = ap->a_td;
1371 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1372 struct fuse_filehandle *fufh, *fufh_tmp;
1375 panic("FUSE: no vnode data during recycling");
1377 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1378 printf("FUSE: vnode being reclaimed with open fufh "
1379 "(type=%#x)", fufh->fufh_type);
1380 fuse_filehandle_close(vp, fufh, td, NULL);
1383 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1384 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1387 fuse_vnode_setparent(vp, NULL);
1389 vfs_hash_remove(vp);
1390 vnode_destroy_vobject(vp);
1391 fuse_vnode_destroy(vp);
1397 struct vnop_remove_args {
1398 struct vnode *a_dvp;
1400 struct componentname *a_cnp;
1404 fuse_vnop_remove(struct vop_remove_args *ap)
1406 struct vnode *dvp = ap->a_dvp;
1407 struct vnode *vp = ap->a_vp;
1408 struct componentname *cnp = ap->a_cnp;
1412 if (fuse_isdeadfs(vp)) {
1415 if (vnode_isdir(vp)) {
1420 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1423 fuse_internal_vnode_disappear(vp);
1428 struct vnop_rename_args {
1429 struct vnode *a_fdvp;
1430 struct vnode *a_fvp;
1431 struct componentname *a_fcnp;
1432 struct vnode *a_tdvp;
1433 struct vnode *a_tvp;
1434 struct componentname *a_tcnp;
1438 fuse_vnop_rename(struct vop_rename_args *ap)
1440 struct vnode *fdvp = ap->a_fdvp;
1441 struct vnode *fvp = ap->a_fvp;
1442 struct componentname *fcnp = ap->a_fcnp;
1443 struct vnode *tdvp = ap->a_tdvp;
1444 struct vnode *tvp = ap->a_tvp;
1445 struct componentname *tcnp = ap->a_tcnp;
1446 struct fuse_data *data;
1450 if (fuse_isdeadfs(fdvp)) {
1453 if (fvp->v_mount != tdvp->v_mount ||
1454 (tvp && fvp->v_mount != tvp->v_mount)) {
1455 SDT_PROBE2(fuse, , vnops, trace, 1, "cross-device rename");
1462 * FUSE library is expected to check if target directory is not
1463 * under the source directory in the file system tree.
1464 * Linux performs this check at VFS level.
1466 data = fuse_get_mpdata(vnode_mount(tdvp));
1467 sx_xlock(&data->rename_lock);
1468 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1471 fuse_vnode_setparent(fvp, tdvp);
1473 fuse_vnode_setparent(tvp, NULL);
1475 sx_unlock(&data->rename_lock);
1477 if (tvp != NULL && tvp != fvp) {
1480 if (vnode_isdir(fvp)) {
1481 if ((tvp != NULL) && vnode_isdir(tvp)) {
1502 struct vnop_rmdir_args {
1503 struct vnode *a_dvp;
1505 struct componentname *a_cnp;
1509 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1511 struct vnode *dvp = ap->a_dvp;
1512 struct vnode *vp = ap->a_vp;
1516 if (fuse_isdeadfs(vp)) {
1519 if (VTOFUD(vp) == VTOFUD(dvp)) {
1522 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1525 fuse_internal_vnode_disappear(vp);
1530 struct vnop_setattr_args {
1532 struct vattr *a_vap;
1533 struct ucred *a_cred;
1534 struct thread *a_td;
1538 fuse_vnop_setattr(struct vop_setattr_args *ap)
1540 struct vnode *vp = ap->a_vp;
1541 struct vattr *vap = ap->a_vap;
1542 struct ucred *cred = ap->a_cred;
1543 struct thread *td = curthread;
1544 struct fuse_dispatcher fdi;
1545 struct fuse_setattr_in *fsai;
1546 pid_t pid = td->td_proc->p_pid;
1550 int sizechanged = 0;
1551 uint64_t newsize = 0;
1552 accmode_t accmode = 0;
1554 if (fuse_isdeadfs(vp)) {
1557 fdisp_init(&fdi, sizeof(*fsai));
1558 fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1562 if (vap->va_uid != (uid_t)VNOVAL) {
1563 fsai->uid = vap->va_uid;
1564 fsai->valid |= FATTR_UID;
1567 if (vap->va_gid != (gid_t)VNOVAL) {
1568 fsai->gid = vap->va_gid;
1569 fsai->valid |= FATTR_GID;
1572 if (vap->va_size != VNOVAL) {
1574 struct fuse_filehandle *fufh = NULL;
1576 /*Truncate to a new value. */
1577 fsai->size = vap->va_size;
1579 newsize = vap->va_size;
1580 fsai->valid |= FATTR_SIZE;
1583 fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
1585 fsai->fh = fufh->fh_id;
1586 fsai->valid |= FATTR_FH;
1589 if (vap->va_atime.tv_sec != VNOVAL) {
1590 fsai->atime = vap->va_atime.tv_sec;
1591 fsai->atimensec = vap->va_atime.tv_nsec;
1592 fsai->valid |= FATTR_ATIME;
1595 * TODO: only require VWRITE if UTIMENS_NULL is set. PR 237181
1598 if (vap->va_mtime.tv_sec != VNOVAL) {
1599 fsai->mtime = vap->va_mtime.tv_sec;
1600 fsai->mtimensec = vap->va_mtime.tv_nsec;
1601 fsai->valid |= FATTR_MTIME;
1604 * TODO: only require VWRITE if UTIMENS_NULL is set. PR 237181
1607 if (vap->va_mode != (mode_t)VNOVAL) {
1608 fsai->mode = vap->va_mode & ALLPERMS;
1609 fsai->valid |= FATTR_MODE;
1615 vtyp = vnode_vtype(vp);
1617 if (fsai->valid & FATTR_SIZE && vtyp == VDIR) {
1621 if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) {
1625 err = fuse_internal_access(vp, accmode, td, cred);
1629 if ((err = fdisp_wait_answ(&fdi)))
1631 vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1633 if (vnode_vtype(vp) != vtyp) {
1634 if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1635 SDT_PROBE2(fuse, , vnops, trace, 1, "FUSE: Dang! "
1636 "vnode_vtype is VNON and vtype isn't.");
1639 * STALE vnode, ditch
1641 * The vnode has changed its type "behind our back".
1642 * There's nothing really we can do, so let us just
1643 * force an internal revocation and tell the caller to
1644 * try again, if interested.
1646 fuse_internal_vnode_disappear(vp);
1651 struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ;
1652 fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid,
1653 fao->attr_valid_nsec, NULL);
1657 fdisp_destroy(&fdi);
1658 if (!err && sizechanged) {
1659 fuse_vnode_setsize(vp, cred, newsize);
1660 VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1666 struct vnop_strategy_args {
1672 fuse_vnop_strategy(struct vop_strategy_args *ap)
1674 struct vnode *vp = ap->a_vp;
1675 struct buf *bp = ap->a_bp;
1677 if (!vp || fuse_isdeadfs(vp)) {
1678 bp->b_ioflags |= BIO_ERROR;
1679 bp->b_error = ENXIO;
1683 if (bp->b_iocmd == BIO_WRITE) {
1686 err = fuse_vnode_refreshsize(vp, NOCRED);
1688 bp->b_ioflags |= BIO_ERROR;
1696 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1697 * fuse_io_strategy sets bp's error fields
1699 (void)fuse_io_strategy(vp, bp);
1706 struct vnop_symlink_args {
1707 struct vnode *a_dvp;
1708 struct vnode **a_vpp;
1709 struct componentname *a_cnp;
1710 struct vattr *a_vap;
1715 fuse_vnop_symlink(struct vop_symlink_args *ap)
1717 struct vnode *dvp = ap->a_dvp;
1718 struct vnode **vpp = ap->a_vpp;
1719 struct componentname *cnp = ap->a_cnp;
1720 const char *target = ap->a_target;
1722 struct fuse_dispatcher fdi;
1727 if (fuse_isdeadfs(dvp)) {
1731 * Unlike the other creator type calls, here we have to create a message
1732 * where the name of the new entry comes first, and the data describing
1733 * the entry comes second.
1734 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1735 * but put together the message manually and just call the core part.
1738 len = strlen(target) + 1;
1739 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1740 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1742 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1743 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1744 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1746 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1747 fdisp_destroy(&fdi);
1752 struct vnop_write_args {
1756 struct ucred *a_cred;
1760 fuse_vnop_write(struct vop_write_args *ap)
1762 struct vnode *vp = ap->a_vp;
1763 struct uio *uio = ap->a_uio;
1764 int ioflag = ap->a_ioflag;
1765 struct ucred *cred = ap->a_cred;
1766 pid_t pid = curthread->td_proc->p_pid;
1769 if (fuse_isdeadfs(vp)) {
1772 err = fuse_vnode_refreshsize(vp, cred);
1776 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1777 ioflag |= IO_DIRECT;
1780 return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1783 SDT_PROBE_DEFINE1(fuse, , vnops, vnop_getpages_error, "int");
1785 struct vnop_getpages_args {
1793 fuse_vnop_getpages(struct vop_getpages_args *ap)
1795 int i, error, nextoff, size, toff, count, npages;
1804 pid_t pid = curthread->td_proc->p_pid;
1807 KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1808 td = curthread; /* XXX */
1809 cred = curthread->td_ucred; /* XXX */
1811 npages = ap->a_count;
1813 if (!fsess_opt_mmap(vnode_mount(vp))) {
1814 SDT_PROBE2(fuse, , vnops, trace, 1,
1815 "called on non-cacheable vnode??\n");
1816 return (VM_PAGER_ERROR);
1820 * If the last page is partially valid, just return it and allow
1821 * the pager to zero-out the blanks. Partially valid pages can
1822 * only occur at the file EOF.
1824 * XXXGL: is that true for FUSE, which is a local filesystem,
1825 * but still somewhat disconnected from the kernel?
1827 VM_OBJECT_WLOCK(vp->v_object);
1828 if (pages[npages - 1]->valid != 0 && --npages == 0)
1830 VM_OBJECT_WUNLOCK(vp->v_object);
1833 * We use only the kva address for the buffer, but this is extremely
1834 * convenient and fast.
1836 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1838 kva = (vm_offset_t)bp->b_data;
1839 pmap_qenter(kva, pages, npages);
1840 VM_CNT_INC(v_vnodein);
1841 VM_CNT_ADD(v_vnodepgsin, npages);
1843 count = npages << PAGE_SHIFT;
1844 iov.iov_base = (caddr_t)kva;
1845 iov.iov_len = count;
1848 uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1849 uio.uio_resid = count;
1850 uio.uio_segflg = UIO_SYSSPACE;
1851 uio.uio_rw = UIO_READ;
1854 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred, pid);
1855 pmap_qremove(kva, npages);
1857 uma_zfree(fuse_pbuf_zone, bp);
1859 if (error && (uio.uio_resid == count)) {
1860 SDT_PROBE1(fuse, , vnops, vnop_getpages_error, error);
1861 return VM_PAGER_ERROR;
1864 * Calculate the number of bytes read and validate only that number
1865 * of bytes. Note that due to pending writes, size may be 0. This
1866 * does not mean that the remaining data is invalid!
1869 size = count - uio.uio_resid;
1870 VM_OBJECT_WLOCK(vp->v_object);
1871 fuse_vm_page_lock_queues();
1872 for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1875 nextoff = toff + PAGE_SIZE;
1878 if (nextoff <= size) {
1880 * Read operation filled an entire page
1882 m->valid = VM_PAGE_BITS_ALL;
1883 KASSERT(m->dirty == 0,
1884 ("fuse_getpages: page %p is dirty", m));
1885 } else if (size > toff) {
1887 * Read operation filled a partial page.
1890 vm_page_set_valid_range(m, 0, size - toff);
1891 KASSERT(m->dirty == 0,
1892 ("fuse_getpages: page %p is dirty", m));
1895 * Read operation was short. If no error occurred
1896 * we may have hit a zero-fill section. We simply
1897 * leave valid set to 0.
1902 fuse_vm_page_unlock_queues();
1904 VM_OBJECT_WUNLOCK(vp->v_object);
1909 return (VM_PAGER_OK);
1913 struct vnop_putpages_args {
1919 vm_ooffset_t a_offset;
1923 fuse_vnop_putpages(struct vop_putpages_args *ap)
1929 int i, error, npages, count;
1937 pid_t pid = curthread->td_proc->p_pid;
1940 KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1941 fsize = vp->v_object->un_pager.vnp.vnp_size;
1942 td = curthread; /* XXX */
1943 cred = curthread->td_ucred; /* XXX */
1945 count = ap->a_count;
1946 rtvals = ap->a_rtvals;
1947 npages = btoc(count);
1948 offset = IDX_TO_OFF(pages[0]->pindex);
1950 if (!fsess_opt_mmap(vnode_mount(vp))) {
1951 SDT_PROBE2(fuse, , vnops, trace, 1,
1952 "called on non-cacheable vnode??\n");
1954 for (i = 0; i < npages; i++)
1955 rtvals[i] = VM_PAGER_AGAIN;
1958 * When putting pages, do not extend file past EOF.
1961 if (offset + count > fsize) {
1962 count = fsize - offset;
1967 * We use only the kva address for the buffer, but this is extremely
1968 * convenient and fast.
1970 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1972 kva = (vm_offset_t)bp->b_data;
1973 pmap_qenter(kva, pages, npages);
1974 VM_CNT_INC(v_vnodeout);
1975 VM_CNT_ADD(v_vnodepgsout, count);
1977 iov.iov_base = (caddr_t)kva;
1978 iov.iov_len = count;
1981 uio.uio_offset = offset;
1982 uio.uio_resid = count;
1983 uio.uio_segflg = UIO_SYSSPACE;
1984 uio.uio_rw = UIO_WRITE;
1987 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred, pid);
1989 pmap_qremove(kva, npages);
1990 uma_zfree(fuse_pbuf_zone, bp);
1993 int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
1995 for (i = 0; i < nwritten; i++) {
1996 rtvals[i] = VM_PAGER_OK;
1997 VM_OBJECT_WLOCK(pages[i]->object);
1998 vm_page_undirty(pages[i]);
1999 VM_OBJECT_WUNLOCK(pages[i]->object);
2005 static const char extattr_namespace_separator = '.';
2008 struct vop_getextattr_args {
2009 struct vop_generic_args a_gen;
2011 int a_attrnamespace;
2015 struct ucred *a_cred;
2016 struct thread *a_td;
2020 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2022 struct vnode *vp = ap->a_vp;
2023 struct uio *uio = ap->a_uio;
2024 struct fuse_dispatcher fdi;
2025 struct fuse_getxattr_in *get_xattr_in;
2026 struct fuse_getxattr_out *get_xattr_out;
2027 struct mount *mp = vnode_mount(vp);
2028 struct thread *td = ap->a_td;
2029 struct ucred *cred = ap->a_cred;
2035 if (fuse_isdeadfs(vp))
2038 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2042 /* Default to looking for user attributes. */
2043 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2044 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2046 prefix = EXTATTR_NAMESPACE_USER_STRING;
2048 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2049 strlen(ap->a_name) + 1;
2051 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2052 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2054 get_xattr_in = fdi.indata;
2056 * Check to see whether we're querying the available size or
2057 * issuing the actual request. If we pass in 0, we get back struct
2058 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2059 * that much data, without the struct fuse_getxattr_out header.
2062 get_xattr_in->size = 0;
2064 get_xattr_in->size = uio->uio_resid;
2066 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2067 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2070 err = fdisp_wait_answ(&fdi);
2073 fsess_set_notimpl(mp, FUSE_GETXATTR);
2077 get_xattr_out = fdi.answ;
2079 if (ap->a_size != NULL)
2080 *ap->a_size = get_xattr_out->size;
2083 err = uiomove(fdi.answ, fdi.iosize, uio);
2086 fdisp_destroy(&fdi);
2091 struct vop_setextattr_args {
2092 struct vop_generic_args a_gen;
2094 int a_attrnamespace;
2097 struct ucred *a_cred;
2098 struct thread *a_td;
2102 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2104 struct vnode *vp = ap->a_vp;
2105 struct uio *uio = ap->a_uio;
2106 struct fuse_dispatcher fdi;
2107 struct fuse_setxattr_in *set_xattr_in;
2108 struct mount *mp = vnode_mount(vp);
2109 struct thread *td = ap->a_td;
2110 struct ucred *cred = ap->a_cred;
2116 if (fuse_isdeadfs(vp))
2119 if (vfs_isrdonly(mp))
2122 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2123 if (ap->a_uio == NULL)
2126 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2131 /* Default to looking for user attributes. */
2132 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2133 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2135 prefix = EXTATTR_NAMESPACE_USER_STRING;
2137 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2138 strlen(ap->a_name) + 1;
2140 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2141 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2143 set_xattr_in = fdi.indata;
2144 set_xattr_in->size = uio->uio_resid;
2146 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2147 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2150 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2151 uio->uio_resid, uio);
2156 err = fdisp_wait_answ(&fdi);
2160 fsess_set_notimpl(mp, FUSE_SETXATTR);
2165 fdisp_destroy(&fdi);
2170 * The Linux / FUSE extended attribute list is simply a collection of
2171 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2172 * byte length followed by a non-NUL terminated string. So, this allows
2173 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2174 * Linux attribute names are reported with the namespace as a prefix (e.g.
2175 * "user.attribute_name"), but in FreeBSD they are reported without the
2176 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2178 * user.attr_name1\0user.attr_name2\0
2182 * <num>attr_name1<num>attr_name2
2184 * Where "<num>" is a single byte number of characters in the attribute name.
2187 * prefix - exattr namespace prefix string
2188 * list, list_len - input list with namespace prefixes
2189 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2192 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2193 char *bsd_list, int *bsd_list_len)
2195 int len, pos, dist_to_next, prefix_len;
2199 prefix_len = strlen(prefix);
2201 while (pos < list_len && list[pos] != '\0') {
2202 dist_to_next = strlen(&list[pos]) + 1;
2203 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2204 list[pos + prefix_len] == extattr_namespace_separator) {
2205 len = dist_to_next -
2206 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2207 if (len >= EXTATTR_MAXNAMELEN)
2208 return (ENAMETOOLONG);
2210 bsd_list[*bsd_list_len] = len;
2211 memcpy(&bsd_list[*bsd_list_len + 1],
2212 &list[pos + prefix_len +
2213 sizeof(extattr_namespace_separator)], len);
2215 *bsd_list_len += len + 1;
2218 pos += dist_to_next;
2225 struct vop_listextattr_args {
2226 struct vop_generic_args a_gen;
2228 int a_attrnamespace;
2231 struct ucred *a_cred;
2232 struct thread *a_td;
2236 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2238 struct vnode *vp = ap->a_vp;
2239 struct uio *uio = ap->a_uio;
2240 struct fuse_dispatcher fdi;
2241 struct fuse_listxattr_in *list_xattr_in;
2242 struct fuse_listxattr_out *list_xattr_out;
2243 struct mount *mp = vnode_mount(vp);
2244 struct thread *td = ap->a_td;
2245 struct ucred *cred = ap->a_cred;
2249 char *bsd_list = NULL;
2255 if (fuse_isdeadfs(vp))
2258 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2263 * Add space for a NUL and the period separator if enabled.
2264 * Default to looking for user attributes.
2266 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2267 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2269 prefix = EXTATTR_NAMESPACE_USER_STRING;
2271 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2273 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2274 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2277 * Retrieve Linux / FUSE compatible list size.
2279 list_xattr_in = fdi.indata;
2280 list_xattr_in->size = 0;
2281 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2282 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2284 err = fdisp_wait_answ(&fdi);
2287 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2291 list_xattr_out = fdi.answ;
2292 linux_list_len = list_xattr_out->size;
2293 if (linux_list_len == 0) {
2294 if (ap->a_size != NULL)
2295 *ap->a_size = linux_list_len;
2300 * Retrieve Linux / FUSE compatible list values.
2302 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2303 list_xattr_in = fdi.indata;
2304 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2305 list_xattr_in->flags = 0;
2306 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2307 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2309 err = fdisp_wait_answ(&fdi);
2313 linux_list = fdi.answ;
2314 linux_list_len = fdi.iosize;
2317 * Retrieve the BSD compatible list values.
2318 * The Linux / FUSE attribute list format isn't the same
2319 * as FreeBSD's format. So we need to transform it into
2320 * FreeBSD's format before giving it to the user.
2322 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2323 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2324 bsd_list, &bsd_list_len);
2328 if (ap->a_size != NULL)
2329 *ap->a_size = bsd_list_len;
2332 err = uiomove(bsd_list, bsd_list_len, uio);
2335 free(bsd_list, M_TEMP);
2336 fdisp_destroy(&fdi);
2341 struct vop_deleteextattr_args {
2342 struct vop_generic_args a_gen;
2344 int a_attrnamespace;
2346 struct ucred *a_cred;
2347 struct thread *a_td;
2351 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2353 struct vnode *vp = ap->a_vp;
2354 struct fuse_dispatcher fdi;
2355 struct mount *mp = vnode_mount(vp);
2356 struct thread *td = ap->a_td;
2357 struct ucred *cred = ap->a_cred;
2363 if (fuse_isdeadfs(vp))
2366 if (vfs_isrdonly(mp))
2369 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2374 /* Default to looking for user attributes. */
2375 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2376 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2378 prefix = EXTATTR_NAMESPACE_USER_STRING;
2380 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2381 strlen(ap->a_name) + 1;
2383 fdisp_init(&fdi, len);
2384 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2386 attr_str = fdi.indata;
2387 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2390 err = fdisp_wait_answ(&fdi);
2393 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2396 fdisp_destroy(&fdi);
2401 struct vnop_print_args {
2406 fuse_vnop_print(struct vop_print_args *ap)
2408 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2410 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2411 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2412 (uintmax_t)fvdat->nlookup,