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;
155 struct vop_vector fuse_fifoops = {
156 .vop_default = &fifo_specops,
157 .vop_access = fuse_vnop_access,
158 .vop_close = fuse_fifo_close,
159 .vop_fsync = fuse_vnop_fsync,
160 .vop_getattr = fuse_vnop_getattr,
161 .vop_inactive = fuse_vnop_inactive,
162 .vop_pathconf = fuse_vnop_pathconf,
163 .vop_print = fuse_vnop_print,
164 .vop_read = VOP_PANIC,
165 .vop_reclaim = fuse_vnop_reclaim,
166 .vop_setattr = fuse_vnop_setattr,
167 .vop_write = VOP_PANIC,
170 struct vop_vector fuse_vnops = {
171 .vop_allocate = VOP_EINVAL,
172 .vop_default = &default_vnodeops,
173 .vop_access = fuse_vnop_access,
174 .vop_advlock = fuse_vnop_advlock,
175 .vop_close = fuse_vnop_close,
176 .vop_create = fuse_vnop_create,
177 .vop_deleteextattr = fuse_vnop_deleteextattr,
178 .vop_fsync = fuse_vnop_fsync,
179 .vop_fdatasync = fuse_vnop_fdatasync,
180 .vop_getattr = fuse_vnop_getattr,
181 .vop_getextattr = fuse_vnop_getextattr,
182 .vop_inactive = fuse_vnop_inactive,
183 .vop_link = fuse_vnop_link,
184 .vop_listextattr = fuse_vnop_listextattr,
185 .vop_lookup = fuse_vnop_lookup,
186 .vop_mkdir = fuse_vnop_mkdir,
187 .vop_mknod = fuse_vnop_mknod,
188 .vop_open = fuse_vnop_open,
189 .vop_pathconf = fuse_vnop_pathconf,
190 .vop_read = fuse_vnop_read,
191 .vop_readdir = fuse_vnop_readdir,
192 .vop_readlink = fuse_vnop_readlink,
193 .vop_reclaim = fuse_vnop_reclaim,
194 .vop_remove = fuse_vnop_remove,
195 .vop_rename = fuse_vnop_rename,
196 .vop_rmdir = fuse_vnop_rmdir,
197 .vop_setattr = fuse_vnop_setattr,
198 .vop_setextattr = fuse_vnop_setextattr,
199 .vop_strategy = fuse_vnop_strategy,
200 .vop_symlink = fuse_vnop_symlink,
201 .vop_write = fuse_vnop_write,
202 .vop_getpages = fuse_vnop_getpages,
203 .vop_putpages = fuse_vnop_putpages,
204 .vop_print = fuse_vnop_print,
207 static u_long fuse_lookup_cache_hits = 0;
209 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
210 &fuse_lookup_cache_hits, 0, "number of positive cache hits in lookup");
212 static u_long fuse_lookup_cache_misses = 0;
214 SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
215 &fuse_lookup_cache_misses, 0, "number of cache misses in lookup");
218 * XXX: This feature is highly experimental and can bring to instabilities,
219 * needs revisiting before to be enabled by default.
221 static int fuse_reclaim_revoked = 0;
223 SYSCTL_INT(_vfs_fusefs, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
224 &fuse_reclaim_revoked, 0, "");
226 uma_zone_t fuse_pbuf_zone;
228 #define fuse_vm_page_lock(m) vm_page_lock((m));
229 #define fuse_vm_page_unlock(m) vm_page_unlock((m));
230 #define fuse_vm_page_lock_queues() ((void)0)
231 #define fuse_vm_page_unlock_queues() ((void)0)
233 /* Check permission for extattr operations, much like extattr_check_cred */
235 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
236 struct thread *td, accmode_t accmode)
238 struct mount *mp = vnode_mount(vp);
239 struct fuse_data *data = fuse_get_mpdata(mp);
242 * Kernel-invoked always succeeds.
248 * Do not allow privileged processes in jail to directly manipulate
252 case EXTATTR_NAMESPACE_SYSTEM:
253 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
254 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
257 case EXTATTR_NAMESPACE_USER:
258 return (fuse_internal_access(vp, accmode, td, cred));
264 /* Get a filehandle for a directory */
266 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
267 struct ucred *cred, pid_t pid)
269 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
271 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
274 /* Send FUSE_FLUSH for this vnode */
276 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
278 struct fuse_flush_in *ffi;
279 struct fuse_filehandle *fufh;
280 struct fuse_dispatcher fdi;
281 struct thread *td = curthread;
282 struct mount *mp = vnode_mount(vp);
285 if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
288 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
292 fdisp_init(&fdi, sizeof(*ffi));
293 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
295 ffi->fh = fufh->fh_id;
297 * If the file has a POSIX lock then we're supposed to set lock_owner.
298 * If not, then lock_owner is undefined. So we may as well always set
301 ffi->lock_owner = td->td_proc->p_pid;
303 err = fdisp_wait_answ(&fdi);
305 fsess_set_notimpl(mp, FUSE_FLUSH);
312 /* Close wrapper for fifos. */
314 fuse_fifo_close(struct vop_close_args *ap)
316 return (fifo_specops.vop_close(ap));
320 struct vnop_access_args {
322 #if VOP_ACCESS_TAKES_ACCMODE_T
327 struct ucred *a_cred;
332 fuse_vnop_access(struct vop_access_args *ap)
334 struct vnode *vp = ap->a_vp;
335 int accmode = ap->a_accmode;
336 struct ucred *cred = ap->a_cred;
338 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
342 if (fuse_isdeadfs(vp)) {
343 if (vnode_isvroot(vp)) {
348 if (!(data->dataflags & FSESS_INITED)) {
349 if (vnode_isvroot(vp)) {
350 if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
351 (fuse_match_cred(data->daemoncred, cred) == 0)) {
357 if (vnode_islnk(vp)) {
361 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
366 * struct vop_advlock_args {
367 * struct vop_generic_args a_gen;
368 * struct vnode *a_vp;
371 * struct flock *a_fl;
376 fuse_vnop_advlock(struct vop_advlock_args *ap)
378 struct vnode *vp = ap->a_vp;
379 struct flock *fl = ap->a_fl;
380 struct thread *td = curthread;
381 struct ucred *cred = td->td_ucred;
382 pid_t pid = td->td_proc->p_pid;
383 struct fuse_filehandle *fufh;
384 struct fuse_dispatcher fdi;
385 struct fuse_lk_in *fli;
386 struct fuse_lk_out *flo;
390 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
392 if (fuse_isdeadfs(vp)) {
396 if (!(dataflags & FSESS_POSIX_LOCKS))
397 return vop_stdadvlock(ap);
399 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
403 fdisp_init(&fdi, sizeof(*fli));
419 fdisp_make_vp(&fdi, op, vp, td, cred);
421 fli->fh = fufh->fh_id;
422 fli->owner = fl->l_pid;
423 fli->lk.start = fl->l_start;
425 fli->lk.end = fl->l_start + fl->l_len - 1;
427 fli->lk.end = INT64_MAX;
428 fli->lk.type = fl->l_type;
429 fli->lk.pid = fl->l_pid;
431 err = fdisp_wait_answ(&fdi);
434 if (err == 0 && op == FUSE_GETLK) {
436 fl->l_type = flo->lk.type;
437 fl->l_pid = flo->lk.pid;
438 if (flo->lk.type != F_UNLCK) {
439 fl->l_start = flo->lk.start;
440 if (flo->lk.end == INT64_MAX)
443 fl->l_len = flo->lk.end - flo->lk.start + 1;
444 fl->l_start = flo->lk.start;
452 struct vop_close_args {
455 struct ucred *a_cred;
460 fuse_vnop_close(struct vop_close_args *ap)
462 struct vnode *vp = ap->a_vp;
463 struct ucred *cred = ap->a_cred;
464 int fflag = ap->a_fflag;
465 struct thread *td = ap->a_td;
466 pid_t pid = td->td_proc->p_pid;
469 if (fuse_isdeadfs(vp))
473 if (fflag & IO_NDELAY)
476 err = fuse_flush(vp, cred, pid, fflag);
477 /* TODO: close the file handle, if we're sure it's no longer used */
478 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
479 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
485 fdisp_make_mknod_for_fallback(
486 struct fuse_dispatcher *fdip,
487 struct componentname *cnp,
493 enum fuse_opcode *op)
495 struct fuse_mknod_in *fmni;
497 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
499 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
503 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
505 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
508 struct vnop_create_args {
510 struct vnode **a_vpp;
511 struct componentname *a_cnp;
516 fuse_vnop_create(struct vop_create_args *ap)
518 struct vnode *dvp = ap->a_dvp;
519 struct vnode **vpp = ap->a_vpp;
520 struct componentname *cnp = ap->a_cnp;
521 struct vattr *vap = ap->a_vap;
522 struct thread *td = cnp->cn_thread;
523 struct ucred *cred = cnp->cn_cred;
525 struct fuse_open_in *foi;
526 struct fuse_entry_out *feo;
527 struct fuse_open_out *foo;
528 struct fuse_dispatcher fdi, fdi2;
529 struct fuse_dispatcher *fdip = &fdi;
530 struct fuse_dispatcher *fdip2 = NULL;
534 struct mount *mp = vnode_mount(dvp);
535 uint64_t parentnid = VTOFUD(dvp)->nid;
536 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
541 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a
542 * writable mode makes sense, and we might as well include readability
547 if (fuse_isdeadfs(dvp)) {
550 bzero(&fdi, sizeof(fdi));
552 if ((vap->va_type != VREG && vap->va_type != VSOCK))
555 if (!fsess_isimpl(mp, FUSE_CREATE)) {
556 /* Fallback to FUSE_MKNOD/FUSE_OPEN */
557 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
560 /* Use FUSE_CREATE */
562 fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
563 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
566 foi->flags = O_CREAT | flags;
567 memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
569 ((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
572 err = fdisp_wait_answ(fdip);
575 if (err == ENOSYS && op == FUSE_CREATE) {
576 fsess_set_notimpl(mp, FUSE_CREATE);
577 fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
578 parentnid, td, cred, mode, &op);
579 err = fdisp_wait_answ(fdip);
587 if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
591 if (op == FUSE_CREATE) {
592 foo = (struct fuse_open_out*)(feo + 1);
594 /* Issue a separate FUSE_OPEN */
596 fdisp_init(fdip2, sizeof(*foi));
597 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
602 err = fdisp_wait_answ(fdip2);
607 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
609 struct fuse_release_in *fri;
610 uint64_t nodeid = feo->nodeid;
611 uint64_t fh_id = foo->fh;
613 fdisp_init(fdip, sizeof(*fri));
614 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
618 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
619 fuse_insert_message(fdip->tick, false);
622 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
623 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
624 feo->attr_valid_nsec, NULL);
626 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
627 fuse_vnode_open(*vpp, foo->open_flags, td);
628 cache_purge_negative(dvp);
632 fdisp_destroy(fdip2);
638 struct vnop_fdatasync_args {
639 struct vop_generic_args a_gen;
641 struct thread * a_td;
645 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
647 struct vnode *vp = ap->a_vp;
648 struct thread *td = ap->a_td;
649 int waitfor = MNT_WAIT;
653 if (fuse_isdeadfs(vp)) {
656 if ((err = vop_stdfdatasync_buf(ap)))
659 return fuse_internal_fsync(vp, td, waitfor, true);
663 struct vnop_fsync_args {
664 struct vop_generic_args a_gen;
667 struct thread * a_td;
671 fuse_vnop_fsync(struct vop_fsync_args *ap)
673 struct vnode *vp = ap->a_vp;
674 struct thread *td = ap->a_td;
675 int waitfor = ap->a_waitfor;
678 if (fuse_isdeadfs(vp)) {
681 if ((err = vop_stdfsync(ap)))
684 return fuse_internal_fsync(vp, td, waitfor, false);
688 struct vnop_getattr_args {
691 struct ucred *a_cred;
696 fuse_vnop_getattr(struct vop_getattr_args *ap)
698 struct vnode *vp = ap->a_vp;
699 struct vattr *vap = ap->a_vap;
700 struct ucred *cred = ap->a_cred;
701 struct thread *td = curthread;
706 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
708 /* Note that we are not bailing out on a dead file system just yet. */
710 if (!(dataflags & FSESS_INITED)) {
711 if (!vnode_isvroot(vp)) {
712 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
719 err = fuse_internal_getattr(vp, vap, cred, td);
720 if (err == ENOTCONN && vnode_isvroot(vp)) {
721 /* see comment in fuse_vfsop_statfs() */
728 bzero(vap, sizeof(*vap));
729 vap->va_type = vnode_vtype(vp);
735 struct vnop_inactive_args {
741 fuse_vnop_inactive(struct vop_inactive_args *ap)
743 struct vnode *vp = ap->a_vp;
744 struct thread *td = ap->a_td;
746 struct fuse_vnode_data *fvdat = VTOFUD(vp);
747 struct fuse_filehandle *fufh, *fufh_tmp;
751 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
752 if (need_flush && vp->v_type == VREG) {
753 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
754 fuse_vnode_savesize(vp, NULL, 0);
756 if (fuse_data_cache_invalidate ||
757 (fvdat->flag & FN_REVOKED) != 0)
758 fuse_io_invalbuf(vp, td);
760 fuse_io_flushbuf(vp, MNT_WAIT, td);
763 fuse_filehandle_close(vp, fufh, td, NULL);
766 if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
773 struct vnop_link_args {
774 struct vnode *a_tdvp;
776 struct componentname *a_cnp;
780 fuse_vnop_link(struct vop_link_args *ap)
782 struct vnode *vp = ap->a_vp;
783 struct vnode *tdvp = ap->a_tdvp;
784 struct componentname *cnp = ap->a_cnp;
786 struct vattr *vap = VTOVA(vp);
788 struct fuse_dispatcher fdi;
789 struct fuse_entry_out *feo;
790 struct fuse_link_in fli;
794 if (fuse_isdeadfs(vp)) {
797 if (vnode_mount(tdvp) != vnode_mount(vp)) {
802 * This is a seatbelt check to protect naive userspace filesystems from
803 * themselves and the limitations of the FUSE IPC protocol. If a
804 * filesystem does not allow attribute caching, assume it is capable of
805 * validating that nlink does not overflow.
807 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
809 fli.oldnodeid = VTOI(vp);
812 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
813 FUSE_LINK, &fli, sizeof(fli), &fdi);
814 if ((err = fdisp_wait_answ(&fdi))) {
819 err = fuse_internal_checkentry(feo, vnode_vtype(vp));
821 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
822 feo->attr_valid_nsec, NULL);
828 struct fuse_lookup_alloc_arg {
829 struct fuse_entry_out *feo;
830 struct componentname *cnp;
835 /* Callback for vn_get_ino */
837 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
839 struct fuse_lookup_alloc_arg *flaa = arg;
841 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
845 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
846 "int", "struct timespec*", "struct timespec*");
848 struct vnop_lookup_args {
849 struct vnodeop_desc *a_desc;
851 struct vnode **a_vpp;
852 struct componentname *a_cnp;
856 fuse_vnop_lookup(struct vop_lookup_args *ap)
858 struct vnode *dvp = ap->a_dvp;
859 struct vnode **vpp = ap->a_vpp;
860 struct componentname *cnp = ap->a_cnp;
861 struct thread *td = cnp->cn_thread;
862 struct ucred *cred = cnp->cn_cred;
864 int nameiop = cnp->cn_nameiop;
865 int flags = cnp->cn_flags;
866 int wantparent = flags & (LOCKPARENT | WANTPARENT);
867 int islastcn = flags & ISLASTCN;
868 struct mount *mp = vnode_mount(dvp);
872 struct vnode *vp = NULL;
874 struct fuse_dispatcher fdi;
875 bool did_lookup = false;
876 struct fuse_entry_out *feo = NULL;
877 enum vtype vtyp; /* vnode type of target */
878 off_t filesize; /* filesize of target */
882 if (fuse_isdeadfs(dvp)) {
886 if (!vnode_isdir(dvp))
889 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
892 if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
895 if (flags & ISDOTDOT) {
896 nid = VTOFUD(dvp)->parent_nid;
899 /* .. is obviously a directory */
902 } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
904 /* . is obviously a directory */
908 struct timespec now, timeout;
910 err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
912 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
914 case -1: /* positive match */
915 if (timespeccmp(&timeout, &now, >)) {
916 atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
919 atomic_add_acq_long(&fuse_lookup_cache_misses,
931 case 0: /* no match in cache */
932 atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
935 case ENOENT: /* negative match */
937 if (timespeccmp(&timeout, &now, <=)) {
939 cache_purge_negative(dvp);
948 fdisp_init(&fdi, cnp->cn_namelen + 1);
949 fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
951 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
952 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
953 lookup_err = fdisp_wait_answ(&fdi);
957 /* lookup call succeeded */
958 nid = ((struct fuse_entry_out *)fdi.answ)->nodeid;
959 feo = (struct fuse_entry_out *)fdi.answ;
961 /* zero nodeid means ENOENT and cache it */
962 struct timespec timeout;
964 fdi.answ_stat = ENOENT;
966 if (cnp->cn_flags & MAKEENTRY) {
967 fuse_validity_2_timespec(feo, &timeout);
968 cache_enter_time(dvp, *vpp, cnp,
971 } else if (nid == FUSE_ROOT_ID) {
974 vtyp = IFTOVT(feo->attr.mode);
975 filesize = feo->attr.size;
977 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
982 /* lookup_err, if non-zero, must be ENOENT at this point */
985 /* Entry not found */
986 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
987 err = fuse_internal_access(dvp, VWRITE, td, cred);
990 * Set the SAVENAME flag to hold onto the
991 * pathname for use later in VOP_CREATE or
994 cnp->cn_flags |= SAVENAME;
1002 /* Entry was found */
1003 if (flags & ISDOTDOT) {
1004 struct fuse_lookup_alloc_arg flaa;
1010 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1013 } else if (nid == VTOI(dvp)) {
1017 struct fuse_vnode_data *fvdat;
1019 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1025 fuse_vnode_setparent(vp, dvp);
1028 * In the case where we are looking up a FUSE node
1029 * represented by an existing cached vnode, and the
1030 * true size reported by FUSE_LOOKUP doesn't match
1031 * the vnode's cached size, fix the vnode cache to
1032 * match the real object size.
1035 * * following attribute cache expiration, or
1036 * * due a bug in the daemon, or
1037 * * the first time that we looked up the file.
1040 if (vnode_isreg(vp) &&
1041 filesize != fvdat->filesize) {
1043 * The FN_SIZECHANGE flag reflects a dirty
1044 * append. If userspace lets us know our cache
1045 * is invalid, that write was lost. (Dirty
1046 * writes that do not cause append are also
1047 * lost, but we don't detect them here.)
1049 * XXX: Maybe disable WB caching on this mount.
1051 if (fvdat->flag & FN_SIZECHANGE)
1052 printf("%s: WB cache incoherent on "
1054 vnode_mount(vp)->mnt_stat.f_mntonname);
1056 (void)fuse_vnode_setsize(vp, cred, filesize);
1057 fvdat->flag &= ~FN_SIZECHANGE;
1061 fuse_internal_cache_attrs(*vpp, &feo->attr,
1062 feo->attr_valid, feo->attr_valid_nsec, NULL);
1064 if ((nameiop == DELETE || nameiop == RENAME) &&
1067 struct vattr dvattr;
1069 err = fuse_internal_access(dvp, VWRITE, td,
1074 * if the parent's sticky bit is set, check
1075 * whether we're allowed to remove the file.
1076 * Need to figure out the vnode locking to make
1079 fuse_internal_getattr(dvp, &dvattr, cred, td);
1080 if ((dvattr.va_mode & S_ISTXT) &&
1081 fuse_internal_access(dvp, VADMIN, td,
1083 fuse_internal_access(*vpp, VADMIN, td,
1091 (nameiop == DELETE) ||
1092 (nameiop == RENAME && wantparent))) {
1093 cnp->cn_flags |= SAVENAME;
1100 if (vp != NULL && dvp != vp)
1102 else if (vp != NULL)
1107 fdisp_destroy(&fdi);
1113 struct vnop_mkdir_args {
1114 struct vnode *a_dvp;
1115 struct vnode **a_vpp;
1116 struct componentname *a_cnp;
1117 struct vattr *a_vap;
1121 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1123 struct vnode *dvp = ap->a_dvp;
1124 struct vnode **vpp = ap->a_vpp;
1125 struct componentname *cnp = ap->a_cnp;
1126 struct vattr *vap = ap->a_vap;
1128 struct fuse_mkdir_in fmdi;
1130 if (fuse_isdeadfs(dvp)) {
1133 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1135 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1136 sizeof(fmdi), VDIR));
1140 struct vnop_mknod_args {
1141 struct vnode *a_dvp;
1142 struct vnode **a_vpp;
1143 struct componentname *a_cnp;
1144 struct vattr *a_vap;
1148 fuse_vnop_mknod(struct vop_mknod_args *ap)
1151 struct vnode *dvp = ap->a_dvp;
1152 struct vnode **vpp = ap->a_vpp;
1153 struct componentname *cnp = ap->a_cnp;
1154 struct vattr *vap = ap->a_vap;
1155 struct fuse_mknod_in fmni;
1157 if (fuse_isdeadfs(dvp))
1160 fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1161 fmni.rdev = vap->va_rdev;
1162 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni,
1163 sizeof(fmni), vap->va_type));
1167 struct vnop_open_args {
1170 struct ucred *a_cred;
1171 struct thread *a_td;
1172 int a_fdidx; / struct file *a_fp;
1176 fuse_vnop_open(struct vop_open_args *ap)
1178 struct vnode *vp = ap->a_vp;
1179 int a_mode = ap->a_mode;
1180 struct thread *td = ap->a_td;
1181 struct ucred *cred = ap->a_cred;
1182 pid_t pid = td->td_proc->p_pid;
1183 struct fuse_vnode_data *fvdat;
1185 if (fuse_isdeadfs(vp))
1187 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1188 return (EOPNOTSUPP);
1189 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1194 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1195 fuse_vnode_open(vp, 0, td);
1199 return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1203 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1206 switch (ap->a_name) {
1207 case _PC_FILESIZEBITS:
1211 *ap->a_retval = NAME_MAX;
1214 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1216 case _PC_SYMLINK_MAX:
1217 *ap->a_retval = MAXPATHLEN;
1223 return (vop_stdpathconf(ap));
1228 struct vnop_read_args {
1232 struct ucred *a_cred;
1236 fuse_vnop_read(struct vop_read_args *ap)
1238 struct vnode *vp = ap->a_vp;
1239 struct uio *uio = ap->a_uio;
1240 int ioflag = ap->a_ioflag;
1241 struct ucred *cred = ap->a_cred;
1242 pid_t pid = curthread->td_proc->p_pid;
1244 if (fuse_isdeadfs(vp)) {
1248 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1249 ioflag |= IO_DIRECT;
1252 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1256 struct vnop_readdir_args {
1259 struct ucred *a_cred;
1266 fuse_vnop_readdir(struct vop_readdir_args *ap)
1268 struct vnode *vp = ap->a_vp;
1269 struct uio *uio = ap->a_uio;
1270 struct ucred *cred = ap->a_cred;
1271 struct fuse_filehandle *fufh = NULL;
1272 struct fuse_iov cookediov;
1274 pid_t pid = curthread->td_proc->p_pid;
1276 if (fuse_isdeadfs(vp)) {
1279 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */
1280 (uio_resid(uio) < sizeof(struct dirent))) {
1284 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1287 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1288 fiov_init(&cookediov, DIRCOOKEDSIZE);
1290 err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1292 fiov_teardown(&cookediov);
1298 struct vnop_readlink_args {
1301 struct ucred *a_cred;
1305 fuse_vnop_readlink(struct vop_readlink_args *ap)
1307 struct vnode *vp = ap->a_vp;
1308 struct uio *uio = ap->a_uio;
1309 struct ucred *cred = ap->a_cred;
1311 struct fuse_dispatcher fdi;
1314 if (fuse_isdeadfs(vp)) {
1317 if (!vnode_islnk(vp)) {
1320 fdisp_init(&fdi, 0);
1321 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1325 if (((char *)fdi.answ)[0] == '/' &&
1326 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1327 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1329 err = uiomove(mpth, strlen(mpth), uio);
1332 err = uiomove(fdi.answ, fdi.iosize, uio);
1335 fdisp_destroy(&fdi);
1340 struct vnop_reclaim_args {
1342 struct thread *a_td;
1346 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1348 struct vnode *vp = ap->a_vp;
1349 struct thread *td = ap->a_td;
1350 struct fuse_vnode_data *fvdat = VTOFUD(vp);
1351 struct fuse_filehandle *fufh, *fufh_tmp;
1354 panic("FUSE: no vnode data during recycling");
1356 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1357 printf("FUSE: vnode being reclaimed with open fufh "
1358 "(type=%#x)", fufh->fufh_type);
1359 fuse_filehandle_close(vp, fufh, td, NULL);
1362 if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1363 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1366 fuse_vnode_setparent(vp, NULL);
1368 vfs_hash_remove(vp);
1369 vnode_destroy_vobject(vp);
1370 fuse_vnode_destroy(vp);
1376 struct vnop_remove_args {
1377 struct vnode *a_dvp;
1379 struct componentname *a_cnp;
1383 fuse_vnop_remove(struct vop_remove_args *ap)
1385 struct vnode *dvp = ap->a_dvp;
1386 struct vnode *vp = ap->a_vp;
1387 struct componentname *cnp = ap->a_cnp;
1391 if (fuse_isdeadfs(vp)) {
1394 if (vnode_isdir(vp)) {
1399 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1402 fuse_internal_vnode_disappear(vp);
1407 struct vnop_rename_args {
1408 struct vnode *a_fdvp;
1409 struct vnode *a_fvp;
1410 struct componentname *a_fcnp;
1411 struct vnode *a_tdvp;
1412 struct vnode *a_tvp;
1413 struct componentname *a_tcnp;
1417 fuse_vnop_rename(struct vop_rename_args *ap)
1419 struct vnode *fdvp = ap->a_fdvp;
1420 struct vnode *fvp = ap->a_fvp;
1421 struct componentname *fcnp = ap->a_fcnp;
1422 struct vnode *tdvp = ap->a_tdvp;
1423 struct vnode *tvp = ap->a_tvp;
1424 struct componentname *tcnp = ap->a_tcnp;
1425 struct fuse_data *data;
1429 if (fuse_isdeadfs(fdvp)) {
1432 if (fvp->v_mount != tdvp->v_mount ||
1433 (tvp && fvp->v_mount != tvp->v_mount)) {
1434 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1441 * FUSE library is expected to check if target directory is not
1442 * under the source directory in the file system tree.
1443 * Linux performs this check at VFS level.
1445 data = fuse_get_mpdata(vnode_mount(tdvp));
1446 sx_xlock(&data->rename_lock);
1447 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1450 fuse_vnode_setparent(fvp, tdvp);
1452 fuse_vnode_setparent(tvp, NULL);
1454 sx_unlock(&data->rename_lock);
1456 if (tvp != NULL && tvp != fvp) {
1459 if (vnode_isdir(fvp)) {
1460 if ((tvp != NULL) && vnode_isdir(tvp)) {
1481 struct vnop_rmdir_args {
1482 struct vnode *a_dvp;
1484 struct componentname *a_cnp;
1488 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1490 struct vnode *dvp = ap->a_dvp;
1491 struct vnode *vp = ap->a_vp;
1495 if (fuse_isdeadfs(vp)) {
1498 if (VTOFUD(vp) == VTOFUD(dvp)) {
1501 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1504 fuse_internal_vnode_disappear(vp);
1509 struct vnop_setattr_args {
1511 struct vattr *a_vap;
1512 struct ucred *a_cred;
1513 struct thread *a_td;
1517 fuse_vnop_setattr(struct vop_setattr_args *ap)
1519 struct vnode *vp = ap->a_vp;
1520 struct vattr *vap = ap->a_vap;
1521 struct ucred *cred = ap->a_cred;
1522 struct thread *td = curthread;
1524 struct fuse_data *data;
1525 struct vattr old_va;
1528 accmode_t accmode = 0;
1530 bool drop_suid = false;
1533 mp = vnode_mount(vp);
1534 data = fuse_get_mpdata(mp);
1535 dataflags = data->dataflags;
1536 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1537 if (cred->cr_ngroups > 0)
1538 cr_gid = cred->cr_groups[0];
1542 if (fuse_isdeadfs(vp)) {
1546 if (vap->va_uid != (uid_t)VNOVAL) {
1548 /* Only root may change a file's owner */
1549 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1551 /* As a special case, allow the null chown */
1552 err2 = fuse_internal_getattr(vp, &old_va, cred,
1556 if (vap->va_uid != old_va.va_uid)
1566 if (vap->va_gid != (gid_t)VNOVAL) {
1567 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1569 if (checkperm && !groupmember(vap->va_gid, cred))
1572 * Non-root users may only chgrp to one of their own
1575 err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1577 /* As a special case, allow the null chgrp */
1578 err2 = fuse_internal_getattr(vp, &old_va, cred,
1582 if (vap->va_gid != old_va.va_gid)
1590 if (vap->va_size != VNOVAL) {
1591 switch (vp->v_type) {
1596 if (vfs_isrdonly(mp))
1601 * According to POSIX, the result is unspecified
1602 * for file types other than regular files,
1603 * directories and shared memory objects. We
1604 * don't support shared memory objects in the file
1605 * system, and have dubious support for truncating
1606 * symlinks. Just ignore the request in other cases.
1610 /* Don't set accmode. Permission to trunc is checked upstack */
1613 * TODO: for atime and mtime, only require VWRITE if UTIMENS_NULL is
1616 if (vap->va_atime.tv_sec != VNOVAL)
1618 if (vap->va_mtime.tv_sec != VNOVAL)
1621 if (vap->va_mode != (mode_t)VNOVAL)
1622 vap->va_mode &= ~(S_ISUID | S_ISGID);
1624 err = fuse_internal_getattr(vp, &old_va, cred, td);
1627 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1630 if (vap->va_mode != (mode_t)VNOVAL) {
1631 /* Only root may set the sticky bit on non-directories */
1632 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1633 && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1635 if (checkperm && (vap->va_mode & S_ISGID)) {
1636 err = fuse_internal_getattr(vp, &old_va, cred, td);
1639 if (!groupmember(old_va.va_gid, cred)) {
1640 err = priv_check_cred(cred, PRIV_VFS_SETGID);
1648 if (vfs_isrdonly(mp))
1651 err = fuse_internal_access(vp, accmode, td, cred);
1655 return fuse_internal_setattr(vp, vap, td, cred);
1659 struct vnop_strategy_args {
1665 fuse_vnop_strategy(struct vop_strategy_args *ap)
1667 struct vnode *vp = ap->a_vp;
1668 struct buf *bp = ap->a_bp;
1670 if (!vp || fuse_isdeadfs(vp)) {
1671 bp->b_ioflags |= BIO_ERROR;
1672 bp->b_error = ENXIO;
1676 if (bp->b_iocmd == BIO_WRITE) {
1679 err = fuse_vnode_refreshsize(vp, NOCRED);
1681 bp->b_ioflags |= BIO_ERROR;
1689 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1690 * fuse_io_strategy sets bp's error fields
1692 (void)fuse_io_strategy(vp, bp);
1699 struct vnop_symlink_args {
1700 struct vnode *a_dvp;
1701 struct vnode **a_vpp;
1702 struct componentname *a_cnp;
1703 struct vattr *a_vap;
1708 fuse_vnop_symlink(struct vop_symlink_args *ap)
1710 struct vnode *dvp = ap->a_dvp;
1711 struct vnode **vpp = ap->a_vpp;
1712 struct componentname *cnp = ap->a_cnp;
1713 const char *target = ap->a_target;
1715 struct fuse_dispatcher fdi;
1720 if (fuse_isdeadfs(dvp)) {
1724 * Unlike the other creator type calls, here we have to create a message
1725 * where the name of the new entry comes first, and the data describing
1726 * the entry comes second.
1727 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1728 * but put together the message manually and just call the core part.
1731 len = strlen(target) + 1;
1732 fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1733 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1735 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1736 ((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1737 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1739 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1740 fdisp_destroy(&fdi);
1745 struct vnop_write_args {
1749 struct ucred *a_cred;
1753 fuse_vnop_write(struct vop_write_args *ap)
1755 struct vnode *vp = ap->a_vp;
1756 struct uio *uio = ap->a_uio;
1757 int ioflag = ap->a_ioflag;
1758 struct ucred *cred = ap->a_cred;
1759 pid_t pid = curthread->td_proc->p_pid;
1762 if (fuse_isdeadfs(vp)) {
1765 err = fuse_vnode_refreshsize(vp, cred);
1769 if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1770 ioflag |= IO_DIRECT;
1773 return fuse_io_dispatch(vp, uio, ioflag, false, cred, pid);
1776 SDT_PROBE_DEFINE1(fusefs, , vnops, vnop_getpages_error, "int");
1778 struct vnop_getpages_args {
1786 fuse_vnop_getpages(struct vop_getpages_args *ap)
1788 int i, error, nextoff, size, toff, count, npages;
1797 pid_t pid = curthread->td_proc->p_pid;
1800 KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1801 td = curthread; /* XXX */
1802 cred = curthread->td_ucred; /* XXX */
1804 npages = ap->a_count;
1806 if (!fsess_opt_mmap(vnode_mount(vp))) {
1807 SDT_PROBE2(fusefs, , vnops, trace, 1,
1808 "called on non-cacheable vnode??\n");
1809 return (VM_PAGER_ERROR);
1813 * If the last page is partially valid, just return it and allow
1814 * the pager to zero-out the blanks. Partially valid pages can
1815 * only occur at the file EOF.
1817 * XXXGL: is that true for FUSE, which is a local filesystem,
1818 * but still somewhat disconnected from the kernel?
1820 VM_OBJECT_WLOCK(vp->v_object);
1821 if (pages[npages - 1]->valid != 0 && --npages == 0)
1823 VM_OBJECT_WUNLOCK(vp->v_object);
1826 * We use only the kva address for the buffer, but this is extremely
1827 * convenient and fast.
1829 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1831 kva = (vm_offset_t)bp->b_data;
1832 pmap_qenter(kva, pages, npages);
1833 VM_CNT_INC(v_vnodein);
1834 VM_CNT_ADD(v_vnodepgsin, npages);
1836 count = npages << PAGE_SHIFT;
1837 iov.iov_base = (caddr_t)kva;
1838 iov.iov_len = count;
1841 uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1842 uio.uio_resid = count;
1843 uio.uio_segflg = UIO_SYSSPACE;
1844 uio.uio_rw = UIO_READ;
1847 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
1848 pmap_qremove(kva, npages);
1850 uma_zfree(fuse_pbuf_zone, bp);
1852 if (error && (uio.uio_resid == count)) {
1853 SDT_PROBE1(fusefs, , vnops, vnop_getpages_error, error);
1854 return VM_PAGER_ERROR;
1857 * Calculate the number of bytes read and validate only that number
1858 * of bytes. Note that due to pending writes, size may be 0. This
1859 * does not mean that the remaining data is invalid!
1862 size = count - uio.uio_resid;
1863 VM_OBJECT_WLOCK(vp->v_object);
1864 fuse_vm_page_lock_queues();
1865 for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1868 nextoff = toff + PAGE_SIZE;
1871 if (nextoff <= size) {
1873 * Read operation filled an entire page
1875 m->valid = VM_PAGE_BITS_ALL;
1876 KASSERT(m->dirty == 0,
1877 ("fuse_getpages: page %p is dirty", m));
1878 } else if (size > toff) {
1880 * Read operation filled a partial page.
1883 vm_page_set_valid_range(m, 0, size - toff);
1884 KASSERT(m->dirty == 0,
1885 ("fuse_getpages: page %p is dirty", m));
1888 * Read operation was short. If no error occurred
1889 * we may have hit a zero-fill section. We simply
1890 * leave valid set to 0.
1895 fuse_vm_page_unlock_queues();
1897 VM_OBJECT_WUNLOCK(vp->v_object);
1902 return (VM_PAGER_OK);
1906 struct vnop_putpages_args {
1912 vm_ooffset_t a_offset;
1916 fuse_vnop_putpages(struct vop_putpages_args *ap)
1922 int i, error, npages, count;
1930 pid_t pid = curthread->td_proc->p_pid;
1933 KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1934 fsize = vp->v_object->un_pager.vnp.vnp_size;
1935 td = curthread; /* XXX */
1936 cred = curthread->td_ucred; /* XXX */
1938 count = ap->a_count;
1939 rtvals = ap->a_rtvals;
1940 npages = btoc(count);
1941 offset = IDX_TO_OFF(pages[0]->pindex);
1943 if (!fsess_opt_mmap(vnode_mount(vp))) {
1944 SDT_PROBE2(fusefs, , vnops, trace, 1,
1945 "called on non-cacheable vnode??\n");
1947 for (i = 0; i < npages; i++)
1948 rtvals[i] = VM_PAGER_AGAIN;
1951 * When putting pages, do not extend file past EOF.
1954 if (offset + count > fsize) {
1955 count = fsize - offset;
1960 * We use only the kva address for the buffer, but this is extremely
1961 * convenient and fast.
1963 bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK);
1965 kva = (vm_offset_t)bp->b_data;
1966 pmap_qenter(kva, pages, npages);
1967 VM_CNT_INC(v_vnodeout);
1968 VM_CNT_ADD(v_vnodepgsout, count);
1970 iov.iov_base = (caddr_t)kva;
1971 iov.iov_len = count;
1974 uio.uio_offset = offset;
1975 uio.uio_resid = count;
1976 uio.uio_segflg = UIO_SYSSPACE;
1977 uio.uio_rw = UIO_WRITE;
1980 error = fuse_io_dispatch(vp, &uio, IO_DIRECT, true, cred, pid);
1982 pmap_qremove(kva, npages);
1983 uma_zfree(fuse_pbuf_zone, bp);
1986 int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
1988 for (i = 0; i < nwritten; i++) {
1989 rtvals[i] = VM_PAGER_OK;
1990 VM_OBJECT_WLOCK(pages[i]->object);
1991 vm_page_undirty(pages[i]);
1992 VM_OBJECT_WUNLOCK(pages[i]->object);
1998 static const char extattr_namespace_separator = '.';
2001 struct vop_getextattr_args {
2002 struct vop_generic_args a_gen;
2004 int a_attrnamespace;
2008 struct ucred *a_cred;
2009 struct thread *a_td;
2013 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2015 struct vnode *vp = ap->a_vp;
2016 struct uio *uio = ap->a_uio;
2017 struct fuse_dispatcher fdi;
2018 struct fuse_getxattr_in *get_xattr_in;
2019 struct fuse_getxattr_out *get_xattr_out;
2020 struct mount *mp = vnode_mount(vp);
2021 struct thread *td = ap->a_td;
2022 struct ucred *cred = ap->a_cred;
2028 if (fuse_isdeadfs(vp))
2031 if (!fsess_isimpl(mp, FUSE_GETXATTR))
2034 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2038 /* Default to looking for user attributes. */
2039 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2040 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2042 prefix = EXTATTR_NAMESPACE_USER_STRING;
2044 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2045 strlen(ap->a_name) + 1;
2047 fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2048 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2050 get_xattr_in = fdi.indata;
2052 * Check to see whether we're querying the available size or
2053 * issuing the actual request. If we pass in 0, we get back struct
2054 * fuse_getxattr_out. If we pass in a non-zero size, we get back
2055 * that much data, without the struct fuse_getxattr_out header.
2058 get_xattr_in->size = 0;
2060 get_xattr_in->size = uio->uio_resid;
2062 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2063 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2066 err = fdisp_wait_answ(&fdi);
2068 if (err == ENOSYS) {
2069 fsess_set_notimpl(mp, FUSE_GETXATTR);
2075 get_xattr_out = fdi.answ;
2077 if (ap->a_size != NULL)
2078 *ap->a_size = get_xattr_out->size;
2081 err = uiomove(fdi.answ, fdi.iosize, uio);
2084 fdisp_destroy(&fdi);
2089 struct vop_setextattr_args {
2090 struct vop_generic_args a_gen;
2092 int a_attrnamespace;
2095 struct ucred *a_cred;
2096 struct thread *a_td;
2100 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2102 struct vnode *vp = ap->a_vp;
2103 struct uio *uio = ap->a_uio;
2104 struct fuse_dispatcher fdi;
2105 struct fuse_setxattr_in *set_xattr_in;
2106 struct mount *mp = vnode_mount(vp);
2107 struct thread *td = ap->a_td;
2108 struct ucred *cred = ap->a_cred;
2114 if (fuse_isdeadfs(vp))
2117 if (!fsess_isimpl(mp, FUSE_SETXATTR))
2120 if (vfs_isrdonly(mp))
2123 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2124 if (ap->a_uio == NULL) {
2126 * If we got here as fallback from VOP_DELETEEXTATTR, then
2127 * return EOPNOTSUPP.
2129 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2130 return (EOPNOTSUPP);
2135 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
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(*set_xattr_in) + uio->uio_resid);
2150 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2152 set_xattr_in = fdi.indata;
2153 set_xattr_in->size = uio->uio_resid;
2155 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2156 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2159 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2160 uio->uio_resid, uio);
2165 err = fdisp_wait_answ(&fdi);
2167 if (err == ENOSYS) {
2168 fsess_set_notimpl(mp, FUSE_SETXATTR);
2171 if (err == ERESTART) {
2172 /* Can't restart after calling uiomove */
2177 fdisp_destroy(&fdi);
2182 * The Linux / FUSE extended attribute list is simply a collection of
2183 * NUL-terminated strings. The FreeBSD extended attribute list is a single
2184 * byte length followed by a non-NUL terminated string. So, this allows
2185 * conversion of the Linux / FUSE format to the FreeBSD format in place.
2186 * Linux attribute names are reported with the namespace as a prefix (e.g.
2187 * "user.attribute_name"), but in FreeBSD they are reported without the
2188 * namespace prefix (e.g. "attribute_name"). So, we're going from:
2190 * user.attr_name1\0user.attr_name2\0
2194 * <num>attr_name1<num>attr_name2
2196 * Where "<num>" is a single byte number of characters in the attribute name.
2199 * prefix - exattr namespace prefix string
2200 * list, list_len - input list with namespace prefixes
2201 * bsd_list, bsd_list_len - output list compatible with bsd vfs
2204 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2205 char *bsd_list, int *bsd_list_len)
2207 int len, pos, dist_to_next, prefix_len;
2211 prefix_len = strlen(prefix);
2213 while (pos < list_len && list[pos] != '\0') {
2214 dist_to_next = strlen(&list[pos]) + 1;
2215 if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2216 list[pos + prefix_len] == extattr_namespace_separator) {
2217 len = dist_to_next -
2218 (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2219 if (len >= EXTATTR_MAXNAMELEN)
2220 return (ENAMETOOLONG);
2222 bsd_list[*bsd_list_len] = len;
2223 memcpy(&bsd_list[*bsd_list_len + 1],
2224 &list[pos + prefix_len +
2225 sizeof(extattr_namespace_separator)], len);
2227 *bsd_list_len += len + 1;
2230 pos += dist_to_next;
2237 struct vop_listextattr_args {
2238 struct vop_generic_args a_gen;
2240 int a_attrnamespace;
2243 struct ucred *a_cred;
2244 struct thread *a_td;
2248 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2250 struct vnode *vp = ap->a_vp;
2251 struct uio *uio = ap->a_uio;
2252 struct fuse_dispatcher fdi;
2253 struct fuse_listxattr_in *list_xattr_in;
2254 struct fuse_listxattr_out *list_xattr_out;
2255 struct mount *mp = vnode_mount(vp);
2256 struct thread *td = ap->a_td;
2257 struct ucred *cred = ap->a_cred;
2261 char *bsd_list = NULL;
2267 if (fuse_isdeadfs(vp))
2270 if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2273 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2278 * Add space for a NUL and the period separator if enabled.
2279 * Default to looking for user attributes.
2281 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2282 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2284 prefix = EXTATTR_NAMESPACE_USER_STRING;
2286 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2288 fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2289 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2292 * Retrieve Linux / FUSE compatible list size.
2294 list_xattr_in = fdi.indata;
2295 list_xattr_in->size = 0;
2296 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2297 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2299 err = fdisp_wait_answ(&fdi);
2301 if (err == ENOSYS) {
2302 fsess_set_notimpl(mp, FUSE_LISTXATTR);
2308 list_xattr_out = fdi.answ;
2309 linux_list_len = list_xattr_out->size;
2310 if (linux_list_len == 0) {
2311 if (ap->a_size != NULL)
2312 *ap->a_size = linux_list_len;
2317 * Retrieve Linux / FUSE compatible list values.
2319 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2320 list_xattr_in = fdi.indata;
2321 list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2322 list_xattr_in->flags = 0;
2323 attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2324 snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2326 err = fdisp_wait_answ(&fdi);
2330 linux_list = fdi.answ;
2331 linux_list_len = fdi.iosize;
2334 * Retrieve the BSD compatible list values.
2335 * The Linux / FUSE attribute list format isn't the same
2336 * as FreeBSD's format. So we need to transform it into
2337 * FreeBSD's format before giving it to the user.
2339 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2340 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2341 bsd_list, &bsd_list_len);
2345 if (ap->a_size != NULL)
2346 *ap->a_size = bsd_list_len;
2349 err = uiomove(bsd_list, bsd_list_len, uio);
2352 free(bsd_list, M_TEMP);
2353 fdisp_destroy(&fdi);
2358 struct vop_deleteextattr_args {
2359 struct vop_generic_args a_gen;
2361 int a_attrnamespace;
2363 struct ucred *a_cred;
2364 struct thread *a_td;
2368 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2370 struct vnode *vp = ap->a_vp;
2371 struct fuse_dispatcher fdi;
2372 struct mount *mp = vnode_mount(vp);
2373 struct thread *td = ap->a_td;
2374 struct ucred *cred = ap->a_cred;
2380 if (fuse_isdeadfs(vp))
2383 if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2386 if (vfs_isrdonly(mp))
2389 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2394 /* Default to looking for user attributes. */
2395 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2396 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2398 prefix = EXTATTR_NAMESPACE_USER_STRING;
2400 len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2401 strlen(ap->a_name) + 1;
2403 fdisp_init(&fdi, len);
2404 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2406 attr_str = fdi.indata;
2407 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2410 err = fdisp_wait_answ(&fdi);
2411 if (err == ENOSYS) {
2412 fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2416 fdisp_destroy(&fdi);
2421 struct vnop_print_args {
2426 fuse_vnop_print(struct vop_print_args *ap)
2428 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2430 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2431 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2432 (uintmax_t)fvdat->nlookup,